| Remote sensing modeling of soil salinization information in arid areas |
|
View Fulltext View/Add Comment Download reader |
| |
| DOI：10.7606/j.issn.1000-7601.2018.01.39 |
| Key Words: soil salinization comprehensive spectral index multiple linear regression model partial least squares regression model support vector machine regression model |
| Author Name | Affiliation | | FENG Juan | College of Resource and Environmental Science, Key Laboratory of Oasis Ecology of Education Ministry, Xinjiang University, Urumuqi, 830046, China | | DING Jian-li | College of Resource and Environmental Science, Key Laboratory of Oasis Ecology of Education Ministry, Xinjiang University, Urumuqi, 830046, China | | YANG Ai-xia | College of Resource and Environmental Science, Key Laboratory of Oasis Ecology of Education Ministry, Xinjiang University, Urumuqi, 830046, China | | CAI Liang-hong | College of Resource and Environmental Science, Key Laboratory of Oasis Ecology of Education Ministry, Xinjiang University, Urumuqi, 830046, China |
|
| Hits: 1499 |
| Download times: 735 |
| Abstract: |
| Taking Weigan-Kuqa River Delta Oasis located in the northern margin of the Tarim Basin in Xinjiang as study area. Found out 15 salt indexes and 5 spectral vegetation indexes from the images in GF-1 and Landsat8 OLI data. By means of gray relational analysis, the soil salinity and image spectral index of 0～10 cm soil layer were analyzed and screened comprehensively, and the comprehensive spectral index with high correlation with soil salinity was further determined. Using multivariate linear regression, partial least-squares regression and support vector machine regression to construct the soil salinity comprehensive index estimation model based on the measured data and the image data respectively for GF-1 and Landsat8 OLI images, and then select the optimal model.The results: (1) SR, CSRI, SI, BI, S6, ARVI, SAVI, NDSI were the higher of all 20 spectral indices and the correlation coefficients were above 0.7, and build a composite spectral index based on the eight spectral indices. (2) Three kinds of estimation models: the coefficient R2 of multiple linear regression model based on GF-1 is 0.6856, which is higher than that of the model constructed based on Landsat8 OLI R2 0.5142; 1-8 partial least squares regression principal components, GF-1 coefficient of determination: 2>3>1, 2 principal components reached 0.6104. Landsat, the coefficient of determination: 4>3>2, 4 principal components up to 0.549; support vector machine model of the 3 kinds of function, the GF-1 coefficient ofdetermination RBF>Polynomial>Linear, RBF function is up to 0.7969. Landsat, the coefficient of determination Polynomial>RBF>Linear, Polynomial function is up to 0.7154. Thethree models show that the R2 of support vector machine regression model is the highest, so the model is better than multiple linear regression and partial least squares regression for the estimation of soil salinization. |
|
|
|