Modeling PFAS contamination in private wells

Introduction

Source code for the paper: A Statistical Approach for Identifying Private Wells Susceptible to Perfluoroalkyl Substances (PFAS)Contamination. Environmental Science & Technology Letters Article ASAP DOI: 10.1021/acs.estlett.1c00264

Authors

• Xindi C. Hu
• Beverly Ge
• Bridger Ruyle
• Jennifer Sun
• Elsie M. Sunderland

Order and purpose of scripts (output given in parentheses)

All can be found in the /scripts folder

Data cleaning and processing

• read_csv.R : read raw datasets with slight formatting/removal of extreme values (modeling_data/PFASwells1.rds)
• recode.R : recode detects/results, add binary variable, merge two main data files (modeling_data/PFASwells.rds)
• soildata.R : process soil data (modeling_data/final_soildata.rds)
• point_sources_ucmr.R : business extraction (modeling_data/final_industries.rds)
• impact.py : quantify and aggregate impact from industries (modeling_data/potential_impact_huc12.csv)
• finalize_iv.R: merge all independent variables into single dataset (modeling_data/merged_variables.rds, modeling_data/unique_ivs.rds)
• finalize.R : finalize data for 5 compounds and sumPFAS as separate dataframes, contained in a list (modeling_data/compounds_data.rds)

Modeling

• logreg.R : logistic regression (models/compounds_logreg.rds, models/compounds_glm.rds)
• eval_logreg.R: evaluate logistic regression models
• trees.R : random forest (models/compounds_forest.rds)
• eval_forest.R: evaluate random forests

Necessary packages

• tidyverse
• raster
• rgdal
• maptools
• proj4
• rgeos
• corrplot
• InformationValue
• caret
• MASS
• tree
• randomForest
• reshape2
• ggthemes
• tmap
• sf
• viridis
• broom
• stargazer
• rlang
• docxtools
• flextable
• kabelExtra
• geojsonR
• geojsonsf
• pROC