Issuu

Verwerking van wintertellingen van akkervogels op vogelvoedselpercelen in Haspengouw In opdracht van Vlaamse Landmaatschappij

Prof. dr. Christel Faes dr. Thomas Neyens Oana Petrof I-Biostat Hasselt University May 29, 2018

Contents I

Nederlandstalige Samenvatting

1 Samenvatting en Conclusies 1.1 Doel Analyse . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Welke soorten hebben baat bij vogelvoedselgewassen in de winter? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Vijf soorten in detail geanalyseerd . . . . . . . . . . . . . . . 1.3.1 Karakteristieken van het vogelvoedselperceel . . . . . 1.3.2 Effect van landschapskenmerken op deze soorten . . . 1.4 Invloed van temperatuur en weersituatie tijdens moment van waarneming . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Overzicht per soort . . . . . . . . . . . . . . . . . . . . . . . .

10 10

Introduction, Materials and Methods

2 2 3 4 5 7

2 Introduction 13 2.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3 Materials 3.1 Outcome variables . . . . . . . . . . . . . . . 3.2 Independent variables . . . . . . . . . . . . . 3.2.1 Landscape, food crop and disturbance 3.2.2 Openness and closeness . . . . . . . . 3.2.3 Clustering . . . . . . . . . . . . . . . .

. . . . .

15 15 18 19 20 20

4 Methods 26 4.1 Exploratory analysis . . . . . . . . . . . . . . . . . . . . . . . 27 4.2 Statistical analysis of presence/absence of birds . . . . . . . . 27 4.2.1 Logistic regression . . . . . . . . . . . . . . . . . . . . 27 i

4.3 4.4

III

4.2.2 Generalized Linear Mixed Model (GLMM) Statistical analysis of number of birds . . . . . . . 4.3.1 Negative Binomial regression (NB) . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

Results: bird food crop for all species

5 Results: impact of bird food crop 5.1 Objective . . . . . . . . . . . . . 5.2 Data exploration . . . . . . . . . 5.3 Results . . . . . . . . . . . . . . . 5.3.1 Geelgors species . . . . . 5.3.2 Kneu species . . . . . . . 5.3.3 Rietgors species . . . . . 5.3.4 Torenvalk species . . . . . 5.3.5 Buizerd species . . . . . . 5.3.6 Fazant species . . . . . . 5.3.7 Groenling species . . . . . 5.3.8 Blauwe Reiger species . . 5.4 Conclusion . . . . . . . . . . . .

. . . . . . . . . . . .

28 29 29 29

31 . . . . . . . . . . . .

. . . . . . . . . . . .

32 32 32 33 34 35 35 35 35 36 36 36 36

IV Results: Effect of environmental factors on Grauwe Gors, Geelgors, Veldleeuwerik, Kneu and Rietgors 37 6 Exploratory analysis results 6.1 Exploratory data analysis for categorical variables 6.1.1 Food availability . . . . . . . . . . . . . . . 6.1.2 Environment and disturbance . . . . . . . . 6.1.3 Impact on prevalence of species . . . . . . . 6.2 Exploratory data analysis for continuous variables 6.2.1 Plot size by species . . . . . . . . . . . . . . 6.2.2 Temperature by species . . . . . . . . . . . 6.2.3 Hellmann number by species . . . . . . . . 6.2.4 Nearest Building by species . . . . . . . . . 6.2.5 Nearest built-upon area by species . . . . . 6.2.6 Nearest plot by species . . . . . . . . . . . . 6.2.7 Nearest railroad by species . . . . . . . . . 6.2.8 Nearest road by species . . . . . . . . . . .

. . . . . . . . . . . . .

38 38 38 42 42 43 43 44 45 46 47 48 49 50

6.3

Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7 Results regression models for Grauwe Gors, Geelgors, Veldleeuwerik, Kneu and Rietgors 53 7.1 Results by species for categorical variables . . . . . . . . . . . 53 7.1.1 Grauwe Gors . . . . . . . . . . . . . . . . . . . . . . . 53 7.1.2 Geelgors . . . . . . . . . . . . . . . . . . . . . . . . . . 54 7.1.3 Veldleeuwerik . . . . . . . . . . . . . . . . . . . . . . . 59 7.1.4 Kneu . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 7.1.5 Rietgors . . . . . . . . . . . . . . . . . . . . . . . . . . 73 7.2 Results by species for continuous variables . . . . . . . . . . . 74 7.2.1 Grauwe Gors . . . . . . . . . . . . . . . . . . . . . . . 74 7.2.2 Geelgors . . . . . . . . . . . . . . . . . . . . . . . . . . 74 7.2.3 Veldleeuwerik . . . . . . . . . . . . . . . . . . . . . . . 80 7.2.4 Kneu . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 7.2.5 Rietgors . . . . . . . . . . . . . . . . . . . . . . . . . . 84 7.3 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . 87 7.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 7.4.1 Grauwe Gors species . . . . . . . . . . . . . . . . . . . 98 7.4.2 Geelgors species . . . . . . . . . . . . . . . . . . . . . 98 7.4.3 Veldleeuwerik species . . . . . . . . . . . . . . . . . . 98 7.4.4 Kneu species . . . . . . . . . . . . . . . . . . . . . . . 99 7.4.5 Rietgors species . . . . . . . . . . . . . . . . . . . . . 99

Results: Openness/Closeness

100

8 Results: Statistical analysis of effect of Openness/Closeness101 8.1 Changes in time . . . . . . . . . . . . . . . . . . . . . . . . . 101 8.2 Results based on data from 2012 . . . . . . . . . . . . . . . . 103 8.3 Results based on data from 2016 and 2017 . . . . . . . . . . . 103

Result: cluster variables

115

9 Results: statistical analysis of clustering effect 116 9.1 Effects of number of plots under contract in 2017 within 100, 150, 200, 250, 300 m . . . . . . . . . . . . . . . . . . . . . . . 116 9.2 Results by species for the area size of the buffer of 100m, 150m, 200m, 250m, 300m around the plot . . . . . . . . . . . 121

iii

9.3 9.4

VII

Effect of the number of plots within cluster of 100m, 150m, 200m, 250m, 300m . . . . . . . . . . . . . . . . . . . . . . . . 129 Effect of the sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters . . . . . . . . . . . . . . . . . . . . . 135 9.4.1 Summary of results . . . . . . . . . . . . . . . . . . . . 141

Conclusions and Recommendations

150

List of Tables 1.1 1.2

Samenvattende tabel m.b.t. het voorkomen van de meest abundante soorten. . . . . . . . . . . . . . . . . . . . . . . . . Samenvatting van de belangrijkste resultaten uit de analyses.

4 11

3.1 3.2 3.3

Presence/absence of birds per species . . . . . . . . . . . . . . Explanatory variables (part 1) . . . . . . . . . . . . . . . . . Explanatory variables (part 2) . . . . . . . . . . . . . . . . .

17 21 22

5.1

Significance using Logistic regression of presence/absence of crop for all species . . . . . . . . . . . . . . . . . . . . . . . .

6.1 6.2 6.3

Exploratory data analysis for categorical variables (part1) . . Exploratory data analysis for categorical variables (part 2) . . Exploratory data analysis for categorical variables (part 3) . .

39 40 41

7.1

Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors . . . . . . . . . . . . . . . . . . . . . Pairwise comparisons of the variable Bird food crop and food availability for Geelgors . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species . . . . . . . . . . . . . .

7.2 7.3 7.4 7.5 7.6 7.7 7.8

55 56 57 60 61 62 63 65

7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18 7.19 7.20 7.21 7.22 7.23 7.24 7.25 7.26 7.27 7.28 7.29

Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species . . . . . . . . . . . . . . . . . . Pairwise comparisons of the variable Bird food crop and food availability for Kneu . . . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species . . . . . . . . . . . . . . . . . Pairwise comparisons of the variable Bird food crop and food availability for Rietgors . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Kneu . . . . . . . . . . . . . . . . . . . . . . Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors . . . . . . . . . . . . . . . . . . . . . Significance using Logistic regression for categorical variables per species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Significance using Logistic regression for continuous variables per species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Significance using Negative Binomial regression for categorical variables per species . . . . . . . . . . . . . . . . . . . . . Significance using Negative Binomial regression for continuous variables per species . . . . . . . . . . . . . . . . . . . . . Significance using GLMM for categorical variables per species Significance using GLMM for continuous variables per species vi

66 67 69 70 71 72 75 76 77 78 79 81 83 85 86 90 91 92 93 94 95

7.30 Significance for Food availability for Geelgors, Kneu and Rietgors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 7.31 Significance for Bird food crop for Geelgors, Kneu and Rietgors 97 8.1 8.2 8.3 8.4 8.5 8.6

Results for Grauwe Gors based on the dataset from 2012 . . Results for Geelgors based on the dataset from 2012 . . . . . Results for Veldleeuwerik based on the dataset from 2012 . . Results for Kneu based on the dataset from 2012 . . . . . . . Results for Rietgors based on the dataset from 2012 . . . . . Results for Grauwe Gors based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.7 Results for Geelgors based on the dataset from 2016 and 2017 8.8 Results for Veldleeuwerik based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.9 Results for Kneu based on the dataset from 2016 and 2017 . 8.10 Results Rietgors based on the dataset from 2016 and 2017 . . 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 9.10 9.11 9.12 9.13 9.14 9.15 9.16

Results for Grauwe Gors based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Geelgors based on the dataset from 2016 and 2017 Results for Veldleeuwerik based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Kneu based on the dataset from 2016 and 2017 . Results for Rietgors based on the dataset from 2016 and 2017 Results for Grauwe Gors based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Geelgors based on the dataset from 2016 and 2017 Results for Veldleeuwerik based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Kneu based on the dataset from 2016 and 2017 . Results for Rietgors based on the dataset from 2016 and 2017 Results for Grauwe Gors based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Geelgors based on the dataset from 2016 and 2017 Results for Veldleeuwerik based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results for Kneu based on the dataset from 2016 and 2017 . Results for Rietgors based on the dataset from 2016 and 2017 Results for Grauwe Gors based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

104 105 106 107 108 109 110 111 112 113

117 118 119 120 122 123 124 126 127 128 130 131 132 133 134 136

9.17 Results for Geelgors based on the dataset from 2016 and 2017 137 9.18 Results for Veldleeuwerik based on the dataset from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 9.19 Results for Kneu based on the dataset from 2016 and 2017 . 139 9.20 Results for Rietgors based on the dataset from 2016 and 2017 140 9.21 Significance for clusters using Logistics regression per species using birds from 2012 only . . . . . . . . . . . . . . . . . . . . 142 9.22 Significance for clusters using Negative Binomial regression per species using birds from 2012 only . . . . . . . . . . . . . 143 9.23 Significance for clusters using Logistic regression per species using birds from 2016 and 2017 . . . . . . . . . . . . . . . . . 144 9.24 Significance for clusters using Negative Binomial regression per species using birds from 2016 and 2017 . . . . . . . . . . 145 9.25 Significance for the number of plots under contract in 2017 per species using birds from 2016 and 2017 . . . . . . . . . . 146 9.26 Significance for the area size of the buffer of 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017 147 9.27 Significance for the number of plots within cluster of 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 9.28 Significance for the sum of the area sizes of all plots within 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 9.29 Exploratory table according to the presence/absence of crop on plots for all species . . . . . . . . . . . . . . . . . . . . . . 156 9.30 Exploratory table according to the presence/absence of crop on plots for all species . . . . . . . . . . . . . . . . . . . . . . 157 9.31 Exploratory table according to the presence/absence of crop on plots for all species . . . . . . . . . . . . . . . . . . . . . . 158 9.32 Exploratory table according to the presence/absence of crop on plots for all species . . . . . . . . . . . . . . . . . . . . . . 159 9.33 Exploratory table according to the presence/absence of crop on plots for all species . . . . . . . . . . . . . . . . . . . . . . 160 9.34 Logistic regression results for presence/absence of crop on plots including all species . . . . . . . . . . . . . . . . . . . . 161

viii

9.35 Logistic regression results for presence/absence of crop on plots including all species . . . . . . . . . . . . . . . . . . . . 162

List of Figures 1.1

Visuele weergave van het voorkomen van de meest abundante soorten. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.1 3.2 3.3 3.4 3.5 3.6

Histogram of sampling frequencies including all visits . . . . . Map of sampling frequencies per plot . . . . . . . . . . . . . . Histograms of counts per species . . . . . . . . . . . . . . . . Explanation definition of clusters at a distance of 100m or 300m Explanation definition of buffers at a distance of 100m or 300m Map of clustering of parcels at a distance of 100m or 300m .

16 17 18 23 23 24

5.1

Estimated prevalence with/without food crop. . . . . . . . . .

6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8

Proportion Proportion Proportion Proportion Proportion Proportion Proportion Proportion

plot plot plot plot plot plot plot plot

7.1 7.2 7.3 7.4

Significant Significant Significant Significant

effects effects effects effects

8.1

Index landscape Biological Evaluation Map 500 (upper left), 200 (upper right), Index BB Biological Evaluation Map 200 (middle left), 500 (middle right), Index BB Map Flanders 200 102

of of of of of of of of

plot size variable . . . temperature variable . Hellmann number . . . Nearest Building . . . Nearest built-upon area Nearest plot . . . . . . Nearest railroad . . . . Nearest road . . . . . .

on on on on

prevalence prevalence prevalence prevalence

estimates estimates estimates estimates

. . . . . . . .

for for for for

. . . . . . . .

44 45 46 47 48 49 50 51

Geelgors . . . Veldleeuwerik Kneu . . . . . Rietgors . . .

. . . .

87 88 89 89

Part I

Nederlandstalige Samenvatting

Chapter 1

Samenvatting en Conclusies 1.1

Doel Analyse

Akkervogels zoals de veldleeuwerik, grauwe gors en Europese patrijs kennen een sterke afname in Vlaanderen, net zoals in de rest van West-Europa. VLM investeert in het platteland met behulp van beheerovereenkomsten. Deze vijfjarige overeenkomsten bieden landbouwers de kans om op vrijwillige basis mee te werken aan de realisatie van natuur-en milieudoelstellingen in Vlaanderen. In het kader van akkervogelbescherming kan dit onder andere door bijvoorbeeld het aanleggen van gemengde grasstroken, vogelvoedselpercelen,... In dit project onderzoeken we het effect van beheersovereenkomst â€™wintervoedselgewasâ€™ op de aanwezigheid van een aantal akkervogels. De uitgevoerde analyses zijn gebaseerd op observationele vogeltellingen in de winterperiode van november 2010 tot maart 2017 in Haspengouw. De soorten die hier onderzocht werden, zijn: grauwe gors, geelgors, veldleeuwerik, kneu en rietgors. Onderzocht werd in hoeverre de landschappelijke context van het vogelvoedselperceel en specifieke eigenschappen van het vogelvoedselperceel zelf (oppervlakte, zaadmengsel, ...) invloed hebben op het voorkomen (en aantallen) van specifieke akkervogelsoorten. Verschillende variabelen werden daarbij in ogenschouw genomen. Deze omvatten variabelen welke de beschikbaarheid van vogelvoedselgewassen beschrijven, lokale landschapselementen (zoals gemengde grasstrook, hagen, hoge bomen in omgeving, enz), klimaat- en weervariabelen (zoals temperatuur en Hellmann getal), de openheid van het landschap, afstand tot bebouwing en de clustering van vo2

gelvoedselpercelen. Aangezien dit een observationele studie is, moeten we rekening houden met mogelijke fouten. Enerzijds is er confounding. Dit probleem ontstaat snel in een niet-gerandomiseerde studie en slaat op de moeilijkheid om een causaal verband aan te tonen tussen de uitkomst (voorkomen van soort) en de predictor (bv. voedselbeschikbaarheid), omdat er een derde element bestaat dat zowel op de uitkomst als op de predictor een effect heeft (bv. periode in de winter). Typisch ontstaan er ook problemen door selectiefouten, zoals een overschatting van het voorkomen van soorten wanneer vooral locaties worden bezocht waar verwacht wordt om de soorten te observeren. In een exploratieve analyse gingen we daarom op zoek naar onbalans in de steekproef, alsook uitschieters. In een volgende fase analyseerden we de data a.d.h.v. statistische modellen. Via deze modellen trachtten we een verband te zoeken tussen een uitkomst, zoals bv. de aanwezigheid van een vogelsoort, enerzijds en een set van verklarende variabelen (predictors) anderzijds. Om een binaire uitkomst, zoals aan- en afwezigheid van een soort te modelleren, werd gebruik gemaakt van een logistische regressie en bij uitbreiding van een logistisch mixed model welk rekening houdt met associatie van metingen op eenzelfde perceel. Daarnaast werd het aantal vogels per soort gemodelleerd met behulp van een negatief binomiaal model, welke een uitbreiding is van een Poisson regressie voor teldata en welk eveneens rekening houdt met extra heterogeniteit in de aantallen.

1.2

Welke soorten hebben baat bij vogelvoedselgewassen in de winter?

Over de ganse studieperiode (2012-2017; 874 bezoeken) werden 67 verschillende soorten waargenomen. 743 bezoeken vonden plaats op momenten dat er een vogelvoedselgewas op het perceel aanwezig was; 94 bezoeken op momenten dat er nog geen vogelvoedselgewas aanwezig was (nulmeting). Tabel 1.1 en Fig. 1.1 geven een overzicht van de 24 vaakst waargenomen soorten op vogelvoedselpercelen in Haspengouw. Merk op er meer percelen waren met dan zonder teelt en dat de locaties met teelten frequenter werden bezocht. De aanwezigheid van het vogelvoedselgewas is daarbij een significante factor voor het voorkomen van geelgors, rietgors, torenvalk, buizerd, kneu, fazant, blauwe reiger en groenling. De impact is het grootst voor 3

kneu, groenling en blauwe reiger. Kneu en groenling verkiezen een zaaddieet. Kneu eet liefst zaden van gewassen en onkruiden, terwijl groenling een gevarieerder dieet heeft, waaronder naast zaden, bessen en insecten worden geprefereerd. Blauwe reiger wordt aangetrokken door de grote hoeveelheden prooidieren, meer bepaald muizen, die tijdens de wintermaanden op deze graanakkers aanwezig zijn. Table 1.1: Samenvattende tabel m.b.t. het voorkomen van de meest abundante soorten. Soort

Effect teelt

Geelgors Rietgors Torenvalk Buizerd Vink Kneu Veldleeuwerik Fazant Blauwe Kiekendief Blauwe Reiger Groenling Grauwe Gors Merel Kramsvogel Winterkoning Houtduif Koolmees Pimpelmees Kievit Patrijs Zwarte Kraai Grote Bonte Specht Roek Ekster

positief positief positief positief geen effect positief geen effect positief geen effect positief positief geen effect geen effect geen effect geen effect geen effect geen effect geen effect geen effect geen effect negatief geen effect geen effect geen effect

1.3

Totaal aantal geobserveerde individuen teelt geen teelt 8008 217 2363 22 273 14 295 9 3847 65 7067 60 1754 184 529 11 151 5 200 1 1735 1 1651 152 122 6 1246 50 42 1 1183 0 98 20 93 2 799 4 61 10 8 29 5 1 226 70 2 1

Aantal perceelbezoeken teelt 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743 743

geen teelt 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94 94

Aantal succesvolle bezoeken (soort is aanwezig) teelt geen teelt 63 (8.4%) 2 (2.1%) 241 (32.4%) 8 (8.5%) 230 (30.9%) 12 (12.7%) 221(29.7%) 9 (9.5%) 156 (20.9%) 13 (13.8%) 169 (22.7%) 1 (1%) 156 (20.9%) 13 (13.8%) 140 (18.8%) 7 (7.4%) 131 (17.6%) 7 (7.4%) 125 (16.8%) 1 (1%) 96 (12.9%) 1 (1%) 63 (8.4%) 2 (2.1%) 46 (6.2%) 3 (3.2%) 37 (4.9%) 3(3.1%) 33 (4.4%) 1 (1%) 32 (4.3%) 4 (4.2%) 31 (4.1%) 6 (6.3%) 21 (2.8%) 1 (1%) 17 (2.2%) 3 (3.1%) 12 (1.6%) 1 (1%) 5 (0.6%) 5 (5.3%) 5 (0.6%) 1 (1%) 4 (0.5%) 1 (1%) 1 (0.1%) 1 (0.1%)

Vijf soorten in detail geanalyseerd

Vijf soorten (grauwe gors, geelgors, veldleeuwerik, kneu en rietgors) werden in detail geanalyseerd. Onderzocht werd in hoeverre de landschappelijke context van het vogelvoedselperceel en specifieke eigenschappen van het vogelvoedselperceel (oppervlakte, zaadmengel, ...) invloed hebben op het succes van vogelvoedselpercelen voor deze soorten. Daarnaast werd onderzocht of de weersituatie een effect had op het voorkomen van deze soorten. Er werden statistische modellen opgesteld om zowel de aanwezigheid van 4

Figure 1.1: Visuele weergave van het voorkomen van de meest abundante soorten. soorten als het aantal individuen per soort te onderzoeken. Door het beperkt aantal waarnemingen voor grauwe gors (65 keer waargenomen op 837 perceelsbezoeken) moeten we voorzichtig zijn met de conclusies voor deze soort.

1.3.1

Karakteristieken van het vogelvoedselperceel

Een nadruk binnen deze studie ligt op het effect van karakteristieken van de vogelvoedselgewassen op het voorkomen van de doelsoorten. Er werd meer bepaald onderzocht in welke mate keuze van zaadmengsel, oppervlakte van het perceel en het clusteren van percelen een invloed hadden op het voorkomen van deze soorten. 1.3.1.1

Keuze van het zaadmengsel

Drie teeltcombinaties werden onderzocht: - Graan (tarwe of haver) zonder bijmenging 5

- Graan (tarwe of haver) met bijmenging vlas - Graan (tarwe of haver) met bijmenging bladrammanas De volgende resultaten werden gevonden: • Geelgors (aanwezigheid en aantal exemplaren) correleert positief met percelen ingezaaid met graan (tarwe of haver) en percelen ingezaaid met graan (tarwe of haver) in combinatie met vlas. • Kneu heeft een uitgesproken voorkeur voor percelen ingezaaid met graan (tarwe of haver) in combinatie met bladrammanas. • Rietgors correleert positief met de drie teelcombinaties; maar met een voorkeur voor graan zonder bijmenging. • Voor grauwe gors kon geen specifieke voorkeur worden waargenomen, wat zoals eerder aangegeven te wijten kan zijn aan het lage aantal observaties van deze soort. Het type teeltcombinatie lijkt niet van belang voor veldleeuwerik. Dit ligt in lijn met resultaten eerder in dit verslag, die niet konden aantonen dat deze soort meer gezien wordt op percelen met teelt dan zonder teelt. Voor enkele soorten (rietgors, geelgors, kneu) blijken de teeltcombinaties in vogelvoedselpercelen belangrijk te zijn. Het is duidelijk dat geelgors en rietgors niet erg kieskeurig zijn wat betreft het type inmenging, i.t.t. kneu, die een combinatie van zaden (gewassen) prefereert. Rietgors prefereert de zaden van graan voor zijn dieet en voor deze soort zal dit waarschijnlijk al genoeg zijn als overwintergewas. Wanneer de resultaten voor alle soorten naast elkaar worden gelegd, is er geen eenduidige conclusie over welke teeltcombinatie het beste resultaat oplevert. Een overkoepelend advies is om een combinatie te blijven handhaven, met een mogelijke toename in graan, wat goed is voor soorten die een eenvoudige inmenging verkiezen.

1.3.1.2

Oppervlakte van het vogelvoedselperceel

In een analyse van het mogelijk effect van oppervlakte op het voorkomen van soorten werden de volgende resultaten gevonden: • Voor rietgors werd in alle modellen een positief verband gevonden met de oppervlakte van het perceel. Op grotere percelen wordt de soort meer gezien en er komen ook meer exemplaren van deze soort voor. 6

• In de meeste modellen werd een soortgelijk effect van oppervlakte van het perceel op het voorkomen van geelgors gevonden, zowel op de frequentie van voorkomen als het aantal exemplaren. • Voor kneu werd een positief verband gevonden tussen de oppervlakte van het perceel en de aanwezigheid van de soort, maar niet op het aantal exemplaren. • Voor grauwe gors en veldleeuwerik kon dit verband niet worden aangetoond. Er is een indicatie dat grauwe gors grote percelen verkiest, maar er is meer onderzoek nodig om hier duidelijke uitspraken over te kunnen doen. De aanwezigheid en de aantallen van veldleeuwerik zijn niet afhankelijk van perceeloppervlakte. 1.3.1.3

Effect van (ruimtelijk) clusteren van vogelvoedselpercelen

Onderzocht werd of het clusteren van vogelvoedselgewassen (vogelvoedselpercelen gelegen binnen een afstand van 100 tot 300 meter van elkaar) een effect heeft op het voorkomen en aantallen van akkervogels. • Een duidelijk effect van clustering werd enkel waargenomen voor rietgors. Hoe dichter en hoe meer vogelvoedselpercelen in cluster bij elkaar liggen hoe vaker en talrijker rietgors wordt waargenomen. Naarmate vogelvoedselpercelen meer geı̈soleerd liggen, wordt rietgors minder waargenomen. • Het logistisch regressiemodel toont een positief verband tussen het voorkomen van grauwe gors en het aantal vogelvoedselpercelen binnen een straal van 100 tot 300 meter maar dit effect wordt niet bevestigd door de andere modellen. Bovendien is de analyse gebaseerd op een beperkt aantal waarnemingen waardoor voorzichtigheid geboden is. Er is dus een indicatie dat grauwe gors het goed doet wanneer meerdere percelen geclusterd worden, maar er is meer onderzoek nodig om dit verband juister te kwantificeren. • Voor geelgors, veldleeuwerik en kneu werd geen (eenduidig) verband aangetoond met clusteren van vogelvoedselpercelen.

