Developing geographic weighted regression (GWR) technique for monitoring soil salinity using sentinel-2 multispectral imagery
- Publisher:
- SPRINGER
- Publication Type:
- Journal Article
- Citation:
- Environmental Earth Sciences, 2021, 80, (3)
- Issue Date:
- 2021-02-01
Closed Access
| Filename | Description | Size | |||
|---|---|---|---|---|---|
| Developing geographic weighted regression (GWR) technique.pdf | Published version | 21 MB |
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
Soil salinity is a widespread natural hazard that negatively influences soil fertility and crop productivity. Using the potential of earth observation data and remote sensing technologies provides an opportunity to address this environmental issue and makes it possible to identify salt-affected regions accurately. While most of the utilized methods and model development techniques for monitoring soil salinity to date have been globally considered and tried to detect salinity and create predictive maps with a single regression algorithm, fewer studies have investigated the potential of local models and weighted regression techniques for estimating soil salinity. Accordingly, this research deals with monitoring surface soil salinity by the potential of Sentinel-2 multispectral imagery using the geographic weighted regression (GWR) technique. The field study was conducted in an area that has suffered from salinization, and the salinity of several soil samples was measured to be used as a source of ground truth data. The most efficient satellite features, which accurately predict surface soil salinity by its higher spectral reflectance, were derived from the Sentinel-2 data to be used as explanatory variables in the analysis. The GWR algorithm was then implemented with a fixed Gaussian kernel, and the optimized bandwidth was calculated in a calibration process using the cross-validation score (CV score). The results of the analysis proved that the GWR method has a great capability to predict soil salinity with an accuracy of two decimal places. The visual interpretation of the local estimates of coefficients and local t-values for each predictor variable has also been provided, which highlights the local variations in the study site. Finally, the achieved results were compared with the outcomes obtained from implementing two global regression techniques, Support Vector Machines (SVM), and Multiple Linear Regression (MLR), which confirmed the higher performance of the GWR algorithm.
Please use this identifier to cite or link to this item: