Predicting Sea Breeze on Long Island based on attributes of prior nights using D-Basis on data from 2017-2019 Summers Wenxin Liu: Computer Science & Mathematics Undergraduate Dr. Kira Adaricheva: Mathematics Dr. Jase Bernhardt: Geology, Environment and Sustainability Dr. Oren Segal: Computer Science Briana Schmidt: Graduated Hofstra 2020 Justin Cabot-Miller: Graduated Hofstra 2020 Hofstra University
November, 2020 1 / 15
Sea Breeze
What is a sea breeze?
weather.gov/jetstream/seabreeze
Our definition for Sea breeze on Long Island 1. Highest temperature happens between sunrise and 3:30pm (subject to change per location) 2. During the two hours after the highest temperature, the wind is onshore (south in our case) for at least 75% of the time * Happens in absence of larger scale weather events * Happens more often in Summer when the temperature difference between land and sea is larger 2 / 15
Sea Breeze
Why study the sea breeze?
Reasons Aviation
Air Pollution
Energy
Weather prediction
Surplus of data
Storm development
Population on coast
Temperature
Insect protection
Heat related mortality
3 / 15
Data
Where did the data come from?
Hofstra Soccer Stadium https://nassau-ny.weatherstem.com/hofstra 4 / 15
Data
What does the raw data look like?
Hofstra data https://nassau-ny.weatherstem.com/hofstra 5 / 15
Data
Data cleaning: Make 5 minute average data
5 min data Credited to Briana Schmidt 6 / 15
D-Basis Introduction
What to do with the 5 minute data?
Day
Pressure +
1 2 3 ...
0 0 1 ...
Hours 19-21 Wind N Dew point +
1 1 0 ...
1 0 1 ...
... ... ... ... ...
... ... ... ... ... ...
SB
1 0 0 ...
Target Normal Stormy
0 1 0 ...
0 0 1 ...
Actual binary input table to D-Basis Credited to Briana Schmidt 7 / 15
D-Basis Introduction
What to do now with the cleaned data? D-Basis! Procedure: 1. Run D-Basis on the 2017-2019 data 2. Analyze results from D-Basis 3. Come up with a hypothesis for prediction method 4. Test our prediction method on 2020 data What is D-basis? A relatively new method, after Apriori Algorithm, to find (strong) association rules and implications for denser binary data Takes in binary data as input Was previously used for medical studies for papers on ovarian cancer in 2015 & on stomach cancer in 2020 Just went public at https://gitlab.com/npar/dbasis 8 / 15
D-Basis Introduction
Support of an Attribute or Association Rule Transaction 1 2 3 4 5
Milk X X X
Eggs X X X X
Butter X X X
Support The set of rows where an attribute (or a set of attributes) appears. Confidence of Association Rule Y → Z |support(Y ∪ Z )| |support(Y )| |support({E,B,M})| 2 |support({E,B})| = 2 = 1 |support({E,M})| = 43 = 0.75 |support(E)|
confidence({E, B} → M) = confidence(E → M) =
9 / 15
D-Basis Introduction
How do we choose the attributes? Relevance! Using implication: Y → b Using basis (all implications in the input): β Total support tsupb (a) =
X |support(Y )| |Y |
: a ∈ Y , (Y → b) ∈ β
Relevance relb (a) =
tsupb (a) tsupÂŹb (a) + 1
10 / 15
Use D-Basis for Sea Breeze Data
Using the D-Basis Algorithm for the Sea Breeze data
Results of attributes with the highest relevance During the night (7pm - 7am) before the predicted day: South wind -> later refined to south wind during at least 60% of the time Rising pressure Rising or constant dew point Method to predict For the three parameters above, if more than two parameters fail, we predict the day would not have sea breeze.
11 / 15
Result
Result of prediction for 2020 summer Days 1 2 3 4 5 ...
Wind 0 0 0 0 1 ...
Pressure Dew Point Prediction SB or not 1 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 1 1 1 1 ... ... ... ... predicted SB predicted Non-SB 20 SB 17 3 11 non-SB 5 6 17 + 6 23 = = 74.2% 31 31 5 False positive days 3 False negative days 23 correct predicted days
Success rate =
12 / 15
Result
Future work to be done Further adjust wind direction coordinate system based on the coastline orientation when determining onshore & offshore for the binary table rather than appointing straight south to be the onshore wind. (the determination of onshore & offshore for 2020 data was based on the 20° coastline of Long Island ) Inspired by Frysinger and Lindner (2001)
Adjust the criteria to determine parameters increasing, constant, or dropping to a more relative approach; the determination is done solely based on numerical difference at the moment. Implement a better method to average wind directions for generating the 5 minute data Run D-Basis again with all data of the four years with adjustments above 13 / 15
References
References Adaricheva K. et al. Measuring the Implications of the D-Basis in Analysis of Data in Biomedical Studies. In: Baixeries J., Sacarea C., Ojeda-Aciego M. (eds) Formal Concept Analysis. ICFCA 2015. Lecture Notes in Computer Science, Springer, vol 9113, 39-57, 2015 Adaricheva K., Nation J.B., Discovery of the D-Basis in Binary Tables Based on Hypergraph Dualization, Theoretical Computer Science, 2017 Frysinger J., Lindner L., A Statistical Sea-Breeze Prediction Algorithm for Charleston, South Carolina, Weather and Forecasting vol 18 pp.614-625, 2003 Miller S.T.K., Keim B.D., Talbot R.W., Mao H., Sea Breeze: Structure, Forecasting, and Impacts, American Geophysical Union, Reviews of Geophysics, 2003 Nation J.B., Cabot-Miller J., Segal O., Lucito R., Adaricheva K., Combining Algorithms to Find Signatures that Predict Risk in Early Stage Stomach Cancer, submitted to Journal of Computational Biology, Nov. 2020, 1-32 Segal O., Cabot Miller J., Adaricheva K., Nation J.B., The D-basis algorithm for association rules of high confidence, IT in Industry, v.6 N3, 2018. Simpson J.E.,Sea Breeze and Local Wind, Cambridge University Press, 1994
14 / 15
References
Pictures
Kotamarthi R., Mearns L., Hayhoe K., Castro C.L., Wuebbles D., Use of Climate Information for Decision-Making and Impacts Research: State of our Understanding, Prepared for the Department of Defense, Strategic Environmental Research and Development Program, 2016 weather.gov/jetstream/seabreeze https://nassau-ny.weatherstem.com/hofstra https://en-gb.topographic-map.com/maps/islb/Long-Island/
15 / 15