Identifying microbial predictors of soil nitrate pools across tropical land uses with machine learning

Nitrification is a key control on soil nitrate (NO₃⁻) pools, and yet the dominant microbial taxa driving the process may vary with land use and land management practices. In this study we test whether dominant nitrifiers (eg: autotrophic vs heterotrophic; bacterial vs fungal) differ between heavily managed tropical soils (urban farms, golf courses) and natural tropical forests in Singapore,  using machine learning to identify the microbe groups most strongly associated with soil NO₃^- pools across sites.

We collected soils across multiple sites in each land-use and quantified soil NO₃⁻ using ion chromatography. To estimate taxon-level abundances, we combined qPCR-derived total bacterial and fungal abundances (16S/18S) with ribosomal DNA Amplicon sequencing relative abundances, using their product as a proxy for genus-level absolute abundance. We compiled a list of canonical ammonia oxidisers and microbes with reported heterotrophic nitrifying strains, and evaluated their ability to predict spatial variation of NO₃⁻ within each land-use type. This was done using three flexible models (generalised additive model, support-vector regression and random forest), where model performance was assessed using R² obtained from leave-one-out and repeated 5-fold cross-validation (200 repeats).

In managed soils, bacterial genera were consistently the strongest predictors of NO₃⁻ (across all models), including the canonical AOB genus Nitrosomonas and bacteria with reported heterotrophic nitrifying strains (Paenibacillus, Rhodococcus). Predictive performance was high across all model types (R² ≈ 0.6–0.85). In forests, fungal genera (notably Aspergillus and Fusarium) ranked highest, but overall predictive performance was lower (R² ≈ 0.3–0.5), suggesting that functional groups not captured by the current candidate set (e.g., ammonia-oxidising archaea) might potentially be driving nitrification in these sites. Further analysis on this is currently in progress

Overall, our results suggest that contrasting nitrifier niches exist in different land uses with bacteria-dominated predictors in managed soils and fungal predictors in forests, which highlights how management may restructure microbial pathways that govern nitrate formation in tropical soils.

Acknowledgements

This research grant is funded by the Singapore National Research Foundation under its Competitive Funding for Water Research (CWR) initiative and administered by PUB, Singapore’s National Water Agency. We also acknowledge NParks, for providing us site access to conduct the measurements.