1.3.2

Effect van landschapskenmerken op deze soorten

Onderzocht werd in hoeverre de ligging van diverse landschapselementen (houtkant, haag, bomenrij), bos, fruitplantages, woningen/tuinen ’grenzend 7

aan’ of ’in de onmiddellijke omgeving (50 meter)’ van het vogelvoedselperceel invloed heeft op het voorkomen (en aantallen) van akkervogels. Het effect van tuinen/woningen, hoogstamboomgaarden, spoorwegen werd niet verder onderzocht omwille van de lage frequentie waarin deze elementen in de dataset voorkwamen, met een grote foutenmarge op de inschatting van hun effecten als gevolg. 1.3.2.1

Houtkanten/hoge bomen(rij) langs VVG-perceel

Opvallend is dat in veel analyses houtkanten en bomenrijen een aanzienlijke correlatie met het voorkomen en/of aantallen van soorten vertoonden. • Voor geelgors (zowel aanwezigheid als aantal exemplaren) werd een positief effect waargenomen indien het vogelvoedselperceel gelegen is binnen een straal van 50 meter van een houtkant. • Voor grauwe gors is er een effect, maar het is minder eenduidig aantoonbaar: In een aantal modellen wordt een positief effect op aanwezigheid/afwezigheid van de soort waargenomen op percelen gelegen langs of in de onmiddellijke omgeving (50 meter) van een houtkant. Wanneer er echter wordt gecorrigeerd voor een structuur in de data, die ontstaat omdat sommige percelen meerdere malen werden bezocht, is dit verband minder eenduidig aantoonbaar (borderline positief effect voor houtkanten gelegen binnen een straal van 50 meter). Ook wordt een positief effect vastgesteld op het aantal waargenomen exemplaren grauwe gors (vermeerdering met factor 4,47) bij VVG-percelen gelegen langs een houtkant. Dit effect vermindert bij verder afgelegen houtkanten (binnen straal van 50 meter). Deze resultaten geven aan dat houtkanten waarschijnlijk van belang zijn voor deze soort. Er moeten meer data verzameld worden over deze soort om een juistere uitspraak te doen. • Het voorkomen (zowel aanwezigheid als aantal exemplaren) van veldleeuwerik wordt (significant) negatief beı̈nvloed indien de vogelvoedselpercelen gelegen zijn in de onmiddellijke omgeving (50 meter) van een houtkant of hoge bomenrij. • Het voorkomen van rietgors wordt negatief beı̈nvloed door het voorkomen van hoge bomenrijen in de omgeving van het VVG-perceel.

• Er werd geen verband vastgesteld tussen het voorkomen van kneu en de aanwezigheid van houtkanten en/of hoge bomenrijen in de onmiddellijke omgeving van het vogelvoedselperceel. 1.3.2.2

Fruitplantages/bos langs VVG-perceel

Het voorkomen van bos en/of fruitplantages kon minder vaak worden gecorreleerd met het voorkomen van de doelsoorten: • Het voorkomen (zowel aanwezigheid als aantal exemplaren) van veldleeuwerik wordt (significant) negatief beı̈nvloed indien de vogelvoedselpercelen gelegen zijn in de onmiddellijke omgeving (50 meter) van een bos of fruitplantage. • Verder kon geen enkel verband worden aangetoond met het voorkomen bossen en fruitplantages in de onmiddellijke omgeving van het vogelvoedselperceel en het voorkomen (en talrijkheid) van geelgors, grauwe gors, rietgors en kneu. 1.3.2.3

Ligging langs wegen

Er werden geen eenduidige resultaten gevonden over een mogelijk effect van de nabijheid van wegen op het voorkomen van soorten. Enkele verbanden werden gezien, maar deze geven enigszins tegenstrijdige boodschappen en dienen dus voorzichtig geı̈nterpreteerd te worden. • Ligging langs onverharde landbouwwegen heeft een significant positief effect op het voorkomen van geelgors. Anderzijds is er een negatieve invloed van wegen met doorgaand verkeer op het aantal geelgorzen. • Voor grauwe gors, kneu, rietgors en veldleeuwerik kon geen effect van onverharde/verharde wegen aangetoond worden. 1.3.2.4

Afstand tot bebouwing

Een effect van de nabijheid van bebouwing werd enkel waargenomen bij veldleeuwerik: • Toenemende afstand tot bebouwing heeft een positief effect op het voorkomen van veldleeuwerik. Dit ligt in lijn met wat eerder werd gevonden over deze soort: veldleeuwerik wordt niet aangetrokken door teelten en verkiest een open gebied, zonder te veel landschapselementen, zoals houtkanten, woningen, etc. 9

• Voor geelgors, kneu, rietgors en grauwe gors werd geen duidelijk verband aangetoond met de afstand tot bebouwing. 1.3.2.5

Landschapstype (open/gesloten)

Op basis van BWK en ALV-perceelsregistraties werd het actuele bodemgebruik in kaart gebracht en omgezet naar open (lage gewassen, . . . ) en gesloten (bos, fruitplantages, . . . ) karakter. Op basis hiervan werd voor ieder vogelvoedselgewas een index berekend die de verhouding open/gesloten weergeeft binnen een straal van 200 en 500 meter. De analyse bracht geen enkel verband aan het licht tussen het voorkomen van soorten en de mate van openheid/geslotenheid in de omgeving van het vogelvoedselgewas. Mogelijk is dit ook te wijten aan de manier waarop de index werd opgebouwd. Landschapstypering op basis van openheid/geslotenheid is moeilijk kwantitatief te bepalen.

1.4

Invloed van temperatuur en weersituatie tijdens moment van waarneming

We stelden ons de vraag: Hebben temperatuur en wintertype (Hellmanngetal = maat voor het aantal vriesdagen, op maandbasis) invloed op de waarnemingen? Dit zegt iets over het voorkomen van de soort, maar ook over de kans op waarneming? • Veldleeuwerik en rietgors worden significant vaker waargenomen op koude dagen. • Er zijn aanwijzingen dat de aanwezigheid of kans op waarneming van geelgors en kneu correleert met hogere temperaturen; veldleeuwerik en kneu worden bovendien minder waargenomen in periodes met veel vriesdagen (Hellmanngetal). • Waarnemingen van grauwe gors vertonen geen verband met temperatuur of Hellmanngetal.

1.5

Overzicht per soort

Als conclusie kan worden gesteld dat geelgors, rietgors en kneu soorten zijn waarvoor op verschillende manieren werd aangetoond dat zij baat hebben 10

bij het inrichten van vogelvoedselpercelen. Geelgors en rietgors zijn minder kieskeurig wat betreft het type teelt, terwijl kneu een duidelijke voorkeur voor een mengsel van graan en bladrammenas vertoont. Voor veldleeuwerik zijn deze percelen minder belangrijk, maar een landschap met genoeg open elementen is voor deze soort vooral essentieel. Dit blijkt uit verschillende analyses, hoewel er geen rechtsreeks verband met de geslotenheid van het landschap kon worden aangetoond. Dit heeft mogelijks te maken met een de maat waarmee geslotenheid wordt uitgedrukt. Verder onderzoek hieromtrent is vereist. Voor grauwe gors zijn er aanwijzingen dat de soort het goed doet op plaatsen met houtkanten en wanneer percelen geclusterd voorkomen. Het is belangrijk om hierbij te herhalen dat de data over deze soort beperkt waren, waardoor de power van de modellen ondermaats was. Dit resulteert veelal in het onopgemerkt blijven van significante effecten; m.a.w. het resultaat dat nu suggereert dat vogelvoedselgewassen geen positief effect hebben op het voorkomen van grauwe gors kan mogelijk weerlegd worden wanneer er meer data zijn. In Tabel 1.2 worden tenslotte de belangrijkste bevindingen kort samengevat. Table 1.2: Samenvatting van de belangrijkste resultaten uit de analyses. Grauwe Gors Aanwezigheid VVG Keuze zaadmengsel Oppervlakte vvg-perceel Clustering vvg-percelen Houtkant Bomen(rij) Fruitplantage/bos Onverharde weg Doorgaand verkeer Bebouwing

Geelgors + + +

Veldleeuwerik

(+) +

(+) (-) -

Kneu + + (+)

Rietgors + + + + -

Part II

Introduction, Materials and Methods

Chapter 2

Introduction Populations of farmland birds such as skylark (Veldleeuwerik, Alauda arvensis), cornbunting (grauwe gors, Emberiza calandra), partridge (Europese patrijs, Perdix perdix) show a strong decrease in Flanders, just as in the rest of Western Europe. Increase in scale and intensification in agriculture changed the landscape, making it difficult for farmland birds to find sufficient food and breeding opportunity in rural areas. VLM informs and sensitizes farmers, stimulating them to adopt sustainable management practices by implementing agro-environmental (management) agreements (AEM). These give farmers the opportunity to integrate nature conservation measures in their farming practices. For the protection of farmland birds, following agreements (AEM) are possible: â€˘ Mixed grass strips: providing insects (summer food) and nesting place in summer â€˘ Birdfeed crops: providing winter food This research focuses on bird observations in the winter period (between mid-November to mid-March) on parcels with bird feed crop. The bird observations took place during the period from November 2010 to March 2017.

2.1

Objective

The objective of this project is to investigate the effectiveness of agroenvironmental agreements (AEM) for farmland birds, and in particular in-

vestigate which bird food crop agreements can be used in a more targeted way. This project is targeted to five common farmland bird species, namely • Grauwe Gors (Emberiza calandra), • Geelgors (Emberiza citrinella), • Veldleeuwerik (Alauda arvensis), • Kneu (Linaria cannabina) and • Rietgors (Emberiza schoeniclus). We investigate the impact of different environmental schemes on both the presence/absence of birds on parcels, as well as the number of birds on the parcels.

2.2

Outline

In Chapter 2, we explain the data to be used in this study, while Chapter 3 explains the statistical methods used. Part II gives the results of the effect of the availability of bird food on presence/absence of all investigated farmland birds. Part III gives the results of environmental factors on both presence/absence and number of birds, for the five common farmland birds. Part IV focuses on the effect of openness and closeness of the landscape on presence/absence and number of birds, also for the five common farmland birds. In Part V, the effect of clustering of plots under contract is investigated. Finally, in Part VI, a summary is provided, with overall conclusions and recommendations.

Chapter 3

Materials This project focuses on bird observations in the winter period (between midNovember to mid-March) on parcels with bird feed crop. The relationship with the implementation of agro-environmental (management) agreements is of interest. In this section, we specify in detail the response variable of interest (the outcome variable), as well as the environmental variables related to the parcel and time of visit of the parcel which could have influenced the bird observations (the independent variables).

3.1

Outcome variables

For each of the five species considered, the number of birds of a specific species observed during a visit of a parcel was recorded. In this report, we will investigate the impact of management of parcels and environmental schemes on both â€˘ the number of birds of a specific species observed (a count variable) â€˘ the presence/absence of a specific species (a binary variable). The binary variable is defined as: Ybinary outcome (present/not present) =

0 if Y = 0 1 otherwise,

(3.1)

where Y is the number of birds observed. A total of 177 plots were visited from 2012 to 2017. The number of visits per year is different for the different parcels, and not all plots were visited every year. The number of visits per plot is denoted as the sampling 15

frequency. The sampling frequencies are presented in Figure 3.1 and 3.2. Figure 3.1 is a histogram plot of the sampling frequencies, while Figure 3.2 shows the number of samples per parcel on a map. There is a large amount of singletons present in the data, namely plots which were visited only once.

Figure 3.1: Histogram of sampling frequencies including all visits

Figure 3.2: Map of sampling frequencies per plot A histogram of the number of birds observed per species is presented in figure 3.3. The presence/absence of the birds is presented in table 3.1 Table 3.1: Presence/absence of birds per species Total visits 873 100% Total visits 873 100%

Grauwe Gors Present Absent 68 805 7.8% 92.2% Veldleeuwerik Present Absent 174 699 19.9% 80.1%

Geelgors Present Absent 393 480 45% 55% Kneu Present Absent 172 701 19.7% 80.3%

Rietgors Present Absent 252 621 28.9% 71.1%

Figure 3.3: Histograms of counts per species Remark 1: Note that two location covariates were provided in the original data set, Perceel ID and Object ID. Since we considered a relationship of one Perceel ID to one Object ID for each plot, five duplicate locations were found. A duplicate location was defined as a plot that was recorded twice (with the same Perceel ID and Object ID) in the Access data set “startperceelgegevens”.

3.2

Independent variables

A selection of explanatory variables which might influence the farmland birds were investigated. These variables describe the effect of small landscape elements (for instance forest, fruit cultivation) in the proximity of the plots, the landscape in general, the bird food crop, the disturbance (for instance traffic, build up area’s), and the effect of clustering of bird food plots.

3.2.1

Landscape, food crop and disturbance

These independent variables can be grouped according to the type of data: categorical variables and continuous variables. The categorical variables are variables in which the observations can take on only a limited number of values. For example, presence or absence. The categorical variables of interest in this study which described landscape, food availability and disturbance are: • availability of food on the plots, • presence of bird food crop on the plots, • adjacency to hedge, • adjacency to tree colonnade, • adjacency to garden, • adjacency to fruit cultivation, • adjacency to tall cone orchard, • faunal grass stretch, • ongoing traffic, • unpaved road, • rail road and • local farm traffic. Also, the presence of tall cone orchard, food availability, forest, hedge, fruit cultivation, garden and tree collonade at 50 meters away from the plot were investigated. The continuous variables are variables in which observations are measured on a continuum scale. They can have almost any numerical value. In this study, the continuous variables considered which described environment and disturbance are: • the plot size, • the temperature at the sampling moment, • the type of winter characterized by the Hellmann number, and 19

• the distance from a specific plot to the nearest building, built-upon area, another plot, railroad and road. The Hellmann number is obtained by summing the absolute values of all mean day temperature that were below zero within a time period. Note that the last mentioned continuous variables were collected in 2017 only, and it was assumed that these characteristics did not change over time. All the variables are summarized in Table 3.2 and 3.3.

3.2.2

Openness and closeness

Another five variables characterize the openness and closeness of the plots, where a value of 0 represents an open area, while a value of 1 corresponds to a closed plot. The data was available for two years only, namely for 2012, in which characteristics were measured on 111 plots, and for 2017, on 305 plots. For the two years, 90 plots were common in both datasets. We separately did an analysis for data obtained in 2012 and 2017. For the dataset including the characteristics of the environment of 2017, we observed that not many visits were done in 2017. Therefore we did not have enough information to test these variables for the year 2017. As an alternative, we assumed that the openness/closeness of the plots remained constant in 2016 and 2017, and performed the analysis for all visits made in 2016 and 2017. The variables are summarized in Table 3.3.

3.2.3

Clustering

Lastly, information about cluster variables measured in 2017 were provided. The following variables are defined: • The number of plots and the area size of the buffer around the plot. • The number of plots within a cluster is recorded. • The sum of area sizes of all plots, all measured within 100, 150, 200, 250 and 300 meters. The definition of clusters and buffers is graphically presented in Figures 3.4 and 3.5. The map of clustering or parcels at a distance of 100m and 300m is presented in 3.6. Since we consider that the cluster’s dimension changed considerably over the study period, these variables were evaluated taking into account only the visits made in 2016 and 2017. The variables are summarized in Table 3.3. 20

Variable name (Dutch) voedselbeschikbaarheid (voedsel)gewas houtkant bos bomen tuin fruitteelt hoogstamboomgaard gemengde grasstrook doorgaand verkeer onverarde weg spoorweg lokaal landbouwverkeer hoogstamboomgaard 50m VVG 50m bos 50m houtkant 50m fruitteelt 50m tuin/serre/gebouw 50m bomen(rij) 50m oppervlakte temperatuur Helmanngetal

Variable name (English) food availability of growth bird food crop adjacency to a hedge adjacency to a forest adjacency to a tree colonnade adjacency to a garden adjacency to a fruit cultivation adjacency to a tall cone orchard adjacency to a faunal grass stretch adjacency to an ongoing traffic adjacency to an unpaved road adjacency to a rail road adjacency to a local farm traffic tall cone orchard within 50 meters bird food availability within 50 meters forest within 50 meters hedge within 50 meters fruit cultivation within 50 meters garden within 50 meters tree collonade within 50 meters plot size weather at the sampling moment type of winter

Table 3.2: Explanatory variables (part 1) Type of variable categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical categorical continuous continuous continuous

Variable name (Dutch) afstand in meter tot dichtsbijgelegen gebouw (laag GRB gbg Informatie Vlaanderen) afstand in meter tot dichtsbijgelegen bebouwde zone (polygoonaggregatie van GRB gbg Informatie Vlaanderen) afstand in meter tot dichtsbijgelegen VVG perceel actief in 2017 afstand in meter tot dichtsbijgelegen treinspoor (laag GRB sbn Informatie Vlaanderen) afstand in meter tot dichtsbijgelegen weg (laag Wegenregister Informatie Vlaanderen) Index bebouwing BWK (Biologische WaarderingsKaart) 200m Index bebouwing BWK (Biologische WaarderingsKaart) 500m Index landschap BWK (Biologische WaarderingsKaart) 200m Index landschap BWK (Biologische WaarderingsKaart) 500m Index bebouwing GRB (Basiskaart Vlaanderen) 200m aantal VVG percelen actief in 2017 binnen een afstand van 100m, 150m, 200m, 250m, 300m oppervlakte in m2 van de buffer van 100m, 150m, 200m, 250m, 300m rond het VVG perceel som van de oppervlakte in m2 van alle VVG percelen binnen een afstand van 100m, 150m, 200m, 250m, 300m aantal VVG percelen actief in 2017 gelegen in de cluster van 100m, 150m, 200m, 250m, 300m

continuous continuous continuous continuous continuous continuous continuous continuous continuous continuous

distance to the built-upon area distance to another plot distance to a railroad distance to a road Index BB Biological Evaluation Map 200 Index BB Biological Evaluation Map 500 Index landscape Biological Evaluation Map 200 Index landscape Biological Evaluation Map 500 Index BB Map Flanders 200 the number of plots (under contract in 2017) within 100, 150, 200, 250 and 300 meters the area size of the buffer around the plot (including the plot) within 100, 150, 200, 250 and 300 meters the sum of area sizes of all plots within 100, 150, 200, 250 and 300 meters the number of plots (under contract in 2017) within cluster of 100, 150, 200, 250 and 300 meters

continuous

Type of variable continuous

Variable name (English) distance to the nearest building

Table 3.3: Explanatory variables (part 2)

Figure 3.4: Explanation definition of clusters at a distance of 100m or 300m

Figure 3.5: Explanation definition of buffers at a distance of 100m or 300m

Figure 3.6: Map of clustering of parcels at a distance of 100m or 300m

In part II of this report, we investigate the effect of bird food crop on presence/absence of birds of any species. In part III of this report, we investigate the effects of landscape, food availability and disturbance on the birds. Part IV investigates the impact of openness and closeness, while part V investigate clustering of plots.

Chapter 4

Methods Randomized controlled trials are the best tool to investigate the association between outcome and explanatory variables. However, in many settings, this cannot or is difficult to perform. As an alternative, observational studies are an important source of information. But, data should be analyzed and interpreted with special attention to bias. Bias is the tendency for under- or overestimation of a trend, most likely due to unbalanced sampling designs, data of poor quality or incorrect statistical model assumptions. A first assumption which is made is that the sample of plots is a representative sample in the environment. In the statistical analysis, we draw inference about a larger population of plots with similar conditions. Selection bias could arise from the way plots are selected in the study. For example, if only plots are selected in which it is known beforehand that there is high change of seeing birds of a particular species, this would lead to preferential sampling bias. Such information bias distorts parameter estimation in the analysis. A second issue in observational studies is confounding. A confounding variable is a variable that influences both the outcome variable (e.g. the number of birds) and the independent variable (the environmental variable). This can cause a spurious association. In a controlled design, the randomization process makes it possible to ascribe a difference in outcome between two groups to the factor differentiates between groups, as within the context of a controlled experiment, these groups are assumed to be different only in the factor under investigation. In an observational study, as this one, groups lack comparability. Therefore, interpretation of effects should be done with care, without the interpretation of a direct association or correlation. Two types of outcomes were investigated for each species, a binary outcome (presence/absence) and a count outcome, which was originally 26

recorded. A logistic regression and a generalized linear mixed model (GLMM) were used for the binary outcome. A negative binomial regression model was used for the count data.

4.1

Exploratory analysis

In the exploratory analysis, we will look at possible issues that might influence the statistical analysis, in particular because of the observational nature of the data. We will look at (1) imbalance in the data, and (2) outlying observations. For the setting of categorical independent variables, a cross-tabulation of the presence/absence of the birds will be made. In this way we can check whether or not there is a large imbalance in the variable, meaning that e.g. very few parcels are observed with presence of a certain characteristic. In such a case, the statistical analysis will be very poor, and will not be performed. For the setting of continuous independent variable, outliers might determine the final results and conclusions. A graphical tool to detect outliers is a scatter plot of the number of birds as compared to independent variable. We will check whether there are outliers in the independent variables, in the sense that there are observations which have a relatively large (or small) value as compared to the majority of observations. If such observations exist, the statistical analysis might be dominated by these outliers. Therefore, as a sensitivity analysis, the statistical analysis will then be performed again without these outliers.

4.2 4.2.1

Statistical analysis of presence/absence of birds Logistic regression

The logistic regression model was used for a binary response variable, which is linked to the predictor by a logit link function as follows (Agresti, 2002) : Yi ∼ Bernoulli(πi ),

logit(πi ) = log

π i = β0 + β1 ∗ Xi 1 − πi

(4.1)

where i = 1,...,n is an index for the visit, Xi is the covariate corresponding to this visit and Yi is the binary outcome reflecting presence or absence of 27

the species of interest. The parameter of interest, πi is the probability to observe at least one bird of a particular species during a visit. As mentioned by (Agresti, 2002), the sign of β1 determines whether πi is increasing or decreasing as Xi increases. We will interpret eβ1 as an odds ratio, meaning that the odds of the presence of a bird increases multiplicatively by eβ1 for every 1-unit increase in Xi . When the independent variable Xi is binary, reflecting presence/absence of some characteristic (and coded as 1/0), the probability of observing a exp(β0 +β1 ) bird when the characteristic is present is given by 1+exp(β , while the 0 +β1 ) probability of observing a bird when the characteristic is absent is given exp(β0 ) by 1+exp(β . This means that the odds of observing the bird (i.e. ratio of 0) presence to absence of bird), with or without the characteristic of the plot, is exp(β0 + β1 ) to exp(β0 ). It is important to mention that the logistic regression neglects the correlation structure within plots due to repeated visits within the same plot, which occurs in this dataset. In general, ignoring this correlation leads to an underestimation of the p-values, potentially resulting in effects being deemed significant when they should not be. However, a logistic regression is relatively easy to fit, even to problematic data, and therefore was used as starting point. In order to take plot-specific correlation into account, we further include a generalized linear mixed model which takes into account this characteristic of the data.

4.2.2

Generalized Linear Mixed Model (GLMM)

The generalized linear mixed model is the most frequently used random effects model in the context of discrete repeated measurements (Molenberghs and Verbeke, 2005). The generalized linear mixed model is an extension of the generalized linear model, the logistic regression in our case, to the context of clustered measurements. Clustering occurs in this project as some measurements are taken on the same plots. Observations within a cluster tend to be more alike than observations from different clusters, thus, they can be positively correlated (Agresti, 2002). We let Yij be the binary outcome (absent/present) for visit j on plot i: Yij |bi ∼ Bernoulli(πij ), logit(πij ) = log

π ij = β0 + bi + β1 ∗ Xi , 1 − πij

(4.2)

where bi is a plot-specific intercept. Thus, every plot has its own intercept β0 + bi . In order to make estimation possible, it is assumed that the plot specific intercepts follow a normal distribution with mean 0 and variance σb2 . The larger σb2 , the more deviation amongst plots, and the stronger the correlation of measurements of the same plot. There were situations when GLMM did not converge. This can be caused by a large number of plots which were visited only once, therefore they had only one sampling moment. Interpretation of parameters is similar as in the logistic regression, but now interpreted as the effect, given that the measurement is taken on a specific plot.

4.3 4.3.1

Statistical analysis of number of birds Negative Binomial regression (NB)

The simplest distribution for the count data is the Poisson (Agresti, 2002). The Poisson distribution has identical mean and variance. However, in our case, the variances are much larger than the means reflecting overdispersion. Count data often show overdispersion, with the variance of the response exceeding the mean. Causes of overdispersion are: multiple measurements taken at the same parcels, zero inflation of the outcome variable, missing covariates. Ignoring such heterogeneity could cause serious bias in the parameter estimates. Therefore, the negative binomial model used in this project has the following form: Yi ∼ N egBin(µi , k) log(µi ) = β0 + β1 ∗ Xi

(4.3)

This model does in a flexible way account for overdispersion. The negative binomial distribution has mean E(Yi ) = µi and the variance is a function of its mean and the dispersion parameter k, var(Y ) = µ + µ2 /k. For all species a large amount of non-events were found, which in some situations, made the convergence hard to achieve for the negative binomial model.

4.4

Remarks

In this project, we analyzed the effect of all explanatory variables on the outcome through a univariate analysis only. No multivariate analysis was performed, because we wanted to avoid the problem of multicollinearity 29

which is caused by covariates being correlated with other covariates from the model. Some convergence issues appeared in the negative binomial regression. This is due to the highly imbalanced nature of the data and zero-inflation of the counts. Also, for some of the species, limited positive counts are observed. When no convergence is obtained, results are not reliable, and are therefore not presented. By translating the response count variables into binary ones, we were able to estimate the relationship with some variables that were difficult to model in the count data setting, due to convergence issues caused by zero-inflation. The level of significance used is 0.01 in order to avoid a multiplicity problem, since a large amount of tests were performed. A 99% confidence interval is calculated for each estimate, where a very wide interval suggests a less precise estimate. A borderline situation occurs when the p-value of the test statistic is equal or slightly higher than the significance level used. The analysis was entirely performed using R version 3.4.2.

