# Building the Inundation Model calgary % select(-geometry) %>% mutate(slope_cat = as.factor(slope_cat)) set.seed(3456) #random set of numbers to replicate sample trainIndex <- createDataPartition(calgary$slope_cat, p = .70, list = FALSE, times = 1) #include all categorical variables as first arguments calgaryTrain <- calgary[ trainIndex,] calgaryTest <- calgary[-trainIndex,] 2 med calgaryModel <- glm(inundation ~ ., family=“binomial”(link=“logit”), data = calgaryTrain %>% as.data.frame() %>% select(-geometry)) # Testing the Model classProbs <- predict(calgaryModel, calgaryTest, type=”response”) testProbs <- data.frame(obs = as.numeric(calgaryTest$inundation), pred = classProbs) high 3 1 distance to water bodies 0 km flow accumulation low testProbs$predClass = ifelse(testProbs$pred > .25,1,0) caret::confusionMatrix(reference = as.factor(testProbs$obs), data = as.factor(testProbs$predClass), positive = “1”) # Training the Model auc(testProbs$obs, testProbs$pred) ANTICIPATING THE NEXT FLOOD predictive modeling academic - 2022 UPenn, Weitzman School of Design project scale: S M L XL group project skills: ArcGIS, R Studio ctrl <- trainControl(method = “cv”, number = 100, savePredictions = TRUE) cvFit <- train(as.factor(inundation) ~ ., data = calgary %>% as.data.frame() %>% select(-geometry, -slope_cat), #remove categorical variables method=”glm”, family=”binomial”, trControl = ctrl) cvFit allPredictions % mutate(allPredictions = round(allPredictions * 100)) #Plot Map with Inundation Predictions for Calgary calgary %>% ggplot()+ geom_sf(aes(fill = allPredictions), color = “transparent”, apha = 0.75)+ geom_sf(data=calgary_boundary, fill = “transparent”, 10 6 7.5 4 5 2 2.5 0 km distance to steep slopes distance to water bodies 0 km With climate change intensifying, the average 100-year floodplain in Canadian river cities is projected to expand by 45% by 2100. In 2013, Calgary experienced its worst flood since 1932, primarily due to heavy rainfall. For this project, we developed a binomial logistic regression model using data from the 2013 Calgary flood to predict inundation in comparable river cities. We tested the model’s accuracy using Edmonton as a case study. To build the model, we used ArcGIS to convert raster data into fishnet grid cells and included six variables: (1) distance to all water bodies, (2) distance to stream network, (3) flow accumulation per cell, (4) distance to steep slopes, (5) slope classification (flat to very steep), and (6) percent impervious surface. The four most statistically significant features are shown to the right. We trained and validated the model in R, comparing predicted inundation with actual 2013 flood data. The model performed strongly, with an ROC curve area close to 1 and prediction scores clustering around 0.9—indicating high accuracy. View the full code on my GitHub. 100% 100% false negative false positive true negative true positive 50 ML Portfolio 2025 Methods & Tangents inundation prediction for Calgary 0% inundation prediction for Edmonton 0% 51 the Next Flood Anticipating