Predicting spatiotemporal mean air temperature using MODIS satellite surface temperature measurements across the Northeastern USA

Air temperature (Ta) stations have limited spatial coverage, particularly in rural areas. Since temperature can vary greatly both spatially and temporally, Ta stations are often inadequate for studies on the health effects of extreme temperature and climate change. Satellites can provide us with daily physical surface temperature (Ts) measurements, enabling us to estimate daily Ta. In this study, we aimed to extend our previous work on predicting Ta from Ts in Massachusetts by predicting 24 h Ta means on a 1 km grid across the Northeast and Mid-Atlantic states, extending both the temporal and spatial coverage, improving upon the methodology and validating our model in other geographical regions across the Northeastern part of the USA. We used mixed model regressions to first calibrate Ts and Ta measurements, regressing Ta measurements against day-specific random intercepts, and fixed and random Ts slopes. Then to capture the ability of neighboring cells to fill in the cells with missing Ts values, we regress the Ta predicted from the first mixed effects model against the mean of the Ta measurements on that day, separately for each grid cell. Out-of-sample tenfold cross-validation was used to quantify the accuracy of our predictions. Our model performance was excellent for both days with available Ts and days without Ts observations (mean out-of-sample R-2 = 0.95 and R-2 = 0.94 respectively). We demonstrate how Ts can be used reliably to predict daily Ta at high resolution in large geographical areas even in non-retrieval days while reducing exposure measurement error. (C) 2014 Elsevier Inc. All rights reserved.