Part III

Results: bird food crop for all species

Chapter 5

Results: impact of bird food crop 5.1

Objective

In this chapter, we investigate how bird species benefited from one of the agro-environmental agreements (AEM): bird food crop. A logistic regression model was used in order to analyze the influence of “bird food crop” on the presence/absence of a specific species.

5.2

Data exploration

The data set contained 67 different species. However, only the more common 24 species were analyzed by the logistic regression model. The sampling frequency (the total number of visits) according to the “bird food crop” type for each species are presented in Table 9.29. Over the entire study period (from 2012 until 2017), a total of 874 visits were made. From 874 visits, on 37 of these no information was recorded about the “bird food crop” presence or absence on plots. Thus, we have 837 visits with the information about “bird food crop” presence or absence. Irrespective of the species, on 743 visits a “bird food crop” was present on a plot, while in 94 visits “bird food crop” was not present on a plot. We calculated for each species the number of times a bird for a specific species is present on a plot according to the sampling frequency. In order to be able to estimate the effect of this covariate, the birds should be present on both types of plots, with and without “bird food crop”. For 43 species, the birds were not seen on both situations, making a statistical analysis impossible.

5.3

Results

In Table 5.1, we present the significance found by the logistic regression model for the 24 most common species. A positive statistical significant influence of the â€œbird food cropâ€? variable was found for Geelgors, Kneu, Rietgors, Torenvalk, Buizerd, Fazant, Groenling and Blauwe Reiger, while a negative statistical significant effect is found for Zwarte Kraai species. The detailed analysis results are presented in Tables 9.34 and 9.35 in Appendix. A graphical display of the results is given in Figure 5.1. Note that the prevalence of several of the species is so rare (in table, all species below Grauwe Gors), that interpretation is limited. Table 5.1: Significance using Logistic regression of presence/absence of crop for all species Species Geelgors Rietgors Torenvalk Buizerd Vink Kneu Veldleeuwerik Fazant Blauwe Kiekendief Blauwe Reiger Groenling Grauwe Gors Merel Kramsvogel Winterkoning Houtduif Koolmees Pimpelmees Kievit Patrijs Zwarte Kraai Grote Bonte Specht Roek

Effect crop positive positive positive positive no effect positive no effect positive no effect positive positive no effect no effect no effect no effect no effect no effect no effect no effect no effect negative no effect no effect To be continued

Figure 5.1: Estimated prevalence with/without food crop. Table 5.1 (continued) Species Crop Ekster no effect

5.3.1

Geelgors species

We find that the odds to observe a Geelgors on a plot with â€œbird food cropâ€? available is e1.033 = 2.8(1.4; 5.36) times the odds of Geelgors if a plot does 34

not have “bird food crop” available. This means that the odds of Geelgors at a plot with no “bird food crop” is approximately 1 to 3, while the odds of Geelgors at a plot on which “bird food crop” is present is almost 3 times larger, 3 to 3.

5.3.2

Kneu species

The odds to observe a Kneu on a plot with “bird food crop” is e3.310 = 27.38(2.03; 368.03) times the odds of Kneu if a plot does not have “bird food crop”. For plots with no “bird food crop”, for every 93 visits where a Kneu bird is not observed, only at 1 visit a Kneu bird is observed. Whereas, for plots with “bird food crop”, for every 93 visits where Kneu is absent, there will be 27 visits where Kneu is observed.

5.3.3

Rietgors species

The odds to observe a Rietgors on a plot with “bird food crop” is e1.641 = 5.16(1.95; 13.66) times the odds of Rietgors if a plot does not have “bird food crop”. This means that the odds of Rietgors at a plot with no “bird food crop” available is approximately 1 to 11, while the odds of Rietgors at a plot on which “bird food crop” is present is almost 5 times larger, 5 to 11.

5.3.4

Torenvalk species

The odds to observe a Torenvalk on a plot with “bird food crop” is e1.117 = 3.05(1.34; 6.95) times the odds of Torenvalk if a plot does not have “bird food crop”. This means that the odds of Torenvalk at a plot with no “bird food crop” is approximately 1 to 7, while the odds of Torenvalk at a plot on which “bird food crop” is present is almost 3 times larger, 3 to 7.

5.3.5

Buizerd species

The odds to observe a Buizerd on a plot with “bird food crop” is e1.117 = 3.99(1.58; 10.06) times the odds of Buizerd if a plot does not have “bird food crop”. This means that the odds of Buizerd at a plot with no “bird food crop” is approximately 1 to 9, while the odds of Buizerd at a plot on which “bird food crop” is present is almost 4 times larger, 4 to 9.

5.3.6

Fazant species

The odds to observe a Fazant on a plot with “bird food crop” is e1.117 = 2.88(1.01; 8.15) times the odds of Fazant if a plot does not have “bird food crop”. This means that the odds of Fazant at a plot with no “bird food crop” is approximately 1 to 12, while the odds of Fazant at a plot on which “bird food crop” is present is almost 3 times larger, 3 to 12.

5.3.7

Groenling species

The odds to observe a Groenling on a plot with “bird food crop” is e2.623 = 13.77(1.01; 186.1) times the odds of Groenling if a plot does not have “bird food crop”. This means that the odds of Groenling at a plot with no “bird food crop” is approximately 1 to 93, the odds of Groenling at a plot on which “bird food crop” is present is almost 14 times larger, 14 to 93.

5.3.8

Blauwe Reiger species

The odds to observe a Blauwe Reiger on a plot with “bird food crop” is e2.933 = 18.78(1.39; 253.09) times the odds of Blauwe Reiger if a plot does not have “bird food crop”. This means that the odds of Blauwe Reiger at a plot with no “bird food crop” is approximately 1 to 93, the odds of Blauwe Reiger at a plot on which “bird food crop” is present is almost 18 times larger, 18 to 93.

5.4

Conclusion

For all species that have a sufficiently large prevalence, we observe a significantly positive impact on the presence of birds if bird food crop is present on the parcels. The odds of observing a bird increases by of at least factor 3 if bird food crop is available. The impact for Kneu, Groenling and Blauwe Reiger is biggest.

Part IV

Results: Effect of environmental factors on Grauwe Gors, Geelgors, Veldleeuwerik, Kneu and Rietgors

Chapter 6

Exploratory analysis results Initially, an exploratory data analysis was performed for both categorical and continuous covariates in order to gain insight of the data and detect possible influential observations which can drive the results of the analysis. It is of interest to mention that the exploratory analysis is not reflecting our final decision, but is only used to investigate features of the covariates. More powerful statistical tools will be used in later sections.

6.1

Exploratory data analysis for categorical variables

For the categorical variables, we present in Tables 6.1 until 6.3 the sampling frequency (the number of times each covariate category was seen in the data), and the total number of birds observed per species according to each category. This was done to investigate the balance in the data. It can indeed be problematic when specific categories were recorded rarely. Unbalance does not only complicate the interpretation of results, it can lead to problems in model convergence. It is therefore wise to coalesce categories to obtain groups with considerably high sampling frequencies.

6.1.1

Food availability

Different variables represent the food availability on a parcel. These are summarized in Table 6.1. The variable “Food availability” has three categories, namely: • eten -: <25% available food, • eten +/-: between 25% and 75% available food and 38

Bird food crop 2 categories

Bird food crop 4 categories

Bird food crop

Covariate Food availability

Category braak eten eten + eten +/Bladrammenas geen teelt Haver haver/bladrammenas h/b/t haver/vlas luzerne Tarwe Tarwe/bladrammenas t/h/b Tarwe/triticale Tarwe/vlas T/v/b/k first category second category third category fourth category geen teelt teelt

Sampling frequency 1 304 314 189 2 94 5 90 2 5 5 271 79 1 3 279 1 94 175 284 284 94 743

Grauwe Gors 0 514 541 757 0 152 6 323 5 7 4 706 62 0 0 538 0 152 716 545 390 152 1651

Geelgors 0 1532 4513 1848 0 217 3 410 30 33 40 2845 455 10 105 4077 0 217 2993 4110 905 217 8008

Table 6.1: Exploratory data analysis for categorical variables (part1) Veldleeuwerik 0 520 814 544 0 184 23 80 1 0 0 895 159 8 0 577 11 184 918 577 259 184 1754

Kneu 0 1044 3733 2326 23 60 42 2463 70 20 66 1804 1585 60 3 926 5 60 1915 946 4206 60 7067

Rietgors 0 381 1002 965 1 22 2 232 10 0 2 1115 290 0 25 681 5 22 1144 681 538 22 2363

Local farm traffic

Rail road

Unpaved road

Ongoing traffic

Faunal grass stretch

Tall cone orchard

Fruit cultivation

Garden

Tree collonade

Forest

Covariate Hedge

Category 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

Sampling frequency 626 240 740 126 773 93 856 10 640 226 865 1 396 470 805 61 703 163 859 7 237 629

Grauwe Gors 700 1138 1770 68 1485 353 1838 0 1437 401 1838 0 1356 482 1459 379 1452 386 1838 0 765 1073

Geelgors 3878 4655 6534 1999 7788 745 8439 94 6157 2376 8503 30 4375 4158 8251 282 6302 2231 8376 157 2606 5927

Veldleeuwerik 1745 204 1739 210 1854 95 1949 0 1807 142 1949 0 854 1095 1867 82 1670 279 1949 0 361 1588

Table 6.2: Exploratory data analysis for categorical variables (part 2) Kneu 5306 1935 6531 710 6838 403 7171 70 5031 2210 7241 0 2293 4948 6893 348 6353 888 7153 88 1341 5900

Rietgors 1787 592 2008 371 2311 68 2377 2 1821 558 2379 0 873 1506 2170 209 1999 380 2378 1 597 1782

Covariate Tall cone orchard 50 meter Food availability 50 meter Forest 50 meter Hedge 50 meter Fruit cultivation 50 meter Garden 50 meter Tree collonade 50 meter

Category 0 1 0 1 0 1 0 1 0 1 0 1 0 1

Sampling frequency 849 17 558 308 748 118 589 268 533 333 822 44 741 125

Grauwe Gors 1838 0 729 1109 1770 68 678 1160 1264 574 1665 173 1781 57

Geelgors 8446 87 4612 3912 6711 1822 3826 4707 5440 3093 8141 392 6980 1553

Veldleeuwerik 1938 11 1142 807 1739 210 1641 308 1698 251 1920 29 1868 81

Table 6.3: Exploratory data analysis for categorical variables (part 3) Kneu 7135 106 4105 3136 6625 616 5189 2052 3945 3296 6905 336 5887 1354

Rietgors 2367 12 838 1541 2012 367 1695 684 1617 762 2252 127 1831 548

• eten +: >75% available food on the plot at the sampling moment. The one plot having category braak is joint with the category eten -. “Bird food crop” had many categories in the original dataset, some of them recorded many times and some of them only a few times as presented in Table 6.1, which would lead to difficulties in the statistical analysis. Therefore, we performed the analysis for this covariate according to both two different categorization schemes: a binary variable with 2 categories and a 4-category variable. The binary variable is defined as: • no crop (geen teelt and • crop teelt), while the four-categories variable is defined as: • first category : No birdfeed crop present • second category : Mainly grain without admixture of other crops (Tarwe/triticale, tarwe, luzerne, haver) • third category : grain in combination with flax (Tarwe/ vlas, haver/ vlas) • fourth category : Grain in combination of bladrammannas (Tarwe/ vlas/ bladrammenas/ koolzaak, haver/ bladrammenas/ tarwe, tarwe/ haver/ bladrammenas, tarwe/ bladrammenas, haver/ bladrammenas, bladrammenas).

6.1.2

Environment and disturbance

Different variables are available describing the direct environment of the parcel and possible disturbance near the parcel. These are summarized in Tables 6.2 and 6.3. The presence of “Garden”, “tall cone orchard”, “rail road” and “tall cone orchard 50 meter” were very rare on the plots, making the statistical analysis impossible for these four covariates. We did not model these variables.

6.1.3

Impact on prevalence of species

We observe that the number of birds for Grauwe Gors is the smallest as compared to the other species. Close in terms of the number of birds reported are Veldleeuwerik and Rietgors. Geelgors is the species with the largest number of birds registered, followed by Kneu. 42

There seems to be a positive effect of “food availability”, especially for Geelgors, Kneu and Rietgors. A similar trend is seen for “bird food crop” (geen teelt versus teelt). Grauwe Gors, Veldleeuwerik and Rietgors seems to prefer plots with bird food crop in second category, while Geelgors is often present on plots of grain in combination with flax category and Kneu on plots of grain in combination of bladrammannas category. For other outcomes, there is no clear effect, as the sampling times is also quite different amongst different categories.

6.2

Exploratory data analysis for continuous variables

Exploratory plots are made, by presenting the occurrence probability of the birds as a function of the covariate of interest. The occurrence probability is calculated as the number of times each species was seen on a plot (the total number of successes) divided by the total number of times a plot was visited (sampling frequency). The size of the circles correspond to the sampling frequency. The larger the circle, more often the plots were visited and thus, the larger the influence of that specific covariate on the response in the logistic regression. In addition, we present the fitted line of the logistic regression model in order to visualize the increasing or decreasing effect. These plots are used as a check for outliers.

6.2.1

Plot size by species

It can be observed that on a large number of plots the Grauwe Gors species was never observed, seen by the many zero-values. The number of occurrences from this species is very limited in our dataset. For the other four species there seems to be a positive increasing trend with “plot size”. Figures show three outlying plots with respect to the “plot size”, with a value of the plot size larger than 1.5.

Figure 6.1: Proportion plot of plot size variable

6.2.2

Temperature by species

As â€œTemperatureâ€? is not constant over the entire study period for each plot, we split the range of this covariate, with a minimum of -11 degrees and a maximum of 14 degrees, in small intervals of 2 degrees. For each temperature interval, we calculate the proportion of visits with birds of particular species present. No outlying observations are observed. Results seem to differ amongst species. For Grauwe Gors, there seems to be no effect of this covariate. There seems to be a positive trend for Geelgors, Veldleeuwerik and Kneu, while for Rietgors there seems to be a negative trend.

Figure 6.2: Proportion plot of temperature variable

6.2.3

Hellmann number by species

In this case, it is a similar situation as with the covariate “Temperature”. “Hellmann number” is not constant over the study period, we used the same method for calculating the proportions as described for the variable “Temperature”. For Grauwe Gors and Veldleeuwerik species, a negative trend is seen on the graphs, while for Geelgors, Kneu, and Rietgors a constant trend is observed. Note however that most observations have “Hellmann number” close to 0, with only some variables larger than 10. In terms of the regression, observations with larger covariate value will therefore have huge impact, and the validity of the observed trend is limited. In order to get a more stable result, the “Hellmann number” was transformed, such that observations are

more equally spread over the covariate range. Since the range of this variable is large, between 0 and 104.7, a log transformation of the “Hellmann number” is considered and a binary version of this covariate was further explored. The results of these analyses are presented in the Appendix.

Figure 6.3: Proportion plot of Hellmann number

6.2.4

Nearest Building by species

The variable “Nearest Building”, representing the distance from the plot to the nearest building, seems to have a positive impact for Kneu, Rietgors and Veldleeuwerik species, a negative one on Geelgors, and no effect on Grauwe Gors. However, this effects can be influenced by one outlying observation, corresponding to a plot with nearest building further then 1.5 km away. There46

fore, we performed the analysis on a dataset excluding this plot in order to see if our final conclusion changes.

Figure 6.4: Proportion plot of Nearest Building

6.2.5

Nearest built-upon area by species

There seems to be a negative impact on Geelgors species and a positive one on Veldleeuwerik species given by the nearest built-upon area. But again, these two effects seem to be influenced greatly by one outlying observation, corresponding to one plot for which the nearest built-upon area is more than 1.5 km away. No other important effect is seen for the other three species.

Figure 6.5: Proportion plot of Nearest built-upon area

6.2.6

Nearest plot by species

The variable â€œnearest plotâ€? seems to have a negative impact for the Geelgors species and a positive one on Veldleeuwerik and Kneu species. Once again, there is a possible outlying observation present on plots which can drive the results, one plot which is more than 1 km away from any of the other plots, therefore the analysis was done without this observation and presented in the Appendix section of this report.

Figure 6.6: Proportion plot of Nearest plot

6.2.7

Nearest railroad by species

There seems to be no important effect of the covariate â€œNearest railroadâ€? on the response of interest of each species. It is important to note that this variable has values between 0 and 10000 meters, which is a very wide range.

Figure 6.7: Proportion plot of Nearest railroad

6.2.8

Nearest road by species

The covariate â€œNearest roadâ€? appear to have a positive effect on Veldleeuwerik species and a negative one on Geelgors, and Rietgors. However, there does not seems to be an effect on Grauwe Gors and Kneu. Three possible outliers are present on all plots, therefore we performed the analysis without these three plots as well.

Figure 6.8: Proportion plot of Nearest road

6.3

Conclusions

The variable “bird food crop” availability on the plots will be analyzed in two ways: as a binary covariate (bird food crop present/ no bird food crop present on the plots) and as a categorical covariate with four categories, namely: • no bird food crop present on the plots • grain without admixture of other crops • grain in combination with flax and • grain in combination of bladrammannas. 51

We will not model the binary variables “garden”, “tall cone orchard”, “rail road” (the binary version) and “tall cone orchard 50 meters”, since these were very rarely present on the plots. For the continuous variables, we wanted to investigate the presence of outliers, observations that lie at an unusual distance from other observations, as these can have serious impact on the statistical analysis. When outlying observations are discovered, these observations are deleted and the statistical analysis was performed once again in order to check if the results were influenced by these observations. The results of this analysis, excluding the outlying plots are presented in the Appendix section of this report. Outliers are found for the variables plot size, nearest building, nearest built-upon area, nearest plot, nearest road. A transformation of the variable Hellmann number will be considered. The usefulness of the variable nearest railroad is questionable, as distances are up to 10 km from the parcels. Such large distances are probably not useful for investigation of impact on the particular parcel.

Chapter 7

Results regression models for Grauwe Gors, Geelgors, Veldleeuwerik, Kneu and Rietgors Results of the logistic regression, negative binomial regression and GLMM regression are presented in Tables 7.1-7.23. The logistic regression and GLMM investigate the effect of the covariates on presence/absence of the birds, while the negative binomial regression investigates the effect on the number of birds observed. Significant (p-value < 0.01) are indicated by a star(*). We will discuss the significant results, per species. An overview of all effect is given in Tables 7.24-7.31, where the strongly positive significant results are indicated by the indicator positive, the strongly negative significant results by the indicator negative and the borderline situation are represented by the color yellow, for both positive and negative results.

7.1 7.1.1

Results by species for categorical variables Grauwe Gors

Results of the statistical analysis for Grauwe Gors are given in Tables 7.1-7.3. According to the logistic regression results, â€œhedgeâ€? had a significant positive influence on Grauwe Gors species. There is a e1.393 = 4.02(2.07; 7.82) times increased possibility to discover the presence of a bird

from this species when a hedge is a adjacent to a plot. However, this effect is not important in the GLMM model. The negative binomial regression indicated a significant positive effect for this covariate as well. The expected number of Grauwe Gors birds on a plot adjacent to a hedge is e1.497 = 4.47(1.01; 19.75) times higher than on a plot which is not adjacent to a hedge. At the same time, the adjacency to faunal grass stretch has a strong negative effect on this species indicated by the logistic regression model. The odds of observing a bird from the Grauwe Gors species are about 63% lower (e−0.992 = 0.370(0.18; 0.75)) if there is a faunal grass stretch adjacent to a plot. The same negative effect is also indicated by the GLMM model on this covariate. However, this effect is close to a borderline situation, thus we consider that more research should be done in this case, after collecting more data about this species. The negative binomial model did not find any effect of this covariate for the number of Grauwe Gors birds observed. Furthermore, the covariate “hedge 50 meters” has a positive significant result according to logistic regression, with a e1.310 = 3.7(1.90; 7.22) times increased probability to discover the presence of a bird from the Grauwe Gors species when a hedge is present at a 50 meters distance from a plot. The negative binomial model found a positive borderline result in this case. GLMM did not find any effect of this covariate on Grauwe Gors species. The logistic regression model indicated that “adjacency to ongoing traffic” had a positive significant effect on the response of interest. But GLMM model and negative binomial did not find any effect of this covariate. The odds are e1.106 = 3.02(1.28; 7.12) increased to find a Grauwe Gors when ongoing traffic is adjacent to a plot. Lastly, the covariate “tree colonnade 50 meters” had a negative borderline effect on Grauwe Gors species, found by logistic regression and negative binomial models. No effect of this covariate was found by GLMM. In conclusion, some significant effects were found on presence/absence of Grauwe Gors based on the logistic regression. But when accounting for count data within plots, none of this remain significant. For the number of birds observed, there is a positive effect of adjacency of “hedge”.

7.1.2

Geelgors

Results of the statistical analysis for Geelgors are given in Tables 7.4-7.7. The variables “Hedge”, “Hedge 50 meters”, “Food availability of growth”, “Bird food crop 4 categories” and “Bird food crop 2 categories” showed a 54

Intercept eten + eten +/-

Intercept teelt

Intercept fourth second third

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food crop 4 categories -3.829 0.714 <0.001* 0.481 0.799 0.548 1.220 0.775 0.115 0.321 0.991 0.746 1.575 0.743 0.034 0.444 0.922 0.630 1.446 0.746 0.053 0.171 0.922 0.853 Bird food crop 2 categories -3.829 0.715 <0.001* 0.481 0.800 0.548 1.450 0.727 0.046 0.318 0.849 0.708 Food availability of growth -2.769 0.243 0.980 0.522 0.434 0.229 0.441 0.316 0.163 0.054 0.614 0.929 0.578 0.343 0.092 0.866 0.700 0.216 -4.967 0.273 0.668

-5.648 1.106

-5.644 0.774 1.286 1.055

0.641 0.445 0.464

0.986 0.851

0.985 0.940 0.877 0.874

GLMM Estimates SE

<0.001* 0.539 0.150

<0.001* 0.194

<0.001* 0.410 0.142 0.227

P-value

Table 7.1: Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors

Intercept local farm traffic

Intercept unpaved road

Intercept ongoing traffic

Intercept faunal grass stretch

Intercept fruit cultivation

Intercept tree colonnade

Intercept forest

Intercept hedge

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Adjacent to hedge -3.206 0.189 <0.001* -0.062 0.299 0.835 1.393 0.258 <0.001* 1.497 0.577 0.009* Adjacent to forest -2.594 0.134 <0.001* 0.718 0.283 0.011 -0.444 0.409 0.278 -1.529 0.738 0.038 Adjacent to a tree colonnade -2.697 0.137 <0.001* 0.497 0.280 0.076 0.365 0.358 0.308 0.642 0.839 0.444 Adjacent to short-cone fruit cultivation -2.610 0.144 <0.001* 0.638 3.026 0.833 -0.171 0.302 0.572 -0.193 6.016 0.974 Adjacent to faunal grass stretch -2.241 0.153 <0.001* 1.019 0.374 0.006* -0.992 0.275 <0.001* -1.109 0.520 0.033 Adjacent to disturbance roads: ongoing traffic -2.789 0.140 <0.001* 0.447 0.275 0.103 1.106 0.333 <0.001* 1.072 0.958 0.263 Adjacent to disturbance roads: unpaved road -2.686 0.143 <0.001* 0.577 0.296 0.051 0.168 0.304 0.58 0.075 0.665 0.910 Adjacent to disturbance roads: local farm traffic -2.274 0.199 <0.001* 0.936 0.485 0.054 -0.581 0.258 0.024 -0.532 0.578 0.358 -4.127 -0.884

-4.918 0.742

-4.795 1.344

-3.968 -1.484

-4.636 -0.421

-4.673 -0.976

-4.680 -0.311

-4.868 0.790

Estimates

0.687 0.683

0.647 0.765

0.599 1.111

0.579 0.665

0.604 0.774

0.589 1.281

0.601 0.895

0.610 0.666

GLMM SE

<0.001* 0.196

<0.001* 0.332

<0.001* 0.227

<0.001* 0.026

<0.001* 0.586

<0.001* 0.446

<0.001* 0.728

<0.001* 0.236

P-value

Table 7.2: Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors

Intercept tree collonade 50 meter

Intercept garden/builing 50 meter

Intercept Fruit cultivation 50 meter

Intercept hedge 50 meter

Intercept Forest 50 meter

Intercept bird food cultivation 50 meter

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food cultivation 50 meter -2.822 0.172 <0.001* 0.128 0.329 0.696 0.415 0.254 0.102 0.955 0.541 0.077 Forest 50 meter -2.590 0.134 <0.001* 0.718 0.283 0.011* -0.468 0.409 0.253 -1.529 0.738 0.038 Hedge 50 meter -3.223 0.196 <0.001* -0.039 0.310 0.899 1.310 0.259 <0.001* 1.355 0.557 0.015 Fruit cultivation 50 meter -2.483 0.149 <0.001* 0.687 0.334 0.039 -0.518 0.284 0.068 -0.274 0.547 0.616 Garden 50 meter -2.628 0.129 <0.001* 0.561 0.274 0.040 -0.417 0.605 0.491 0.403 1.073 0.708 Tree collonade 50 meter -2.523 0.129 <0.001* 0.726 0.283 0.010 -1.395 0.597 0.019 -1.719 0.736 0.019 -4.526 -1.386

-4.750 0.301

-4.484 -0.702

-4.942 0.901

-4.703 -0.184

-5.011 0.848

0.580 1.138

0.599 1.301

0.607 0.702

0.627 0.643

0.603 0.903

0.659 0.659

GLMM Estimates SE

Table 7.3: Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors

<0.001* 0.223

<0.001* 0.817

<0.001* 0.317

<0.001* 0.161

<0.001* 0.838

<0.001* 0.198

P-value

strong positive impact for the Geelgors species, found by all three models. According to the logistic regression model, the odds of detecting a Geelgors bird are e1.193 = 3.29(2.25; 4.89) times larger on a plot which had an adjacent hedge. The negative binomial model and GLMM indicate a similar significant positive effect of this covariate. The covariate “Hedge 50 meters” is statistically significant, conforming to logistic regression. Thus, the odds of finding a bird from the Geelgors species on a plot with a hedge at a distance of 50 meters over the odds of finding a bird from the Geelgors species on a plot without a hedge at a distance of 50 meters are e0.958 = 2.6(1.82; 3.74). These results are in agreement with the ones found by the negative binomial and GLMM. “Food availability of growth” was found to be a significant positive effect by all the models used. We performed a multiple comparisons analysis for this variable in order to determine which percentage of food present on the plots is the one which has the greatest impact on the response of interest (Table 7.4). All three models indicate that the presence of 75% or more food on the plot leads to an increase in the number of Geelgors birds as compared to less than 25% of food. Furthermore, “Bird food crop 2 categories” shows a positive significant result given by all the models used. Thus, bird food crop present on a plot leads to an increase in the number of Geelgors birds. In order to find which of the four categories has the highest influence on the response of interest, a multiple comparison analysis was used (7.7). According to the logistic regression and GLMM, there is no difference between the second and the third category, but these two categories perform better than the first and the fourth. Therefore, the presence of Tarwe/triticale, Tarwe, luzerne, Haver as well as Tarwe/vlas, haver/vlas on the plots contribute to an increment of the Geelgors species. Conforming to logistic regression, the odds to observe a bird from Geelgors species for a plot adjacent to an unpaved road are 93% (e0.659 = 1.93(1.28; 2.91)) higher than on a plot without an unpaved road nearby. A positive influence of this covariate is found by GLMM. The odds ratio is equal to e1.081 = 2.94(1.27; 6.82), thus an increased chance by 12% to see a Geelgors bird on a plot adjacent to an unpaved road, with a wide confidence interval. No effect of this covariate was detected by the negative binomial model. A positive effect was found for the covariates “Forest” by the logistic regression, but no effect was further established by GLMM or negative binomial model. “Tree colonnade 50 meters” was found to be in a borderline situation according to logistic regression, while the other two models did not 58

find any effect of this variable on Geelgors species. “Ongoing traffic” seemed to negatively influence the Geelgors species, effect found by the negative binomial model, while the logistic regression indicated a borderline situation. No effect was found by GLMM. In conclusion, the presence of “hedge on a plot at 50 meters”, increases both the presence and the number of Geelgors. The presence of “bird food crop” (especially grain without admixture of other crops and grain in combination with flax), increases presence and amount of Geelgors. While, the adjacency to “unpaved roads” increases the presence of Geelgors, “ongoing traffic” reduces the number of Geelgors.

