摘要
针对如何克服不同传感器地表温度的时空分辨率矛盾的问题,该文利用随机森林(RF)算法、BP神经网络(BP)算法和多元回归(MLR)算法,直接将原始1km分辨率的MODIS LST降尺度至250m分辨率,并评估地理要素对不同降尺度算法的多维响应。结果表明:①在不同海拔、坡度、坡向和土地利用类型中,RF算法的降尺度效果均为最佳;②在降尺度模型中考虑经纬度、地形因子等地理要素能显著提升降尺度效果;③降尺度效果随海拔升高先增后减,随坡度增加先增后降,从不同坡向看,降尺度效果分异明显,从不同土地利用类型看,林地、草地的降尺度精度最高,耕地次之,水域和建筑用地降尺度精度最低。
In view of the problem how to overcome the contradictions between spatial and temporal resolution in land surface temperature(LST)derived from different satellite sensors,three downscaling algorithms namely random forest,BP neural network and multiple linear regression models were applied to downscale LST of MODIS images form 1km resolution to 250 mresolution respectively,and multidimensional response of geographic elements to different downscaling algorithms was evaluated.The main conclusions include:①RF algorithm outperform at all elevations,slopes,aspects and land use types;②Taking into account the geographical factors in downscaling model could significantly improve the downscaling accuracy,such as latitude and longitude,topographic factors and so on;③As the elevation and slope increased,downscaling accuracy first increased and then decreased.Downscaling result had different performance at different aspects and land use types.To be specific,downscaling results of three models in the west,the northwest,the north,the northeast,the southwest were superior to those in the east,the southeast and the south at aspects.In accuracy of downscaling,forest land was highest,which was followed by grassland,cultivated land,waters and building-up area in land use types.
作者
夏晓圣
王军红
程先富
XIA Xiaosheng;WANG Junhong;CHENG Xianfu(College of Geography and Tourism,Anhui Normal University,Wuhu,Anhui 241002,China;Anhui Key Laboratory of Natural Disaster Process and Prevention,Wuhu,Anhui 241002,China;Huoshan Soil and Water Conservation Experimental Station,Liu’an,Anhui 237266,China)
出处
《测绘科学》
CSCD
北大核心
2019年第12期134-140,共7页
Science of Surveying and Mapping
基金
国家自然科学基金项目（41271516）
安徽师范大学研究生科研创新与实践项目（2018kycx051）
关键词
地表温度
降尺度
地理要素
多维响应
MODIS
随机森林
land surface temperature
downscaling
geographical factors
multi-dimensional response
MODIS
random forest
