山地地貌地表温度的深度学习空间模拟  

Deep Learning Spatial Simulation of Land Surface Temperature in Mountainous Terrain

作　　者：鲍舒琪 张成福 冯霜 贺帅 苗林 BAO Shuqi;ZHANG Chengfu;FENG Shuang;HE Shuai;MIAO Lin(College of Desert Management,Inner Mongolia Agricultural University,Hohhot 010000,China)

机构地区：[1]内蒙古农业大学沙漠治理学院,呼和浩特010000

出　　处：《遥感信息》2023年第3期25-31,共7页Remote Sensing Information

基　　金：内蒙古自治区自然科学基金项目(2022MS03039)。

摘　　要：针对在山地地貌下环境因子与地表温度(land surface temperature,LST)之间存在的空间特征问题,提出利用深度学习方法分析LST在不同植被情景下与环境因子的关系特征。结合山地LST的影响因素和模型特点,构建大青山LST预测模拟模型,利用模型做LST与环境因子变化分析。结果表明:构建的LST深度学习模型预测值与观测值空间分布特征吻合度高(R 2为0.89,MAE为0.60℃,MSE为0.65℃);LST随NDVI、海拔和坡度的增加而降低,随平均气温和地表反照率的增加而增加;随NDVI的增大,LST随各环境因子变化的速率变化不同。研究表明,利用深度学习方法预测山地地貌LST的空间分布具有可行性,该方法有助于理解山地环境因子与LST的空间分布关系。To address the problem of spatial characteristics between environmental factors and land surface temperature(LST)in mountainous landscapes,we analyze the relationship characteristics between LST and environmental factors under different vegetation scenarios using deep learning methods.Combining the mountain factors of LST and model characteristics,we construct a predictive model of Daqing Mountain’s LST and use it to analysis changes between LST and environmental factors.The results show that the predicted values of the constructed LST deep learning model match well with the spatial distribution characteristics of the observed values(R 2 is 0.89,MAE is 0.60°C,and MSE is 0.65°C).The LST decreases with increasing NDVI,elevation and slope,and increases with increasing average temperature and surface albedo.The analysis of different scenarios shows that the rate of change of LST with each environmental factor varies with the increase of NDVI.It is feasible that using deep learning methods to predict the spatial distribution of mountain landscape LST,which helps understand the relationship between mountain environmental factors and the spatial distribution of LST.

关 键 词：深度学习 地表温度 山地地貌 归一化植被指数 内蒙古大青山 环境因子 

分 类 号：P338.9[天文地球—水文科学] TP79[水利工程—水文学及水资源]