基于地理加权回归模型的博斯腾湖湖滨绿洲土壤盐分离子含量高光谱估算

以博斯腾湖湖滨绿洲为研究区,分析HCO_3~–、Cl~–、SO_4~(2–)、Ca~(2+)、Mg~(2+)、Na~++K~+等主要土壤盐分离子含量与土壤高光谱反射率的分数阶微分光谱变换与RSI、DSI、NDSI等二维土壤指数的相关性优选特征波段,构建基于地理加权回归模型的土壤盐分离子含量估算模型。研究结果表明:Na~++K~+的微分变换特征波段集中在468～724 nm与1 182～1 539 nm,二维土壤指数的特征波段集中在1 742～2 395 nm,基于RSI的特征波段优选下地理加权回归模型对Na~++K~+含量的估算效果较好,建模集R~2=0.94,RMSE=0.22,验证集R~2=0.74,RMSE=0.19;SO_4~(2–)含量在1.2阶优选的位于469～636 nm波段估算效果较佳,建模集R~2=0.91,RMSE=0.02,验证集R~2=0.75,RMSE=0.33;Ca~(2+)、Mg~(2+)优选的特征波段主要集中在912～2 340 nm的近红外波段;Cl~–含量在1阶的近红外波段建模效果较好,建模集R~2=0.74,RMSE=0.03,验证集R~2=0.93,RMSE=0.11;含量相对较高的Na~++K~+、SO_4~(2–)、Cl~–的地理加权回归模型精度高于含量较低的Ca~(2+)、Mg~(2+)。

Hyperspectral Estimation of Soil Salt Ion Contents in Lakeside Oasis of Bosten Lake Based on Geographical Weighted Regression Model

In this paper, the lakeside oasis of Bosten Lake was taken as the study area, the contents of main soil salt ions (HCO-3, Cl^-, SO₄^2-, Ca²⁺, Mg²⁺, Na⁺+K⁺) were measured, soil hyperspectral reflectance, fractional differential spectral transformation, and 2D soil indexes such as RSI, DSI and NDSI were obtained, and then the estimation models of soil salt ion contents were constructed based on the geographically weighted regression (GWR) model. The results showed that the feature bands of Na⁺+K⁺were concentrated in 468-724 nm and 1 182-1 539 nm under the differential spectral transformation, and the feature bands of 2D soil indexes were concentrated in the near-infrared band (1 742-2 395 nm). GWR model based on RSI feature band optimization estimated Na⁺+K⁺ content well, in which the modeling set R² was 0.94 and RMSE was 0.22, the validation set R²was 0.74 and RMSE was 0.19. The optimal band of SO₄^2- content in order 1.2 was 469-636 nm, in which the modeling set R² was 0.91 and RMSE was 0.02, the validation set R² was 0.75 and RMSE was 0.33. The preferred feature bands of Ca²⁺ and Mg²⁺ were mainly concentrated in the near-infrared band (912-2 340 nm). The modeling effect of the near-infrared band with Cl^- content in the first order was better, the modeling set R² = 0.74, RMSE = 0.03, verification set R² = 0.93, RMSE = 0.11. The accuracies of GWR model of Na⁺+K⁺, SO₄^2-and Cl^- with higher contents were higher than those of Ca²⁺ and Mg²⁺ with lower contents.