基于星载高光谱数据的南京新济洲湿地土壤有机质估测研究

在高光谱数据预处理、土壤有机质高光谱敏感波段提取基础上, 建立多元线性回归、最邻近法、装袋算法、多元感知器、随机森林5种遥感估测模型。用10折交叉验证方法, 借助相关系数、绝对误差、均方根误差、相对误差、相对均方根误差5个指标, 对遥感估测模型结果进行精度评价, 选择精度最高的模型进行湿地土壤有机质遥感估测和空间分析。结果表明:土壤有机质高光谱敏感波段主要集中在925、1 144、1 477、1 780 nm 4个波段; 在预测土壤有机质的5种模型中, 多元线性回归模型预测精度最高, 随机森林次之; 土壤有机质空间分布呈现由洲滩中间向四周逐渐增加的带状分布格局; 新济洲沼泽地土壤有机质含量最高, 为2.22%;靠近沼泽的林地次之; 植被覆盖度较低的农地和裸地的土壤有机质最低, 为0.43%;这种土壤有机质空间分布格局与研究区土壤类型的带状分布存在密切联系。

Estimation of Wetland Soil Organic Matter Based on Spaceborne Hyperspectral Image in Xinjizhou of Nanjing

Based on hyperspectral data pretreatment and hyperspectral spectral bands extraction of soil organic matter, remote sensing estimation models of multiple linear regression, multilayer perception, k-nearest neighbor, bagging and random forest were built.Cross-validation method was applied to evaluate the accuracy of 5 models with the help of 5 indexes:correlation coefficient establishment, mean absolute error, root mean squared error, relative absolute error, and root relative squared error.The most accurate model was selected for remote sensing and spatial analysis of soil organic matter in wetlands.Research results showed that the sensitive hyper-spectral band wavelength to soil organic matter was mainly concentrated in 925 nm, 1 144 nm, 1 477 nm, and 1 780 nm 4 bands.Among 5 prediction models, the performance of the multiple linear regression was the best, followed by random forest.The spatial distribution of soil organic matter presents a zonal distribution pattern gradually increasing from the middle of the bank to the periphery.The soil organic matter content of Xinjizhou wetland was the highest, 2.22%, followed by the forestry near wetland.The soil organic matter of the agricultural land and bare land with the lowest vegetation coverage was the lowest, 0.43%.The spatial distribution pattern of soil organic matter is closely related to the zonal distribution of soil types in the study area.