Model development for the estimation of seasonal urban air temperature patterns using MODIS satellite data, a case study in Szeged

Abstract

<p>Urban air temperature (T<sub>air</sub>) is an essential variable for a variety of urban issues, and analyzing the spatial patterns of T<sub>air</sub> is of great importance for urban planning and management. However, it is difficult to obtain T<sub>air</sub> data with a high spatial resolution because of the absence of weather stations within a heterogeneous city. In this research, T<sub>air</sub> predictive models were developed in Szeged by the help of an urban climate monitoring system which is a part of COST FAIRNESS project. Our investigation focused on the whole year instead of just the summer which has been the most studied season for urban heat issues. Two statistical mthods, multiple linear regression and random forest regression, were used for developing T<sub>air </sub>estimation models. T<sub>air</sub> data obtained from in situ meteorological stations from 2014 to 2017 were used as the dependent variable for models training. Land surface temperature data from numerous MODIS satellite images and 7 auxiliary variables were used as the independent variables. The auxiliary data are open source, including local climate zone (LCZ) lassification data containing a wide range of urban surface information and atmospheric parameters in the urban boundary layer associated with near-surface urban climate. The atmospheric parameters calculated from the ERA5 reanalysis data. We calculated RMSE based on 10-fold cross-validation to valadite the models. The results indicated that the random forest models performed better than multiple linear models with lower RMSE in four seasons. According to the importance analysis in random forest, both LCZ classification and atmospheric parameters are effective in reducing model errors. LCZ parameters affect the models significantly during the day, while atmospheric parameters affects the models significantly at night. Both effects of these two auxiliary variables showed their maximum in summer. In the end of this research, we used the final models to estimate T<sub>air</sub> and mapped the seasonal average T<sub>air</sub> patterns from 2018 to 2019. Our overall aim is to develop a generalized methodology using globally available land surface temperature and auxiliary data capable to estimate the urban air temperature.</p>