Bioclimatic variables as important spatial predictors of soil hydraulic properties across Australia's agricultural region

Abstract

Soil water directly or indirectly affects almost all ecological processes. Soil available water capacity (AWC), the difference between field capacity, or drained upper limit (DUL), and wilting point, or lower limit (LL15), and saturated water content (SAT) are among the most important soil hydraulic properties controlling soil water dynamics. These properties vary across space and are expensive to measure directly. It is difficult to obtain reliable estimates of soil hydraulic properties at an appropriate scale for water and land management. Here we modelled LL15, DUL, SAT and AWC measurements from 1127 whole-soil profiles across Australian agricultural areas with the Random Forest machine learning model using 19 bioclimatic and 15 topographical covariates. The amount of variance explained by the model reached up to R2 = 0.69 depending on the property and soil depth assessed. For all soil hydraulic properties, the bioclimatic variables alone contributed to more than 90% of the explained variance. Particularly, temperature of driest and wettest quarter, and precipitation of warmest month were the three most influential variables. Using the derived models, we also mapped the four hydraulic properties across Australian agricultural areas in six sequential depths down to 2 m at a spatial resolution of 90 m. Moreover, we combined our mapping of AWC with existing products via an ensemble model averaging approach which proved to be more accurate than each of the three contributing products. Our results uncover the significant role of bioclimatic variables in regulating soil hydraulic properties, providing a benchmark assessment of soil hydraulic properties in agricultural regions for efficient water-related land management.