Imputing Satellite-Derived Aerosol Optical Depth Using a Multi-Resolution Spatial Model and Random Forest for PM2.5 Prediction

A task for environmental health research is to produce complete pollution exposure maps despite limited monitoring data. Satellite-derived aerosol optical depth (AOD) frequently used as a predictor in various models improve PM2.5 estimation, significant gaps coverage. We analyze and AOD from July 2011 the contiguous United States. examine two methods aid gap-filling AOD: (1) lattice kriging, spatial statistical method adapted handle large amounts data, (2) random forest, tree-based machine learning method. First, we evaluate each model’s performance prediction of AOD, additionally consider ensemble combining predictors. In order accurately assess predictive these methods, construct spatially clustered holdouts mimic observed patterns missing Finally, whether through one proposed can forest model. Our results suggest that kriging gap-filling. Based on summary metrics performance, predictions based were largely similar regardless inclusion gap-filled but there was some variability daily model predictions.