Harnessing the power of categorical maps for spatial modelling – a case study for soil type using Maxent

Many potentially useful environmental maps (e.g. of vegetation or soil type) exist in the form of categorical maps that partition the landscape using polygons or raster cells with categorical descriptions. However, the richness of this map content is often unrealisable in environmental modelling due to the data models for the categories being unable to reflect continuous change in the environment. The objective of this study was to derive continuous raster replacements for a categorical map that would allow its content to be used for environmental modelling. The case study demonstrates the development of soil layers as predictors for Maxent that are synthesised from a geological map of soil type combined with global rasters of soil properties. Using data fusion, the categorical soil map is first transformed into a set of discrete ordinal rasters by numerically quantifying the soil categories for each soil property. Uncertainty is then introduced by adding Gaussian noise. The method produced high-resolution soil property rasters that each incorporated the detailed local environmental knowledge embodied in the categorical map. The new predictors proved highly effective for the Maxent modelling task and also produced similar results when used with other species distribution modelling algorithms. This approach of synthesizing continuous predictors using a detailed categorical map as a guide is likely to be effective for a wide range of categorical maps in combination with Earth observation and related data products and provides a new method by which categorical map content can be effectively incorporated into environmental models.