Model development for the estimation of urban air temperature based on surface temperature and NDVI

Predictive models for urban air temperature (T air ) were developed by using urban land surface temperature (LST) retrieved from Landsat-8 and MODIS data, NDVI retrieved from Landsat-8 data and T air measured by 24 climatological stations in Szeged. The investigation focused on summer period (June−September) during 2016−2019 in Szeged. The relationship between T air and LST was analyzed by calculating Pearson correlation coefficient, root-mean-square error and mean-absolute error using the data of 2017−2019, then unary (LST) and binary (LST and NDVI) linear regression models were developed for estimating T air . The data in 2016 were used to validate the accuracy of the models. Correlation analysis indicated that there were strong correlations during the nighttime and relatively weaker ones during the daytime. The errors between T air and LST MODIS-Night was the smallest, followed by LST MODIS-Day and LST Landsat-8 respectively. The validation results showed that all models could perform well, especially during nighttime with an error of less than 1.5 o . However, the addition of NDVI into the linear regression models did not significantly improve the accuracy of the models, and even had a negative effect. Finally, the influencing factors and temporal and spatial variability of the correlation between T air and LST were analyzed. LST Landsat-8 had a larger original error with T air , but the regression model based on Landsat-8 had a stronger ability to reduce errors.