Chenzhou and Guilin 30 meters of GDP spatialized data products in 2010, 2015 and 2020

Based on the spatial downscaling model of large selective nuclear fusion, the feature information of high-resolution guidance images and 100-meter GDP spatialized data is deeply integrated to achieve 30-meter high-resolution GDP estimation.
(1) GDP spatialization based on multi-source data fusion
In order to realize GDP spatialization, this study first constructs a multi-source dataset of three GDP industries, and then proposes a multi-channel weighted fusion network (SVMFN) based on spatial feature vectors. The model solves the spatial autocorrelation problem by introducing spatial feature vectors and uses multi-source datasets to achieve sub-industry modeling. The spatialization results of the three industries are linearly corrected and superimposed to generate GDP spatialization data with a resolution of 100 meters. In order to verify the effectiveness of the model, SVMFN is compared with other machine learning and deep learning methods, and the accuracy of spatialization results is verified by using township statistical GDP data, which provides data support for subsequent spatial downscaling. In addition, the spatial distribution characteristics of GDP and its three industries are analyzed in detail, and the characteristic importance of the multi-source datasets used by each industry is evaluated.
(2) High-resolution GDP estimation based on spatial downscaling
In order to achieve high-resolution GDP estimation, this study proposes a spatial downscaling model based on Large Selective Kernel Fusion Network for Spatial Downscaling (LSKF-Net) based on the spatial downscaling method of guided images. The model is composed of three parts: feature extraction module, large selection nuclear fusion module and image reconstruction module, which fully integrates 100 meters of GDP spatialization data and high-resolution guidance images to achieve 10-meter resolution GDP estimation. In order to verify the advantages of the model, it is compared with the traditional interpolation method and other spatial downscaling models. At the same time, the accuracy of the model results is further verified by comparing the accuracy and spatial distribution with the original 100m GDP spatialization data and other 1km GDP spatialization products."