| 基于多光谱遥感的典型绿洲棉田春季土壤盐分反演及验证 |
| |
| 引用本文： | 刘旭辉,白云岗,柴仲平,张江辉,丁邦新,江柱.基于多光谱遥感的典型绿洲棉田春季土壤盐分反演及验证[J].干旱区地理,2022,45(4):1165-1175. |
| |
| 作者姓名： | 刘旭辉 白云岗 柴仲平 张江辉 丁邦新 江柱 |
| |
| 作者单位： | 1.新疆农业大学资源与环境学院,新疆 乌鲁木齐 8300522.新疆水利水电科学研究院,新疆 乌鲁木齐 8300493.西北农林科技大学水利与建筑工程学院,陕西 杨凌 712100 |
| |
| 摘 要： | 为探索快速提取典型绿洲棉田土壤盐分的有效方法,获取区域尺度的土壤盐渍化特征及空间分布,进而为土壤盐渍化防治提供参考。以新疆兵团农二师31团为研究区域,2019、2021年春季Landsat 8 OLI多光谱影像和野外实测土壤含盐量为数据源,将波段组、光谱指数组和全变量组作为模型输入变量组,采用多元逐步回归（Multiple stepwise regression, MSR）、偏最小二乘回归（Partial least squares regression, PLSR）、极限学习机（Extreme learning machine, ELM）、支持向量机（Support vector machine, SVM）和BP神经网络（Back propagation neural network, BPNN）构建基于3个输入变量组的土壤盐分遥感反演模型,探究输入变量和建模方法对模型精度的影响效果,通过对比确定春季土壤盐分最优反演模型,定量反演地表土壤含盐量。结果表明：（1） 研究区主要为非盐化土和轻度盐化土,总样本变异系数为0.67,呈中等变异性;光谱反射率与土壤盐渍化程度的关系表现为土壤盐渍化越重,光谱反射率越高。（2） 海岸波段（b1）、蓝波段（b2）、绿波段（b3）、红波段（b4）和盐分指数（SI1、SI2、SI3、SI4、S3、S4、S5）均通过显著性检验P<0.01,相关系数均达到0.4以上。（3） 所有模型中,基于全变量组建立的BPNN反演模型精度最高,建模集R2为0.705;验证集R2为0.556。（4） 由反演结果可知,2019、2021年春季耕作区土壤主要为非盐化土,分别占耕作区总面积的55.55%和64.62%,其次为轻度盐化土,分别占44.31%和35.17%;2021年土壤盐渍化程度较2019年有所减轻。
|
| 关 键 词： | 多光谱遥感反演 土壤盐分 光谱反射率 变量组 机器学习 |
| 收稿时间： | 2021-10-15 |
Inversion and validation of soil salinity based on multispectral remote sensing in typical oasis cotton field in spring |
| |
| LIU Xuhui,BAI Yungang,CHAI Zhongping,ZHANG Jianghui,DING Bangxin,JIANG Zhu.Inversion and validation of soil salinity based on multispectral remote sensing in typical oasis cotton field in spring[J].Arid Land Geography,2022,45(4):1165-1175. |
| |
| Authors: | LIU Xuhui BAI Yungang CHAI Zhongping ZHANG Jianghui DING Bangxin JIANG Zhu |
| |
| Institution: | 1. College of Resources and Environment, Xinjiang Agricultural University, Urumqi 830052, Xinjiang, China2. Xinjiang Research Institute of Water Resources and Hydropower, Urumqi 830049, Xinjiang, China3. College of Water Resources and Architectural Engineering, Northwest A & F University, Yangling 712100, Shaanxi, China |
| |
| Abstract: | This study aimed to explore an effective method for extracting soil salinity from cotton fields in oasis and determine the characteristics and spatial distribution of soil salinization on a regional scale to provide reference for soil salinization control. The 31st Regiment of the 2nd Division of Xinjiang Production and Construction Corps was taken as the research area; Landsat 8 OLI multispectral images and field measurements of soil salinity in spring 2019 and 2021 were taken as data sources; and the band group, spectral index group, and total variable group were taken as the model input variable group. Multiple stepwise regression (MSR), partial least squares regression, extreme learning machine, support vector machine, and back propagation neural network (BPNN) were used to construct a remote sensing inversion model of soil salt based on the three input variable groups. Precision evaluation was conducted, and the effects of input variables and modeling methods on model accuracy were explored. The best inversion model of soil salt in spring was determined through comparison and quantitatively inverted surface soil salt content. Results showed that (1) the study area was mainly composed of non-salinized soil and slightly salinized soil, and the coefficient of variation of total samples was 0.67, implying moderate variability. The relationship between spectral reflectance and soil salinization is as follows: serious soil salinization leads to great spectral reflectance. (2) Significance tests were conducted on coastal band(b1), blue(b2), green(b3), and red(b4) and the salinity indices of SI1, SI2, SI3, SI4, S3, S4, and S5 (P<0.01). The obtained correlation coefficients were all above 0.4, which can represent soil salinity to a certain extent. (3) Among the six linear regression models, the MSR model established on the band group had the best inversion effect, and the BPNN inversion model established on the full variable group had the highest accuracy. (4) According to the inversion results, the soil in spring 2019 and 2021 was mainly nonsalinized soil accounting for 55.55% and 64.62% of the total tillage area, respectively, followed by mild salinized soil accounting for 44.31% and 35.17%, respectively. Soil salinization in 2021 was reduced compared with that in 2019. |
| |
| Keywords: | multispectral remote sensing inversion soil salinity spectral reflectance variable group machine learning |
|
| 点击此处可从《干旱区地理》浏览原始摘要信息 |
| 点击此处可从《干旱区地理》下载免费的PDF全文 |
|
