Digital soil mapping of soil salinity using EM38 and quasi‐3d modelling software (EM4Soil)

Abstract

AbstractLand development in the form of irrigation has led to increased agricultural productivity, but natural stores of connate salt have led to salinization. To manage salinity, baseline information about the electrical conductivity of a saturated soil paste extract (ECe – dS/m) is necessary. To value add to the limited ECe that can be collected, proximal sensed data from electromagnetic (EM) induction instruments (e.g. EM38) are increasingly being used because the measured apparent soil electrical conductivity (ECa – mS/m) can be correlated with measured topsoil (0–0.3 m), subsurface (0.3–0.6 m), subsoil (0.6–0.9 m) and deep subsoil (0.9–1.2 m) ECe. However, errors may be introduced in prediction, given an EM38 measures ECa at depths of 0–1.5 m in vertical (EM38v) and 0–0.75 m in horizontal (EM38h) mode. To overcome this, we develop a linear regression (LR) between estimates of electrical conductivity (σ – mS/m), inverted from EM38v and EM38h using a quasi‐3d algorithm and measured ECe at the same depths. First, the LR was established (using R2) between estimates of σ inverted from ECa at heights of 0.5 (EM38v0.5 and EM38h0.5) and 0 m (EM38v0 and EM38h0), either alone or in combination, as well as in vertical mode (i.e. EM38v0.5 and EM38v0). ECa were also degraded (100%, 80%, 60% and 40%) to compare loss of prediction agreement (Lin's concordance) and accuracy (root mean square error). We use Normalized Difference Vegetation Index (NDVI) to qualitatively indicate crop growth. Moderate coefficient of determination (R2) was achieved between σ and ECe when we use the EM38v0.5 and EM38h0.5 (0.65) and EM38v0.5 and EM38v0 (0.69), but strong R2 was achieved using EM38v0 and EM38h0 (0.78) and in combination with the EM38v0.5 and EM38h0.5 (0.71). However, while good agreement (Lin's > 0.8) was achieved, during a leave‐one‐out cross‐validation for most EM38 combinations, the best result was achieved using EM38v0 and EM38h0 (Lin's = 0.87). There was also loss in prediction agreement and accuracy using any of the degraded ECa data sets, however. The final 3d map of ECe, as well as NDVI, showed where highly saline (>8 dS/m) areas in the west of most fields resulted as a function of leakage from the Kham‐rean Canal and topography (i.e. lower lying areas). We conclude the approach has broad application to map, potentially manage and monitor large areas of north‐east Thailand.