MODIS地表温度产品在青藏高原冻土模拟中的适用性评价

The Applicability of MODIS Land Surface Temperature Products to Simulating the Permafrost Distribution over the Tibetan Plateau

利用MODIS地表温度反演产品,以青藏高原为研究区域,通过单点、区域、模型3方面来验证MODIS地表温度产品在青藏高原冻土模拟中的适用性.通过69个气象站点观测的地表0cm温度数据与所在位置的MODIS地表温度数据比较,二者在时间序列上的变化趋势基本一致,但是平均误差较大.在区域验证中,实测地表温度数据采用经纬度海拔回归方法和克里金插值地统计方法得到空间温度分布,与MODIS地表温度产品进行面状要素的时间序列比较,其结果与MODIS地表温度产品在时间和空间上变化趋势是稳定的,但也存在一定的差异.单点和区域验证都揭示两个数据集在暖季(5—8月)比冷季(9月至翌年4月)差异大,相关系数小.模型验证中,选用半经验半物理的TTOP模型,分别应用实测地表温度数据和MODIS地表温度产品模拟青藏高原多年冻土分布情况,MODIS地表温度产品的模拟结果比实测数据的模拟结果更接近青藏高原冻土图.

As a main parameter of energy balance on the earth surface,the land surface temperature（LST） is one of the necessary boundary conditions of a numerical permafrost model,and it is also an important observational item of any permafrost monitoring.Traditionally land surface temperature data are collected from meteorological stations.However,the point observation often cannot represent an areal situation well especially in mountainous areas.With its development,the remote sensing technique increasingly becomes an applicable way to observe land surface temperature.The limitations here are generally caused by complex terrain conditions which consequently bring large uncertainties to the derivation of LST.Applicability analysis is required prior to formal application of the derived LST data.This paper will investigate the applicability of MODIS LST products to permafrost simulation over the Tibetan Plateau,through single-point,area,and model based analyses.The single-point analysis,which compares 0-cm temperature data from 69 meteorological stations over the plateau to the MODIS LSTs of the corresponding grid cells,shows the MODIS LSTs have similar trends as measured 0-cm temperature data,but with obvious difference in terms of magnitude.The areal analysis,comparing the time series MODIS LST data to two interpolated areal 0-cm temperature data sets derived respectively by a LAT/LON/ELEV regression approach and a Kriging interpolation,reveals that there are a fairly good consistency between MODIS LST and regressed data set in terms of both space and time regardless of some differences.Meanwhile poor agreement is found between MODIS LST and Kriging interpolated data set as shown a small correlation and large mean differences.Both the single-point and areal analyses present poor agreements in the warm seasons（May-August） than the cold seasons（September-April）.For model validation,the semi-physical and semi-empirical TTOP model were chosen,and the measured temperature data and the MODIS LST products were used,respectively,to simulate the permafrost distribution on the Tibetan Plateau.Area statistics and Kappa agreements calculated from the model approach show that the simulation with MODIS LST is closer to the published permafrost map.