2026 Advancing Precision Conservation through Machine Learning for Field-Level Prioritization of Agricultural Best Management Practices

The graphical abstract. It depicts the ACPF geospatial mapping, machine learning workflow, and the resulting prioritized BMPs.

Resource link: https://www.sciencedirect.com/science/article/pii/S2095633926000602

Bodrud-Doza et al. (2026) integrate ACPF with machine learning to address a core limitation of ACPF: ACPF maps BMP opportunities everywhere but provides no quantitative prioritization guidance, flagging 95.9% of fields as needing intervention in a 205 km² tile-drained Ontario watershed. They developed a field-surveyed Composite BMP Need Score (CBNS) and trained XGBoost and Random Forest models on 19 ACPF-derived geospatial predictors to predict conservation need across all 627 fields in the watershed. Hydrologic connectivity and erosive force were the strongest drivers. The result is a transparent, transferable framework that identifies the top 20% of fields for priority BMP targeting and pairs predictions with uncertainty maps so practitioners know where to act with confidence.

_{ACPF HUB is supported through Cooperative Agreement # NR213A750008C007 with the USDA Natural Resources Conservation Service. Funding for the development of new tools and capabilities for the ACPF is provided, in part, by the USDA Natural Resource Conservation Service (NRCS) Conservation Effects Assessment Project- Watershed Assessment Studies under the ACPF Enhancement Project, NRCS Interagency Agreement # NRC21IRA0010883 with USDA ARS.}