High-resolution mapping of soil pH for Denmark

Soil pH exhibits significant spatiotemporal variability due to natural factors (e.g., climate, terrain) and human activities (e.g., land use, soil management). Key drivers include soil texture, carbon content, vegetation, slope, and climate. Understanding this variability is essential for soil management. High-quality and readily interpretable soil pH maps are also needed as a covariate to investigate their relationships with complex soil properties at finer resolutions

Existing soil pH maps often lack the fine resolution required for field-scale applications, typically providing resolutions between 90 m and 250 m. To address this gap, we aimed to: (1) generate a high-resolution (10 m) soil pH map of Denmark with uncertainty estimates using machine learning, and (2) identify the main factors influencing soil pH variability across different land uses.

We analyzed 7000 topsoil samples (0–20 cm) collected from natural and agricultural sites. Soil pH was measured in H₂O and 0.01 M CaCl₂, with the delta pH calculated as their difference. Environmental layers at 10-m resolution, representing soil properties, climate, vegetation, and geomorphology, were harmonized as model inputs. We employed quantile regression forest models, splitting the dataset into 70% training and 30% testing for validation. Model accuracy was assessed using normalized root mean square error, concordance correlation coefficient, and R². We analyzed how environmental factors control the pH and the delta pH using the SHAP (SHapley Additive exPlanations) algorithm.

The pH measured in CaCl₂ achieved the highest model performance, followed by H₂O pH and delta pH. The SHAP analysis highlighted the factors driving pH variability in natural versus agricultural settings. Soil texture, climate variables, and oxalate-extractable Fe and Al were the strongest predictors. Topographical parameters related to hydrological properties also impacted the spatial distribution of the response variables. Our results indicate that soil properties and topographical features had a higher contribution than remote sensing indices representing vegetation growth patterns. This hierarchy of influence suggests that while remote sensing data is valuable, it should be complemented by high-quality topographical and soil data for optimal pH mapping outcomes.

The methodology used in this research allowed us to establish the environmental covariates affecting pH variation. These fine-resolution maps serve as valuable tools for mapping other soil properties (e.g. ion exchange capacity, soil microbial diversity, etc.), enhancing therefore, agricultural management and planning to achieve healthy soils.