采煤拉张裂隙区土壤有效磷高光谱估测OA
Hyperspectral estimation of soil available phosphorus in coal mining-induced tensile fissure zones
[目的]采煤活动引发的地表裂隙破坏土壤结构,影响土壤理化性质与养分分布,严重制约了矿区农业可持续发展.开展采煤拉张裂隙区土壤有效磷含量快速估测,对于矿区复垦土壤肥力评价和养分管理具有重要意义.[方法]文章以淮北朱庄煤矿3522工作面采动裂隙区为研究区,采集90个土壤样本,测定其有效磷含量与高光谱数据;采用Savitzky-Golay平滑(SG)、一阶微分(FD)和标准正态变量变换(SNV)设计了9种光谱预处理组合,并以此分别建立偏最小二乘回归(PLSR)、随机森林(RF)和支持向量回归(SVR)3种全波段估测模型.同时,基于各模型的最优预处理组合,采用竞争性自适应重加权采样算法(CARS)进行特征波段筛选,对比分析全波段与特征筛选模型的预测性能.[结果](1)研究区土壤有效磷含量整体处于中低水平,相关光谱信息呈弱响应和多波段分散特征.(2)全波段建模中,PLSR模型表现最优,最优预处理为FD-SG-SNV,验证集决定系数R²为0.694,预测残差偏差比(RPD)为1.843;RF和SVR模型的最优R²分别为0.476、0.665.(3)经CARS特征筛选后,PLSR模型验证集R²降至0.525,而SVR和RF精度略有提升,CARS-SVR表现最佳,R²为0.669,RPD为1.770.[结论]在该研究样本条件下,全波段PLSR模型表现相对较优,说明其对小样本高维光谱数据中分散弱信息具有一定整合能力.CARS-SVR模型在特征筛选后仍保持了接近全波段模型的预测效果,表明特征筛选后非线性模型预测精度有小幅改善.研究结果可为采煤拉张裂隙区土壤有效磷高光谱估测及建模策略选择提供依据.
[Purpose]Coal mining activities can induce surface tensile fissures and damage soil structure,thereby altering soil physicochemical properties and nutrient distribution.These changes restrict ecological restoration and sustainable agricultural production in mining areas.Soil available phosphorus(AP),as an important indicator of soil phosphorus supply capacity and reclaimed soil fertility,requires rapid and non-destructive estimation in coal mining-induced tensile fissure zones.[Method]The mining-induced tensile fissure zone of the 3522 working face in Zhuzhuang Coal Mine,Huaibei,was selected as the study area.A total of 90 soil samples were collected,and their AP contents and hyperspectral reflectance were measured.Nine spectral preprocessing combinations were applied using savitzky-golay smoothing(SG),first-order differentiation(FD),and standard normal variable transformation(SNV).Three regression models—partial least squares regression(PLSR),random forest(RF),and support vector regression(SVR)—were used to construct full-band AP estimation models.Competitive adaptive reweighted sampling(CARS)was then used to select feature bands based on the optimal preprocessing scheme for each model.The prediction performance of the full-band models and feature-selection-based models was then compared.[Result](1)Soil AP content in the study area was generally at a medium-to-low level,and the AP-related spectral information was characterized by weak responses distributed across multiple bands.(2)The full-band PLSR model with FD-SG-SNV preprocessing achieved the best validation performance,with a validation R² of 0.694 and an RPD of 1.843.The optimal validation R² values of the RF and SVR full-band models were 0.476 and 0.665,respectively.(3)After CARS feature selection,the validation R² of PLSR decreased to 0.525,whereas the validation R² values of SVR and RF increased slightly.Among the CARS-based models,CARS-SVR performed best,with a validation R² of 0.669 and an RPD of 1.770.[Conclusion]Under the sample conditions of this study,the full-band PLSR model shows relatively stable performance,indicating its ability to integrate dispersed weak spectral information from small-sample and high-dimensional spectral data.CARS feature selection has different effects on linear and nonlinear models.It reduces the prediction accuracy of PLSR but slightly improves the performance of SVR and RF.Therefore,full-band PLSR can be used as a preferred model for preliminary hyperspectral estimation of soil AP in the study area,while CARS-SVR can serve as an alternative when dimensionality reduction is required.
蒋慧敏;郭辉;戴志林
安徽理工大学空间信息与测绘工程学院,淮南 232001||矿山环境与灾害协同监测煤炭行业工程研究中心,安徽 淮南 232001安徽理工大学空间信息与测绘工程学院,淮南 232001||矿山环境与灾害协同监测煤炭行业工程研究中心,安徽 淮南 232001泉州海洋职业学院人工智能学院,福建 泉州 362700
土壤有效磷采煤拉张裂隙竞争性自适应重加权采样算法(CARS)高光谱遥感估测模型
soil available phosphoruscoal mining-induced tensile fissurecompetitive adaptive reweighted sampling(CARS)algorithmhyperspectral remote sensingestimation model
《中国农业信息》 2026 (2)
55-68,14
矿山环境与灾害协同监测煤炭行业工程研究中心开放基金"煤炭开采裂隙区土壤质量变异性特征与高光谱监测方法"(KSXTJC202202)
评论