| Land surface temperature(LST)is a result of surface energy balance and a key factor of regional and global surface physical processes.The natural environment of karst areas in southern China is complex and changeable,and most areas have been affected by unreasonable human factors for a long time,resulting in serious damage to the ecosystem,aggravated water and soil loss,serious land degradation,and large areas of exposed bedrock.Through the construction of highly heterogeneous landscape factors and downscale models to obtain high-resolution land surface temperature and explore the driving and differences of highly heterogeneous landscape factors on land surface temperature at different scales,the research basis is provided to clarify the water and heat balance of this special geomorphic area and improve ecological problems such as desertification,water and soil loss and resource imbalance in the region,so as to serve ecological restoration and the development of small agriculture.In this study,GF-5,GF-2,Sentinel-2,Landsat-8,MODIS LST,ALOS DEM and ASTER GDEM were used as data sources,and the upper reaches of Chishu River Basin and Bijie City were selected as the study area.Based on watershed scale and local scale,random forest model was used to study the driving of land surface temperature and high-resolution land surface temperature estimation of highly heterogeneous landscape.The four seasons diurnal LST of 30 m in the upper reaches of Chishui River Basin and 4m LST in two areas of Bijie City were obtained,and the driving factors and their contributions were obtained in the process of high-resolution land surface temperature estimation.The driving difference of different scale and height heterogeneous landscape factors on land surface temperature was discussed.The research main results of this paper are as follows:(1)In the watershed scale high-resolution land surface temperature estimation,the 30 m LST estimation results of four seasons show the same spatial distribution as the original 1km LST,which better preserves the original LST information.The spatial distribution of LST after scaling down has a more detailed expression,which significantly improves the spatial detail texture of the original LST.The LST error of the accuracy evaluation based on the measured sites in four seasons is less than 4K,the highest accuracy is 0.25 K in summer,less than 3K in autumn and winter,and 3.51 K in spring.The accuracy evaluation results based on upscaling showed that the LST RMSE of four seasons were all less than 1K,and the error order of four seasons was also summer < autumn < winter < spring,and the high-resolution estimation effect of four seasons at night was better than that of daytime.In summary,high-resolution land surface temperature estimates for four seasons in day and night show good results,which can provide reliable support for the subsequent analysis of driver differences.(2)In the estimation of local high-resolution land surface temperature,considering the small number of GF-2 bands,it is unable to provide all factors required for downscaling in karst areas.Therefore,Sentinel-2 data(10m)and ALOS DEM data(12.5m)are used to add scale factors with 10 m resolution.The downscaling model was constructed by secondary indirect downscaling method,and 4m high-resolution LST was estimated.The results show that the direct downscaling results of the two local study areas have different degrees of noise points,outliers,temperature dispersion and other problems,and the peak signal-to-noise ratio(PSNR)index is 22.756 and 27.584,respectively.Combined with the intermediate scale,more band data,terrain factors and high-resolution texture information are introduced to carry out the secondary indirect downscaling of 40m-10m-4m,which can better solve the disadvantages of direct downscaling.The PSNR index of the downscaling results in the two local study areas increased by 6.075 and 2.778 respectively.The local mutability of LST is shown in more detail,and the low temperature and high temperature map spots are captured more accurately,and the high-resolution land surface temperature estimation effect in karst areas with frequent relief is significantly improved.(3)The driving factors and contribution of land surface temperature at the catchment scale are different at night and day.During the daytime,the driving factors of the four seasons include building,bare land and topographic humidity.The contribution of bare land to land surface temperature is higher than that of building,and the driving effect is most obvious in spring(32%).The contribution of bare rock to LST in spring,summer and autumn is more than 3%.The contribution of topographic humidity to LST in spring,summer,autumn and winter was 7%,2.5%,3.8% and 17.4%,respectively.At night,DEM is the most important driving factor for LST in all seasons(contributing more than 37%).In addition,the water body is particularly driven in winter(contribution up to 14%).(4)The driving factors and contribution of land surface temperature at local scale are different in different relief regions.At the local scale,DEM,vegetation and bare rock were the main driving factors of LST,and DEM contributed more to LST in the study area with high average relief(29.5% and 10.7%,respectively).Vegetation contributed more than 10% to LST in the study area with small average relief.The contribution of bare rock to LST is above 20%in both study areas.(5)Driving differences of land surface temperature at different scales.At basin and local scale,highly heterogeneous landscape factors such as elevation,vegetation,bare ground(bare rock)and topographic humidity are important factors driving the spatial distribution of LST in karst areas.However,elevation and topographic humidity were more significantly driven at the catchment scale,with the average contribution of 49.8% and 5.8% at the catchment scale,but only 20.1% and 2.5% at the local scale.The driving effects of bare land and vegetation were more obvious at local scale,with the average contribution of 47.7% and 7.4%,respectively,and 20.33% and 4.52% at catchment scale. |
PDF Full Text Request