Estimation of soil salt content using machine learning techniques based on remote-sensing fractional derivatives, a case study in the Ebinur Lake Wetland National Nature Reserve, Northwest China
Abstract
Soil salinity is a common global environmental problem that severely restricts industrial, agricultural and human development. In Northwest China, soil salinity is a problem affecting the Lake Ebinur area and needs to be monitored and addressed. The use of optical remote sensing technology to for timely and accurate soil salinity monitoring has great potentials and can be crucial for industrial and agricultural development. Optical remote sensing technology is an important data source for monitoring of soil salinization because of its rich spectral information and high-resolution. Based on the HJ-1/HSI data of fractional derivative transformation, this paper calculated the parameters, such as the difference index (DI), normalized difference index (NDI), ratio index (RI) and salt index (SI), that can invert the spatial distribution of salinization and then calculates a particle swarm optimization support vector machine (PSO-SVM). These index parameters were used to invert the spatial distribution information of salinized soil to explore the influence of fractional derivatives on the estimation accuracy of the spatial distribution of salinized soil. The results showed the following: (1) The optimal model based on different fractional derivatives and exponential transformations was the DI transformation under the fractional derivative of 1.2-order. R2, RMSE and RPD were 0.66, 13.81 g/kg and 2.59, respectively. These data show that the model had a good prediction effect. (2) The application of fractional derivatives in the processing of hyperspectral images could highlight the spectral details of the images and enrich their pretreatment methods, which is of great significance for the inversion of the distribution of salinized soil by hyperspectral images. (3) In the lakeside area of Lake Ebinur, the soil is mainly saline to severely saline. In the northern mountainous region of Lake Ebinur, the soil is mainly non-saline soil to mildly saline soil. Finally, in the Kuitun riverbank area in the southeast of Lake Ebinur, the soils are mainly non-saline soils to mildly saline soils. The results of this study demonstrate the potential of spectral index optimization and prediction models to monitor soil salinization based on fractional derivatives. This technique highlights the spectral detail information and enriches the preprocessing methods of hyperspectral imagery, resulting in the mapping of the spatial distribution of the degree of soil salinization. This has practical importance for the prevention and control of soil salinization and its spatial extent in arid and semi-arid regions.
Publication Title
Ecological Indicators
Recommended Citation
Wang, Z., Zhang, X., Zhang, F., Chan, N., Kung, H., Liu, S., & Deng, L. (2020). Estimation of soil salt content using machine learning techniques based on remote-sensing fractional derivatives, a case study in the Ebinur Lake Wetland National Nature Reserve, Northwest China. Ecological Indicators, 119 https://doi.org/10.1016/j.ecolind.2020.106869
