The thickness of Australia's regolith – the weathered mantle overlying bedrock – varies markedly, from near-zero on residual uplands to over a kilometre in palaeovalleys. Although knowledge of regolith thickness is critical for land and water management, it remains poorly constrained at decision-making scales. Predicting regolith depth is challenging due to spatial heterogeneity, legacy surface-process imprints, subsurface weathering histories and limited high-resolution subsurface data. However, machine learning approaches applied to increasingly available terrain, lithology, geochemistry, vegetation and climate datasets offer new opportunities to estimate regolith depth.
This study used Gradient Boosted Machine (GBM) to model regolith thickness across ∼28,800 km2 of the Southern Zone of Rejuvenated Drainage (SZRD) in Western Australia. A compilation of 1568 regolith thickness observations from drillholes and passive seismic surveys was used to train and test the model, generating a spatially explicit prediction. The model incorporated 37 environmental covariates spanning geological, climatic and terrain-related domains. Predicted regolith thickness varied from 0.00 to 27.53 m, with the deepest profiles concentrated in broad palaeovalleys in the east. Regolith thickness was most strongly influenced by radiometric potassium, lithological age, rainfall, slope height and Height Above Nearest Drainage. Model accuracy was weak (R2 = 0.27; RMSE = 11.76 m), reflecting the region's complex geomorphic and weathering history. Nonetheless, the map delineates major zones of regolith deposition and stripping, identifies key environmental drivers and provides a first-order framework for groundwater recharge modelling, mineral exploration and land suitability assessment. Future work should aim to obtain more spatially balanced regolith depth samples to address persistent challenges posed by biased and correlated ground data.
扫码关注我们
求助内容:
应助结果提醒方式:
