Estimating soil salinity with different fractional vegetation cover using remote sensing

Soil salinization is a serious restrictive factor affecting sustainable agricultural development. In order to explore the effect of Fractional Vegetation Cover (FVC), we monitored soil salinization in sites different vegetation coverage in Jiefangzha Irrigation District in Inner Mongolia using satellite remote sensing. From May to August 2018, we carried out field sampling at different depths in each month, and calculated FVC and spectral covariates using GF-1 satellite images in the corresponding sampling period. Based on the FVC division criteria for Inner Mongolia, we took the following steps: (a) setting up a control treatment A (the full data with undivided FVC, TA) and experimental treatments B (bare land, TB), C (mid-low FVC, TC), D (mid FVC, TD) and E (high FVC, TE); (b) conducting the Best Subset Selection (BSS) for all spectral covariates at different depths of each treatment; and (c) constructing the Soil Salt Content (SSC) inversion models using partial least square regression (PLSR), Cubist, and Extreme Learning Machine (ELM). The results indicated that (a) classifying FVC could improve the stability and predictive ability of the models; (b) the performance of the three modeling methods were different (Cubist was the best, ELM next and PLSR the poorest); (c) the optimal inversion models for TB, TC and TE were constructed by Cubist at 0-20, 0-40 and 0-20 cm, and for TD was constructed by ELM at 0-60 cm, respectively. The results can provide references for soil salinization prevention and agricultural production in Jiefangzha Irrigation District and other areas with the similar vegetation cover.

Automated summary

This study investigated the impact of Fractional Vegetation Cover (FVC) on soil salinization using satellite remote sensing in Inner Mongolia. The results showed that classifying FVC improved the stability and predictive ability of the models, with Cubist performing the best, followed by ELM and PLSR. Optimal inversion models were constructed for different FVC categories and depths, providing valuable references for soil salinization prevention and agricultural production in the study area.