7.1.3

Veldleeuwerik

Results of the statistical analysis for Veldleeuwerik are given in Tables 7.8-7.10. All models applied indicated a significant negative effect on the Veldleeuwerik species by the covariates “Hedge”, “Hedge 50 meters” and “Fruit cultivation 50 meters”. The odds of identifying a Veldleeuwerik bird on a plot adjacent to a hedge are 51% less than the odds of seeing a Veldleeuwerik bird on a plot which is not adjacent to a hedge. The expected log count for a plot next to a hedge is 1.135 lower than the expected log count for a plot without a hedge indicated by the negative binomial model. According to logistic regression results, the odds are 56% lower to identify a bird from Veldleeuwerik species on a plot with a hedge at a distance of 50 meters as compared with a plot with no hedge. A similar effect was discovered by negative binomial and GLMM. “Fruit cultivation at 50 meters” has a significant negative impact on the response of interest, thus the results of logistic regression reflect that the odds of detecting a bird from Veldleeuwerik species is about e−0.629 = 0.53(0.33; 0.86) times less for a plot localized at a distance of 50 meters away from a fruit cultivation as compared with a plot with no fruit cultivation. A similar effect is obtained by GLMM, where the odds ratio of e−0.976 = 0.37(0.17; 0.80) represents a decreased chance of identifying a Veldleeuwerik bird by 63% on a plot which has a fruit cultivation 50 meters away. From the results of the negative binomial model, the expected number of Veldleeuwerik bird on a plot localized at a distance of 50 meters away from a fruit cultivation is e−1.397 = 0.24(0.11; 0.51) times lower than the plot with no fruit cultivation. “Fruit cultivation” has no effect on presence/absence of Veldleeuwerik 59

Intercept eten + eten +/-

Intercept teelt

Intercept fourth second third

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food crop 4 categories -1.127 0.240 <0.001* 0.836 0.270 0.001* 0.211 0.293 0.471 0.806 0.333 0.015 1.000 0.268 <0.001* 1.519 0.309 <0.001* 1.542 0.269 <0.001* 1.836 0.309 <0.001* Bird food crop 2 categories -1.127 0.239 <0.001* 0.836 0.274 0.002* 1.033 0.251 <0.001* 1.541 0.289 <0.001* Food availability of growth -0.558 0.119 <0.001* 1.614 0.149 <0.001* 0.664 0.165 <0.001* 1.084 0.211 <0.001* 0.378 0.188 0.045 0.666 0.241 0.006* -0.710 0.786 0.602

-1.554 1.483

-1.558 0.381 1.463 2.013

0.185 0.213 0.237

0.312 0.304

0.315 0.360 0.328 0.329

GLMM Estimates SE

<0.001* <0.001* 0.011

<0.001* <0.001*

<0.001* 0.29 <0.001* <0.001*

P-value

Table 7.4: Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors

Intercept local farm traffic

Intercept unpaved road

Intercept ongoing traffic

Intercept faunal grass stretch

Intercept fruit cultivation

Intercept tree collonade

Intercept forest

Intercept hedge

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Adjacent to hedge -0.812 0.079 <0.001* 1.649 0.105 <0.001* 1.193 0.147 <0.001* 1.194 0.202 <0.001* Adjacent to forest -0.554 0.071 <0.001* 2.024 0.100 <0.001* 0.515 0.177 0.003* 0.546 0.257 0.034 Adjacent to a tree collonade -0.499 0.069 <0.001* 2.155 0.098 <0.001* 0.233 0.202 0.249 -0.269 0.295 0.363 Adjacent to short-cone fruit cultivation -0.531 0.075 <0.001* 2.093 0.107 <0.001* 0.225 0.147 0.125 0.131 0.213 0.54 Adjacent to faunal grass stretch -0.343 0.092 <0.001* 2.191 0.133 <0.001* -0.255 0.129 0.048 -0.126 0.185 0.498 Adjacent to disturbance roads: ongoing traffic -0.424 0.067 <0.001* 2.179 0.096 <0.001* -0.658 0.261 0.011 -0.957 0.340 0.004* Adjacent to disturbance roads: unpaved road -0.609 0.073 <0.001* 2.045 0.103 <0.001* 0.659 0.159 <0.001* 0.361 0.232 0.119 Adjacent to disturbance roads: local farm traffic -0.295 0.117 0.011 2.162 0.171 <0.001* -0.254 0.140 0.069 -0.048 0.203 0.813 0.138 -0.529

-0.454 1.081

-0.189 -0.794

-0.952 -0.078

-0.289 0.194

-0.285 0.425

-0.341 0.609

-0.627 1.382

Estimates

0.243 0.289

0.146 0.326

0.137 0.563

0.190 0.247

0.155 0.300

0.141 0.426

0.145 0.349

0.144 0.268

GLMM SE

Table 7.5: Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors

0.570 0.068

0.002* <0.001*

0.167 0.158

<0.001* 0.751

0.062 0.518

0.044 0.318

0.019 0.081

<0.001* <0.001*

P-value

Intercept tree collonade 50 meter

Intercept garden/builing 50 meter

Intercept Fruit cultivation 50 meter

Intercept hedge 50 meter

Intercept Forest 50 meter

Intercept bird food cultivation 50 meter

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food cultivation 50 meter -0.547 0.082 <0.001* 1.973 0.116 <0.001* 0.197 0.133 0.14 0.374 0.192 0.051 Forest 50 meter -0.513 0.071 <0.001* 2.055 0.100 <0.001* 0.255 0.176 0.147 0.403 0.256 0.116 Hedge 50 meter -0.784 0.081 <0.001* 0.116 0.109 <0.001* 0.958 0.140 <0.001* 1.026 0.196 <0.001* Fruit cultivation 50 meter -0.528 0.082 <0.001* 2.146 0.117 <0.001* 0.144 0.133 0.278 -0.050 0.192 0.796 Garden 50 meter -0.458 0.066 <0.001* 2.148 0.096 <0.001* -0.235 0.269 0.384 -0.367 0.377 0.331 Tree collonade 50 meter -2.523 0.129 <0.001* 2.092 0.100 <0.001* -1.395 0.597 0.019 0.219 0.258 0.395

-0.313 0.539

-0.245 0.103

-0.297 0.156

-0.629 1.195

-0.303 0.408

-0.339 0.289

0.143 0.379

0.138 0.564

0.169 0.275

0.153 0.264

0.145 0.359

0.165 0.278

GLMM Estimates SE

Table 7.6: Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors

0.028 0.155

0.076 0.855

0.079 0.569

<0.001* <0.001*

0.036 0.256

0.040 0.299

P-value

Estimates 0.211 1.000 1.542 0.789 1.331 0.541 Estimates 0.663 0.377 -0.285

Coefficients fourth - first second - first third - first second - fourth third - fourth third - second

Coefficients eten + - eten eten +/- - eten eten +/- - eten +

SE 0.165 0.188 0.186

SE 0.293 0.268 0.269 0.205 0.207 0.169 P-value <0.001* 0.110 0.272

P-value 0.885 <0.001* <0.001* <0.001* <0.001* 0.007*

Logistic regression

Negative Binomial Bird food crop Estimates SE P-value 0.806 0.333 0.070 1.519 0.309 <0.001* 1.836 0.309 <0.001* 0.246 2.889 0.019 1.029 0.246 <0.001* 0.317 0.214 0.442 Food availability Estimates SE P-value 1.084 0.210 <0.001* 0.666 0.241 0.016 -0.417 0.240 0.190 Estimates 0.785 0.601 -0.184

Estimates 0.381 1.463 2.013 1.082 1.632 0.550

SE 0.213 0.237 0.230

SE 0.360 0.328 0.329 0.266 0.271 0.216

GLMM

P-value <0.001* 0.030 0.703

P-value 0.709 <0.001* <0.001* <0.001* <0.001* 0.051

Table 7.7: Pairwise comparisons of the variable Bird food crop and food availability for Geelgors

species, but a significant negative effect is found on the expected number of Veldleeuwerik birds on a plot adjacent to a fruit cultivation. The effect is e−1.461 = 0.23(0.10; 0.53) times lower than on a plot which is not adjacent to a fruit cultivation. This covariate is found to be a borderline situation by the logistic model and the GLMM. Since “Forest” variable was found to be a negatively significant variable by the logistic regression, we interpret it as the odds of observing a Veldleeuwerik bird on a plot adjacent to a forest are 70% less than the odds of seeing a Veldleeuwerik bird on a plot which is not adjacent to a forest. A similar negative effect was found by GLMM as well. “Forest 50 meters” has a negative effect found by the logistic regression model. There is a e−1.228 = 0.29(0.12; 0.70) times decreased probability to see a bird from this species when a forest is located at a distance of 50 meters away from the a plot. However, the effect is not important when accounting for repeated measurements at same plot. The variable “Tree colonnade 50 meters” were found to be significant by the logistic regression model and negative binomial model, but it is a borderline situation according to GLMM. The odds to see a bird from Veldleeuwerik species are lower by 64% (e−0.999 = 0.36(0.16; 0.82)) for a plot which has a tree colonnade located at a distance of 50 meters away. The expected number of Veldleeuwerik birds on a plot which has a tree colonnade located at a distance of 50 meters away is e−1.416 = 0.24(0.08; 0.67) times lower than on a plot with no tree colonnade. The negative binomial model found a negative significant effect of the variable “garden 50 meters” on the number of Veldleeuwerik species. The expected number of Veldleeuwerik bird on a plot localized at a distance of 50 meters away from a garden is e−1.526 = 0.22(0.04; 0.99) times lower than the plot with no garden. The logistic regression and GLMM did not find any effect of this covariate on presence/ absence of Veldleeuwerik species. Lastly, the logistic regression model found a borderline situation for the covariate “local farm traffic”, while the negative binomial and GLMM did not find any effect. In summary, the presence of “hedge”, “forest” and “fruit cultivation” reduces both the presence and the number of Veldleeuwerik, especially in 50 meters from plot. Also “garden” and “tree colonnade” at a distance of 50 meters reduces the amount of Veldleeuwerik.

Intercept eten + eten +/-

Intercept teelt

Intercept fourth second third

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food crop 4 categories -1.829 0.299 <0.001* 0.672 0.419 0.109 0.769 0.345 0.026 -0.279 0.520 0.591 0.654 0.329 0.047 0.502 0.483 0.299 0.133 0.341 0.6963 0.037 0.483 0.939 Bird food crop 2 categories -1.829 0.298 <0.001* 0.671 0.422 0.112 0.504 0.312 0.106 0.187 0.448 0.676 Food availability of growth -1.449 0.146 <0.001* 0.533 0.227 0.019 0.087 0.204 0.669 0.451 0.321 0.159 0.371 0.222 0.094 0.523 0.366 0.153 -1.795 0.076 0.397

-2.029 0.331

-2.012 0.534 0.492 0.036

0.207 0.234 0.250

0.345 0.344

0.343 0.386 0.364 0.372

GLMM Estimates SE

<0.001* 0.745 0.113

<0.001* 0.336

<0.001* 0.167 0.176 0.924

P-value

Table 7.8: Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species

Intercept local farm traffic

Intercept unpaved road

Intercept ongoing traffic

Intercept faunal grass stretch

Intercept fruit cultivation

Intercept tree collonade

Intercept forest

Intercept hedge

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Adjacent to hedge -1.437 0.093 <0.001* 0.851 0.161 <0.001* -0.646 0.215 0.002* -1.135 0.316 <0.001* Adjacent to forest -1.457 0.087 <0.001* 0.701 0.152 <0.001* -1.203 0.338 <0.001* -0.384 0.395 0.331 Adjacent to a tree collonade -1.535 0.087 <0.001* 0.719 0.148 <0.001* -0.505 0.307 0.101 -0.893 0.452 0.048 Adjacent to short-cone fruit cultivation -1.469 0.093 <0.001* 0.867 0.158 <0.001* -0.518 0.213 0.015 -1.461 0.322 <0.001* Adjacent to faunal grass stretch -1.756 0.128 <0.001* 0.558 0.203 0.006* 0.317 0.169 0.061 0.174 0.282 0.537 Adjacent to disturbance roads: ongoing traffic -1.560 0.086 <0.001* 0.694 0.147 <0.001* -0.319 0.335 0.342 -0.706 0.519 0.173 Adjacent to disturbance roads: unpaved road -1.514 0.091 <0.001* 0.717 0.157 <0.001* -0.393 0.229 0.0864 -0.390 0.354 0.271 Adjacent to disturbance roads: local farm traffic -1.961 0.176 0.011 0.185 0.259 0.476 0.511 0.199 0.011 0.612 0.308 0.047 -2.013 0.377

-1.699 -0.225

-1.736 -0.139

-1.787 0.084

-1.538 -0.843

-1.693 -0.478

-1.565 -1.100

-1.496 -0.901

Estimates

0.276 0.304

0.171 0.343

0.162 0.548

0.209 0.269

0.165 0.331

0.163 0.461

0.157 0.425

0.165 0.317

GLMM SE

<0.001* 0.215

<0.001* 0.513

<0.001* 0.8

<0.001* 0.755

<0.001* 0.011

<0.001* 0.3

<0.001* 0.009*

<0.001* 0.004*

P-value

Table 7.9: Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species

Intercept tree collonade 50 meter

Intercept garden/builing 50 meter

Intercept Fruit cultivation 50 meter

Intercept hedge 50 meter

Intercept Forest 50 meter

Intercept bird food cultivation 50 meter

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food cultivation 50 meter -1.585 0.105 <0.001* 0.577 0.177 0.001* 0.004 0.173 0.98 0.189 0.292 0.517 Forest 50 meter -1.453 0.087 <0.001* 0.704 0.153 <0.001* -1.228 0.338 <0.001* -0.407 0.392 0.299 Hedge 50 meter -1.377 0.094 <0.001* 0.845 0.166 <0.001* -0.823 0.211 <0.001* -0.855 0.304 0.004* Fruit cultivation 50 meter -1.378 0.099 <0.001* 0.982 0.171 <0.001* -0.629 0.187 <0.001* -1.397 0.284 <0.001* Garden 50 meter -1.538 0.085 <0.001* 0.704 0.144 <0.001* -0.963 0.473 0.041 -1.526 0.590 0.009* Tree collonade 50 meter -1.471 0.087 <0.001* 0.773 0.150 <0.001* -0.999 0.313 0.001* -1.416 0.397 <0.001*

-1.603 -0.972

-1.710 -0.584

-1.397 -0.976

-1.427 -0.994

-1.590 -0.994

-1.750 0.014

0.159 0.433

0.160 0.645

0.169 0.294

0.164 0.302

0.159 0.431

0.184 0.282

GLMM Estimates SE

Table 7.10: Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik species

<0.001* 0.025

<0.001* 0.365

<0.001* <0.001*

<0.001* 0.021

<0.001* 0.96

P-value

7.1.4

Kneu

Results of the statistical analysis for Geelgors are given in Tables 7.11-7.13. Since “Food availability of growth” had a significant positive effect found by all the models used, a multiple comparisons analysis was used (7.14). The GLMM results implies that the odds of seeing a Kneu bird on a plot is about e1.193 = 3.29(1.77; 6.13) times greater for a plot with 75% or more available food than a plot with less than 25% food. In the same time, the odds of observing a Kneu bird on a plot is about e1.119 = 3.06(1.56; 5.99) times greater for a plot with 25% up to 75% food than a plot with less than 25% food. No difference was found between the categories 25% up to 75% food and 75% or more food. Analogous results were found by the negative binomial as well as GLMM. A strong positive effect was obtained for “Bird food crop 2 categories”, indicating an increase in the number of birds of Kneu species when a type of bird food crop is present on the plot. As it can be observed in 7.11 for the variable “Bird food crop 4 categories”, the fourth category is preferred over the first category as indicated by all models. The negative binomial model shows a positive effect of the second category over the first one, but this effect is a borderline situation according to logistic regression and GLMM. A positive borderline effect is found by the negative binomial for the third category when compared to the first one. However, all models illustrate that the fourth category is preferred as compared to the second and the third category. Thus, the Kneu species benefits the most from the fourth category. The variable “Faunal grass stretch” has a strong positive effect on this species indicated by the logistic regression model. The odds of seeing a bird from the Kneu species are about e0.801 = 2.22(1.41; 3.52) times higher in the presence of a faunal grass stretch adjacent to a plot. These results are consistent with the ones found by GLMM. However, in the case of count data, no effect of this covariate was found by the negative binomial model. Lastly, the logistic regression model found a borderline situation for the covariate “local farm traffic”, while the negative binomial and GLMM did not find any effect. In conclusion, the presence and number of Kneu increases with the availability of growth. It benefits most from grain in combination with bladrammanas category of bird food. “Faunal grass stretch” also increases the presence of Kneu (though not the amount of Kneu).

Intercept eten + eten +/-

Intercept teelt

Intercept fourth second third

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food crop 4 categories -4.533 1.005 <0.001* -0.449 0.471 0.341 4.613 1.017 <0.001* 3.628 0.577 <0.001* 2.781 1.019 0.006* 2.357 0.539 <0.001* 2.603 1.021 0.011 1.652 0.539 0.002* Bird food crop 2 categories -4.533 1.005 <0.001* -0.449 0.491 0.361 3.310 1.009 0.001* 2.701 0.519 <0.001* Food availability of growth -2.109 0.184 <0.001* 1.231 0.264 <0.001* 1.113 0.225 <0.001* 1.277 0.372 <0.001* 1.087 0.247 <0.001* 1.279 0.425 0.002* -2.283 1.193 1.119

-4.673 3.323

-4.634 4.688 2.802 2.619

0.214 0.241 0.261

1.013 1.014

1.012 1.023 1.023 1.024

GLMM Estimates SE

<0.001* <0.001* <0.001*

<0.001* 0.001*

<0.001* <0.001* 0.006* 0.010

P-value

Table 7.11: Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Adjacent to hedge Intercept -1.670 0.100 <0.001* 1.963 0.194 <0.001* hedge 0.256 0.183 0.162 0.002 0.376 0.995 Adjacent to forest Intercept -1.567 0.090 <0.001* 2.024 0.179 <0.001* forest -0.217 0.247 0.38 -0.489 0.464 0.292 Adjacent to a tree collonade Intercept -1.535 0.087 <0.001* 2.025 0.176 <0.001* tree collonade -0.692 0.329 0.035 -0.753 0.528 0.154 Adjacent to short-cone fruit cultivation Intercept -1.577 0.096 <0.001* 1.891 0.192 <0.001* fruit cultivation -0.085 0.196 0.664 0.260 0.382 0.495 Adjacent to faunal grass stretch Intercept -2.066 0.143 <0.001* 1.545 0.238 <0.001* faunal grass stretch 0.801 0.178 <0.001* 0.694 0.330 0.035 Adjacent to disturbance roads: ongoing traffic Intercept -1.568 0.087 <0.001* 1.999 0.173 <0.001* ongoing traffic -0.420 0.348 0.227 -0.566 0.607 0.351 Adjacent to disturbance roads: unpaved road Intercept -1.522 0.091 <0.001* 2.053 0.185 <0.001* unpaved road -0.429 0.232 0.065 -0.568 0.416 0.173 Adjacent to disturbance roads: local farm traffic Intercept -1.961 0.176 <0.001* 1.497 0.305 <0.001* local farm traffic 0.493 0.200 0.014 0.611 0.363 0.092 Bird food cultivation 50 meter Intercept -1.630 0.107 <0.001* 1.857 0.209 <0.001* bird food cultivation 50 meter 0.087 0.173 0.616 0.267 0.344 0.438

GLMM SE 0.142 0.233 0.128 0.301 0.121 0.390 0.132 0.248 0.167 0.206 0.123 0.439 0.127 0.279 0.215 0.242 0.145 0.219

Estimates -1.630 0.281 -1.529 -0.099 -1.474 -0.683 -1.494 -0.208 -1.895 0.683 -1.533 -0.161 -1.478 -0.333 -1.782 0.335 -1.616 0.197

Table 7.12: Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species

<0.001* 0.369

<0.001* 0.167

<0.001* 0.234

<0.001* 0.714

<0.001* <0.001*

<0.001* 0.401

<0.001* 0.080

<0.001* 0.74

<0.001* 0.227

P-value

Intercept tree collonade 50 meter

Intercept garden/builing 50 meter

Intercept Fruit cultivation 50 meter

Intercept hedge 50 meter

Intercept Forest 50 meter

Intercept bird food cultivation 50 meter

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food cultivation 50 meter -1.630 0.107 <0.001* 1.857 0.209 <0.001* 0.087 0.173 0.616 0.267 0.344 0.438 Forest 50 meter -1.547 0.089 <0.001* 2.024 0.179 <0.001* -0.370 0.256 0.148 -0.489 0.464 0.292 Hedge 50 meter -1.604 0.101 <0.001* 1.996 0.199 <0.001* 0.022 0.181 0.905 -0.109 0.360 0.762 Fruit cultivation 50 meter -1.632 0.107 <0.001* 1.825 0.209 <0.001* 0.091 0.172 0.598 0.335 0.343 0.328 Garden 50 meter -1.582 0.086 <0.001* 1.984 0.172 <0.001* -0.264 0.369 0.474 -0.356 0.675 0.598 Tree collonade 50 meter -1.574 0.090 <0.001* 1.921 0.180 <0.001* -0.166 0.243 0.495 0.253 0.463 0.585

-1.535 -0.071

-1.545 0.0009

-1.553 0.021

-1.576 0.088

-1.523 -0.146

-1.616 0.197

0.127 0.307

0.121 0.466

0.145 0.218

0.144 0.229

0.127 0.311

0.145 0.219

GLMM Estimates SE

Table 7.13: Results of Logistic regression, Negative Binomial regression and GLMM for Kneu species

<0.001* 0.817

<0.001* 0.998

<0.001* 0.925

<0.001* 0.699

<0.001* 0.639

<0.001* 0.369

P-value

Estimates 4.613 2.781 2.603 -1.831 -2.010 -0.179 Estimates 1.113 1.087 -0.026

Coefficients fourth - first second - first third - first second - fourth third - fourth third - second

Coefficients eten + - eten eten +/- - eten eten +/- - eten +

SE 0.225 0.247 0.209

SE 1.017 1.019 1.021 0.225 0.234 0.244 P-value <0.001* <0.001* 0.991

P-value <0.001* 0.027 0.044 <0.001* <0.001* 0.871

Logistic regression

Negative Binomial Bird food crop Estimates SE P-value 3.628 0.577 <0.001* 2.357 0.539 <0.001* 1.652 0.539 0.011 -1.271 0.423 0.014 -1.976 0.424 <0.001* -0.705 0.371 0.223 Food availability Estimates SE P-value 1.277 0.372 0.002* 1.279 0.425 0.007* 0.002 0.425 0.999 Estimates 1.193 1.119 -0.074

Estimates 4.688 2.802 2.619 -1.886 -2.068 -0.183

SE 0.241 0.261 0.223

SE 1.023 1.023 1.025 0.240 0.250 0.251

GLMM

P-value <0.001* <0.001* 0.94

P-value <0.001* 0.026 0.043 <0.001* <0.001* 0.874

Table 7.14: Pairwise comparisons of the variable Bird food crop and food availability for Kneu

7.1.5

Rietgors

Results of the statistical analysis for Rietgors are given in Tables 7.15-7.17. A multiple comparison analysis results are presented in 7.18 for the “Food availability of growth” covariate, since this was found to have a positive significant effect on the Rietgors species. All three models used indicated a strong positive effect of the 75% or more available food and between 25% up to 75% available food categories as compared with less than 25% food category. According to GLMM, the odds of seeing a Rietgors bird on a plot is about e1.014 = 2.75(1.58; 4.79) times greater for a plot with 75% or more available food than a plot with less than 25% food, while the odds of observing a Rietgors bird on a plot is about e0.934 = 2.54(1.37; 4.71) times greater for a plot with 25% up to 75% food than a plot with less than 25% food. There does not seem to be a difference between the categories 25% up to 75% food and 75% or more food. Furthermore, all models indicated a positive statistically significant influence of any type of bird food crop on Rietgors species as compared with a plot with no bird food crop. Thus, the odds of seeing a Rietgors bird on a plot with bird food crop is about e1.730 = 5.64(1.97; 16.13) times greater than for a plot with no bird food crop. A multiple comparison analysis indicates the bird food crop category which influences the most the Rietgors species. As seen in Table 7.18, the second, third and fourth category are all positively influencing the Rietgors species as compared with no bird food crop on the plot. There seems to be no difference between the second and the fourth category as well as between the third and the fourth category. The logistic regression and GLMM found that the second category is favored over the third one. According to GLMM, the odds of finding a Rietgors bird is 51% e−0.669 = 0.51(0.31; 0.82) lower for a plot with the third category of bird food crop as compared to a plot which has the second category. However, no effect was found by the negative binomial model. All models used found a strong positive effect on this species of the covariate “Bird food cultivation 50 meter”. According to logistic regression results, the odds of seeing a bird from the Rietgors species increase by e0.762 = 2.14(1.16; 3.95) in the presence of bird food cultivation at a distance of 50 meters away from a plot. The negative binomial model found a strong negative effect of the covariate “tree colonnade” on the Rietgors species. The expected number of Rietgors birds on a plot adjacent to a tree colonnade is e−1.447 = 0.23(0.09; 0.59) times lower than on a plot which is not adjacent to a tree colonnade. This effect of the tree colonnade was found to be a borderline situation by the 73

logistic regression model and no effect was established by GLMM. In summary, for Rietgors, the availability of any food (> 50%) is important for both presence and amount of birds. It prefers any bird food crop. Also, “bird food cultivation 50 meter” increases presence and amount of bird. “Tree colonnade” reduces the number of birds.

