全球变化知识资源中心

Large-scale spatiotemporal variations in land surface temperature (LST) can be derived from satellite observations, particularly for remote high-elevated mountain glaciers where in situ observations are scarce. Our previous LST retrieval study on Qiyi Glacier of China employed a generalized single-channel algorithm and Landsat thermal infrared data, but a clear LST summertime overestimation was observed, particularly on the glacier tongue. Here we extend our previous study by calculating the thermal radiance on Qiyi Glacier that is emitted from the surrounding terrain, and incorporating this thermal radiation term into two generalized single-channel LST algorithms to analyze its influence on the errors of glacier surface temperatures retrieved from remote sensing data. In addition, we also discuss other possible causes of this overestimation in LST retrieval. The key findings of our study are as follows. (1) The surrounding terrain provides nonnegligible summertime thermal radiation to the glacier in the Landsat7 band 6 spectrum range (10.4-12.5 mu m) that is 2.6 times larger than the atmospheric down-welling thermal radiance on the glacier tongue. The retrieved LSTs yield an overestimation of up to 0.28 K when the effect of thermal radiation from the surrounding slopes is not considered. (2) The average errors due to the surrounding topographic thermal radiation and the cubic convolution resampling method (CC method) employed in the Landsat7 thermal infrared data processing are only 0.07 K and - 0.01 K, respectively. These are much smaller than the average LST errors (1.79 K), indicating that the neither the surrounding topographic thermal radiation nor the CC method are primary factors in the retrieved LST overestimation. (3) The sole dependence of the generalized single-channel algorithm on the atmospheric water vapor content may therefore be the most likely cause of this overestimation, as high-altitude regions generally possess low atmospheric water contents. Radiative transfer models (e.g., MODTRAN) may provide a solution to this problem by simulating the influence of other atmospheric constituents on LST retrieval under conditions of low water vapor, which can then indicate any additional input variables that should be incorporated into the LST algorithm.