Combining proximal and remote sensors for regional soil characterization in rural Haiti

Agronomic optimization is critical in developing countries, especially where soil resources are constrained. This research, the first of its kind in Haiti, used predictive modeling to relate laboratory-derived physical and chemical soil data to proximal and remotely sensed data collected on 32,949 georeferenced surface soil (0–20 cm) samples in the Arcahaie region. A representative subset of collected samples ( n = 300) was then tested using a litany of predictive models (e.g., random forest, gradient boosting, stacking ensemble, XGBoost) relating the lab-derived to proximally sensed data for the prediction of soil pH, sand, silt, clay, soil organic matter, cation exchange capacity, soil organic carbon, and plant available P, K, Si, Fe, and Cu. Results showed that sand, silt, clay, soil organic carbon, soil organic matter and cation exchange capacity all have predictive R² of ≥0.80; predictions of soil texture components and soil organic carbon/organic matter were particularly strong. Other parameters, while still significant, were less robust. The models were used to predict the physical and chemical properties of the full dataset, then spatially interpolated to provide parameter variability maps across the region in support of agronomic optimization. Future research should work to extend the methodology successfully demonstrated herein to other regions of agronomic importance in Haiti and other developing countries. Furthermore, the approaches could be extended to three-dimensional modeling of subsoil properties elucidating optimal soil fertility in the rooting zone.