Estimation of daily maximum and minimum air temperatures in urban landscapes using MODIS time series satellite data

Urban air temperature is considered a significant variable for a variety of urban issues, and analyzing the spatial patterns of air temperature is important for urban planning and management. However, insufficient weather stations limit accurate spatial representation of temperature within a heterogeneous city. This study used a random forest machine learning approach to estimate daily maximum and minimum air temperatures (T_max and T_min) for two megacities with different climate characteristics: Los Angeles, USA, and Seoul, South Korea. This study used eight time-series land surface temperature (LST) data from Moderate Resolution Imaging Spectroradiometer (MODIS), with seven auxiliary variables: elevation, solar radiation, normalized difference vegetation index, latitude, longitude, aspect, and the percentage of impervious area. We found different relationships between the eight time-series LSTs with T_max/T_min for the two cities, and designed eight schemes with different input LST variables. The schemes were evaluated using the coefficient of determination (R²) and Root Mean Square Error (RMSE) from 10-fold cross-validation. The best schemes produced R² of 0.850 and 0.777 and RMSE of 1.7 °C and 1.2 °C for T_max and T_min in Los Angeles, and R² of 0.728 and 0.767 and RMSE of 1.1 °C and 1.2 °C for T_max and T_min in Seoul, respectively. LSTs obtained the day before were crucial for estimating daily urban air temperature. Estimated air temperature patterns showed that T_max was highly dependent on the geographic factors (e.g., sea breeze, mountains) of the two cities, while T_min showed marginally distinct temperature differences between built-up and vegetated areas in the two cities.