Thermal Infrared Image Downscaling Method And Scale Effect Analysis

Thermal infrared images are mainly used for the inversion of land surface temperature,and land surface temperature is an important indicator of surface energy flow.Land surface temperature is closely related to plant growth,crop yield,surface water evaporation and circulation,global climate change,and urban expansion,natural and human activities.Increasing demand for high spatial resolution thermal infrared imaging.Due to the limitations of thermal infrared imaging technology,high temporal resolution and high temporal resolution are not directly available.Downscaling is an important means to obtain high spatial resolution land surface temperature under existing conditions.Scale effects are widely found in geographical factors.As a typical geo-ecological factor,land surface temperature has obvious scale effects.This paper takes Beijing area as the research area,uses FY-3/MERSI data and Landsat 8 data to study the spatial downscaling method of thermal infrared image,and quantitatively evaluates the downscaling results.test.To the existence of scale effects,the semi-variogram and Moran’s index at the pixel level are used to study the scale effects of surface features such as surface temperature,normalized difference vegetation index,urban index,biophysical composition index,and normalized difference building index.At the feature level,the landscape ecological index is used to study the scale effect of urban thermal landscape.The main research contents and innovations of this paper are as follows:(1)The three-layer decomposition model method uses the guided filtering and low-pass filtering to decompose the image into a detail layer,an edge layer and a low frequency layer,wherein the detail layer and the edge layer are proportionally increased to low-resolution thermal infrared data,thereby realizing Downscaling of thermal infrared images.Through the analysis of the downscaling results,the root mean square error of the downscaled land surface temperature of the three-layer decomposition model is 0.913 K,which is improved by 0.937 K and 0.832 K respectively compared with the DisTrad method and the TsHARP algorithm.In this paper,sensitivity analysis of the selection of window size for guided filtering and low-pass filtering is performed.(2)An improved three-layer decomposition model for thermal infrared data downscaling.In order to make the three-layer decomposition model suitable for the thermal infrared band,this paper improves the weight image,and proposes to use the variance to divide the auxiliary data into homogeneous and heterogeneous regions.Different regions use different weights;quantitative evaluation of image downscaling effect In this respect,the normalized RMSE(W-RMSE)is proposed.Using the ratio of the small-area variance to the global variance as the weight,it is difficult for RMSE to express the relationship between the pixel and the neighborhood,nor can it express the structural difference of the LST.The problem has been improved.Considering that different surface feature factors have different ability to describe surface feature types,this paper proposes a multi-factor three-layer decomposition model with different feature factors in different regions.The results show that the downscaling accuracy of combination factors is higher than single factor.The combination of the vegetation index and the town index has the highest precision.The study also found that the source of the downscaling error is mainly the architectural area,both single-factor and multi-factor methods.Through the analysis of multi-level downscaling results,it is found that the greater the span of the downscaling,the worse the accuracy of the downscaling results,and the error increases gradually as the scale decreases.(3)Scale effect analysis.Correlation analysis method is used to analyze the correlation coefficient of surface temperature and surface feature factors at different scales.It is found through experiments that the influence of scale on correlation coefficient is very obvious,and the influence of scale effect on correlation coefficient can be fitted by model.Using the semivariogram in the geostatistics and the Moran index to study the parameters,the variation of the nugget value,the abutment value,the nugget abutment and the range indicate the existence of the scale effect and reach it in the small scale space.The maximum value of the scale effect.Using the fractal dimension calculations of different angles,it is found that the fractal dimension of Beijing area is “W” type,that is,the spatial distribution is similar and structurally similar in the northwest-southeast and northeast-southwest directions.(4)Select thermal landscape level patches as the evaluation index of urban heat islands.The area change between different thermal levels is closely related to the change of scale.With the increase of the scale of the study area,the weak thermal landscape level plaque area decreases sharply,mainly to the adjacent dominant landscape level patches.Using the landscape evaluation indicators in landscape ecology,select the landscape level and the landscape level to evaluate the landscape price index,and the law of the thermal landscape level plaques in Beijing.Studies have shown that the changes in scale have a significant impact on the thermal landscape in Beijing,and there are obvious scale effects in the thermal landscape grade patches.Using the inflection point and semi-variogram curve characteristics in the landscape index variation curve,this paper proposes to divide the scale domain space into different research domains,and the scale of about 100 meters is the largest scale to study the internal heterogeneity of urban features..