Disaggregating conventional soil maps with limited descriptive data: A knowledge-based approach in Serra Gaucha, Brazil

Conventional soil surveys remain as a major source of information about soils and their properties, but are no longer suited for new demands requiring finer soil data. Spatial disaggregation has been proposed as a way to enrich the resolution of such inventories, in order to better fit the needs for each application. In general, this involves the allocation of individual soil types or their properties within map units based on rules developed from contextual information stored in digital soil geographical databases. However, many countries do not have such soil data infrastructures, thus requiring another strategy for spatially distinguishing individual soils within the polygons of conventional soil maps. Focusing this reality, we developed a disaggregation exercise in a study area located in the wine production zone Serra Gaucha, Northeast of State Rio Grande do Sul, southern Brazil. We proposed a knowledge-based approach to go around the scarcity of descriptive information, using a heuristic selection of typical sites for each map unit component, based on the survey report and on field observations. A subset of these typical locations was then employed to build a decision tree and to predict individual soil types in the whole study area, using a set of 21 environmental covariates. The spatial detail in predicted maps was considerably improved in relation to the original, while number, proportions and extents of predicted soil types per map unit agreed with the correspondent in the survey report at rates above 70%. Total extent of each soil type was close to the computed from the original, and results have revealed probable inconsistencies in informed proportions of two similar soils. In comparison to a set of 233 independent pedons, the predicted soil types matched that of the validation points better than each component of the original map units. Furthermore, when applying a framework for assessing soil suitability for viticulture, the predicted maps appeared much more coherent with current land use than when using the original polygons. We concluded that the proposed method is a promising option for disaggregating conventional soil maps with poor descriptive data, and that estimates made by experienced soil surveyors could easily fill in contextual information gaps. (c) 2016 Elsevier B.V. All rights reserved.