SELECTION OF PREDICTOR VARIABLES IN DOWNSCALING LAND SURFACE TEMPERATURE USING RANDOM FOREST ALGORITHM

文献类型： 外文期刊

第一作者： Li, Wan

作者： Li, Wan;Wu, Hua;Liu, Qingsheng;Li, Wan;Wu, Hua;Liu, Qingsheng;Duan, Si-Bo;Li, Zhao-Liang

作者机构：

关键词： Land surface temperature; downscaling; variable selection

期刊名称：2019 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2019)

ISSN： 2153-6996

年卷期： 2019 年

页码：

收录情况： SCI

摘要： In this work, land surface temperature (LST) was downscaled by statistical regression model based on the nonlinear relationship with environment variables, including land surface reflectance, spectral indices, terrain factors, land cover type, reanalysis data and geolocation information. The correlation between predictor variables and LST was examined and compared with each other, in which 16 variables were finally selected into model, the variable dataset was credited to have relatively best performance with the trade-off between algorithm accuracy and computational complexity. With the optimal variable dataset, the LST of Moderate Resolution Imaging Spectroradiometer (MODIS) was downscaled from 990m to 90m by using random forest (RF) regression algorithm. Results of visual and quantitative analysis showed the satisfied downscaling results on 13 May, 2017 in Qinyang City, with the bias, coefficient of determination (R-2) and root mean square error (RMSE) of -0.02, 0.9 and 2.18 K, respectively. Comparison with the algorithm for sharpening thermal imagery (TsHARP) also demonstrated the accuracy and robustness of RF model with selected variable dataset.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序