IMPROVEMENT OF TRADITIONAL AND HYBRID INTERPOLATION TECHNIQUES USING SUPPORT VECTOR MACHINE FOR LAND SURFACE TEMPERATURE ANALYSIS IN URBAN AREAS

Titipong PHOOPHATHONG , Teerawong LAOSUWAN , Satith SANGPRADID , Yannawut UTTARUK , Thinnakon ANGKAHAD

DOI: 10.21163/GT_2025.201.21

ABSTRACT: Interpolation techniques are highly effective numerical methods for achieving comprehensive spatial data coverage without the need to measure data at every location within the study area. Traditional interpolation methods, such as Inverse Distance Weighted (IDW) and Kriging, are numerical computations that rely on mathematical models without considering environmental influences on the spatial factors being interpolated. On the other hand, Support Vector Machines (SVM) are machine learning algorithms designed to enhance the accuracy of numerical computations. This research aims to improve and compare traditional interpolation techniques, specifically IDW, Ordinary Kriging (OK), and OK + SVM. The latter technique combines the OK interpolation concept with SVM learning to classify land cover and weight the interpolation of spatial data related to Land Surface Temperature (LST) in urban areas. The study revealed that the IDW technique produced values of T_max and T_min that were the closest to the actual measured values, followed by OK + SVM and OK, respectively. Furthermore, when assessing the interpolated data from 1,650 points extracted from LST using each technique against statistical tests such as MAE, MSE, and RMSE, it was found that OK + SVM provided better results than IDW and OK. Therefore, OK + SVM enhances interpolation accuracy by incorporating land cover classification, outperforming IDW and OK in LST estimation.

Keywords: Mathematical Method; Support Vector Machines; Inverse Distance Weighted; Ordinary Kriging; Land Surface Temperature; Geostatistic; Non-geostatistic; Digital Image Processing

Full article here