Soil salinization resulting from shallow saline groundwater is a major global environmental issue causing land degradation, especially in semi-arid regions such as Australia. The adverse impact of shallow saline groundwater on soil salinization varies in space and time due to the variation in groundwater levels and salt concentration. Understanding the spatio-temporal variation is therefore vital to develop an effective salinity management strategy. In New South Wales, Australia, a hydrogeological landscape unit approach is generally applied, based on spatial information and expert operators, classifying the landscape in relation to landscape and climate. In this paper, a data science approach (random forest model) is introduced, based on historical groundwater quality and quantity data providing predictions in a 4-dimensional space. As a case study, we demonstrate the spatio-temporal factors impacting standing water levels (SWL) and associated salinity and predict the spatial and temporal variability in the Muttama catchment (1059 km2), in NSW, south eastern Australia. The random forest model explains 77% of the variance in the groundwater salinity (electrical conductivity) and 65% of the SWL. Spatial factors were the most significant variables determining the space-time variation in groundwater salinity and the occurrence of groundwater at the surface. Drilled piezometer depth and elevation are dominant factors controlling SWL, while salinity is mainly determined by underlying geology. The methodology in this study predicts salinity and SWL in the landscape at fine scales, through time, improving options for salinity management.
Keywords: Landscape model; Random forests modelling; South East Australia; Variability.
Copyright © 2021 Elsevier B.V. All rights reserved.