A possibility theory approach for estimating available water capacity from imprecise information contained in soil databases

Soil survey data are required for different kinds of environmental and agronomic studies, especially for estimating general soil quality indicators such as Available Water Capacity (AWC) over large areas. Generally, AWC estimates take the form of a unique value per soil class, and the uncertainty pervading these estimates is not assessed, although it is of crucial importance for the users. To overcome this problem without requiring expensive-to-measure data, we propose an alternative approach which estimates AWC from two different types of information currently available in soil databases: (i) imprecise descriptions of soil classes, i.e. synthesis profiles or qualitative descriptions of soil classes, (ii) pedotransfer functions and other statistical relations between input soil properties. The imprecise descriptions of soil classes were considered as sets of fuzzy intervals of values of soil properties and represented by possibility distributions. Statistical relations were translated into fuzzy relations of soil properties by applying a “probability-to-possibility” transformation. AWC was then estimated by means of a fuzzy interval calculus that was implemented as a Constraint Satisfaction Problem (CSP), allowing the representation of relations between the variables. AWC estimates were expressed by possibility distributions that were translated into a set of prediction intervals corresponding to different possibility levels. The approach was applied to the estimation of the AWC from soil class descriptions from three detailed soil maps in the Languedoc plain (southern France). The validation study demonstrated that the proposed approach provided reliable but fairly imprecise estimates, which allows the soil classes to be ranked with respect to the AWC. The best reliability–precision compromise was obtained at possibility level of 0.8 by using qualitative descriptions as a data source. Because it only uses currently available information of soil databases, the proposed approach can be extensively used in the future for regional applications which allow for a certain amount of (known) imprecision, or in studies aimed at defining improvements to existing soil databases.