|
| 典型绿洲灌区棉田土壤盐分多光谱遥感反演与季节差异性研究 |
| Multispectral remote sensing inversion and seasonal difference in soil salinity of cotton field in typical oasis irrigation area |
| 投稿时间：2022-04-28 修订日期：2022-07-21 |
| DOI：10.13254/j.jare.2022.0248 |
| 中文关键词: 多光谱遥感，土壤含盐量，变量组，机器学习，空间分布 |
| 英文关键词: multispectral remote sensing, soil salinity content, variable group, machine learning, spatial distribution |
| 基金项目:国家重点研发计划项目（2021YFD1900805-4）；新疆维吾尔自治区重大科技专项（2020A01002-1） |
| 作者 | 单位 | E-mail | | 刘旭辉 | 新疆农业大学资源与环境学院, 乌鲁木齐 830052
新疆水利水电科学研究院, 乌鲁木齐 830049 | | | 白云岗 | 新疆水利水电科学研究院, 乌鲁木齐 830049 | xjbaiyg@sina.com | | 柴仲平 | 新疆农业大学资源与环境学院, 乌鲁木齐 830052 | | | 张江辉 | 新疆水利水电科学研究院, 乌鲁木齐 830049 | | | 江柱 | 新疆水利水电科学研究院, 乌鲁木齐 830049 | | | 丁邦新 | 新疆水利水电科学研究院, 乌鲁木齐 830049
西北农林科技大学水利与建筑工程学院, 陕西 杨凌 712100 | | | 张超 | 新疆农业大学资源与环境学院, 乌鲁木齐 830052
新疆水利水电科学研究院, 乌鲁木齐 830049 | |
|
| 摘要点击次数: 887 |
| 全文下载次数: 966 |
| 中文摘要: |
| 为探究区域尺度的土壤含盐量空间分布、季节性变化和年际变化特征，本研究以新疆第二师31团棉田为研究区，以2019年和2021年春、夏、秋季实测土壤含盐量和Landsat 8 OLI多光谱影像为基础，将波段组、盐分指数组、植被指数组和全变量组作为模型输入变量组，通过相关性分析优选特征光谱参量，采用极限学习机（Extreme Learning Machine，ELM）、支持向量机（SupportVector Machine，SVM）和BP神经网络（Back Propagation Neural Network，BPNN）构建基于各变量组的不同季节土壤盐分反演模型，通过实测数据评价精度筛选确定各季节最优模型，定量反演地表土壤含盐量。结果表明：研究区两年春、夏、秋季土壤含盐量总样本变异系数分别为 0.67、0.56、0.67，呈中等变异性；中度盐化土的光谱反射率高于轻度盐化土和非盐化土；基于全变量组的BPNN模型均为各季节最优的土壤盐分反演模型，精度由高到低依次为夏季>春季>秋季；两年各季节土壤含盐量由大到小顺序均为秋季>春季>夏季，说明灌排及农业耕作措施对土壤盐分动态变化影响较大；2019—2021年各季节土壤含盐量均有所减小，说明灌区灌排措施对盐碱地治理效果明显。研究表明，基于多光谱影像建立的机器学习模型可定量反演土壤含盐量，为南疆典型绿洲灌区棉田土壤盐渍化监测提供参考。 |
| 英文摘要: |
| To explore the spatial distribution, seasonal variation, and interannual variation characteristics of soil salinity content on a regional scale, this study took the cotton fields of the 31st Crops of the 2nd Division in Xinjiang, China as the research area, based on soil salinity content measured in spring, summer, and autumn of 2019 and 2021 and Landsat 8 OLI multispectral images. The band, salinity indices, vegetation indices, and total variable groups were used as the input variable group of the model, and the characteristic spectral parameters were optimized through correlation analysis. Extreme learning machine（ELM）, support vector machine（SVM）, and back propagation neural network（BPNN）were used to construct soil salinity inversion models in different seasons based on each variable group. By evaluating the accuracy of measured data, the optimal model in each season was selected to quantitatively invert the surface soil salt content. The results showed that the variation coefficients of soil salinity content in spring, summer, and autumn were 0.67, 0.56, and 0.67, respectively, showing moderate variability. The spectral reflectance of moderately salinized soil was higher than that of light salinized soil and non-salinized soil. The BPNN model based on the total variable group was the optimal soil salinity inversion model in each season, and the accuracy was in the order as follow：summer, spring, and autumn from high to low. The soil salt content in each season in the two years was in this order as follow：autumn > spring > summer, indicating that irrigation and drainage and agricultural tillage measures had a great influence on the dynamic change in soil salinity. Soil salinity content decreased in each season from 2019 to 2021, indicating that irrigation and drainage measures in irrigated areas had significant effects on saline–alkali land management. The results of this study showed that the machine learning model based on multispectral image can quantitatively invert soil salinity content and can be used as a reference for soil salinization monitoring of cotton field in typical oasis irrigated areas in southern Xinjiang. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
