基于XGBoost的LST空间降尺度方法  

A Spatial Downscaling Method of Land Surface Temperature Based on XGBoost Algorithm

作　　者：颜佳楠 陈虹 姚光林 吴骅[1] YAN Jianan;CHEN Hong;YAO Guanglin;WU Hua(State Key Laboratory of Resources and Environment Information System,Institute of Geographic Sciences and Resources Research,Chinese Academy of Sciences,Beijing 100101,China;China Aero Geophysical Survey & Remote Sensing Center for Nature Resources,Beijing 100083,China;The 8th Institute of Geology and Mineral Exploration of Shandong Province,Rizhao 276826,China)

机构地区：[1]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101 [2]中国自然资源航空物探遥感中心,北京100083 [3]山东省第八地质矿产勘查院,山东日照276826

出　　处：《无线电工程》2021年第12期1508-1516,共9页Radio Engineering

基　　金：国家自然科学基金资助项目(41901308);国家重点研究发展计划(2018YFB0504800)。

摘　　要：地表温度(Land Surface Temperature,LST)是热环境监测中的重要参数之一。针对LST降尺度,开展了极端梯度提升树(Extreme Gradient Boost,XGBoost)用于LST的空间降尺度研究,对比分析了地表反射率、光谱指数、地形因子、大气再分析数据、经纬度以及地表覆盖类型6种回归核在不同组合形式条件下将1 km分辨率的MODISLST产品降尺度至100 m分辨率的LST的模型性能。研究结果表明,选取group2(光谱指数)、group4(光谱指数、地形)、group7(光谱指数、经纬度)作为回归核时,模型表现能力最好,此时均方根误差在2 K左右,且能保持清晰的图像纹理。光谱指数对于LST降尺度的作用较为重要。单个局部研究区数据训练的降尺度模型泛化能力还是不够,为保证降尺度图像的准确性和相似性,XGBoost模型在训练过程中仍需选择更具代表性的训练数据。Land Surface Temperature(LST)is one of the important parameters in thermal environment monitoring.The downscaling method of land surface temperature by using the Extreme Gradient Boosting Tree(XGBoost)model is studied.The performance of the LST model for downscaling MODISLST product of 1 km resolution to 100 m resolution is compared and analyzed under different combinations of six regression kernels including surface reflectance,spectral index,topographic factors,atmospheric reanalysis data,longitude and latitude information,and surface coverage types.The results show that when group2(spectral index),group4(spectral index,terrain)and group7(spectral index,longitude and latitude information)are selected as the regression kernel,the performance of the model is the best,the RMSE is about 2 K,and clear texture can be maintained.The spectral index plays an important role in the downscaling of surface temperature.However,the generalization capability of downscaling model trained by local data itself is still insufficient.To ensure the accuracy and similarity of the downscaled images,training data with higher representativeness should be selected for the XGBoost model in the training process.

关 键 词：地表温度 降尺度 机器学习 XGBoost 

分 类 号：P407.6[天文地球—大气科学及气象学]