| 喀斯特地区地表温度空间降尺度方法初探 |
| |
| 引用本文： | 尹枷愿,蔡宏,田鹏举,唐敏.喀斯特地区地表温度空间降尺度方法初探[J].地理与地理信息科学,2021,37(2):38-46,99. |
| |
| 作者姓名： | 尹枷愿 蔡宏 田鹏举 唐敏 |
| |
| 作者单位： | 贵州大学矿业学院,贵州 贵阳550025;贵州省生态气象和卫星遥感中心,贵州 贵阳550025 |
| |
| 摘 要： | 喀斯特地区地形起伏大,常规的降尺度方法及所选择的因子对其不适用。该文根据喀斯特地区的特点,选取反射率、遥感指数及高程因子为尺度因子,通过随机森林模型建立MODIS第31、32波段辐射亮度数据和尺度因子之间的非线性关系,构建适合喀斯特地区的随机森林(Karst Random Forest,KRF)模型,成功将空间分辨率为1 km的热红外辐射亮度降至100 m,最后利用劈窗算法反演得到空间分辨率为100 m的地表温度。将KRF方法与仅考虑常规因子的多因子随机森林回归模型(MTVRF)和热锐化算法(TsHARP)对比,结果表明:1)在不同高差的喀斯特地区,KRF方法可较大程度提高地表温度降尺度精度,均方根误差(RMSE)在遵义市西北部和贵阳市以南地区分别为2.46 K和1.45 K,较MTVRF模型分别降低了0.1419 K和0.2928 K,较TsHARP算法分别降低了0.6204 K和0.6953 K,且在地形起伏度较低的喀斯特山区城市(贵阳市以南)表现更好;2)在喀斯特地区不同地类上,KRF方法效果也较好,其中植被区域最优,RMSE为1.41 K,破碎的裸土区域RMSE为1.84 K。研究显示,考虑特殊尺度因子的KRF方法可提高喀斯特地区地表温度的降尺度精度,为该地区以地表温度为基础的研究提供更精细可靠的地表温度产品。
|
| 关 键 词： | 地表温度 热红外辐射 降尺度 随机森林 喀斯特地区 |
Spatial Downscaling Research of the Land Surface Temperature in Karst Region |
| |
| YIN Jia-yuan,CAI Hong,TIAN Peng-ju,TANG Min.Spatial Downscaling Research of the Land Surface Temperature in Karst Region[J].Geography and Geo-Information Science,2021,37(2):38-46,99. |
| |
| Authors: | YIN Jia-yuan CAI Hong TIAN Peng-ju TANG Min |
| |
| Institution: | (Mining College of Guizhou University,Guiyang 550025;Guizhou Ecological Meteorology and Satellite Remote Sensing Center,Guiyang 550025,China) |
| |
| Abstract: | Due to the large topographic relief in the Karst region,conventional land surface temperature(LST)downscaling methods and choice of scale factors are not applicable.Considering the surface characteristics of the Karst region,we have proposed the Karst random forest regression model(KRF),utilizing surface reflectance,spectral indices and terrain factors as scale factors.The nonlinear relationship between MODIS thermal infrared radiance bands(band 31 and band 32)and scale factors was established based on random forest model.And the spatial resolution of thermal infrared radiances was successfully downscaled from 1 km to 100 m.Finally,the split window algorithm was used to invert the thermal infrared radiance into the surface temperature with spatial resolution of 100 m.A comparison was also made with two other conventional downscaling methods,namely the random forest regression model with multitype predictor variables(MTVRF)and the thermal sharpening algorithm(TsHARP).The results are summarized as follows.1)For different altitude difference in the Karst region,the KRF method can greatly improve the precision of LST downscaling and the root mean square errors(RMSE)in two regions(northwest of Zunyi City and south of Guiyang City)were reduced to 2.46 K and 1.45 K,respectively.Compared with the MTVRF method these values decreased by 0.1419 K and 0.2928 K and compared with the TsHARP method these values decreased by 0.6204 K and 0.6953 K,respectively.In addition,the KRF method showed better results for the Karst mountain city areas with relatively low topographic relief(south of Guiyang City).2)For different land types in the Karst region,the KRF method also achieved good results,where it performed the best for the vegetation areas with a RMSE value of 1.41 K,while the RMSE value for the bare soil areas was 1.84 K due to the fragmentation.It shows that KRF method considering special scale factors can improve the downscaling precision of surface temperature in Karst region,therefore,it can provide more precise and reliable surface temperature products for the research based on the land surface temperature in the Karst region. |
| |
| Keywords: | land surface temperature thermal infrared radiances downscaling random forest Karst region |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
