Relationships Between Maximum Air Temperature and Remotely Sensed Data Across Biomes of the São Francisco River Basin

We propose and validate three linear regression models, using maximum air temperature and retrievals of LST of the MYD21A1D data product (LSTMYD21A1D) as response and predictor variables, for the three major biomes in the São Francisco River Basin as defined by the Instituto Brasileiro de Geografia e Estatística (IBGE) in the Map of Biomes of Brazil (MBB), which is often used to support public policies. A dataset of 94,544 observations of ground-based measurements and retrievals of LST is used to build the models. The efficacy and performance of the models are tested by means of cross-validation and of benchmarking with the linear regression model for the entire basin, regardless the type of biomes. The three proposed linear regression models indicated that LSTMYD21A1D explains well the variations in air Tmax for all the types of biomes in the SFRB (R2: from 0.46 to 0.54). The results of the cross-validation revealed that the three models were equally good at providing estimates of air Tmax across the basin (RMSE < 2.66oC and MAE < 2.14oC), even though they were built with subsets of data sampled in a specific biome. The benchmark test indicated that the models, when applied to their specific biome, provide better estimates of air temperature than the regression model for the entire basin.