Estimating daily air temperature across the Southeastern United States using high-resolution satellite data: A statistical modeling study

Abstract

Accurate estimates of spatio-temporal resolved near-surface air temperature (Ta) are crucial for environmental epidemiological studies. However, values of Ta are conventionally obtained from weather stations, which have limited spatial coverage. Satellite surface temperature (Ts) measurements offer the possibility of local exposure estimates across large domains. The Southeastern United States has different climatic conditions, more small water bodies and wetlands, and greater humidity in contrast to other regions, which add to the challenge of modeling air temperature. In this study, we incorporated satellite Ts to estimate high resolution (1km×1km) daily Ta across the southeastern USA for 2000–2014. We calibrated Ts–Ta measurements using mixed linear models, land use, and separate slopes for each day. A high out-of-sample cross-validated R2 of 0.952 indicated excellent model performance. When satellite Ts were unavailable, linear regression on nearby monitors and spatio-temporal smoothing was used to estimate Ta. The daily Ta estimations were compared to the NASA's Modern-Era Retrospective Analysis for Research and Applications (MERRA) model. A good agreement with an R2 of 0.969 and a mean squared prediction error (RMSPE) of 1.376°C was achieved. Our results demonstrate that Ta can be reliably predicted using this Ts-based prediction model, even in a large geographical area with topography and weather patterns varying considerably.