Surface temperature downscaling method based on data mining

Abstract

The invention discloses a surface temperature downscaling method based on data mining. The surface temperature downscaling method comprises the following steps: selecting proper low-spatial-resolution surface temperature data and high-spatial-resolution visible light data as data sources according to an experimental area; a global model result and a local model result are obtained through the steps of data preprocessing, sample collection and the like; based on a model weight index (Weighti), combining a global model result and a local model result; calculating a temperature residual error in a scale * scale window based on a modified energy balance formula, and applying an adjustment result to each pixel in the window to complete a low-resolution surface temperature data downscaling process; and calculating MAE, R and RMSE between the downscaled surface temperature and other high-resolution surface temperature products, and verifying the precision and applicability of the algorithm for surface temperature downscaling. According to the method, the spatial resolution of the low-spatial-resolution surface temperature data is improved on the basis of full mining of spectral information of the multi-source remote sensing data.

Description

technical field [0001] The invention relates to the technical field of agricultural remote sensing, in particular to a method for downscaling (improving the spatial resolution) of low spatial resolution surface temperature products. Background technique [0002] Land Surface Temperature (LST) is often defined as the temperature of the Earth's surface. As an important parameter in environmental research and resource management, it is widely used in drought monitoring, evapotranspiration, soil moisture estimation and forest fire detection and other fields. In the 1970s, remote sensing technology began to be applied to surface temperature retrieval. Its macroscopic, fast, and economical advantages made up for the shortcomings of traditional ground monitoring in terms of spatial distribution. Advanced Very High Resolution Radiometer (AVHHR) and Moderate Resolution Imaging Spectrometer (MODIS) are two thermal infrared sensors widely used at present. All of them can achieve at l...

AI Technical Summary

Problems solved by technology

[0003] However, due to the limitations of existing spaceborne thermal infrared sensor technology, the spatial resolution of current spaceborne thermal infrared images is generally low (sixty meters to tens of kilometers), which limits the high spatial resolution of surface temperature information in some Applications in the study of the thermal environment at smaller scales, and the analysis of the thermal properties of land cover at small scales can contribute to a deeper understanding of changes in small-scale weather systems

[0004] At present, there are two ways to improve the spatial resolution of thermal infrared images: one is to essentially improve the spatial resolution of thermal infrared images by improving the imaging system performance of thermal infrared detectors. In recent years, some airborne high-resolution The high-resolution thermal infrared sensor has been successfully developed, but it is difficult to obtain practical application and promotion due to problems such as difficulty in acquisition, high cost, long acquisition cycle, and complicated preprocessing process; another method is based on image processing methods, using existing The visible-near-infrared high-spatial-resolution data of the visible light-near-infrared data is used, and a suitable algorithm is used to downscale the low-spatial-resolution thermal infrared image so that the spatial resolution reaches the level of the visible-near-infrared data. Compared with the first method, this One method is more efficient and more operable

[0005] At present, the widely used surface temperature downscaling methods are mainly based on the statistical regression of surface parameters, such as the classic Disaggregation procedures for radiometric surface temperature model, TsHARP (an algorithm for sharpening thermal imagery) algorithm and E-DisTrad , NL-DisTrad and other improved algorithms of Dis Trad and TsHARP algorithms, and geographically weighted regression models such as GWR and GTWR that have emerged in recent years, but in general, there are two common problems in the current commonly used surface temperature downscaling algorithms: (1 ) relies on the input of high-order remote sensing products, such as various band indices (NDVI, NDBI, EVI, SAVI, etc.), land use types, and DEM information, and most algorithms only use a small number of indices, which leads to algorithms in some Areas with complex land cover have lower accuracy; (2) Some new algorithms that have emerged in recent years have applied machine learning methods such as artificial neural networks and support vector machines, and adopted more indices. Compared with traditional modeling methods , these machine learning methods have higher requirements for sample data. Sample collection based on traditional downscaling modes usually collect samples by manually selecting regions of interest or supervised classification. These methods are difficult to collect a large number of samples with complete attribute information, and cannot identify spatial Details, ignoring the interaction of adjacent elements, this defect restricts the accuracy of the results of these downscaling algorithms

Images