| 基于WorldView-2影像的土壤含盐量反演模型 |
| |
| 引用本文： | 吾木提·艾山江,买买提·沙吾提,依力亚斯江·努尔麦麦提,茹克亚·萨吾提,王敬哲.基于WorldView-2影像的土壤含盐量反演模型[J].农业工程学报,2017,33(24):200-206. |
| |
| 作者姓名： | 吾木提·艾山江 买买提·沙吾提 依力亚斯江·努尔麦麦提 茹克亚·萨吾提 王敬哲 |
| |
| 作者单位： | 1. 新疆大学资源与环境科学学院，乌鲁木齐 830046; 2. 新疆绿洲生态教育部重点实验室，乌鲁木齐 830046;,1. 新疆大学资源与环境科学学院，乌鲁木齐 830046; 2. 新疆绿洲生态教育部重点实验室，乌鲁木齐 830046; 3. 新疆智慧城市与环境建模普通高校重点实验室，乌鲁木齐 830046;,1. 新疆大学资源与环境科学学院，乌鲁木齐 830046; 2. 新疆绿洲生态教育部重点实验室，乌鲁木齐 830046;,1. 新疆大学资源与环境科学学院，乌鲁木齐 830046; 2. 新疆绿洲生态教育部重点实验室，乌鲁木齐 830046;,1. 新疆大学资源与环境科学学院，乌鲁木齐 830046; 2. 新疆绿洲生态教育部重点实验室，乌鲁木齐 830046; |
| |
| 基金项目： | 国家自然科学基金资助项目（41361016、41561089、40901163、41761077）共同资助 |
| |
| 摘 要： | 针对WorldView-2影像高空间分辨率评价其定量反演土壤含盐量的能力,以盐渍化现象较为明显的新疆克里雅河流域为研究对象,基于WorldView-2影像和实测高光谱数据,利用偏最小二乘回归(partial least squares regression,PLSR)和BP人工神经网络(back propagation artificial neural networks,BP ANN)方法建立定量反演该流域土壤含盐量模型并做出研究区高空间分辨率土壤含盐量分布图。结果表明:1)利用实测高光谱数据和影像数据分别建立的2种模型中BP神经网络模型预测精度都高于PLSR模型,其中基于影像数据建立的6:8:1结构的3层BP神经网络模型决定系数R2、均方根误差RMSE、相对分析误差RPD分别为0.851、0.979、2.337,模型的稳定性和预测能力都优于PLSR模型(R2、RMSE、RPD分别为0.814、1.139、2.007)。2)利用WorldView-2影像提高了土壤含盐量制图的空间分辨率,归一化植被指数NDVI和比例植被指数RVI较有效降低了植被覆盖与土壤水分对预测精度的影响。该文建立的考虑植被覆盖与土壤水分定量反演土壤含盐量的模型不需要复杂的参数,一定程度上满足了干旱、半干旱地区的盐渍化监测需求,可以促进WorldView-2等高空间分辨率卫星在盐渍化监测中的进一步应用。
|
| 关 键 词： | 遥感 土壤 盐分测量 WorldView-2影像 克里雅河流域 实测高光谱 神经网络 反演模型 |
| 收稿时间： | 2017/7/27 0:00:00 |
| 修稿时间： | 2017/11/30 0:00:00 |
Inversion model of soil salt content based on WorldView-2 image |
| |
| Umut Hasan,Mamat Sawut,Ilyas Nurmamat,Rukiya Sawut and Wang Jingzhe.Inversion model of soil salt content based on WorldView-2 image[J].Transactions of the Chinese Society of Agricultural Engineering,2017,33(24):200-206. |
| |
| Authors: | Umut Hasan Mamat Sawut Ilyas Nurmamat Rukiya Sawut and Wang Jingzhe |
| |
| Affiliation: | 1. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China; 2. Key Laboratory of Oasis Ecology of Ministry of Education, Urumqi 830046, China;,1. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China; 2. Key Laboratory of Oasis Ecology of Ministry of Education, Urumqi 830046, China; 3. Key Laboratory of Xinjiang Wisdom City and Environment Modeling, Xinjiang University, Urumqi 830046, China;,1. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China; 2. Key Laboratory of Oasis Ecology of Ministry of Education, Urumqi 830046, China;,1. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China; 2. Key Laboratory of Oasis Ecology of Ministry of Education, Urumqi 830046, China; and 1. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China; 2. Key Laboratory of Oasis Ecology of Ministry of Education, Urumqi 830046, China; |
| |
| Abstract: | Abstract: Soil salinization has become one of the global environmental issues, especially in arid and semi-arid areas. In order to prevent its further deterioration, it is important to monitor soil salinity timely, quantitatively and dynamically. Remote sensing technique has become a promising method to detect and monitor the soil salinity due to its many advantages. The aim of this study was to evaluate the ability of quantitative inversion of soil salt content based on the WorldView-2 images with high spatial resolution. In this paper, Keriya River basin, Xinjiang, China was selected as the study area. Based on the WorldView-2 image data and soil salt content, this paper used 2 kinds of methods including the partial least squares regression (PLSR) and back propagation artificial neural network (BP ANN) to establish the quantitative inversion models of soil salt content. Soil salinity information was extracted from the WorldView-2 data, which was synchronized with field sampling