7.2 7.2.1

Results by species for continuous variables Grauwe Gors

A positive effect of the plot size on Grauwe Gors species was found by the negative binomial model when all plots are included in the analysis. In the Appendix section (Table 8.37), the analysis is performed again without the outlying observations and no effect is found by any of the models used for the “plot size” variable. Thus, the outlying observation highly influenced the results presented in Table 7.19. According to the results presented in Appendix section, the expected count of Grauwe Gors species decrease by e−0.581 = 0.55(0.42; 0.73) for each 1 meter increase in the distance to the nearest road, thus this represents a 44% decrease. Nevertheless, this effect is not found by the GLMM, nor by the logistic regression model. The logistic regression model indicates a negative impact of the distance to the nearest railroad covariate on the Grauwe Gors species. We infer that a 1 km increase in the distance to the nearest railroad of a plot has around 13.8% (e−1.486e−04∗1000 = 0.862(0.75; 0.98)) decrease in the odds of identifying a Grauwe Gors bird. However, no effect was seen by the GLMM and negative binomial model. In conclusion, only distance to the nearest railroad has an impact on presence/absence of Grauwe Gors. The number of birds is however impacted by distance to the nearest road.

7.2.2

Geelgors

According to the results presented in the Appendix, the “plot size” variable has a positive significant effect on the Geelgors species according to the negative binomial model. Increasing the plot size by one m2 multiplies the expected number of the Geelgors birds by e1.096 = 2.99(1.30; 6.87). Thus, according to the negative binomial model, larger plots attract more Geelgors

Intercept eten + eten +/-

Intercept teelt

Intercept fourth second third

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food crop 4 categories -2.375 0.369 <0.001* -1.452 0.369 <0.001* 1.673 0.403 <0.001* 2.575 0.432 <0.001* 1.946 0.389 <0.001* 2.845 0.409 <0.001* 1.276 0.394 0.001* 2.326 0.409 <0.001* Bird food crop 2 categories -2.375 0.370 <0.001* -1.452 0.371 <0.001* 1.641 0.378 <0.001* 2.609 0.386 <0.001* Food availability of growth -1.428 0.145 <0.001* 0.222 0.174 0.201 0.973 0.187 <0.001* 0.970 0.243 <0.001* 0.805 0.211 <0.001* 1.408 0.276 <0.001* -1.631 1.014 0.935

-2.630 1.730

-2.640 1.825 2.096 1.332

0.188 0.215 0.240

0.412 0.408

0.413 0.445 0.425 0.426

GLMM Estimates SE

<0.001* <0.001* <0.001*

<0.001* <0.001*

<0.001* <0.001* <0.001* 0.002*

P-value

Table 7.15: Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species

Intercept local farm traffic

Intercept unpaved road

Intercept ongoing traffic

Intercept faunal grass stretch

Intercept fruit cultivation

Intercept tree collonade

Intercept forest

Intercept hedge

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Adjacent to hedge -1.151 0.086 <0.001* 0.875 0.129 <0.001* 0.077 0.164 0.638 -0.093 0.250 0.709 Adjacent to forest -1.109 0.078 <0.001* 0.844 0.120 <0.001* -0.143 0.209 0.495 0.041 0.309 0.894 Adjacent to a tree collonade -1.068 0.076 <0.001* 0.939 0.115 <0.001* -0.663 0.274 0.015 -1.447 0.362 <0.001* Adjacent to short-cone fruit cultivation -1.048 0.083 <0.001* 0.875 0.128 <0.001* -0.348 0.177 0.049 -0.099 0.255 0.695 Adjacent to faunal grass stretch -1.134 0.105 <0.001* 0.579 0.159 <0.001* 0.008 0.146 0.958 0.470 0.220 0.032 Adjacent to disturbance roads: ongoing traffic -1.135 0.076 <0.001* 0.844 0.115 <0.001* 0.052 0.264 0.844 0.079 0.403 0.844 Adjacent to disturbance roads: unpaved road -1.123 0.081 <0.001* 0.897 0.123 <0.001* -0.036 0.184 0.845 -0.260 0.278 0.35 Adjacent to disturbance roads: local farm traffic -1.153 0.135 0.011 0.688 0.204 <0.001* 0.032 0.161 0.844 0.224 0.243 0.357 -1.133 0.064

-1.079 -0.031

-1.086 -0.003

-1.015 -0.138

-0.951 -0.521

-1.011 -0.696

-1.078 -0.046

-1.069 -0.061

Estimates

0.229 0.262

0.138 0.295

0.127 0.482

0.172 0.235

0.139 0.274

0.128 0.406

0.135 0.322

0.144 0.261

GLMM SE

<0.001* 0.805

<0.001* 0.916

<0.001* 0.996

<0.001* 0.558

<0.001* 0.057

<0.001* 0.086

<0.001* 0.886

<0.001* 0.816

P-value

Table 7.16: Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species

Intercept tree collonade 50 meter

Intercept garden/builing 50 meter

Intercept Fruit cultivation 50 meter

Intercept hedge 50 meter

Intercept Forest 50 meter

Intercept bird food cultivation 50 meter

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Bird food cultivation 50 meter -1.340 0.097 <0.001* 0.268 0.136 0.049 0.524 0.148 <0.001* 1.145 0.221 <0.001* Forest 50 meter -1.093 0.079 <0.001* 0.844 0.120 <0.001* -0.262 0.213 0.219 0.041 0.309 0.894 Hedge 50 meter -1.139 0.088 <0.001* 0.877 0.133 <0.001* 0.027 0.158 0.863 -0.089 0.240 0.711 Fruit cultivation 50 meter -1.016 0.089 <0.001* 0.933 0.139 <0.001* -0.321 0.155 0.038 -0.237 0.228 0.299 Garden 50 meter -1.135 0.076 <0.001* 0.863 0.114 <0.001* 0.077 0.292 0.792 -0.209 0.450 0.643 Tree collonade 50 meter -1.077 0.078 <0.001* 0.753 0.119 <0.001* -0.384 0.221 0.081 0.516 0.306 0.092

-1.032 -0.388

-1.122 0.624

-0.918 -0.439

-1.075 -0.032

-1.079 -0.037

-1.355 0.762

0.132 0.351

0.129 0.489

0.149 0.242

0.149 0.252

0.135 0.331

0.155 0.238

GLMM Estimates SE

Table 7.17: Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors species

<0.001* 0.269

<0.001* 0.202

<0.001* 0.071

<0.001* 0.898

<0.001* 0.909

<0.001* 0.001*

P-value

Estimates 1.673 1.946 1.276 0.273 -0.397 -0.669 Estimates 0.972 0.805 -0.167

Coefficients fourth - first second - first third - first second - fourth third - fourth third - second

Coefficients eten + - eten eten +/- - eten eten +/- - eten +

SE 0.186 0.211 0.193

SE 0.403 0.389 0.394 0.201 0.211 0.183 P-value <0.001* <0.001* 0.659

P-value <0.001* <0.001* 0.005* 0.510 0.222 0.001*

Logistic regression

Negative Binomial Bird food crop Estimates SE P-value 2.575 0.432 <0.001* 2.845 0.409 <0.001* 2.327 0.410 <0.001* 0.270 0.286 0.774 -0.248 0.287 0.817 -0.519 0.250 0.155 Food availability Estimates SE P-value 0.970 0.243 <0.001* 1.408 0.276 <0.001* 0.438 0.273 0.243 Estimates 1.014 0.934 -0.079

Estimates 1.825 2.096 1.331 0.270 -0.493 -0.764

SE 0.215 0.239 0.219

SE 0.445 0.425 0.426 0.236 0.249 0.212

GLMM

P-value <0.001* <0.001* 0.930

P-value <0.001* <0.001* 0.008* 0.645 0.182 0.001*

Table 7.18: Pairwise comparisons of the variable Bird food crop and food availability for Rietgors

Intercept nearest road

Intercept nearest railroad

Intercept nearest plot

Intercept nearest built-upon area

Intercept nearest building

Intercept Hellmann number

Intercept temperature

Intercept OPP

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Plot size -3.328 0.252 <0.001* -1.456 0.546 0.007* 0.542 0.385 0.16 3.390 0.943 <0.001* Weather at the sampling moment -2.436 0.174 <0.001* 1.112 0.355 0.001* 0.002 0.027 0.948 -0.090 0.056 0.110 Weather/type of winter -2.402 0.140 <0.001* 0.702 0.291 0.016 -0.008 0.008 0.33 0.005 0.013 0.721 Nearest building -2.606 0.275 <0.001* 0.720 0.559 0.198 0.00046 0.00056 0.412 -0.001 0.001 0.624 Nearest built-upon area -2.796 0.334 <0.001* 0.694 0.654 0.289 0.00067 0.0005 0.182 -0.0003 0.001 0.743 Nearest plot -2.277 0.227 <0.001* 1.017 0.462 0.027 -0.00087 0.00077 0.258 -0.003 0.001 0.029 Nearest railroad -2.068e+00 2.172e-01 <0.001* 1.110 0.498 0.026 -1.486e-04 5.209e-05 0.004* -0.0001 0.0001 0.188 Nearest road -2.356 0.20528 <0.001* -0.0058 0.00678 0.391 -3.559 0.0005

-4.669 0.001

-4.716 -0.011

-5.517 1.138

Estimates

0.689 0.009

0.591 0.010

0.641 0.039

1.100 1.316

GLMM SE

Table 7.19: Results of Logistic regression, Negative Binomial regression and GLMM for Grauwe Gors

<0.001* 0.96

<0.001* 0.915

<0.001* 0.757

<0.001* 0.387

P-value

birds. However, the logistic regression and GLMM did not find the same effect. The logistic regression model shows a negative impact of the distance to the nearest plot on the Geelgors species. We infer that a 1 km increase in the distance to the nearest plot has around 75% (e−0.0014∗1000 = 0.25(1.11; 14.70)) decrease in the odds of finding a Geelgors bird. The negative binomial model indicates a negative effect of this covariate on the Geelgors species. However, it can be seen in the Appendix section of this report that after log transforming the variable “Nearest plot” based on the results of the exploratory data analysis, this covariate is no longer significant for the negative binomial model. A borderline situation is encountered by the logistic regression for the covariate “nearest building” when all plots are included in the analysis. In the Appendix section of this report, it can be seen that the logistic regression indicates a strong negative effect of this variable on the Geelgors species. No effect was found by the negative binomial, while the GLMM did not converge for this variable. The variable “nearest road” is found to be a borderline situation by the negative binomial regression in the Appendix section, but no other model finds an effect of this covariate. In the original analysis, no effect was found by any of the models used. “Temperature” shows a positive significant results according to GLMM. The odds ratio is e0.056 = 1.06(1.01; 1.11), thus the odds of finding a Geelgors bird increase by 6% with each additional degree Celsius. The results in Table 16 of the three models applied show no effect of the variable “Nearest build-upon area”, but we observed a possible outlying observation in the exploratory data analysis. After proceeding with the calculations without that observation, it can be seen in Table 67 that the logistic regression model indicates a negative statistically significant effect. In conclusion, the “plot size” has a positive effect on the amount of birds, while “temperature” has an increasing effect on presence of this species. Therefore, higher temperature at the sampling moment leads to a higher presence of Geelgors. An increasing distance to the nearest building, nearest built-upon area and nearest plot have a negative effect on presence of Geelgors species.

7.2.3

Veldleeuwerik

The negative binomial model indicates a negative impact of the “temperature” on the Veldleeuwerik species, thus the expected count of Veldleeuwerik 80

Intercept nearest road

Intercept nearest railroad

Intercept nearest plot

Intercept nearest built-upon area

Intercept nearest building

Intercept Hellmann number

Intercept temperature

Intercept OPP

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Plot size -1.447 0.123 <0.001* 0.825 0.200 <0.001* 0.748 0.199 <0.001* 1.511 0.345 <0.001* Weather at the sampling moment -0.335 0.097 <0.001* 2.376 0.125 <0.001* 0.035 0.015 0.023 -0.018 0.019 0.351 Weather/type of winter -0.175 0.076 <0.001* 2.280e+00 9.975e-02 <0.001* -0.002 0.003 0.456 8.934e-05 4.398e-03 0.984 Nearest building -0.503 0.174 0.0039* 2.397 0.268 <0.001* -0.001147 0.00047 0.015 -0.001 0.0006 0.027 Nearest built-upon area -0.525 0.194 0.0067* 2.277 0.314 <0.001* -0.00069 0.00035 0.047 -0.0006 0.0005 0.26 Nearest plot -0.582 0.136 <0.001* 2.302 0.223 <0.001* -0.00136 0.00046 0.0029* -0.002 0.0006 0.006* Nearest railroad -6.701e-01 1.250e-01 <0.001* 2.054e+00 1.772e-01 <0.001* 4.752e-05 2.500e-05 0.057 1.766e-05 3.587e-05 0.622 Nearest road -0.805 0.121 <0.001* 2.116 0.197 <0.001* -0.0027 0.00319 0.384 -0.008 0.005 0.108 -0.268 -0.002

-0.183 -0.005

-0.446 0.056

-0.516 0.683

Estimates

Table 7.20: Results of Logistic regression, Negative Binomial regression and GLMM for Geelgors

0.239 0.005

0.139 0.004

0.161 0.019

0.313 0.540

GLMM SE

0.263 0.710

0.189 0.195

0.005* 0.003*

0.099 0.207

P-value

birds reduces by e−0.103 = 0.90(0.83; 0.97) for every additional one degree Celsius. No effect of “temperature” is found by the logistic regression and GLMM. The results presented in Table 71 show that the “Hellmann number” variable is found to have a negative effect on the Veldleeuwerik species given by the logistic regression and GLMM, by both transformations applied to this covariate in the Appendix section. In the case of the dichotomized version of the variable, the logistic regression found a negative borderline significance. However, the logarithmic transformation of the “Hellmann number” variable gives a positive effect by the negative binomial model. The logistic regression model shows a positive impact of the distance to the nearest building variable on the Veldleeuwerik species. The odds ratio is e0.002∗1000 = 7.38(1.57; 34.66), thus the odds of finding a Veldleeuwerik bird increase over 7-fold with each additional 1 km increase in the distance to the nearest building. However, the confidence interval is very wide in this situation. No effect was found by the negative binomial and GLMM. Furthermore, a positive impact of the distance to the nearest build-upon area on the Veldleeuwerik species was indicated by the logistic regression model. The odds ratio is e0.0015∗1000 = 4.48(1.23; 16.24), thus the odds of finding a Veldleeuwerik bird increase over 4-fold with each additional 1 km increase in the distance to the nearest build-upon area. However, no effect was found by the negative binomial and GLMM. In the same time, in Appendix section it can be seen that the results are not influenced by the outlying observation for Veldleeuwerik species. The negative binomial model indicates a negative borderline effect of the covariate “nearest road” on the Veldleeuwerik species in the Appendix section. The negative binomial model did not find any effect, while the GLMM did not converge for this covariate. In summary, for Veldleeuwerik, we find negative effect of “temperature” on the amount of birds, distance to the “nearest building” and “nearest built-upon area” has a positive impact on the presence of this species.

7.2.4

Kneu

A positive borderline effect of the plot size on Kneu species was found by the logistic regression model and GLMM. No effect was found by the negative binomial model. However, after transformation, this covariate is not significant anymore. “Hellmann number” was found to have a negative effect on Kneu species 82

Intercept nearest road

Intercept nearest railroad

Intercept nearest plot

Intercept nearest built-upon area

Intercept nearest building

Intercept Hellmann number

Intercept temperature

Intercept OPP

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Plot size -2.439 0.173 <0.001* -0.623 0.302 0.039 0.562 0.269 0.036 1.170 0.519 0.024 Weather at the sampling moment -1.527 0.127 <0.001* 1.179 0.186 <0.001* 0.040 0.019 0.038 -0.103 0.029 <0.001* Weather/type of winter -1.208 0.095 <0.001* 0.596 0.155 <0.001* -0.025 0.008 0.001* 0.020 0.007 0.003* Nearest building -2.4026 0.228 <0.001* 0.179 0.375 0.634 0.0018 0.00044 <0.001* 0.002 0.0008 0.038 Nearest built-upon area -2.592 0.275 <0.001* 0.018 0.439 0.966 0.0015 0.00039 <0.001* 0.002 0.0007 0.039 Nearest plot -1.939 0.184 <0.001* 0.961 0.316 0.002* 0.00088 0.00044 0.046 -0.0007 0.0008 0.417 Nearest railroad -1.524e+00 1.585e-01 <0.001* 5.803e-01 2.716e-01 0.032 -1.516e-05 3.249e-05 0.641 2.424e-05 5.495e-05 0.659 Nearest road -1.885 0.159 <0.001* 0.872 0.278 0.001* 0.0074 0.0029 0.014 -0.003 0.007 0.650 -1.728 0.007

-1.554 -0.026

-1.868 0.034

-2.048 0.494

Estimates

Table 7.21: Results of Logistic regression, Negative Binomial regression and GLMM for Veldleeuwerik

0.287 0.005

0.164 0.008

0.194 0.022

0.338 0.517

GLMM SE

<0.001* 0.154

<0.001* 0.002*

<0.001* 0.119

<0.001* 0.339

P-value

by the logistic regression model and GLMM. We interpret the results from the Appendix section under both transformations applied (Table 72). The negative binomial model did not show any effect of this covariate on Kneu species. “Temperature” shows a positive significant results according to the logistic regression. The odds ratio is e0.059 = 1.06(1.01; 1.12), thus the odds of finding a Kneu bird increase by 6% with each additional degree Celsius. The negative binomial as well as GLMM indicate no effect of this covariate on the Kneu species. In summary, the “Hellmann number” has a negative impact on the presence of Kneu species. This also corresponds with a positive impact on presence with “temperature”. A higher temperature at the sampling moment seems to result in a higher presence of Kneu.

7.2.5

Rietgors

The covariate “Plot size” was found to have a positive significant effect on the Rietgors species. According to the logistic regression results presented in Table 73 in the Appendix, the estimated odds of seeing a Rietgors bird multiply by e0.907 = 2.47(1.21; 5.07) for each 1 m2 increase in the size of the plot. The odds ratio, given by the results of GLMM, is e1.119 = 3.06(1.01; 9.27), thus the odds of finding a Rietgors bird increase over 3-fold with each additional m2 in the plot size. The expected count of Rietgors birds increases by e1.805 = 6.07(2.18; 16.95) for every additional m2 in the plot size. “Temperature” shows a negative borderline result according to logistic regression and GLMM. The negative binomial model shows a negative impact of “Temperature” on Rietgors species, thus the expected count of Rietgors birds reduces by e−0.063 = 0.93(0.88; 0.99) for every additional one degree Celsius. In Appendix section of this project, it can be seen that the “nearest plot” variable was found to be a significant variable by the logistic regression and the negative binomial model. No effect was found by any model when we included all the plots in the analysis. However, in Appendix section, the logistic regression found a positive effect of “nearest build-upon area” on Rietgors species. In conclusion, larger plots correspond to higher presence and higher counts of Rietgors. An increase in “temperature” reduces the number of birds on a plot. Distance to the “nearest plot” decreases the amount of birds of this species. 84

Intercept nearest road

Intercept nearest railroad

Intercept nearest plot

Intercept nearest built-upon area

Intercept nearest building

Intercept Hellmann number

Intercept temperature

Intercept OPP

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Plot size -2.639 0.175 <0.001* 0.825 0.371 0.026 0.892 0.257 <0.001* 1.217 0.643 0.058 Weather at the sampling moment -1.725 0.137 <0.001* 2.041 0.233 <0.001* 0.059 0.021 0.004* 0.001 0.037 0.972 Weather/type of winter -1.293 0.095 <0.001* 2.129 0.182 <0.001* -0.013 0.006 0.025 -0.0006 0.008 0.938 Nearest building -1.900 0.206 <0.001* 1.887e+00 2.880e-02 <0.001* 0.00063 0.00042 0.136 1.012e-04 6.452e-05 0.117 Nearest built-upon area -1.919 0.2469 <0.001* 0.000468 0.00039 0.233 Nearest plot -1.7673 0.177 <0.001* 0.00036 0.00046 0.431 Nearest railroad -1.485e+00 1.578e-01 <0.001* 2.127e+00 3.172e-01 <0.001* -2.630e-05 3.266e-05 0.421 -3.875e-05 6.424e-05 0.546 Nearest road -1.718 0.154 <0.001* 0.0018 0.0034 0.595 -

-2.110 0.959

Estimates

Table 7.22: Results of Logistic regression, Negative Binomial regression and GLMM for Kneu

0.284 0.422

GLMM SE

<0.001* 0.023

P-value

Intercept nearest road

Intercept nearest railroad

Intercept nearest plot

Intercept nearest built-upon area

Intercept nearest building

Intercept Hellmann number

Intercept temperature

Intercept OPP

Coefficients

Logistic regression Negative Binomial Estimates SE P-value Estimates SE P-value Plot size -2.159 0.146 <0.001* -0.792 0.234 0.0007* 0.969 0.221 <0.001* 2.201 0.400 <0.001* Weather at the sampling moment -0.771 0.101 <0.001* 1.237 0.149 <0.001* -0.037 0.016 0.023 -0.063 0.023 0.008* Weather/type of winter -0.925 0.084 <0.001* 0.912 0.119 <0.001* 0.002 0.003 0.575 0.008 0.005 0.115 Nearest building -1.4466 0.184 0.0603 0.624 0.281 0.026 0.000389 0.0004 0.331 -0.0002 0.0006 0.714 Nearest built-upon area -1.276e+00 2.162e-01 <0.001* 0.804 0.327 0.014 -8.348e-05 3.708e-04 0.822 -0.0005 0.0005 0.3205 Nearest plot -1.201 0.1531 0.0005* 0.804 0.226 <0.001* -0.00054 0.00046 0.24 -0.002 0.0006 0.003* Nearest railroad -1.165e+00 1.400e-01 <0.001* 9.458e-01 2.096e-01 <0.001* 1.299e-05 2.811e-05 0.644 -2.024e-05 4.247e-05 0.634 Nearest road -1.282 0.136 <0.001* 0.647 0.207 0.002* -0.00188 0.0035 0.595 -0.006 0.005 0.247 -0.923 -0.005

-1.109 0.003

-0.924 -0.040

-1.577 1.120

Estimates

Table 7.23: Results of Logistic regression, Negative Binomial regression and GLMM for Rietgors

0.249 0.006

0.132 0.004

0.147 0.018

0.269 0.430

GLMM SE

<0.001* 0.382

<0.001* 0.354

<0.001* 0.027

<0.001* 0.009*

P-value

7.3

Summary of results

Results are summarized in Tables 7.24-7.31. Some significant effects were found on presence/absence of Grauwe Gors based on the logistic regression. But when accounting for the sampling frequency within plots, none of these remain significant. Only distance to the nearest railroad has an impact on presence/absence of Grauwe Gors. For the number of birds observed, there is a positive effect of adjacency to a hedge, as well as distance to the nearest road. The average number of birds increases with a factor of 4.47 (1/01; 19.75) when the plot is adjacent to an hedge. The presence of hedge on a plot at 50 meters, increases both the presence and the number of Geelgors. The presence of bird food crop (especially grain without admixture of other crops and grain in combination with flax), increases presence and amount of Geelgors. While, the adjacency to unpaved roads increases the presence of Geelgors, ongoing traffic reduces the number of Geelgors. Furthermore, the plot size has a positive effect on the amount of birds, while temperature has an increasing effect on presence of this species. Therefore, higher temperature at the sampling moment leads to a higher presence of Geelgors. An increasing distance to the nearest building, nearest built-upon area and nearest plot have a negative effect on presence of Geelgors species. The (binary) significant effects on prevalence of birds is presented in Figure 7.1.

