High-resolution mapping of acid sulfate soils in Northern Australia through predictive models

Acid sulfate soils can form when pyrite-rich marshes are drained or tidal influence decreases, naturally or anthropogenically. Often they are vegetated by mangroves, serve as important aquatic bird habitats, and harbor unique microbial biodiversity. They limit coastal development potential. The current best map of Australian acid sulfate soils is based on expert interpretation of soil surveys, most at 1:100,000 scale. In the current study, we attempted the first predictive spatial modeling to map acid sulfate soils and soil sulfur concentration on a 90-m grid. We use 34 national environmental 90-m grids and 614 observations on the presence/absence of acid sulfate soils to train a machine learning algorithm, random forests, to predict the location of acid sulfate soils across three regions of Northern Australia. We use the same environmental data and 1315 measurements of total elemental sulfur at 60–80 cm depth across Australia to train another random forest model to predict the spatial pattern of sulfur concentrations in the three regions. The model for acid sulfate soils is better than that for sulfur but both related maps showed that high sulfur concentrations generally coincide with the predicted presence of acid sulfate soils. This coincidence between the predicted occurrence of acid sulfate soils and sulfur concentrations confirms the modeling approach presented here was reliable and could be applied to map acid sulfate soils at high resolution globally.