Comparative regional-scale soil salinity assessment with near-ground apparent electrical conductivity and remote sensing canopy reflectance

Soil salinity is recognized worldwide as a major threat to agriculture, particularly in arid and semi-arid regions. Producers and decision makers need updated and accurate maps of salinity in agronomically and environmentally relevant ranges (i.e., <20 dS m⁻¹, when salinity is measured as electrical conductivity of the saturation extract, EC_e). State-of-the-art approaches for creating accurate EC_e maps beyond field scale (i.e., 1 km²) include: (i) Analysis Of Covariance (ANOCOVA) of near-ground measurements of apparent soil electrical conductivity (EC_a) and (ii) regression modeling of multi-year remote sensing canopy reflectance and other co-variates (e.g., crop type, annual rainfall). This study presents a comparison of the two approaches to establish their viability and utility. The approaches were tested using 22 fields (total 542 ha) located in California’s western San Joaquin Valley. In 2013 EC_a-directed soil sampling resulted in the collection of 267 soil samples across the 22 fields, which were analyzed for EC_e, ranging from 0 to 38.6 dS m⁻¹. The ANOCOVA EC_a-EC_e model returned a coefficient of determination (R²) of 0.87 and root mean square prediction error (RMSPE) of 3.05 dS m⁻¹. For the remote sensing approach seven years (2007–2013) of Landsat 7 reflectance were considered. The remote sensing salinity model had R² = 0.73 and RMSPE = 3.63 dS m⁻¹. The robustness of the models was tested with a leave-one-field-out (lofo) cross-validation to assure maximum independence between training and validation datasets. For the ANOCOVA model, lofo cross-validation provided a range of scenarios in terms of RMSPE. The worst, median, and best fit scenarios provided global cross-validation R² of 0.52, 0.80, and 0.81, respectively. The lofo cross-validation for the remote sensing approach returned a R² of 0.65. The ANOCOVA approach performs particularly well at EC_e values <10 dS m⁻¹, but requires extensive field work. Field work is reduced considerably with the remote sensing approach, but due to the larger errors at low EC_e values, the methodology is less suitable for crop selection, and other practices that require accurate knowledge of salinity variation within a field, making it more useful for assessing trends in salinity across a regional scale. The two models proved to be viable solutions at large spatial scales, with the ANOCOVA approach more appropriate for multiple-field to landscape scales (1–10 km²) and the remote sensing approach best for landscape to regional scales (>10 km²).