Mapping soil properties at multiple depths from disaggregated legacy soil maps in the Brittany region, France

Abstract Digital Soil Mapping (DSM) is increasingly needed to improve existing soil information and derive soil property maps at the suitable spatial resolution for sustainable soil landscape management. However, predicting several soil properties while preserving specific pedological process is a great challenge, particularly when only coarse soil information is available over large areas. Spatial disaggregation seems to be an effective technique to extract pedological information by downscaling the original information to produce soil maps at finer resolutions. In a previous study, legacy soil maps of Brittany (France) were disaggregated at a 50 m spatial resolution using the DSMART (Disaggregation and Harmonization of Soil Map Units Through Resampled Classification Trees) algorithm and pedological knowledge. The present study had two main objectives: (i) assess the preservation of the relationships between soil properties when soil properties are estimated at standard depths by applying the equal-area spline method on soil data at pedon scale, and (ii) combine disaggregated soil maps and spline-function results to estimate spatial patterns of nine soil properties for six regular soil-depth intervals down to 200 cm across Brittany, an area of 27,040 km2. To this end, soil properties were first generated for standard soil-depth intervals using spline functions. Then, for mapping soil properties at the six standard depths, weighted mean of each soil attribute was calculated for each grid cell from reference soil-property values of the three most probable predicted soil types. Their associated probabilities of occurrence were used as weights. To assess the ability of spline functions to preserve soil-property relationships, multiple statistical analyses were performed using original and splined soil datasets. Bivariate and multivariate analysis highlighted that spline functions preserved soil-property relationships. Derived digital soil maps showed strong spatial patterns: SOC and silt contents generally decreased with depth, while sand content and coarse fragment percentage consistently increased with depth. In addition, experimental semivariogram analysis of SOC content showed high spatial variability over short distances for all soil-depth intervals except the deepest (100–200 cm), while silt content showed high semivariance for the deepest soil layers. This study can be considered an example of harmonization to common output specifications, which generates a geo-database of quantitative soil properties that describe lateral and vertical soil variation for regular depth intervals. These predictions can be incorporated into environmental models to help decision makers manage landscapes.