Downscaling the Midsummer Temperature-Humidity Index Based on Multiple Machine Learning Methods

Abstract

To improve the finesse of the temperature-humidity index (THI), this study applies four machine learning methods in THI downscaling, including multiple linear regression, random forest (RF), support vector machine, and gradient boosting machine. The temperature data and specific humidity data of the China Meteorological Administration Land Data Assimilation System (CLDAS) are used to establish a downscaling model, and site observational data are used to test the model precision. By taking land surface temperature (LST), vegetation coverage, altitude, and slope as downscaling factors, the monthly average THI calculated by CLDAS-V2.0 data is downscaled from 6 to 1 km in Anhui Province in July and August from 2002 to 2021. The results show that the spatial resolution of THI is improved effectively by the four downscaling models, and there is a significant correlation between the downscaled values and the site values, with a correlation coefficient greater than 0.97. The downscaling effect of RF is slightly better than that of the other three algorithms and better describes the distribution of summer resort resources. Simulated results from RF are piecewise corrected by using the mean variation, and the correlation between corrected values and observations in July and August are both improved (>0.98). According to the estimation of the corrected THI (1 km × 1 km), the proportion of summer resort area in Anhui Province is 9.58% in July and 19.29% in August.

Keywords

• China meteorological administration land data assimilation system (CLDAS)
• downscale
• machine learning
• moderate-resolution imaging spectroradiometer (MODIS)
• temperature-humidity index (THI)