MODIS与Landsat 8地表温度融合拼接——以粤港澳大湾区为例

以粤港澳大湾区为例,基于时空影像融合模型(STI-FM),提出大区域多源LST(Land Surface Temperature,LST)影像融合拼接模型(Multi-source Image Fusion Stitching Model,MI-FSM),消除时间不同、天气差异的影响,生成覆盖大湾区的中高分辨率地表温度数据。首先,应用STI-FM融合MODISLST与Landsat LST,将不同时相的多幅LandsatLST合成为具有统一特定时相的LST影像。然后,对合成的LST影像进行镶嵌处理,实现粤港澳大湾区多幅Landsat 8 LST图像的拼接。为了验证STI-FM在研究区的适用性,选取研究区中心“夏-夏、冬-夏”2组Landsat 8 LST图像,将合成的Landsat LST与验证数据进行对比与评价,结果表明:STI-FM在研究区具有较强的适用性。对精度进行评价,验证模型的可靠性,结果表明:不同时相MODISLST图像间拟合程度较高,其回归分析的确定系数(M)均在0.6-0.9之间,RMSE均<1.5;最后对整体以及局部细节的目视分析表明:融合拼接的成果较为理想。

Surface Temperature Splicing Study Fusing MODIS and Landsat 8: A Case Study in the Guangdong-Hong Kong-Macao Greater Bay Area

Consider the Guangdong-Hong Kong-Macao Greater Bay Area as an example, this paper proposes a Multi-source Image Fusion Stitching Model (MI-FSM) for multi-source Land Surface Temperature (LST) images on a large-area study area; this model is based on the existing Spatio-Temporal Image Fusion Model (STI-FM). MI-FSM can eliminate the influences of different time and weather differences in the Landsat 8 data, realize seamless splicing between images, and generate medium- and high-resolution surface temperature data covering the Greater Bay Area. The data sources selected in the study were the Landsat 8 LST obtained from original Landsat 8 images via a surface temperature inversion algorithm and the MODIS LST 5d synthetic product. First, using the STI-FM fusion MODIS LST and Landsat LST, multiple Landsat LSTs of different phases were synthesized into LST images with uniform specific phases. Then, the synthesized LST images were mosaic-processed to splice multiple Landsat 8 LST images in the Guangdong-Hong Kong-Macao Greater Bay Area. To verify the applicability of STI-FM in the study area, this paper selected two groups of Landsat 8 LST images, “summer-summer” and “winter-summer,” in the center of the study area and compared and evaluated the synthesized Landsat LST concerning the verification data. In the “summer-summer” verification group, the minimum, maximum, mean, and standard deviation between the synthesized LST product and the actual LST were 0.01, 1.02, 0.10, and 0.03, respectively, and the coefficient of determination (R ²), Root Mean Square Error (RMSE), and Absolute Average Difference (AAD) were 0.849 5, 0.655 9, and 0.690 6, respectively. In the “winter-summer” verification group, the minimum, maximum, mean, and standard deviation between the synthesized LST product and the actual LST were 0.07, 0.08, 0.42, and 0.99, respectively, and R ², RMSE, and AAD were 0.681 7, 1.375 3, and 1.012 9, respectively. The results show that STI-FM has a strong applicability in the study area. Finally, the accuracy of the model was further verified via an accuracy evaluation: the degree of fitting between different time-phase MODIS LST images was high, the R ² value was between 0.6 and 0.9, and the RMSE was <1.5. Additionally, we selected three typical land cover types, cities, woodlands, and paddy fields and discussed the spatial distributions of their errors. The analysis results show that a linear relation is the most significant in cities, followed by paddy fields, and that the paddy fields are sensitive to seasonal differences. Woodlands have the poorest adaptability to the model, and there is no obvious relation with the errors for the woodlands. An evaluation was performed to prove that the results were reasonable: a visual analysis of the overall and local details indicated that the results of the fusion stitching were ideal.