Improvement of spatial interpolation accuracy of daily maximum air temperature in urban areas using a stacking ensemble technique

Abstract

The reliable and robust monitoring of air temperature distribution is essential for urban thermal environmental analysis. In this study, a stacking ensemble model consisting of multi-linear regression (MLR), support vector regression (SVR), and random forest (RF) optimized by the SVR is proposed to interpolate the daily maximum air temperature (Tmax) during summertime in a mega urban area. A total of 10 geographic variables, including the clear-sky averaged land surface temperature and the normalized difference vegetation index, were used as input variables. The stacking model was compared to Cokriging, three individual data-driven methods, and a simple average ensemble model, all through leave-one-station-out cross validation. The stacking model showed the best performance by improving the generalizability of the individual models and mitigating the sensitivity to the extreme daily Tmax. This study demonstrates that the stacking ensemble method can improve the accuracy of spatial interpolation of environmental variables in various research fields.

Keywords

• spatial interpolation
• cokriging
• multi-linear regression
• support vector regression
• random forest
• simple average ensemble
• stacking ensemble