H51D-1403
Influence of Scale Effect and Model Performance in Downscaling ASTER Land Surface Temperatures to a Very High Spatial Resolution in an Agricultural Area

Friday, 18 December 2015
Poster Hall (Moscone South)
Ji Zhou1, Guoquan Li1, Shaomin Liu2, Wenfeng Zhan3 and Xiaodong Zhang1, (1)UEST University of Electronic Science & Technology of China, Chengdu, China, (2)Beijing Normal University, State Key Laboratory for Remote Sensing Science, Research Center for Remote Sensing and GIS, and School of Geography, Beijing, China, (3)Nanjing University, Jiangsu Provincial Key Laboratory of Geographic Information Science and Technology, International Institute for Earth System Science, Nanjing, China
Abstract:
At present land surface temperatures (LSTs) can be generated from thermal infrared remote sensing with spatial resolutions from ~100 m to tens of kilometers. However, LSTs with high spatial resolution, e.g. tens of meters, are still lack. The purpose of LST downscaling is to generate LSTs with finer spatial resolutions than their native spatial resolutions. The statistical linear or nonlinear regression models are most frequently used for LST downscaling. The basic assumption of these models is the scale-invariant relationships between LST and its descriptors, which is questioned but rare researches have been reported. In addition, few researches can be found for downscaling satellite LST or TIR data to a high spatial resolution, i.e. better than 100 m or even finer. The lack of LST with high spatial resolution cannot satisfy the requirements of applications such as evapotranspiration mapping at the field scale.

By selecting a dynamically developing agricultural oasis as the study area, the aim of this study is to downscale the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) LSTs to 15 m, to satisfy the requirement of evapotranspiration mapping at the field scale. Twelve ASTER images from May to September in 2012, covering the entire growth stage of maize, were selected. Four statistical models were evaluated, including one global model, one piecewise model, and two local models. The influence from scale effect in downscaling LST was quantified. The downscaled LSTs are evaluated from accuracy and image quality.

Results demonstrate that the influence from scale effect varies according to models and the maize growth stage. Significant influence about -4 K to 6 K existed at the early stage and weaker influence existed in the middle stage. When compared with the ground measured LSTs, the downscaled LSTs resulted from the global and local models yielded higher accuracies and better image qualities than the local models. In addition to the vegetation indices, the surface albedo is an important descriptor for downscaling LST through explaining its spatial variation induced by soil moisture.

<< Previous Abstract | Next Abstract >>