基于FY-3C热红外和被动微波遥感的高分辨率地表温度融合方法研究

Land surface temperature (LST) is an important physical parameter in the earth surface system. The precise estimation of the spatiotemporal continuous LST is essential for the understanding the processes of surface energy balance, soil surface moisture, and evapotranspiration, as well as to understand trends in future climatic change on various spatial scales. Satellite-based remote sensing is the most effective method of measuring LST at the regional and global scales. Thermal infrared remote sensing data are widely used to retrieve LST with high spatial resolution and high precision, but they are limited to cloud-free conditions because of their inability to penetrate clouds. In contrast, passive microware (PMW) remote sensing data can be able to overcome the atmospheric influences while estimating LST, but the retrieved LSTs are constrained by low spatial resolution. The FY-3C satellite, which is China’s second-generation polar-orbiting meteorological satellites, was launched on September 23, 2013. The FY-3C VIRR (Visible and InfraRed Radiometer) and MWRI (Microwave Radiation Imager) provide daily LST products with spatial resolution of 1 km and 25 km, respectively. Based on the random forests and geostatistical theory, a new MWRI -based LST downscaling algorithm is proposed in this study, which considers the nonlinear relationship between LST and land surface variables such as NDVI, Albedo, MNDWI, LAI, and DEM, and the scale effect. Then, integrating the MODIS LST products, the error patterns of the FY-3C VIRR and downscaled MWRI LST products are investigated using triple collocation analysis; with performance assessments to consider different land cover classifications. Based on the above error analysis, a data fusion method is constructed by merging the FY-3C VIRR and downscaled MWRI LST products, which generates the all-weather LST products with fine resolution and high precision. The study is expected to provide valuable references for further improvement of the high-resolution LST products from the domestic meteorological satellites, and is useful for developing new multi-sources data fusion algorithms for domestic satellite remote sensing data products.

时空连续的高精度地表温度数据对于区域及全球地气系统能量平衡和生态系统的研究具有重要的科学意义和实际意义。卫星遥感是获取全球地表温度数据的重要途径。然而，热红外遥感产品因气象因素造成地表温度缺省，被动微波遥感地表温度产品的空间分辨率低，限制了它们在区域气象、水文生态以及气候模式的应用。本项目以国产FY-3C卫星资料为对象，拟基于随机森林算法和地统计学理论发展考虑地表温度与地表参量非线性关系及尺度效应的被动微波遥感地表温度数据空间降尺度方法；研究分析不同土地覆盖类型的热红外和降尺度被动微波遥感地表温度产品误差的时空分布特征；在此基础上，探索建立不同土地覆盖类型的热红外和降尺度被动微波遥感地表温度产品的融合方法，进而估算生成无云和有云情况下的高分辨率地表温度数据。本研究对于提高我国气象卫星高分辨率地表温度产品的精度，推动国产卫星遥感数据产品融合技术的发展具有重要的理论参考价值和现实意义。

时空连续的高精度地表温度数据对于区域及全球地气系统能量平衡和生态系统的研究具有重要的科学意义和实际意义。本项目的主要研究内容包括和结果分4个部分：1）多源地表温度产品评价：利用黑河流域的8个站点地表温度观测数据评价了FY-3C VIRR、MOD11A1、ERA5-Land、中国气象局陆面数据同化系统（CLDAS）的地表温度产品。结果表明，FY-3C VIRR地表温度产品的精度最差，MOD11A1地表温度产品的精度最佳；CLDAS白天地表温度产品的精度优于ERA5-Land白天地表温度产品，然而CLDAS夜间地表温度产品的精度劣于ERA5-Land夜间地表温度产品。2）地表温度数据空间降尺度：发展了基于随机森林和面到点克里金插值方法的地表温度空间降尺度方法（RFATPK），并首先应用于ASTER地表温度的空间降尺度；与TsHARP方法降尺度结果进行比较，RFATPK方法能获取高精度的降尺度地表温度数据，展示降尺度地表温度的更多细节；在这个基础上，将RFATPK方法直接应用于ERA5-Land地表温度产品的空间降尺度，获取高精度的1 km分辨率逐日白天和夜间ERA5-Land降尺度地表温度产品。3）多源地表温度产品的误差分析：利用Triple Collocation（TC）方法分析2015-2020年FY-3C VIRR、MOD11A1和降尺度ERA5-Land白天和夜间地表温度产品的时空误差分布特征。结果表明，不同季节FY-3C VIRR、MOD11A1和降尺度ERA5-Land白天和夜间地表温度呈现不同的误差分布；FY-3C VIRR、MOD11A1和降尺度ERA5-Land三个产品的地表温度误差整体上从南向北逐渐减少；FY-3C VIRR地表温度的误差最大，其次是降尺度ERA5-Land地表温度；三种地表温度产品的冬季地表温度精度相对较高。4）多源地表温度产品的融合：构建了不同季节（春、夏、秋、冬）的多源地表温度产品融合模型，获取了无云和有云情况下2015-2020年1 km分辨率每日白天和夜间地表温度产品。相对3种地表温度产品，基于TC方法的地表温度融合产品的精度在一定程度上有所提高，特别是夜间地表温度融合产品；相比FY-3C VIRR夜间地表温度，TC夜间地表温度融合结果精度提升了10%以上。成果可以为提高国产气象卫星数据产品的质量提供技术支撑。