Modelling effective soil depth at field scale from soil sensors and geomorphometric indices

Abstract

The effective soil depth (ESD) affects both dynamic of hydrology and plant growth. In the southeast of Buenos Aires province, the presence of petrocalcic horizon constitutes a limitation to ESD. The aim of this study was to develop a statistic model to predict spatial patterns of ESD using apparent electrical conductivity at two depths: 0-30 (ECa_30) and 0-90 (ECa_90) and geomorphometric indices. To do this, a Random Forest (RF) analysis was applied. RF was able to select those variables according to their predictive potential for ESD. In that order, ECa_90, catchment slope, elevation and ECa_30 had main prediction importance. For validating purposes, 3035 ESD measurements were carried out, in five fields. ECa and ESD values showed complex spatial pattern at short distances. RF parameters with lowest error (OOBerror) were calibrated. RF model simplified which uses main predictors had a similar predictive development to it uses all predictors. Furthermore, RF model simplified had the ability to delineate similar pattern to those obtained from in situ measure of ESD in all fields. In general, RF was an effective method and easy to work. However, further studies are needed which add other types of variables importance calculation, greater number of fields and test other predictors in order to improve these results.