Assessing soil nutrient spatial variability in Mexico through digital soil mapping: Implications for sustainable agriculture

The availability and spatial variability of soil nutrients strongly influence ecosystem services, food production, and sustainable land management, yet major information gaps persist at national and global scales. This study developed the first national digital maps of soil properties and nutrients for Mexico, contributing to the FAO's Global Soil Nutrient and Nutrient Budget Map (GSNmap) initiative. Using approximately 7000 harmonized soil observations at 0–30 cm depth and a comprehensive set of environmental covariates, we applied Quantile Regression Forest (QRF) models to predict soil texture (sand, silt, and clay contents), bulk density (BD), pH, cation exchange capacity (CEC), organic carbon (OC) and total nitrogen (N) contents, available phosphorus (P), and exchangeable potassium (K) and calcium (Ca) concentrations at 250 m resolution. Model performance was heterogeneous but overall robust at the national scale. Soil texture fractions, pH, CEC, OC, K, and Ca were predicted with moderate to strong accuracy (R 2 ≈ 0.50–0.77). In contrast, N and P exhibited lower predictive accuracy (R 2 ≤ 0.26), reflecting their strong sensitivity to localized management practices and land-use history, and suggesting that more intensive sampling strategies are required to capture their spatial variability. Independent hold-out validation confirmed the generalizability of model predictions across agricultural lands and grasslands. The resulting maps reveal pronounced regional contrasts. The Yucatán Peninsula is characterized by high CEC and OC but low P availability associated with karstic, carbonate-rich soils; northern regions exhibit low OC and N and increased vulnerability to structural degradation and salinization under arid, irrigation-dependent systems; and Central-South Mexico faces constraints related to soil acidity in predominantly rainfed agriculture. Differences among agricultural systems were also evident, with cultivated pastures showing higher OC and CEC than rainfed croplands, highlighting the influence of land use and management on soil functioning. By delivering spatially explicit predictions together with uncertainty estimates, this study provides a national baseline for diagnosing soil fertility constraints and supports region-specific soil management strategies. The results establish an operational and uncertainty-informed framework for soil nutrient mapping in Mexico, strengthening the linkage between national soil information systems and global initiatives to support sustainable soil governance and food security planning.