Developing machine learning models for air temperature estimation using MODIS data

Air temperature is a key variable in wide range of environmental applications, including land–atmosphere interaction, climate change research and hydrology crop growth models, among others. The objective this study was to estimate daily maximum (Tmax) minimum (Tmin) temperatures, based on MODIS AQUA/TERRA land surface (LST), NDVI, extraterrestrial solar radiation precipitation data. Artificial neural networks (ANN) random forests (RF) models were developed predict these temperatures covering weather stations Córdoba (Argentina) for 2018-2020. results show that RF ANN machine learning algorithms are capable modeling non-linear relationships between registered LST data, very robust way. validation the confirms Tmax Tmin can be accurately estimated using, jointly or separately, AQUA TERRA LST. best present determination coefficients equal 0.81/0.91 root mean square error 2.7/2.1 ºC Tmax/Tmin, when using day/night satellite overpass time respectively. robustness confidence developed, ease free accessibility input data at global scale, suggest methodologies have potential applied other regions.