JAUE2020-034: Downscale Air Temperature Based on the Random Forest Model: a Case Study of Central City in Kanto Major Metropolitan Area

Abstract

Air Temperature (AT) is a pivotal indicator in urban heat island (UHI) research. However, remote sensing data still has some shortcomings. For example, air temperature is difficult to obtain from remote sensing, and the air temperature obtained from Nation Land Numerical Information (NLNI) has a resolution with only 1 km. In order to solve the problem of low spatial resolution of temperature data, this research downscaled AT based on random forest model from 1 km to 250 m, based on AT data and Landsat 5 data. This research selected Kanto Central City in Kanto Major Metropolitan Area as study area. R² was introduced to evaluate the conversion accuracy of random forest model. Research indicates random forest model can be used for AT downscaling research, and there is an obvious heat island phenomenon in the study area.