黑河流域ASTER与MODIS融合生成高分辨率地表温度的验证

Verification of high-resolution land surface temperature by blending ASTER and MODIS data in Heihe River Basin

融合多源遥感数据生成高时空分辨率数据具有重要的应用价值。为了解决高空间分辨率数据重访周期长及云雨天气带来的数据短缺问题,该文基于增强自适应的遥感图像时空融合方法(enhanced spatial and temporal adaptive reflectance fusion model,ESTARFM),使用多时相MODIS数据提供地物时间变化信息,结合ASTER影像提供的空间细节信息,选择多波段数据(可见光近红外数据和地表温度数据)共同作为输入变量融合生成高时空地表温度。融合结果分别与地表红外辐射计观测温度和ASTER温度产品进行了验证。验证结果表明:基于ESTARFM方法降尺度地表温度影像清晰,融合结果与地表红外辐射计观测温度呈显著的线性正相关关系,相关系数均高于为0.71,预测得到的地表温度与真实测得的数据的平均绝对偏差均低于2.00 K,均方根误差均低于2.60 K。与ASTER地表温度产品的验证中,整体验证结果的R2均在0.95以上。此外,ESTARFM方法在各个地类中的融合效果较好,均表现出非植被区域的相关性高于植被和水体,尤其在2012年8月27日非植被的R2达到0.91。

Land surface temperature（LST） is a key parameter in investigating environmental, ecological processes and climate change at various scales, and is also valuable in the studies of evapotranspiration, soil moisture conditions, surface energy balance, and urban heat islands. However, it is difficult to acquire satellite LSTs with both high spatial and temporal resolutions due to tradeoffs between these parameters. The Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model（ESTARFM） algorithm was initially designed to predict surface reflectance and is based on the assumption that MODIS and Landsat surface reflectance are highly consistent over homogeneous areas. However, the ESTARFM method prediction results degrade somewhat when the method is used for heterogeneous fine-grained landscapes. This research extended the ESTARFM model from reflectivity range to thermal infrared for estimation of daily temperature at 90 m resolution combined MODIS and ASTER. The implementation of ESTARFM requires input of the search window size, selection of spectrally similar pixels, determination of the weight of similar pixels, and computation of the correction coefficient and temporal weight. The calculation of weights for spectrally similar pixels involves weighing the contribution of neighboring pixels to the computation of a central pixel. Using a local moving window, neighboring spectrally similar pixels were included for the computation of the LST corresponding to a central pixel with the temporal weights of the two dates. In this study, we used multiple bands, i.e., red, NIR, and LST bands, as the input variables and generated high spatial-temporal resolution land surface temperature, combining temporal change information from multi-temporal MODIS with high-resolution spatial resolution from ASTER. The objective of this paper was to evaluate the ESTARFM method using ground measurements coordination with ASTER LST products collected in an arid region of Northwest China during the first thematic Multi-Scale Observation Experiment on Evapotranspiration（MUSOEXE） over heterogeneous land surfaces in 2012, as part of the Heihe Watershed Allied Telemetry Experimental Research（HiW ATER）.We didn＇t modified the model even if the linear hypothesis was directly applied in LST prediction, which may result in uncertainty and errors. The remote sensing data were acquired with from Jun to September in 2012. The results showed that ESTARFM was positively and linearly related with the actual measured TIR. The correlation coefficient values were all found to be greater than 0.71. The mean absolute error and root mean square error were all below 2.00 K and 2.60 K, respectively. From the feature of scattering plots between the predicted and observed LST, the data points fell close to the diagonal line in each panel, indicated that the predictions were all in good agreement with the observations. Overall, the values of mean absolute error and root mean square error between the predicted and the observed LST were quiet small; whereas the correlation coefficient values between the predicted LSTs and ASTER LST products were all found to be greater than 0.95.It should be noted that some pixels in the scatter plots showed differences between the predicted and observed LSTs. These discrepancies revealed a major limitation to the method; i.e., ESTARFM does not capture land cover that has been altered between two imaging dates. Thus, changes in land cover or other surface conditions can lead to prediction errors. In addition, the fusion results showed that value of correlation coefficient was better in non-vegetation area than vegetation and water area, and up to 0.91 especially in the August 27, 2012. However, the application of the ESTARFM and its variants to LST prediction is immature in terms of methodology. Many critical issues have not been solved, especially with respect to the determination of the search window size, conversion coefficient improvement, and thermal landscape heterogeneity.