time, and covered an area of 1.2 km × 1 km. The distance between adjacent sampling points was 100 m in east-west direction, and 200 m in north-south direction. Sixty-six sampling points were designed in the study area, and digging depth in soil was 20 cm. Hand-held GPS (global position system) receiver was used to record the coordinates of sampling points, and the soil salt content and soil spectra were measured in the indoor. Spectral radiometric calibration and atmospheric correction were performed on the WorldView-2 data to match the image data with the measured re?ectance spectra. The measurement of soil spectra was conducted using an ASD (analytical spectral devices) FieldSpec3 portable spectro radiometer (American Analytical Spectral Devices, Inc.) at wavelengths from 350 to 2500 nm with a sampling interval of 1.4 nm from 350 to 1000 nm and 2 nm from 1000 to 2500 nm. The edge bands including 350-399 and 2401-2500 nm were removed from the measured spectral data, and the remaining 400-2400 nm spectrum curve was smoothed with Savitzky-Golay smoothing method in software OriginPro. Original soil spectral data were continuum-removed in ENVI 5.1 to analyze the spectral characteristics of soil. Correlation analysis between the original and two-order derivative of measured reflectance data and the soil salinity was performed by using Pearson correlation analysis method, and the significant bands were used to establish the inversion model. The geographic locations and surface re?ectance of the soil samples were obtained precisely from WorldView-2 multi-spectral data. Spectral re?ectance of each band of WorldView-2 data was simulated by calculating a weighted average of the measured re?ectance spectra to reduce the error resulted from the spectral resolution difference of the image derived spectra and measured re?ectance spectra. PLSR model was established, in which the reflectance of 4 bands i.e. B3, B4, B5 and B7 of WorldView-2 image and NDVI (normalized difference vegetation index) and RVI (ratio vegetation index) were selected as independent variables, and salt content was used as dependent variable. Three-layer BP neural network model was established in which the input layer was made up of the reflectance of 4 bands of WorldView-2 image (B3, B4, B5 and B7) and NDVI and RVI, and the number of net neurons was 6; the output layer was a neuron corresponding to the salt content of sampling point. After a lot of tentative computation, the optimal number of neurons in the hidden layer was selected as 8. The results showed that: 1) The prediction accuracy of BP neural network model based on WorldView-2 image data was higher than the PLSR model in the study area, and the coefficient of determination (R2), root mean square error (RMSE) and residual prediction deviation (RPD) were 0.851, 0.979 and 2.337 respectively for the former and 0.814, 1.139 and 2.007 respectively for the latter. 2) The spatial resolution of salinity mapping could be improved by using WorldView-2 images. The NDVI and the RVI were helpful to reduce the influence of vegetation cover and soil moisture on the prediction accuracy. This inversion model established in this paper can meet the needs of monitoring salinization in arid and semi-arid area and promote the further application of WorldView-2 high spatial resolution satellite in the monitoring of salinization. |
| |
| Keywords: | remote sensing soils salinity measurements WorldView-2 image keriya river basin measured spectral data neural network inversion models |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
|
点击此处可从《农业工程学报》下载全文 |
|