Figure 7.1: Significant effects on prevalence estimates for Geelgors

The presence of hedge, forest and fruit cultivation reduces both the presence and the number of Veldleeuwerik, especially in 50 meters from plot. Also garden and tree colonnade at a distance of 50 meters reduces the amount of Veldleeuwerik. In addition, we find negative effect of temperature on the amount of birds, distance to the nearest building and nearest built-upon area has a positive impact on the presence of this species. The (binary) significant effects on prevalence of birds is presented in Figure 7.2.

Figure 7.2: Significant effects on prevalence estimates for Veldleeuwerik The presence and number of Kneu increases with the availability of growth. It benefits most from grain in combination with bladrammanas category of bird food. Faunal grass stretch also increases the presence of Kneu (though not the amount of Kneu). The Hellmann number has a negative impact on the presence of Kneu species. This also corresponds with a positive impact on presence with temperature. A higher temperature at the sampling moment seems to result in a higher presence of Kneu. The (binary) significant effects on prevalence of birds is presented in Figure 7.3.

Figure 7.3: Significant effects on prevalence estimates for Kneu For Rietgors, the availability of any food (> 50%) is important for both presence and amount of birds. It prefers any bird food crop. Also, bird food cultivation 50 meter increases presence and amount of bird. Tree colonnade reduces the number of birds. Larger plots correspond to higher presence and higher counts of Rietgors. An increase in temperature reduces the number of birds on a plot. Distance to the nearest plot decreases the amount of birds of this species. The (binary) significant effects on prevalence of birds is presented in Figure 7.4.

Figure 7.4: Significant effects on prevalence estimates for Rietgors

Bird food crop (2 categories) Hedge Forest Tree collonade Garden Fruit cultivation Tall cone orchard Faunal grass stretch Ongoing traffic Unpaved road Local farm traffic Rail road Tall cone orchard 50 meter Bird food cultivation 50 meter Forest 50 meter Hedge 50 meter Fruit cultivation 50 meter Garden 50 meter Tree collonade 50 meter

Bird food crop (4 categories)

Variable Food availability eten + eten +/fourth category third category second category teelt

Grauwe Gors no effect no effect no effect no effect no effect no effect positive no effect no effect no effect negative positive no effect no effect no effect no effect positive no effect no effect negative

Geelgors positive no effect no effect positive positive positive positive positive no effect no effect no effect negative positive no effect no effect no effect positive no effect no effect negative

Veldleeuwerik no effect no effect no effect no effect no effect no effect negative negative no effect negative no effect no effect no effect positive no effect negative negative negative no effect negative

Kneu positive positive positive positive positive positive no effect no effect no effect no effect positive no effect no effect positive no effect no effect no effect no effect no effect no effect

Table 7.24: Significance using Logistic regression for categorical variables per species Rietgors positive positive positive positive positive positive no effect no effect negative no effect no effect no effect no effect no effect positive no effect no effect no effect no effect no effect

Variable OPP Temperature Hellmann number Nearest building Nearest built-upon area Nearest plot Nearest railroad Nearest road

Grauwe Gors no effect no effect no effect no effect no effect no effect negative no effect

Geelgors positive no effect no effect negative no effect negative no effect no effect

Veldleeuwerik no effect no effect negative positive positive no effect no effect no effect

Kneu positive positive no effect no effect no effect no effect no effect no effect

Rietgors positive no effect no effect no effect no effect no effect no effect no effect

Table 7.25: Significance using Logistic regression for continuous variables per species

Bird food crop (4 categories)

Variable Food availability eten + eten +/fourth category third category second category teelt

Grauwe Gors no effect no effect no effect no effect no effect no effect positive no effect no effect no effect no effect no effect no effect no effect no effect no effect positive no effect no effect negative

Geelgors positive positive positive positive positive positive positive no effect no effect no effect no effect negative no effect no effect no effect no effect positive no effect no effect no effect

Veldleeuwerik no effect no effect no effect no effect no effect no effect negative no effect no effect negative no effect no effect no effect no effect no effect no effect negative negative negative negative

Kneu positive positive positive positive positive positive no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect

Table 7.26: Significance using Negative Binomial regression for categorical variables per species Rietgors positive positive positive positive positive positive no effect no effect negative no effect no effect no effect no effect no effect positive no effect no effect no effect no effect no effect

Variable OPP Temperature Hellmann number Nearest building Nearest built-upon area Nearest plot Nearest railroad Nearest road

Grauwe Gors positive no effect no effect no effect no effect no effect no effect -

Geelgors positive no effect no effect no effect no effect negative no effect no effect

Veldleeuwerik no effect negative positive no effect no effect no effect no effect no effect

Kneu no effect no effect no effect no effect no effect -

Rietgors positive negative no effect no effect no effect negative no effect no effect

Table 7.27: Significance using Negative Binomial regression for continuous variables per species

Bird food crop (4 categories)

Variable Food availability eten + eten +/fourth category third category second category teelt

Grauwe Gors no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect negative no effect no effect no effect no effect no effect no effect no effect no effect no effect

Geelgors positive positive no effect positive positive positive positive no effect no effect no effect no effect no effect positive no effect no effect no effect positive no effect no effect no effect

Veldleeuwerik no effect no effect no effect no effect no effect no effect negative negative no effect negative no effect no effect no effect no effect no effect negative negative negative no effect negative

Table 7.28: Significance using GLMM for categorical variables per species Kneu positive positive positive positive positive positive no effect no effect no effect no effect positive no effect no effect no effect no effect no effect no effect no effect no effect no effect

Rietgors positive positive positive positive positive positive no effect no effect no effect no effect no effect no effect no effect no effect positive no effect no effect no effect no effect no effect

Variable OPP Temperature Hellmann number Nearest building Nearest built-upon area Nearest plot Nearest railroad Nearest road

Grauwe Gors no effect no effect no effect no effect

Geelgors no effect positive no effect

Veldleeuwerik no effect no effect negative no effect

Kneu positive no effect no effect

Table 7.29: Significance using GLMM for continuous variables per species Rietgors positive negative no effect no effect

96 Rietgors

Kneu

Geelgors

eten + - eten eten +/- - eten eten +/- - eten + eten + - eten eten +/- - eten eten +/- - eten + eten + - eten eten +/- - eten eten +/- - eten +

Logistic regression positive no effect no effect positive positive no effect positive positive no effect

GLMM positive positive no effect positive positive no effect positive positive no effect

Negative binomial positive positive no effect positive positive no effect positive positive no effect

Table 7.30: Significance for Food availability for Geelgors, Kneu and Rietgors

97 Rietgors

Kneu

Geelgors

fourth - first second - first third - first second - fourth third - fourth third - second fourth - first second - first third - first second - fourth third - fourth third - second fourth - first second - first third - first second - fourth third - fourth third - second

Logistic regression no effect positive positive positive positive positive positive positive no effect negative negative no effect positive positive positive no effect no effect negative

GLMM no effect positive positive positive positive no effect positive positive no effect negative negative no effect positive positive positive no effect no effect negative

Negative binomial no effect positive positive positive positive no effect positive positive positive positive negative negative no effect positive positive positive no effect no effect no effect

Table 7.31: Significance for Bird food crop for Geelgors, Kneu and Rietgors

7.4 7.4.1

Conclusion Grauwe Gors species

For the Grauwe Gors species, an effect of the contractual agro-environmental agreements could not be shown, therefore the bird food cultivation has no effect on the presence/absence or on the amount of this species.There is a positive effect of adjacency of “hedge” on this species. Thus, we conclude that this species prefers an undisturbed environment. The “number of plots within cluster of 100, 150, 200, 250 and 300 meters” has a positive effect according to logistic regression on this species, therefore increasing the number of plots within these clusters leads to an increased possibility to find Grauwe Gors. The plot size does not play a role on the presence/absence or amount of the Grauwe Gors species.

7.4.2

Geelgors species

In general, one can say that the agro-environmental agreements have a highly positive effect on Geelgors. The presence of “bird food crop” (especially grain without admixture of other crops and grain in combination with flax), increases presence and amount of Geelgors. The Geelgors species prefers a plot where bird food is available as compared with a plot with no bird food. A large positive effect was found by the negative binomial model of the variable “plot size” on Geelgors birds. Larger plots seem to attract more Geelgors birds. The presence of “hedge on a plot at 50 meters”, increases both the presence and the number of Geelgors. While, the adjacency to “unpaved roads” increases the presence of Geelgors, “ongoing traffic” reduces the number of Geelgors. The “temperature” has a positive effect on this species found by GLMM. It is important to note that it is not sure whether “temperature” affects the occurrence and/or the detection probability of this species. In other words, a higher temperature may lead to an increased chance to detect Geelgors.

7.4.3

Veldleeuwerik species

For Veldleeuwerik, an effect of the “bird food crop” could not be seen, thus this species is not influenced by the presence/absence of “bird food crop” on the plots. The presence of “hedge”, “forest” and “fruit cultivation” reduces both the presence and the number of Veldleeuwerik, especially in 50 meters from plot. Also “garden” and “tree colonnade” at a distance of 50 meters reduces the amount of Veldleeuwerik. The “temperature” has a negative 98

effect as well according to the negative binomial model, thus this species is being observed more often on days with lower temperatures. Therefore, since these landscape elements negatively impact the Veldleeuwerik species, open areas with few elements of the landscape are favored. The plot size has no impact on the presence/absence or amount of this species.

7.4.4

Kneu species

Kneu is a species that highly benefits from the “bird food crop” agreements. The Kneu species seems to strongly benefit from food present on the plot, with the grain in combination of bladrammannas category of food being the most preferred. The “Hellmann number” has a negative impact on the presence of Kneu species. This also corresponds with a positive impact on presence with “temperature”. A higher temperature at the sampling moment seems to result in a higher presence of Kneu. The size of the plots does not influence this species.

7.4.5

Rietgors species

Rietgors is a species that benefits greatly from the “bird food crop” strategies, more specifically when nearby plots form a greater aggregation of specifically cultivated fields. The availability of any food (> 50%) is important for both presence and amount of birds. It prefers any bird food crop, but Rietgors benefits more from the grain without admixture of other crops (Tarwe/triticale, Tarwe, luzerne, Haver) category of food as compared with the grain in combination with flax (Tarwe/vlas, haver/vlas). A larger plot size attracts more birds. Also, “bird food cultivation 50 meter” increases presence and amount of bird. “Tree colonnade” reduces the number of birds present on the plots. Therefore, the conservation strategy for this species should focus more on cultivated plot locations that are closer to each other.

Part V

Results: Openness/Closeness

100

Chapter 8

Results: Statistical analysis of effect of Openness/Closeness The environment around a plot was characterized by five variables according to two maps in order to describe the openness or closeness. We consider that an open plot is not surrounded by any type of building, build-upon area, forest or any other natural barrier. Thus, it was of interest to investigate if an open or a closed environment is preferred by a specific species and if this characteristic leads to an increase or decrease in the number of birds of the five species.

8.1

Changes in time

Since we do not have these variables measured over the entire study period, it is important to see if the 5 covariates change over time. Therefore, the five plots presented below are an exploratory tool to check if the openness/ closeness of the plots changed over time. As can be seen in Figures 8.1, the landscape according to the Biological Evaluation Map over a radius of 200 meters and 500 meters changed in 2017 as compared to 2012. It seems that the plots became more open in 2017 as compared with 2012. The figures further indicate the change in openness and closeness regarding the building and build-upon areas. As can be seen, these variables do not change much over the study period. In order to check the effect of these covariates on the presence and amount of birds, we do three types of 101

analyses: (1) we investigate the relationship of openness and occurrence of birds for measurements collected in 2012 only; (2) we investigate the same for measurements collected in 2016 and 2017; (3) we investigate the relationship for all years, but only for those variables that remain constant over the years. These last results are given in Appendix. In this way, we check the significance of these variables.

Figure 8.1: Index landscape Biological Evaluation Map 500 (upper left), 200 (upper right), Index BB Biological Evaluation Map 200 (middle left), 500 (middle right), Index BB Map Flanders 200

102

8.2

Results based on data from 2012

Results for 2012 are summarized in Tables 8.1-8.5. Almost none of these effects are found to be significant. A positive significant results was found by the logistic regression (presence/absence) for the variable “Index BB Biological Evaluation Map 500 meters” in the case of Rietgors species. No effect of this variable was found for the amount of birds. Note however that there is a serious reduction in power to detect effect, as only one year of data collection is used here.

8.3

Results based on data from 2016 and 2017

Similarly, we conduct the analysis for data of 2016 and 2017 together (Tables 8.6-8.10). Note that this is done because there is too limited information for year 2017 only. For the Geelgors species, a positive significant results was found by the logistic regression model for the variable “Index BB Biological Evaluation Map 200 meters”. No effect of this variable was found by the negative binomial model. According to the negative binomial model, the variable “Index BB Biological Evaluation Map 200 meters” has a significant positive effect on Veldleeuwerik species, which is not determined by the logistic regression. The same situation happens for “Index BB Map Flanders 200 meters” which has a positive impact on Veldleeuwerik species according to the negative binomial, but not according to the logistic regression model. In addition, when we compare the results of “Index BB” with the analysis of all years (and assuming “Index BB” is constant), we again find other results . When we include all the information about the visits made on plots, the effect of these on the Veldleeuwerik species become negative, as it can be seen in the Appendix. This indicates that results have to be interpreted with care. “Index landscape Biological Evaluation Map 500 meters” has a positive effect on Grauwe Gors, Geelgors and Rietgors species as reported by the negative binomial model, but no effect is found by the logistic regression model. When we compare the results of the logistic regression model in 2012 with the ones of the same model from 2016 and 2017, we can see that the effects found are not constant over time and in general no effect of these variables is found.

103

104

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error P-value Index BB Biological Evaluation Map 200 Intercept -3.237 0.507 <0.001* -3.238 0.507 <0.001* Index BB Biological Evaluation Map 200 7.490 7.958 0.347 7.491 7.958 0.347 Index BB Biological Evaluation Map 500 Intercept -3.288 0.585 <0.001* -0.033 1.166 0.978 Index BB Biological Evaluation Map 500 5.460 8.692 0.53 -2.539 22.120 0.909 Index BB Map Flanders 200 Intercept -3.073 0.465 <0.001* -0.017 0.960 0.986 Index BB Map Flanders 200 -5.326 30.882 0.863 -103.477 185.384 0.577 Index landscape Biological Evaluation Map 200 Intercept -4.572 6.459 0.479 -19.47 15.64 0.213 Index landscape Biological Evaluation Map 200 1.553 6.730 0.817 20.16 16.25 0.215 Index landscape Biological Evaluation Map 500 Intercept -4.685 7.156 0.513 -20.50 15.19 0.177 Index landscape Biological Evaluation Map 500 1.721 7.684 0.823 21.76 16.32 0.183

Table 8.1: Results for Grauwe Gors based on the dataset from 2012

105

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error Index BB Biological Evaluation Map 200 Intercept 0.4003 0.2029 0.0485* 2.619 0.208 Index BB Biological Evaluation Map 200 -2.256 5.168 0.6625 -8.490 5.533 Index BB Biological Evaluation Map 500 Intercept 0.055 0.238 0.8182 2.327 0.241 Index BB Biological Evaluation Map 500 11.186 5.603 0.045 5.989 4.544 Index BB Map Flanders 200 Intercept 0.384 0.194 0.0481* 2.573 0.199 Index BB Map Flanders 200 -3.422 9.616 0.722 -17.174 10.619 Index landscape Biological Evaluation Map 200 Intercept 5.943 3.011 0.048* 6.331 2.389 Index landscape Biological Evaluation Map 200 -5.838 3.129 0.062 -4.039 2.508 Index landscape Biological Evaluation Map 500 Intercept 5.272 3.050 0.084 5.849 2.823 Index landscape Biological Evaluation Map 500 -5.294 3.277 0.106 -3.623 3.051

Table 8.2: Results for Geelgors based on the dataset from 2012

0.038 0.235

0.008* 0.107

<0.001* 0.106

<0.001* 0.188

<0.001* 0.125

P-value

106

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error P-value Index BB Biological Evaluation Map 200 Intercept -1.386 0.249 <0.001* 0.398 0.375 0.290 Index BB Biological Evaluation Map 200 0.015 6.426 0.998 3.627 9.622 0.706 Index BB Biological Evaluation Map 500 Intercept -1.284 0.285 <0.001* 0.625 0.432 0.148 Index BB Biological Evaluation Map 500 -3.538 6.041 0.558 -6.488 8.360 0.438 Index BB Map Flanders 200 Intercept -1.389 0.239 <0.001* 0.487 0.360 0.177 Index BB Map Flanders 200 0.630 11.750 0.957 -9.050 18.846 0.631 Index landscape Biological Evaluation Map 200 Intercept -9.985 5.354 0.062 -12.122 5.591 0.030 Index landscape Biological Evaluation Map 200 8.929 5.497 0.104 13.004 5.806 0.025 Index landscape Biological Evaluation Map 500 Intercept -4.543 3.779 0.229 -4.621 5.186 0.373 Index landscape Biological Evaluation Map 500 3.403 4.049 0.401 5.453 5.595 0.330

Table 8.3: Results for Veldleeuwerik based on the dataset from 2012

107

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error P-value Index BB Biological Evaluation Map 200 Intercept -1.683 0.279 <0.001* Index BB Biological Evaluation Map 200 -6.283 10.293 0.542 Index BB Biological Evaluation Map 500 Intercept -1.591 0.319 <0.001* Index BB Biological Evaluation Map 500 -5.852 7.474 0.434 Index BB Map Flanders 200 Intercept -1.672 0.267 <0.001* Index BB Map Flanders 200 -64.875 94.014 0.49 Index landscape Biological Evaluation Map 200 Intercept -3.117 3.586 0.385 Index landscape Biological Evaluation Map 200 1.433 3.743 0.702 Index landscape Biological Evaluation Map 500 Intercept -1.060 3.729 0.776 Index landscape Biological Evaluation Map 500 -0.749 4.038 0.853 -

Table 8.4: Results for Kneu based on the dataset from 2012

108

Estimates Standard Error P-value Estimates Index BB Biological Evaluation Map 200 Intercept -1.342 0.243 <0.001* 1.098 Index BB Biological Evaluation Map 200 4.134 5.502 0.452 -1.369 Index BB Biological Evaluation Map 500 Intercept -1.758 0.307 <0.001* 0.478 Index BB Biological Evaluation Map 500 13.096 4.958 0.008* 15.206 Index BB Map Flanders 200 Intercept -1.313 0.233 <0.001* 1.079 Index BB Map Flanders 200 6.503 10.044 0.517 0.457 Index landscape Biological Evaluation Map 200 Intercept -1.127 2.770 0.684 1.808 Index landscape Biological Evaluation Map 200 -0.161 2.908 0.956 -0.767 Index landscape Biological Evaluation Map 500 Intercept 4.048 3.027 0.1811 8.248 Index landscape Biological Evaluation Map 500 -5.811 3.312 0.079 -7.832

Coefficients

Table 8.5: Results for Rietgors based on the dataset from 2012 P-value 0.005* 0.894 0.286 0.070 0.004* 0.981 0.693 0.873 0.120 0.172

Standard Error 0.396 10.257 0.448 8.400 0.379 19.075 4.577 4.803 5.308 5.738

109

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error P-value Index BB Biological Evaluation Map 200 Intercept -2.190 0.291 <0.001* Index BB Biological Evaluation Map 200 1.842 6.047 0.761 Index BB Biological Evaluation Map 500 Intercept -1.795 0.307 <0.001* 0Index BB Biological Evaluation Map 500 -11.009 7.280 0.13 Index BB Map Flanders 200 Intercept -2.015 0.257 <0.001* Index BB Map Flanders 200 -37.545 35.626 0.292 Index landscape Biological Evaluation Map 200 Intercept -6.504 2.780 0.019* Index landscape Biological Evaluation Map 200 4.797 2.987 0.108 Index landscape Biological Evaluation Map 500 Intercept -7.939 2.718 0.003* -11.453 4.039 0.005* Index landscape Biological Evaluation Map 500 6.618 3.012 0.028 14.088 4.676 0.003*

Table 8.6: Results for Grauwe Gors based on the dataset from 2016 and 2017

110

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error Index BB Biological Evaluation Map 200 Intercept -0.643 0.185 <0.001* 2.583 0.259 Index BB Biological Evaluation Map 200 12.637 4.357 0.003* -3.571 5.762 Index BB Biological Evaluation Map 500 Intercept -0.437 0.2001 0.029* 2.626 0.288 Index BB Biological Evaluation Map 500 2.029 3.017 0.501 -3.171 4.404 Index BB Map Flanders 200 Intercept -0.456 0.166 0.005* 2.539 0.238 Index BB Map Flanders 200 14.772 8.324 0.075 -4.547 11.155 Index landscape Biological Evaluation Map 200 Intercept 2.249 1.173 0.055 6.209 1.602 Index landscape Biological Evaluation Map 200 -2.945 1.315 0.025 -4.283 1.794 Index landscape Biological Evaluation Map 500 Intercept 1.090 1.197 0.363 7.508 1.688 Index landscape Biological Evaluation Map 500 -1.704 1.405 0.225 -6.080 1.976

Table 8.7: Results for Geelgors based on the dataset from 2016 and 2017

<0.001* 0.002*

<0.001* 0.017

<0.001* 0.684

<0.001* 0.472

<0.001* 0.535

P-value

111

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error Index BB Biological Evaluation Map 200 Intercept -0.644 0.186 <0.001* 1.851 0.270 Index BB Biological Evaluation Map 200 -2.642 4.335 0.542 -17.438 6.338 Index BB Biological Evaluation Map 500 Intercept -0.682 0.208 0.001* 1.953 0.304 Index BB Biological Evaluation Map 500 -0.477 3.202 0.881 -10.490 4.747 Index BB Map Flanders 200 Intercept -0.634 0.171 <0.001* 1.761 0.250 Index BB Map Flanders 200 -11.043 9.901 0.265 -38.776 14.509 Index landscape Biological Evaluation Map 200 Intercept -3.527 1.427 0.013* -3.070 1.778 Index landscape Biological Evaluation Map 200 3.160 1.570 0.044 5.121 1.982 Index landscape Biological Evaluation Map 500 Intercept -3.377 1.387 0.015* 1.953 0.304 Index landscape Biological Evaluation Map 500 3.131 1.599 0.0502 -10.490 4.747

Table 8.8: Results for Veldleeuwerik based on the dataset from 2016 and 2017

<0.001* 0.027

0.084 0.010

<0.001* 0.007*

<0.001* 0.027

<0.001* 0.005*

P-value

112

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error P-value Index BB Biological Evaluation Map 200 Intercept -1.101 0.205 <0.001* Index BB Biological Evaluation Map 200 -1.308 4.681 0.78 Index BB Biological Evaluation Map 500 Intercept -0.966 0.225 <0.001* Index BB Biological Evaluation Map 500 -4.284 3.885 0.27 Index BB Map Flanders 200 Intercept -1.158 0.188 <0.001* Index BB Map Flanders 200 3.741 8.283 0.652 Index landscape Biological Evaluation Map 200 Intercept -0.176 1.224 0.885 Index landscape Biological Evaluation Map 200 -1.082 1.381 0.434 Index landscape Biological Evaluation Map 500 Intercept -1.724 1.406 0.220 Index landscape Biological Evaluation Map 500 0.699 1.637 0.669 -

Table 8.9: Results for Kneu based on the dataset from 2016 and 2017

113

Logistic regression Negative Binomial Coefficients Estimates Standard Error P-value Estimates Standard Error Index BB Biological Evaluation Map 200 Intercept -0.269 0.178 0.130 1.205 0.200 Index BB Biological Evaluation Map 200 3.6229 3.941 0.358 2.073 4.410 Index BB Biological Evaluation Map 500 Intercept -0.195 0.197 0.322 1.224 0.222 Index BB Biological Evaluation Map 500 0.208 3.004 0.945 0.767 3.386 Index BB Map Flanders 200 Intercept -0.268 0.163 0.100 1.175 0.182 Index BB Map Flanders 200 11.831 8.175 0.148 8.787 8.413 Index landscape Biological Evaluation Map 200 Intercept -1.254 1.147 0.274 0.273 1.266 Index landscape Biological Evaluation Map 200 1.205 1.281 0.347 1.105 1.416 Index landscape Biological Evaluation Map 500 Intercept -2.786 1.262 0.027* -2.316 1.358 Index landscape Biological Evaluation Map 500 3.057 1.468 0.037 4.154 1.582

Table 8.10: Results Rietgors based on the dataset from 2016 and 2017

0.087 0.009*

0.829 0.435

<0.001* 0.296

<0.001* 0.821

<0.001* 0.638

P-value

In the case of the negative binomial model, no effect was observed for 2012, but using the data from 2016 and 2017 several positive effects were found. Moreover, the analysis presented in Appendix section shows negative effects. However, we consider that the effects observed are not strong enough, thus these variables do not affect any of the species of interest.

