Soil salinity assessment using vegetation indices derived from Sentinel-2 multispectral data. application to Lezíria Grande, Portugal

• Main Author: Ramos, Tiago B.
• Publication Date: 2020
• Other Authors: Castanheira, Nádia, Oliveira, Ana R., Paz, Ana Marta, Darouich, Hanaa, Simionesei, Lucian, Farzamian, Mohammad, Gonçalves, Maria C.
• Format: Article
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10400.5/22004
• topic: canopy response salinity index; soil-environmental covariates; multi-year maximum; multiple regression analysis; rootzone salinity

Lezíria Grande is an important agricultural area in Portugal, prone to waterlogging and salinity problems due to the influence of estuarine tides on groundwater dynamics. Simple, non-invasive, practical approaches are need for monitoring soil salinity in the region and preventing further degradation of soil resources. The objective of this study was to develop regression models for soil salinity assessment in Lezíria Grande based on the relationship between multi-year crop reflectance data derived from Sentinel-2 multispectral imagery and rootzone salinity. Nine vegetation indices (VI), computed from the annual averages of the spectral bands, were tested between 2017 and 2019. The multi-year maximum from each pixel was then used for correlating the VI with the ground-truth dataset. This dataset was composed of average values of the electrical conductivity of the soil saturation paste extract (ECe mean) measured in 80 sampling sites (0–1.5 m depth) located in four agricultural fields representative of the salinity gradient in the region. The Canopy Response Salinity Index (CRSI), which uses the blue (490 nm), green (560 nm), red (665 nm), and infrared (842 nm) bands, provided the strongest correlation with measured data (r=−0.787). Regression models further considered vegetation cover and soil type as explanatory variables, with predictions resulting in a coefficient of determination (R2) ranging from 0.63 to 0.91 and a root mean square error (RMSE) varying from 1.63 to 3.26 dS m−1. The use of remote sensing data for soil salinity assessment showed to be an interesting option to consider in future soil monitoring programs. Nevertheless, more detailed covariates are needed for improving salinity assessment models at the regional scale