基于国产资源一号02D卫星和机器学习算法的钦州湾滨海土壤盐分反演

Inversion of coastal soil salinity in Qinzhou Bay based on domestic ZY1-02D satellite and machine learning algorithm

以钦州湾滨海地区作为研究区,基于国产资源一号02D(ZY1-02D)多光谱卫星提取的相关特征参数,在Ada Booost、Light GBM、XGBoost、RFR以及Cat Boost五种不同机器学习算法的支持下,设置了5种不同的输入变量组合,并基于决定系数(R^(2))和均方根误差(RMSE)对不同模型的性能进行了评价.结果显示:研究区实测土壤盐分含量范围为0.740~10.352 g/kg,均值为1.739 g/kg;Cat Boost相较于Ada Booost、Light GBM、XGBoost、RFR有更好的预测性能,Cat Boost结合全变量组在预测阶段取得了最高精度(R^(2)=0.8317,RMSE=0.3957 g/kg);在全变量组中,纹理特征中的均值对土壤盐分含量最为敏感,贡献度最高;研究区土壤盐分含量预测值为0~8.784g/kg,均值为2.478g/kg,轻度盐渍土分布广泛,主要集中分布在研究区西部,在东部地区分布较零散.国产资源卫星遥感数据结合Cat Boost模型在钦州湾滨海土壤盐分反演中表现出较好的性能,可为大规模估算土壤盐分含量提供一种新的方法和思路.

The relevant feature parameters extracted from the domestic ZY1-02D multispectral satellite were used to characterize the soil salinity over the coastal area of Qinzhou Bay with the support of Ada Boot,LightGBM,XGBoost,RFR,and CatBoost machine learning algorithms.The performance of each model was evaluated with the coefficient of determination(R^(2))and root mean square error(RMSE).The results show that the soil total salt content in the research area was measured to range from 0.740 to 10.352g/kg with an average of 1.739g/kg.Model simulation results demonstrate that CatBoost had the best predictive performance over AdaBoost,LightGBM,XGBoost,and RFR,and combined CatBoost with the highest accuracy(R^(2)=0.8317,RMSE=0.396g/kg);and of all variables in a group,the mean of texture features was most sensitive to soil salinity and made the highest contribution;The soil salt content was simulated to range from 0to 8.784g/kg,with an average of 2.478g/kg,in which mild salinity mainly occurred in the western part of the study area and scattered in the eastern part.The combination of domestic resource satellite remote sensing data and CatBoost model has shown good performance in retrieving soil salinity in the coastal area of Qinzhou Bay,providing a new approach to characterizing coastal soil salinity at a large-scale.