Research On Disaggregation Of Land Surface Temperature And Its Application In GF-1/GF-2 Image

The land surface temperature(LST)is an important environmental factor in characterizing climate change.As an important means of global and regional LST acquisition,satellite thermal infrared(TIR)remote sensing only provides low spatial resolution(sixty meters to tens of kilometers)image,which is inadequate to capture the more complex temperature changes in heterogeneous urban environment.The disaggregation of LST(DLST)based on spectral mixing model is an important method to obtain high-resolution LST,which fuses TIR image and high-resolution visible/near infrared image.The high spatial resolutions of GF-1 and GF-2 provide a good foundation for DLST to estimate very-high-spatial-resolution LST.However,the existing studies have ignored within-class spectral and thermal variability,especially at very high spatial resolution,which greatly affected the accuracy of the very-high-spatial-resolution LST estimation.To solve the above problems,this paper introduces object-oriented image analysis and multiple endmember spectral mixture analysis to study the method and application of estimating very-high-spatial-resolution LST,based on GF-1,GF-2 multispectral image and Landsat-8 TIR image.The main contents are in three folds:(1)A multiple endmember object spectral mixture analysis(MEOSMA)method is proposed.By the introduction of image segmentation and the proposed local Minimum Average Spectral Angle(MASA)method,the MEOSMA solves the shortcomings of traditional methods in endmember extraction and optimization,and further reduces the influence brought by within-class variability.Experimental results prove that the MEOSMA method improves the accuracy of spectral unmixing.(2)A Multiple Endmember Object Spectral Mixture Analysis and Thermal Mixing(MEOSMA-TM)method for DLST is proposed.Based on MEOSMA and linear thermal mixing model,the MEOSMA-TM method solves the problem of within-class spectral and thermal variability effectively,and obtains very-high-resolution LSTs by using GF-1/GF-2 multiple spectral image and Landsat TIR image.In this part,the reliability of the proposed MEOSMA-TM method is verified with the LST data before downscaling and the measured LST data from Hangzhou National Conventional Meteorological Station,respectively.The experimental results show that this method has acceptable accuracy.(3)The influence of transportation on urban land surface temperature is studied.Based on the very-high-resolution LSTs estimated by MEOSMA-TM,this paper uses multi-scale grid analysis,buffer analysis and spatial statistics methods to study the spatial and temporal distribution of LST in Hangzhou and analyze the influence of road density,different grade roads,road greening and transportation hubs over urban LSTs.It turns out that the proposed methods help to improve the accuracy of spectral unmixing and very-high-resolution LST estimation,and can promote the application of GF-1/GF-2 in urban thermal environment studies.The outcomes of this study can be useful in many fields,such as urban land use planning,urban ecological environment management,transportation network planning,urban road design,etc.