114

Part VI

Result: cluster variables

115

Chapter 9

Results: statistical analysis of clustering effect 9.1

Effects of number of plots under contract in 2017 within 100, 150, 200, 250, 300 m

As it can be seen in Tables 9.1 - 9.5, a strong positive effect on the Rietgors species of the variables representing the number of plots within 100, 150, 200, 250 and 300 meters was found by all three models used. No effect is seen for any other species. According to the logistic regression and GLMM, the odds ratio is e0.399 = 1.49(1.04; 2.12), thus the odds of finding a Rietgors bird increases by 49% with each additional plot within 100 meters. Furthermore, the expected count of Rietgors birds increases by e0.430 = 1.54(1.09; 2.15) for every additional plot within 100 meters. A similar situation is found for the variable representing the number of plots within 150 meters by the logistic regression and GLMM, thus the odds of seeing a bird from the Rietgors species increases by e0.331 = 1.39(1.02; 1.88) for every additional plot within the 150 meters. For the negative binomial model, we infer that the expected count of Rietgors birds increases by e0.354 = 1.42(1.08; 1.88) for every additional plot within 150 meters. Moreover, the odds of detecting the presence of a bird from the Rietgors species increases by e0.289 = 1.33(1.02; 1.74) for every additional plot within the 200 meters, while negative binomial regression results indicates a similar impact on the response of interest. The odds ratio for the covariate representing the number of plots within 116

117

-2.949 0.298 -2.874 0.237 -2.602 0.139 -2.836 0.171 -3.135 0.212

Intercept VVG d100

Intercept VVG d150

Intercept VVG d200

Intercept VVG d250

Intercept VVG d300

Coefficient

0.615 0.089

0.568 0.095

0.564 0.124

0.612 0.145

0.662 0.185

<0.001* 0.018

<0.001* 0.074

<0.001* 0.26

<0.001* 0.102

<0.001* 0.108

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value VVG d100 -0.795 1.062 0.454 0.462 0.363 0.204 VVG d150 VVG d200 VVG d250 VVG d300 -5.251 0.373

-4.844 0.311

-4.645 0.297

-5.050 0.455

-5.296 0.574

2.017 0.215

1.886 0.217

1.870 0.264

2.015 0.324

2.141 0.399

0.009* 0.083

0.010 0.150

0.013 0.261

0.012 0.159

0.013 0.149

GLMM Estimate SE P-value

Table 9.1: Results for Grauwe Gors based on the dataset from 2016 and 2017

118

-0.012 -0.057 0.141 -0.109 0.213 -0.122 0.273 -0.125 0.249 -0.107

Intercept VVG d100

Intercept VVG d150

Intercept VVG d200

Intercept VVG d250

Intercept VVG d300

Coefficient

0.322 0.069

0.318 0.074

0.336 0.092

0.346 0.106

0.360 0.125

0.439 0.117

0.392 0.092

0.526 0.182

0.683 0.303

0.973 0.648

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value VVG d100 3.218 0.491 <0.001* -0.172 0.169 0.31 VVG d150 3.230 0.470 <0.001* -0.152 0.141 0.28 VVG d200 3.375 0.452 <0.001* -0.193 0.119 0.104 VVG d250 3.381 0.426 <0.001* -0.173 0.094 0.066 VVG d300 3.092 0.435 <0.001* -0.069 0.088 0.436 -0.100 -0.133

-0.043 -0.163

-0.177 -0.135

-0.591 0.013

-1.299 0.326

1.317 0.300

1.299 0.324

1.390 0.394

1.465 0.462

1.640 0.578

0.939 0.658

0.974 0.614

0.899 0.732

0.687 0.978

0.428 0.573

GLMM Estimate SE P-value

Table 9.2: Results for Geelgors based on the dataset from 2016 and 2017

119

-0.959 0.116 -0.930 0.092 -0.911 0.076 -1.024 0.097 -1.029 0.090

Intercept VVG d100

Intercept VVG d150

Intercept VVG d200

Intercept VVG d250

Intercept VVG d300

Coefficient

0.343 0.066

0.339 0.071

0.354 0.089

0.367 0.106

0.386 0.129

0.002* 0.172

0.002* 0.169

0.010* 0.394

0.011* 0.383

0.013* 0.369

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value VVG d100 1.295 0.585 0.027 0.271 0.201 0.179 VVG d150 1.136 0.552 0.039 0.276 0.164 0.092 VVG d200 1.309 0.538 0.015 0.209 0.141 0.138 VVG d250 1.138 0.498 0.022 0.199 0.109 0.067 VVG d300 1.043 0.503 0.038 0.200 0.101 0.047 -1.259 0.068

-1.252 0.072

-1.143 0.039

-1.154 0.048

-1.180 0.065

0.561 0.109

0.555 0.118

0.586 0.149

0.605 0.177

0.632 0.211

0.025 0.537

0.024 0.539

0.051 0.795

0.056 0.786

0.062 0.758

GLMM Estimate SE P-value

Table 9.3: Results for Veldleeuwerik based on the dataset from 2016 and 2017

120

-1.503 0.146 -1.452 0.111 -1.753 0.189 -1.555 0.113 -1.563 0.105

Intercept VVG d100

Intercept VVG d150

Intercept VVG d200

Intercept VVG d250

Intercept VVG d300

Coefficient

0.385 0.070

0.380 0.075

0.409 0.095

0.408 0.114

0.433 0.139

<0.001* 0.135

<0.001* 0.133

<0.001* 0.047

<0.001* 0.327

<0.001* 0.295

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value VVG d100 VVG d150 VVG d200 VVG d250 VVG d300 -1.594 0.108

-1.588 0.116

-1.753 0.189

-1.509 0.118

-1.551 0.151

0.436 0.075

0.433 0.081

0.409 0.095

0.476 0.123

0.493 0.148

<0.001* 0.151

<0.001* 0.150

<0.001* 0.047

0.002* 0.339

0.002* 0.307

GLMM Estimate SE P-value

Table 9.4: Results for Kneu based on the dataset from 2016 and 2017

250 meters is e0.236 = 1.26(1.02; 1.56) given by the logistic model and GLMM, thus the odds of finding a Rietgors bird increase increase by 26% with each additional plot within 250 meters. A similar positive effect is found by the negative binomial as well. Lastly, according to the logistic regression and GLMM, the estimated odds of seeing a Rietgors bird multiply by e0.243 = 1.27(1.04; 1.55) for each additional plot within 300 meters. The expected count of Rietgors birds increases by e0.202 = 1.22(1.03; 1.45) for every additional plot within 300 meters. In conclusion, plots nearby have a beneficial effect for Rietgors.

9.2

Results by species for the area size of the buffer of 100m, 150m, 200m, 250m, 300m around the plot

Tables 9.6 until 9.10 show the positive impact of the the area size of the buffer on the presence of the Grauwe Gors species. According to the logistic regression results, the odds ratio is e3.191e−05∗1000 = 1.032(1.01; 1.06), thus the odds of finding a Grauwe Gors bird increases by 3.2% for an increase in the area size of 1000 m2 of the buffer of 100 meters around the plot. The odds of seeing a bird from the Grauwe Gors species increases by e2.478e−05∗1000 = 1.025(1.01; 1.05) for an increase in the area size of 1000 m2 of the buffer of 150 meters around the plot. Since the odds ratio is equal to e2.025e−05∗1000 = 1.02(1.003; 1.04), the odds of finding a Grauwe Gors bird grows by 2% for an increase in the area size of 1000 m2 of the buffer of 200 meters around the plot. Similarly, the odds of finding a Grauwe Gors bird increase by 1.7% (e1.712e−05∗1000 = 1.017(1.003; 1.03)) for an increase in the area size of 1000 m2 of the buffer of 250 meters around the plot. Finally, the odds of finding a Grauwe Gors bird grows by 1.5% (e1.482e−05∗1000 = 1.015(1.002; 1.02)) for an increase in the area size of 1000 m2 of the buffer of 300 meters around the plot. It is worth mentioning that the negative binomial model indicates a borderline situation of the covariates representing the area size of the buffer of 100, 150, 200, 250 and 300 meters for the Geelgors species. However, this effect is not found by the logistic regression model. Moreover, the magnitude of the estimate is very small, meaning that the effects of these covariates on the Geelgors species are not strong enough.

121

122

-1.094 0.399 -1.026 0.331 -0.999 0.289 -0.942 0.236 -1.041 0.243

Intercept VVG d100

Intercept VVG d150

Intercept VVG d200

Intercept VVG d250

Intercept VVG d300

Coefficient

0.347 0.078

0.338 0.082

0.361 0.103

0.371 0.118

0.385 0.138

0.002* 0.001*

0.005* 0.003*

0.006* 0.004*

0.005* 0.005*

0.004* 0.003*

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value VVG d100 0.207 0.389 0.595 0.430 0.131 0.001* VVG d150 0.283 0.370 0.445 0.354 0.108 0.001* VVG d200 0.335 0.352 0.341 0.288 0.090 0.001* VVG d250 0.435 0.331 0.188 0.218 0.071 0.002* VVG d300 0.419 0.335 0.211 0.202 0.066 0.002* -1.041 0.243

-0.942 0.236

-0.999 0.289

-1.026 0.331

-1.094 0.399

0.347 0.078

0.338 0.082

0.361 0.103

0.371 0.118

0.385 0.138

0.002* 0.001*

0.005* 0.003*

0.006* 0.004*

0.005* 0.005*

0.004* 0.003*

GLMM Estimate SE P-value

Table 9.5: Results for Rietgors based on the dataset from 2016 and 2017

123

-6.289e+00 2.478e-05 -7.216e+00 2.025e-05 -8.123e+00 1.712e-05 -9.018e+00 1.482e-05

Intercept OPP d100

Intercept OPP d150

Intercept OPP d200

Intercept OPP d250

Intercept OPP d300

2.312e+00 4.691e-06

2.036e+00 5.437e-06

1.758e+00 6.467e-06

1.475e+00 7.979e-06

<0.001* 0.001*

<0.001* 0.002*

-5.326e+00 3.191e-05

Coefficient 1.184e+00 1.043e-05

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value OPP d100 -2.854e+00 1.830e+00 0.119 3.465e-05 1.944e-05 0.074 OPP d150 -3.594e+00 2.288e+00 0.116 2.507e-05 1.451e-05 0.084 OPP d200 -4.328e+00 2.741e+00 0.1144 1.966e-05 1.158e-05 0.089 OPP d250 -5.057e+00 3.193e+00 0.113 1.618e-05 9.631e-06 0.093 OPP d300 -5.785e+00 3.644e+00 0.1124 1.375e-05 8.247e-06 0.095 -

GLMM Estimate SE P-value

Table 9.6: Results for Grauwe Gors based on the dataset from 2016 and 2017

124

-1.364e+00 8.030e-06 -1.579e+00 6.218e-06 -1.791e+00 5.064e-06 -2.003e+00 4.268e-06

Intercept OPP d100

Intercept OPP d150

Intercept OPP d200

Intercept OPP d250

Intercept OPP d300

1.313e+00 2.987e-06

1.153e+00 3.498e-06

9.920e-01 4.222e-06

8.311e-01 5.329e-06

0.127 0.153

0.120 0.148

0.111 0.141

0.101 0.132

0.088 0.121

-1.140e+00 1.123e-05

Coefficient 6.690e-01 7.234e-06

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value OPP d100 6.188e-01 8.211e-01 0.451 2.204e-05 8.759e-06 0.012 OPP d150 1.270e-01 1.030e+00 0.902 1.617e-05 6.556e-06 0.014 OPP d200 -3.586e-01 1.237e+00 0.772 1.278e-05 5.242e-06 0.015 OPP d250 -8.415e-01 1.444e+00 0.560 1.057e-05 4.367e-06 0.016 OPP d300 -1.323e+00 1.650e+00 0.423 9.008e-06 3.743e-06 0.016 -

GLMM Estimate SE P-value

Table 9.7: Results for Geelgors based on the dataset from 2016 and 2017

As it can be observed in the table below, the negative binomial model found a positive effect of the area size of the buffer of 100, 150, 200, 250 and 300 meters on the amount of the Veldleeuwerik birds. No effect was found on presence/absence by the logistic regression model for these covariates. Therefore, we infer that the expected count of Veldleeuwerik birds increases by e3.885e−05∗1000 = 1.039(1.01; 1.07) for an increase in the area size of 1000 m2 of the buffer of 100 meters around the plot. In the same time, the expected count of Veldleeuwerik birds increases by e2.850e−05∗1000 = 1.029(1.01; 1.05) for an increase in the area size of 1000 m2 of the buffer of 150 meters around the plot. An increment in the area size of 1000 m2 of the buffer of 200 meters around the plot results in an increase in the expected count of birds from Veldleeuwerik species by e2.248e−05∗1000 = 1.022(1.01; 1.04), thus a growth of 2.2%. Furthermore, the expected count of Veldleeuwerik birds increases by e1.856e−05∗1000 = 1.019(1.01; 1.03) for an increase in the area size of 1000 m2 of the buffer of 250 meters around the plot, while an increment in the area size of 1000 m2 of the buffer of 300 meters around the plot results in an increase in the expected count of birds from Veldleeuwerik species by e1.580e−05∗1000 = 1.016(1.004; 1.03), thus a growth of 1.6%. In the case of the Rietgors species, the logistic regression model results indicate a positive effect of these variables, while the negative binomial did not find any effect. Thus, according to the logistic regression model, the estimated odds of seeing a Rietgors bird multiply by e2.445e−05∗1000 = 1.025(1.001; 1.05) for an increment in the area size of 1000 m2 of the buffer of 100 meters around the plot, while an increase in the area size of 1000 m2 of the buffer of 150 meters leads to an increase in the estimated odds of seeing a Rietgors bird of 1.8% (e1.794e−05∗1000 = 1.018(1.001; 1.03)). The estimated odds of seeing a Rietgors bird increase of 1.4%, 1.2% and 1% for an increment in the area size of 1000 m2 for the buffer of 200, 250 and 300 meters around the plot. In conclusion, while we find a positive effect on presence of Grauwe Gors and Rietgors with area size of the buffer, we find a positive effect on the amount of birds for Veldleeuwerik. For all of these, the impact is largest in a smaller buffer.

125

126

-1.305e+00 4.114e-06 -1.540e+00 3.727e-06 -1.772e+00 3.355e-06 -2.001e+00 3.031e-06

Intercept OPP d100

Intercept OPP d150

Intercept OPP d200

Intercept OPP d250

Intercept OPP d300

1.324e+00 2.976e-06

1.159e+00 3.471e-06

9.935e-01 4.164e-06

8.276e-01 5.207e-06

0.131 0.308

0.126 0.334

0.121 0.371

0.115 0.429

0.107 0.534

-1.063e+00 4.333e-06

Coefficient 6.605e-01 6.963e-06

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value OPP d100 -1.809e+00 9.697e-01 0.062 3.885e-05 1.030e-05 <0.001* OPP d150 -2.706e+00 1.210e+00 0.025 2.850e-05 7.671e-06 <0.001* OPP d200 -3.573e+00 1.449e+00 0.014 2.248e-05 6.116e-06 <0.001* OPP d250 -4.426e+00 1.686e+00 0.008* 1.856e-05 5.086e-06 <0.001* OPP d300 -5.273e+00 1.924e+00 0.006* 1.580e-05 4.353e-06 <0.001* -

GLMM Estimate SE P-value

Table 9.8: Results for Veldleeuwerik based on the dataset from 2016 and 2017

127

Intercept OPP d300

Intercept OPP d250

Intercept OPP d200

Intercept OPP d150

Intercept OPP d100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP d100 -1.217e+00 7.311e-01 0.096 8.549e-07 7.773e-06 0.912 OPP d150 -1.201e+00 9.175e-01 0.190 3.955e-07 5.834e-06 0.946 OPP d200 -1.187e+00 1.102e+00 0.282 1.988e-07 4.669e-06 0.966 OPP d250 -1.174e+00 1.286e+00 0.361 1.013e-07 3.892e-06 0.979 OPP d300 -1.162e+00 1.470e+00 0.429 4.801e-08 3.337e-06 0.989 -

GLMM Estimate SE P-value

Table 9.9: Results for Kneu based on the dataset from 2016 and 2017

128

-2.848e+00 1.794e-05 -3.396e+00 1.418e-05 -3.939e+00 1.172e-05 -4.480e+00 9.997e-06

Intercept OPP d100

Intercept OPP d150

Intercept OPP d200

Intercept OPP d250

Intercept OPP d300

1.521e+00 3.498e-06

1.337e+00 4.113e-06

1.153e+00 4.992e-06

9.687e-01 6.354e-06

0.003* 0.004*

0.003* 0.005*

-2.289e+00 2.445e-05

Coefficient 7.832e-01 8.759e-06

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value OPP d100 5.774e-01 7.045e-01 0.412 9.264e-06 7.497e-06 0.217 OPP d150 3.402e-01 8.819e-01 0.700 6.966e-06 5.602e-06 0.214 OPP d200 1.089e-01 1.058e+00 0.918 5.586e-06 4.473e-06 0.212 OPP d250 -1.197e-01 1.233e+00 0.923 4.665e-06 3.723e-06 0.210 OPP d300 -3.468e-01 1.407e+00 0.805 4.006e-06 3.189e-06 0.209 -

GLMM Estimate SE P-value

Table 9.10: Results for Rietgors based on the dataset from 2016 and 2017

9.3

Effect of the number of plots within cluster of 100m, 150m, 200m, 250m, 300m

The logistic regression model indicates a positive strong effect of the variables number of plots within the cluster of 100, 150, 200, 250 and 300 meters on Grauwe Gors species, while the GLMM shows a borderline situation for the variables number of plots within the cluster of 100, 150, 200, 250. However, the negative binomial did not find any strong effect of these variables on the Grauwe Gors species. Therefore, we interpret the logistic regression results since those results are found to be strongly significant. The estimated odds of seeing a Grauwe Gors bird increase of 25% (e0.229 = 1.25(1.04; 1.51)), 22% (e0.202 = 1.22(1.04; 1.43)) and 17% (e0.155 = 1.17(1.02; 1.33)) for each additional plot within the cluster of 100, 150 and 200 meters respectively. Moreover, the estimated odds of seeing a bird from Grauwe Gors species multiply by e0.097 = 1.10(1.01; 1.19) for each additional plot within the cluster 250 meters, while an increase in the number of plots within a cluster of 300 meters leads to an increase in the estimated odds of seeing a Grauwe Gors bird of 9% (e0.092 = 1.09(1.007; 1.19)). For Veldleeuwerik species no effect was shown by the logistic regression model and GLMM. However, the negative binomial model indicated a positive borderline significant factor of the covariate representing the number of plots within a cluster of 250 meters, and a positive statistically significant effect of the number of plots within a cluster of 300 meters. Thus, we infer that the expected count of Veldleeuwerik birds increases by e0.076 = 1.08(1.01; 1.15) for an additional plot within a cluster of 300 meters. The logistic regression model shows a positive impact of the variables number of plots within cluster of 100, 150 and 300 meters on the presence of Rietgors species. There is a borderline results found for the variables number of plots within cluster of 200 and 250 meters. Thus, for an additional plot within a cluster of 100, 150 and 300 meters respectively, has a multiplicative effect of e0.123 = 1.13(1.01; 1.26), e0.103 = 1.11(1.01, 1.21) and e0.045 = 1.04(1.001; 1.09) on the odds that a Rietgors bird is observed. These results are confirmed by the GLMM model. For the variables number of plots within cluster of 100 and 150 meters, the negative binomial model found a positive significant effect as well. The expected count of Rietgors birds increases by e0.114 = 1.12(1.008; 1.24) and e0.096 = 1.10(1.008; 1.20) for an extra plot within 100 and 150 meters.

129

130

-3.986 0.229 -4.103 0.202 -3.963 0.155 -3.848 0.097 -3.871 0.092

Intercept sumVG C100

Intercept sumVG C150

Intercept sumVG C200

Intercept sumVG C250

Intercept sumVG C300

Coefficient

0.884 0.033

0.825 0.031

0.869 0.051

0.877 0.061

0.850 0.072

Negative Binomial Estimate SE P-value sumVG C100 <0.001* -1.034 0.815 0.205 0.001* 0.197 0.107 0.065 sumVG C150 <0.001* -1.262 0.798 0.114 <0.001* 0.184 0.087 0.035 sumVG C200 <0.001* -1.133 0.810 0.162 0.002* 0.144 0.072 0.044 sumVG C250 <0.001* -1.021 0.779 0.190 0.002* 0.089 0.045 0.048 sumVG C300 <0.001* -0.909 0.853 0.286 0.006* 0.080 0.046 0.082

Logistic regression Estimate SE P-value

-5.262 0.115

-5.139 0.123

-5.232 0.194

-5.328 0.254

-5.256 0.289

1.961 0.057

1.880 0.056

1.914 0.089

1.952 0.112

1.927 0.131

0.007* 0.042

0.006* 0.029

0.006* 0.028

0.006* 0.024

0.006* 0.027

GLMM Estimate SE P-value

Table 9.11: Results for Grauwe Gors based on the dataset from 2016 and 2017

131

-0.123 -0.005 -0.062 -0.014 -0.017 -0.017 -0.077 -0.006 0.117 -0.021

Intercept sumVG C100

Intercept sumVG C150

Intercept sumVG C200

Intercept sumVG C250

Intercept sumVG C300

Coefficient

0.294 0.016

0.276 0.016

0.287 0.026

0.287 0.032

0.288 0.039

0.691 0.212

0.778 0.705

0.953 0.515

0.829 0.670

0.670 0.904

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value sumVG C100 1.999 0.381 <0.001* 0.118 0.051 0.019 sumVG C150 2.071 0.381 <0.001* 0.092 0.043 0.030 sumVG C200 2.124 0.382 <0.001* 0.070 0.035 0.041 sumVG C250 2.190 0.369 <0.001* 0.044 0.022 0.044 sumVG C300 2.268 0.397 <0.001* 0.036 0.022 0.102 0.002 -0.048

-0.670 0.012

-0.582 0.003

-0.783 0.042

-1.077 0.113

1.179 0.073

1.164 0.076

1.206 0.118

1.244 0.155

1.297 0.191

0.998 0.513

0.565 0.879

0.629 0.977

0.529 0.789

0.406 0.553

GLMM Estimate SE P-value

Table 9.12: Results for Geelgors based on the dataset from 2016 and 2017

132

-0.808 0.024 -0.812 0.021 -0.821 0.018 -0.861 0.016 -1.216 0.039

Intercept sumVG C100

Intercept sumVG C150

Intercept sumVG C200

Intercept sumVG C250

Intercept sumVG C300

Coefficient

0.339 0.017

0.299 0.017

0.308 0.027

0.308 0.034

0.309 0.040

Negative Binomial Estimate SE P-value sumVG C100 0.008* 1.251 0.461 0.006* 0.544 0.110 0.061 0.072 sumVG C150 0.008* 1.133 0.455 0.013 0.527 0.105 0.051 0.038 sumVG C200 0.007* 1.084 0.453 0.017 0.503 0.089 0.041 0.029 sumVG C250 0.003* 1.039 0.432 0.016 0.352 0.061 0.025 0.017 sumVG C300 <0.001* 0.600 0.452 0.184 0.025 0.076 0.025 0.002*

Logistic regression Estimate SE P-value

-1.657 0.051

-1.179 0.014

-1.090 0.008

-1.071 0.006

-1.063 0.006

0.594 0.029

0.515 0.029

0.521 0.046

0.519 0.058

0.519 0.069

0.005* 0.079

0.022 0.622

0.036 0.863

0.039 0.914

0.041 0.933

GLMM Estimate SE P-value

Table 9.13: Results for Veldleeuwerik based on the dataset from 2016 and 2017

133

Intercept sumVG C300

Intercept sumVG C250

Intercept sumVG C200

Intercept sumVG C150

Intercept sumVG C100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value sumVG C100 -1.282 0.344 <0.001* 0.026 0.044 0.554 sumVG C150 -1.279 0.342 <0.001* 0.022 0.037 0.556 sumVG C200 -1.196 0.338 <0.001* 0.007 0.029 0.814 sumVG C250 -1.134 0.321 <0.001* -0.0006 0.019 0.972 sumVG C300 -1.265 0.352 <0.001* 0.009 0.019 0.621 -1.322 0.010

-1.182 -0.001

-1.245 0.007

-1.334 0.023

-1.336 0.028

0.415 0.020

0.376 0.021

0.391 0.032

0.403 0.039

0.405 0.047

0.002* 0.622

0.002* 0.961

0.002* 0.825

<0.001* 0.561

<0.001* 0.559

GLMM Estimate SE P-value

Table 9.14: Results for Kneu based on the dataset from 2016 and 2017

134

-0.792 0.123 -0.783 0.103 -0.685 0.069 -0.613 0.041 -0.741 0.045

Intercept sumVG C100

Intercept sumVG C150

Intercept sumVG C200

Intercept sumVG C250

Intercept sumVG C300

Coefficient

0.308 0.017

0.285 0.017

0.298 0.027

0.301 0.035

0.302 0.042

0.016* 0.008*

0.032* 0.016

0.021* 0.011

0.009* 0.003*

Logistic regression Estimate SE P-value

Negative Binomial Estimate SE P-value sumVG C100 0.645 0.313 0.039 0.114 0.041 0.005* sumVG C150 0.643 0.311 0.039 0.096 0.034 0.005* sumVG C200 0.717 0.313 0.022 0.072 0.028 0.011 sumVG C250 0.831 0.303 0.006* 0.041 0.018 0.019 sumVG C300 0.769 0.325 0.018 0.041 0.017 0.020 -0.741 0.045

-0.613 0.041

-0.685 0.069

-0.783 0.103

-0.792 0.123

0.308 0.017

0.285 0.017

0.298 0.027

0.301 0.035

0.302 0.042

0.016 0.008*

0.032* 0.016

0.021* 0.011

0.009* 0.003*

GLMM Estimate SE P-value

Table 9.15: Results for Rietgors based on the dataset from 2016 and 2017

In conclusion, the number of plots has a positive impact on the presence of Grauwe Gors. It also has a positive impact on the number of birds for Rietgors, for an increase of number of plots within a cluster of 150m. This positive effect for Rietgors is in line with previous results.

