基于地理加权回归模型的绿洲土壤表层有机质含量高光谱估算

Hyperspectral Estimation of Surface Soil Organic Matter Content in the Oasis Based on Geographically Weighted Regression Model

【目的】研究利用高光谱数据估算土壤表层有机质含量,为绿洲区大范围,快速,低成本,监测土壤表层有机质含量提供技术参考。【方法】以新疆博斯腾湖湖滨绿洲为研究区,采用地理加权回归模型,优选高光谱数据与土壤有机质含量的特征波段,构建研究区表层土壤有机质含量的高光谱估算模型。【结果】研究区表层土壤有机质含量变化不大,变化系数为55%,最小值为2.37 g/kg,最大值为51.47 g/kg,平均值为21.20 g/kg。土壤有机质特征波段主要集中在645~1958 nm,其中1/R二阶的相关系数值最大为0.73,且在P=0.05水平下,通过显著性检验的波段数为83。构建的地理加权回归模型中,二维土壤指数1/R RSI建模效果最优,建模集R^(2)=0.91,RMSE=2.56,验证集R^(2)=0.95,RMSE=1.10。【结论】利用地理加权回归模型估算土壤有机质估算,建模效果可以达到一定的精度要求。

【Objective】Taking the lakeside oasis of Bosten Lake in Xinjiang,China as the study area,the content of soil surface organic matter was estimated by hyperspectral data,which provided a technical reference for monitoring the content of soil surface organic matter in the oasis area of lakeside of Bosten Lake in a wide range with rapidity and low cost.【Methods】Based on the geographical weighted regression model,the hyperspectral estimation model of soil organic matter content in the study area was constructed by selecting the characteristic bands of hyperspectral data and soil organic matter content.【Results】The content of organic matter in surface soil in the study area showed moderate variation.The change coefficient was 55%,the minimum value was 2.37 g/kg,the maximum value was 51.47 g/kg,and the average value was 21.20 g/kg.The characteristic bands of soil organic matter were mainly concentrated in the range of 645-1,958 nm,in which the correlation coefficient of the second order of 1/R was 0.73,and P was at 0.05.The number of bands passing the significance test was 83.The modeling effect of two-dimensional soil index 1/R RSI was the optimal in the geographically weighted regression model,with modeling set R^(2)=0.91,RMSE=2.56,and validation set R^(2)=0.95,RMSE=1.10.【Conclusion】Using a geo-weighted regression model to estimate soil organic matter estimates,the modeling results can achieve a certain level of accuracy,and this study provides a reference for modeling soil organic matter when there are few sampling points.