地表温度降尺度优化研究

Research on Optimization of Land Surface Temperature Downscaling

针对地表温度(land surface temperature,LST)降尺度过程存在的误差增加问题,采用改善图像梯度的方式优化降尺度的结果。通过不同强度的滤波处理及温度精度的跟踪验证,成功优化了降尺度后的LST,并提出一种新的指标——梯度异常频率(gradient abnormal frequency,GAF)来评价温度图像的质量。首先,利用光谱指数法对热红外数据反演的LST进行降尺度,此过程GAF、平均绝对误差(mean absolute error,MAE)和均方根误差(root mean square error,RMSE)的值明显增加;通过均值滤波处理后,GAF降低,MAE和RMSE也有小幅下降;加大滤波强度后,GAF继续下降,但下降到10%左右时,MAE和RMSE不再变化。结果表明,GAF的降低能够提高降尺度后LST的精度;GAF到达某个临界值时,LST的精度最大限度改善。该文弥补了光谱指数法降尺度后温度像元连续性差的缺陷,对推进高分辨率城市地表温度研究具有重要应用价值。

For the issue of error increase after downscaling the land surface temperature(LST),the method of improving image gradient is used to optimize the results after downscaling.Through filtering processing with different strengths and tracking verification of temperature accuracy,the downscaled LST is successfully optimized and a new index,gradient abnormal frequency(GAF),is proposed to evaluate the quality of the temperature images.Firstly,the spectral index method is used to downscale the LST retrieved from the thermal infrared data,and the values of GAF,mean absolute error(MAE),and root mean square error(RMSE)increase significantly after downscaling.After the mean filtering,the GAF reduces,and the MAE and RMSE also slightly decrease.After increasing the strength of the filter,GAF continued to decline,but when it drops to around 10%,MAE and RMSE no longer change.The results indicate that the reduction of GAF can improve the accuracy of LST after downscaling;when the GAF reaches a certain critical value,the accuracy of LST is greatly improved.This study makes up the shortcomings of bad continuity of temperature pixels after downscaling by spectral index method,and it has an important application value to promote the research of high-resolution urban LST.