9.4

Effect of the sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters

Strongly positive significant results were found by the logistic regression model on presence of birds, while no effect was detected by the negative binomial on counts for the variables sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters on Grauwe Gors species. The estimated odds of finding a Grauwe Gors bird multiply by e4.083e−05∗1000 = 1.04(1.01; 1.07) for an increment of 1000 m2 in the sum of area sizes within 100 meters, while an increment of 1000 m2 in the sum of area sizes within 150 meters has a e3.765e−05∗1000 = 1.038(1.008; 1.06) effect in the odds of seeing a Grauwe Gors bird. The estimated odds of seeing a Grauwe Gors bird increase of 2.4% 2.376e−05∗1000 (e = 1.024(1.001; 1.04)), 2% (e2.024e−05∗1000 = 1.02(1.001; 1.03)) 2.326e−05∗1000 and 2.3% (e = 1.023(1.005; 1.04)) for an increase of 1000 m2 in the sum of area sizes within 200, 250 and 300 meters. There seems to be a borderline positive results of the variable sum of the area sizes of all plots within 300 meters on Geelgors species, detected by the negative binomial model. The logistic regression model did not find any effect of this variable. Both models used, logistic and negative binomial, showed a positive significant result of the variables sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters on Rietgors species. The results of the logistic regression model imply that the estimated multiplicative effect of 1000 m2 in sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters on the odds of seeing a Rietgors bird is e2.793e−05∗1000 = 1.028(1.002; 1.05), e2.657e−05∗1000 = 1.026(1.003; 1.05), e2.525e−05∗1000 = 1.025(1.004; 1.04), e2.054e−05∗1000 = 1.02(1.003; 1.03) and e1.833e−05∗1000 = 1.018(1.003; 1.03) respectively. Furthermore, the expected count of Rietgors birds increases by e2.890e−05∗1000 = 1.028(1.006; 1.05), e2.641e−05∗1000 = 1.026(1.005; 1.04), e2.126e−05∗1000 = 1.021(1.004; 1.03), e1.620e−05∗1000 = 1.016(1.002; 1.03) and e1.376e−05∗1000 = 1.013(1.001; 1.02) for an increase of 1000 m2 in sum of the area sizes of all plots within 100, 150, 200, 250 and 300 meters. 135

136

Intercept OPP VVG300

Intercept OPP VVG250

Intercept OPP VVG200

Intercept OPP VVG150

Intercept OPP VVG100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP VVG100 -3.531e+00 6.501e-01 <0.001* -6.806e-01 8.199e-01 0.406 4.083e-05 1.195e-05 <0.001* 4.203e-05 2.385e-05 0.078 OPP VVG150 -3.511e+00 6.522e-01 <0.001* -5.950e-01 8.270e-01 0.472 3.765e-05 1.122e-05 <0.001* 3.761e-05 2.248e-05 0.094 OPP VVG200 -3.051e+00 5.557e-01 <0.001* -4.165e-01 8.255e-01 0.614 2.376e-05 8.807e-06 0.006* 3.075e-05 1.936e-05 0.112 OPP VVG250 -3.073e+00 5.556e-01 <0.001* -6.885e-01 7.780e-01 0.376 2.024e-05 7.123e-06 0.004* 2.837e-05 1.508e-05 0.059 OPP VVG300 -3.462e+00 6.421e-01 <0.001* -1.152e+00 7.707e-01 0.135 2.326e-05 6.944e-06 <0.001* 3.227e-05 1.308e-05 0.014 -

GLMM Estimate SE P-value

Table 9.16: Results for Grauwe Gors based on the dataset from 2016 and 2017

137

Intercept OPP VVG300

Intercept OPP VVG250

Intercept OPP VVG200

Intercept OPP VVG150

Intercept OPP VVG100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP VVG100 -1.330e-01 2.782e-01 0.633 2.280e+00 3.778e-01 <0.001* -6.417e-07 8.205e-06 0.938 2.165e-05 1.110e-05 0.051 OPP VVG150 -6.306e-02 2.786e-01 0.821 2.323 0.379 <0.001* -3.464e-06 7.727e-06 0.654 0.00002 0.00001 0.069 OPP VVG200 2.840e-02 2.742e-01 0.918 2.558e+00 3.747e-01 <0.001* -6.416e-06 6.694e-06 0.338 9.621e-06 8.858e-06 0.277 OPP VVG250 1.116e-01 2.678e-01 0.677 2.810e+00 3.657e-01 <0.001* -8.269e-06 5.614e-06 0.141 1.022e-06 7.188e-06 0.887 OPP VVG300 -1.895e-02 2.696e-01 0.944 2.143e+00 3.604e-01 <0.001* -3.492e-06 4.769e-06 0.464 1.477e-05 6.235e-06 0.018 -

GLMM Estimate SE P-value

Table 9.17: Results for Geelgors based on the dataset from 2016 and 2017

138

Intercept OPP VVG300

Intercept OPP VVG250

Intercept OPP VVG200

Intercept OPP VVG150

Intercept OPP VVG100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP VVG100 -8.931e-01 2.996e-01 0.002* 1.208e+00 4.376e-01 0.006* 9.108e-06 8.414e-06 0.279 2.431e-05 1.281e-05 0.058 OPP VVG150 -9.070e-01 3.006e-01 0.002* 1.170e+00 4.364e-01 0.007* 9.017e-06 7.866e-06 0.252 2.328e-05 1.192e-05 0.051 OPP VVG200 -9.347e-01 2.962e-01 0.001* 1.233e+00 4.324e-01 0.004* 8.860e-06 6.647e-06 0.183 1.999e-05 1.017e-05 0.049 OPP VVG250 -9.868e-01 2.911e-01 <0.001* 1.196e+00 4.159e-01 0.004* 9.083e-06 5.388e-06 0.091 1.644e-05 8.116e-06 0.043 OPP VVG300 -8.817e-01 2.907e-01 0.002* 1.231e+00 4.250e-01 0.004* 5.282e-06 4.804e-06 0.272 1.459e-05 7.331e-06 0.047 -

GLMM Estimate SE P-value

Table 9.18: Results for Veldleeuwerik based on the dataset from 2016 and 2017

139

Intercept OPP VVG300

Intercept OPP VVG250

Intercept OPP VVG200

Intercept OPP VVG150

Intercept OPP VVG100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP VVG100 -1.249e+00 3.281e-01 <0.001* 4.580e-06 9.210e-06 0.619 OPP VVG150 -1.265e+00 3.295e-01 <0.001* 4.859e-06 8.587e-06 0.571 OPP VVG200 -1.510e+00 3.366e-01 <0.001* 1.201e-05 6.994e-06 0.086 OPP VVG250 -1.390e+00 3.218e-01 <0.001* 7.101e-06 5.674e-06 0.211 OPP VVG300 -1.428e+00 3.287e-01 <0.001* 7.061e-06 5.095e-06 0.166 -

GLMM Estimate SE P-value

Table 9.19: Results for Kneu based on the dataset from 2016 and 2017

140

Intercept OPP VVG300

Intercept OPP VVG250

Intercept OPP VVG200

Intercept OPP VVG150

Intercept OPP VVG100

Coefficient

Logistic regression Negative Binomial Estimate SE P-value Estimate SE P-value OPP VVG100 -7.684e-01 2.960e-01 0.009* 5.555e-01 2.986e-01 0.063 2.793e-05 9.782e-06 0.004* 2.890e-05 8.553e-06 <0.001* OPP VVG150 -7.814e-01 2.961e-01 0.008* 5.678e-01 2.989e-01 0.057 2.657e-05 9.113e-06 0.003* 2.641e-05 7.994e-06 <0.001* OPP VVG200 -8.194e-01 2.949e-01 0.005* 5.686e-01 2.871e-01 0.048 2.525e-05 8.236e-06 0.002* 2.126e-05 6.600e-06 0.001* OPP VVG250 -7.778e-01 2.846e-01 0.006* 6.485e-01 2.801e-01 0.021 2.054e-05 6.675e-06 0.002* 1.620e-05 5.352e-06 0.003* OPP VVG300 -8.013e-01 2.878e-01 0.005* 6.845e-01 2.861e-01 0.017 1.833e-05 5.795e-06 0.001* 1.376e-05 4.841e-06 0.005* -

GLMM Estimate SE P-value

Table 9.20: Results for Rietgors based on the dataset from 2016 and 2017

In conclusion, a positive effect of area size was observed for presence of Grauwe Gors, while a positive effect was detected on both presence and amount of Rietgors. For all of these, the impact is largest if area size is increased in small region (100m).

9.4.1

Summary of results

141

142

Variable Index BB Biological Evaluation Map 200 Index BB Biological Evaluation Map 500 Index BB Map Flanders 200 Index landscape Biological Evaluation Map 200 Index landscape Biological Evaluation Map 500

Grauwe Gors no effect no effect no effect no effect no effect

Geelgors no effect no effect no effect no effect no effect

Veldleeuwerik no effect no effect no effect no effect no effect

Kneu no effect no effect no effect no effect no effect

Table 9.21: Significance for clusters using Logistics regression per species using birds from 2012 only Rietgors no effect positive no effect no effect no effect

143

Grauwe Gors no effect no effect no effect no effect no effect

Geelgors no effect no effect no effect no effect no effect

Veldleeuwerik no effect no effect no effect no effect no effect

Kneu -

Rietgors no effect no effect no effect no effect no effect

Table 9.22: Significance for clusters using Negative Binomial regression per species using birds from 2012 only

144

Grauwe Gors no effect no effect no effect no effect no effect

Geelgors positive no effect no effect no effect no effect

Veldleeuwerik no effect no effect no effect no effect no effect

Kneu no effect no effect no effect no effect no effect

Rietgors no effect no effect no effect no effect no effect

Table 9.23: Significance for clusters using Logistic regression per species using birds from 2016 and 2017

145

Grauwe Gors positive

Geelgors no effect no effect no effect no effect positive

Veldleeuwerik positive no effect positive no effect no effect

Kneu -

Rietgors no effect no effect no effect no effect positive

Table 9.24: Significance for clusters using Negative Binomial regression per species using birds from 2016 and 2017

146

Grauwe Gors no effect no effect no effect no effect no effect Grauwe Gors no effect Grauwe Gors no effect no effect no effect no effect no effect

Variable VVG d100 VVG d150 VVG d200 VVG d250 VVG d300 Variable VVG d100 VVG d150 VVG d200 VVG d250 VVG d300 Variable VVG d100 VVG d150 VVG d200 VVG d250 VVG d300

Logistic regression Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect Negative Binomial Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect GLMM Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect Kneu no effect no effect no effect no effect no effect

Kneu -

Kneu no effect no effect no effect no effect no effect

Rietgors positive positive positive positive positive

Table 9.25: Significance for the number of plots under contract in 2017 per species using birds from 2016 and 2017

147

Grauwe Gors positive positive positive positive positive Grauwe Gors no effect no effect no effect no effect no effect Grauwe Gors -

Variable OPP d100 OPP d150 OPP d200 OPP d250 OPP d300 Variable OPP d100 OPP d150 OPP d200 OPP d250 OPP d300 Variable OPP d100 OPP d150 OPP d200 OPP d250 OPP d300

Kneu -

Kneu no effect no effect no effect no effect no effect

Rietgors -

Rietgors no effect no effect no effect no effect no effect

Rietgors positive positive positive positive positive

Table 9.26: Significance for the area size of the buffer of 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017

148

Grauwe Gors positive positive positive positive positive Grauwe Gors no effect no effect no effect no effect no effect Grauwe Gors positive positive positive positive no effect

Variable sumVG C100 sumVG C150 sumVG C200 sumVG C250 sumVG C300 Variable sumVG C100 sumVG C150 sumVG C200 sumVG C250 sumVG C300 Variable sumVG C100 sumVG C150 sumVG C200 sumVG C250 sumVG C300

Logistic regression Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect Negative Binomial Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect positive no effect positive GLMM Geelgors Veldleeuwerik no effect no effect no effect no effect no effect no effect no effect no effect no effect no effect Kneu no effect no effect no effect no effect no effect

Kneu -

Kneu no effect no effect no effect no effect no effect

Rietgors positive positive positive positive positive

Table 9.27: Significance for the number of plots within cluster of 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017

149

Grauwe Gors positive positive positive positive positive Grauwe Gors no effect no effect no effect no effect no effect Grauwe Gors -

Variable OPP VVG100 OPP VVG150 OPP VVG200 OPP VVG250 OPP VVG300 Variable OPP VVG100 OPP VVG150 OPP VVG200 OPP VVG250 OPP VVG300 Variable OPP VVG100 OPP VVG150 OPP VVG200 OPP VVG250 OPP VVG300

Kneu -

Kneu no effect no effect no effect no effect no effect

Rietgors -

Rietgors positive positive positive positive positive

Table 9.28: Significance for the sum of the area sizes of all plots within 100m, 150m, 200m, 250m, 300m per species using birds from 2016 and 2017

Part VII

Conclusions and Recommendations

150

The purpose of this study was to evaluate how the effectiveness of agroenvironmental agreements for farmland birds can be increased, how can the bird food crop measures (VVG vogelvoedselgewas) be used in a more targeted way, taking into account environmental factors as well as the development of conservation strategies on each of the five species, Grauwe Gors, Geelgors, Veldleeuwerik, Kneu and Rietgors. We analyzed the data based on a binary and a count outcome (presence/absence). Firstly, we used a logistic regression model for the binary outcome. However, this model does not take into account the repeated measurements on the same location throughout the study period. Thereby, we proceeded with a generalized linear mixed model, which by the use of a random effect for each plot, considered the clustered measurements of this dataset. Secondly, a negative binomial model was used for the count data, since a Poisson regression model was not appropriate due to overdispersion present in the data. The negative binomial model takes into account extra variability present in the data given by multiple visits on plots. It should be noted that the dataset includes a large amount of zeros for the responses of interest. As further research, a ZeroInflated Negative Binomial (ZINB) can be used. If the number of zeros is often much larger than a Poisson or negative binomial regression permit, a ZINB can capture much better this characteristic of the data. When looking at the analyses about the abundance of Grauwe Gors, a general trend towards occurrences in undisturbed areas can be seen. An effect of the contractual agro-environmental agreements could not be shown, therefore the bird food cultivation has no effect on the presence/absence or on the amount of this species. It is important to note that the analyses here are possibly underpowered, due to observations of this species being particularly small, which might result in a non-significant effect, when in reality, a significant effect should be seen. When investigating the results in depth, it can be seen that the logistic regression and negative binomial model found a positive significant effect of “hedge”. The variable “hedge 50 meters” had a positively influence on this species established by the logistic regression only. The “faunal grass stretch” has a negative impact on this species according to the logistic model, an effect which is not found by the other two models used. Furthermore, the negative binomial model detected a negative influence of the “nearest road” predictor on this species. The logistic regression model found a negative effect of the “nearest railroad”, while this is most probably not a relevant effect, due to the fact that most railroads are not within the vicinity of the plots. Thus, we conclude that this species prefers an undisturbed environment. The “number of plots within cluster of 100, 150, 200, 250 and 300 meters” has a positive effect according 151

to logistic regression on this species, therefore increasing the number of plots within these clusters leads to an increased possibility to find Grauwe Gors. Like mentioned before, it is important to mention that the dataset contained little information about this species. Thus, more data should be recorded about this species in order to make more accurate conclusions. For Geelgors, clear trends were seen. In general, one can say that the agro-environmental agreements have a highly positive effect on Geelgors. The presence of hedges have a positive effect on this species occurrence. When looking at it more closely, the variables which positively influenced the Geelgors species, results found by all models used, were “adjacency to hedge”, “hedge 50 meters”, “food availability of growth”, “Bird food crop 2 categories” and “Bird food crop 4 categories”. Therefore, we consider that the adjacency of a hedge to a plot as well as a hedge located at 50 meters away from a plot lead to an increase in the number of Geelgors birds. The Geelgors species prefers a plot where bird food is available as compared with a plot with no bird food. Moreover, this species benefits the most from a plot with 75% or more available food. At the same time, the grain without admixture of other crops (Tarwe/triticale, Tarwe, luzerne, Haver) and grain in combination with flax (Tarwe/vlas, haver/vlas) categories have a highly positive effect on Geelgors species. Furthermore, there was a positive effect found by the logistic model and GLMM of the predictor “unpaved road” on Geelgors species, but no effect was found by the negative binomial model. The adjacency of “ongoing traffic” to a plot has a negative effect on this species according to the negative binomial model only. A large positive effect was found by the negative binomial model of the variable “plot size” on Geelgors birds. Larger plots seem to attract more Geelgors birds. The “temperature” has a positive effect on this species found by GLMM. It is important to note that it is not sure whether “temperature” affects the occurrence and/or the detection probability of this species. In other words, a higher temperature may lead to an increased chance to detect Geelgors. For Veldleeuwerik, an effect of the “bird food crop” could not be seen, thus this species is not influenced by the presence/absence of “bird food crop” on the plots. Elements within the landscape influence this species’ occurrence. A strongly negative influence of the covariates “hedge”, “hedge 50 meters” and “fruit cultivation 50 meters” was found by all three models used in the situation of Veldleeuwerik species. In the same time, another negative effect of the variable “tree colonnade 50 meters” on this species was established by the logistic and negative binomial models. Therefore, since these landscape elements negatively impact the Veldleeuwerik species, open areas with few elements of the landscape are favored. The “temperature” 152

has a negative effect as well according to the negative binomial model, thus this species is being observed more often on days with lower temperatures. Kneu is a species that highly benefits from the “bird food crop” agreements. It is positively influenced by the variables “food availability of growth”, “Bird food crop 2 categories”, “Bird food crop 4 categories” according to all models used. The Kneu species seems to strongly benefit from food present on the plot, with the grain in combination of bladrammannas category of food being the most preferred. Moreover, 25% up to 75% of food available on the plot has a strong positive impact on this species. The adjacency of “faunal grass stretch” to a plot has a positive effect on Kneu species, as found by both logistic regression and GLMM. Under both transformations used, the predictor “Hellmann number” had a negative effect on this species according to the logistic regression and GLMM. Rietgors is a species that benefits greatly from the “bird food crop” strategies, more specifically when nearby plots form a greater aggregation of specifically cultivated fields. A strongly positive effect of the variables “food availability of growth”, “Bird food crop 2 categories”, “Bird food crop 4 categories”, “bird food cultivation 50 meters”, “plot size” and “temperature” found by all models used. Rietgors benefits more from the grain without admixture of other crops (Tarwe/triticale, Tarwe, luzerne, Haver) category of food as compared with the grain in combination with flax (Tarwe/vlas, haver/vlas). In addition, the presence of bird food cultivation at 50 meters away from a plot greatly increases the chance to see a Rietgors bird. A larger plot size attracts more birds. Moreover, this species favors an increased temperature on plots (this species is being observed more often on days with higher temperatures), although this might again be due to detection differences when temperatures change. The negative binomial model found a negative effect of the adjacency of a tree colonnade to a plot on this species. The distance to the nearest plot has a negative effect on Rietgors species, an effect found by the logistic and negative binomial model. An important positive effect of the predictors “number of plots under contract in 2017 within 100, 150, 200, 250 and 300 meters” and “number of plots within cluster of 100, 150, 200, 250 and 300 meters” were found on Rietgors species. Therefore, the conservation strategy for this species should focus more on cultivated plot locations that are closer to each other.

153

Bibliography Agresti, A. (2002). Categorical data analysis, Second Edition New York, Wiley. Molenberghs, G. and Verbeke, G. (2005). Models for Discrete Longitudinal Data. Springer Series in Statistics, Springer, New York.

154

Appendix

155

156

Blauwe Reiger

Heggenmus

Groenling

Fazant

Winterkoning

Blauwe Kiekendief

Buizerd

Torenvalk

Vink

Rietgors

Kneu

Veldleeuwerik

Geelgors

Grauwe Gors

Species

Cultivation variable crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop

Sampling frequency 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94

Number of times a bird is present on a plot 63 = 0.084 = 8.4%) 63 ( 743 2 2 ( 94 = 0.021 = 2.1%) 354 (47.6%) 23 (24.4%) 156 (20.9%) 13 (13.8%) 169 (22.7%) 1 (1%) 241 (32.4%) 8 (8.5%) 156 (20.9%) 13 (13.8%) 230 (30.9%) 12 (12.7%) 221 (29.7%) 9 (9.5%) 131 (17.6%) 7 (7.4%) 33 (4.4%) 1 (1%) 140 (18.8%) 7 (7.4%) 96 (12.9%) 1 (1%) 24 (3.2%) 0 (0%) 125 (16.8%) 1 (1%)

Table 9.29: Exploratory table according to the presence/absence of crop on plots for all species

157

Kramsvogel

Roodborst

Kievit

Graspieper

Velduil

Holenduif

Kauw

Ringmus

Keep

Koolmeens

Pimpelmees

Zwarte Kraai

Patrijs

Grote Bonte Specht

Species

Sampling frequency 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94

Number of times a bird is present on a plot 5 (0.6%) 1 (1%) 12 (1.6%) 1 (1%) 5 (0.6%) 5 (5.3%) 21 (2.8%) 1 (1%) 31 (4.1%) 6 (6.3%) 20 (2.7%) 0 (0%) 40 (5.4%) 0 (0%) 3 (0.4%) 0 (0%) 8 (1%) 0 (0%) 8 (1%) 0 (0%) 19 (2.5%) 0 (0%) 17 (2.2%) 3 (3.1%) 24 (3.2%) 1 (1%) 37 (4.9%) 3 (3.1%)

Table 9.30: Exploratory table according to the presence/absence of crop on plots for all species

158

Postduif

Gele Kwikstaart

Spreeuw

Koperwiek

Witte Kwikstaart

Roek

Putter

Merel

Grote Zilverreiger

Sperwer

Ruigpootbuizerd

Visarend

Houtduif

Huismus

Species

Sampling frequency 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94

Number of times a bird is present on a plot 5 (0.6%) 0 (0%) 32 (4.3%) 4 (4.2%) 2 (0.2%) 0 (0%) 6 (0.8%) 0 (0%) 13 (1.7%) 0 (0%) 15 (2%) 0 (0%) 46 (6.2%) 3 (3.2%) 3 (0.4%) 1 (1%) 4 (0.5%) 1 (1%) 2 (0.2%) 0 (0%) 2 (0.2%) 0 (0%) 5 (0.7%) 0 (0%) 5 (0.7%) 0 (0%) 4 (0.5%) 0 (0%)

Table 9.31: Exploratory table according to the presence/absence of crop on plots for all species

159

Gaai

Boomklever

Ekster

Smelleken

Distelvink

Boerenzwaluw

Aalscholver

Wilde Eend

Waterhoen

Groene Specht

Slechtvalk

Cirlgors

Havik

Kraai

Species

Sampling frequency 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94

Number of times a bird is present on a plot 6 (0.8%) 0 (0%) 2 (0.2%) 0 (0%) 1 (0.1%) 0 (0%) 2 (0.2%) 0 (0%) 2 (0.2%) 0 (0%) 1 (0.1%) 0 (0%) 2 (0.2%) 0 (0%) 3 (0.4%) 0 (0%) 1 (0.1%) 0 (0%) 1 (0.1%) 0 (0%) 1 (0.1%) 0 (0%) 1(0.1%) 1 (1%) 2 (0.2%) 0 (0%) 4 (0.5%) 0 (0%)

Table 9.32: Exploratory table according to the presence/absence of crop on plots for all species

160

Zanglijster

Watersnip

Grote Lijster

Boomkruiper

Tijftjaf

Kokmeeuw

Kraanvogel

Vlaamse Gaai

Canadese Gans onbekend

Grote Gele Kwikstaart

Staartmees

Species

Cultivation variable crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop crop no crop

Sampling frequency 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94 743 94

Number of times a bird is present on a plot 3 (0.4%) 0 (0%) 1 (0.1%) 0 (0%) 2 (0.2%) 0 (0%) 2 (0.2%) 0 (0%) 2 (0.2%) 0 (0%) 1 (0.1%) 0 (0%) 1 (0.1%) 0 (0%) 1 (0.1%) 0 (0%) 2 (0.2%) 0 (0%) 1 (0.1%) 0 (0%) 1 (0.1%) 0 (0%)

Table 9.33: Exploratory table according to the presence/absence of crop on plots for all species

161 Zwarte Kraai

Patrijs

Fazant

Blauwe Kiekendief

Buizerd

Torenvalk

Vink

Rietgors

Kneu

Veldleeuwerik

Geelgors

Grauwe Gors

Species

Coefficients Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop

Estimates -3.829 1.450 -1.127 1.033 -1.829 0.504 -4.533 3.310 -2.375 1.641 -1.661 0.647 -1.921 1.117 -2.245 1.384 -2.520 0.977 -2.520 1.058 -4.532 0.422 -2.879 -2.117

Standard Error 0.715 0.727 0.239 0.251 0.298 0.312 1.005 1.009 0.370 0.378 0.282 0.294 0.309 0.319 0.350 0.359 0.393 0.404 0.393 0.404 1.005 1.046 0.459 0.642

P-value <0.001* 0.046 <0.001* <0.001* <0.001* 0.106 <0.001* 0.001* <0.001* <0.001* <0.001* 0.027 <0.001* <0.001* <0.001* <0.001* <0.001* 0.016 <0.001* 0.008* <0.001* 0.687 <0.001* <0.001*

Table 9.34: Logistic regression results for presence/absence of crop on plots including all species

162 Ekster

Merel

Kievit

Pimpelmees

Grote Bonte Specht

Blauwe Reiger

Groenling

Winterkoning

Roek

Houtduif

Kramsvogel

Koolmees

Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop Intercept Crop

-2.685 -0.449 -3.412 0.462 -3.113 0.011 -4.532 -0.687 -4.533 1.462 -4.533 2.623 -4.533 2.933 -4.532 -0.463 -4.533 0.994 -3.412 -0.343 -3.412 0.693 -4.533 -2.078

0.422 0.460 0.586 0.610 0.511 0.542 1.005 1.123 1.005 1.021 1.005 1.011 1.005 1.010 1.005 1.101 1.005 1.029 0.586 0.636 0.586 0.606 1.005 1.418

<0.001* 0.328 <0.001* 0.449 <0.001* 0.984 <0.001* 0.54 <0.001* 0.152 <0.001* 0.009* <0.001* 0.003* <0.001* 0.674 <0.001* 0.334 <0.001* 0.589 <0.001* 0.253 <0.001* 0.143

Table 9.35: Logistic regression results for presence/absence of crop on plots including all species