Soil type mapping using the generalised linear geostatistical model: A case study in a Dutch cultivated peatland

We present the generalised linear geostatistical model (GLGM) for soil type mapping and investigate if spatial prediction with this model results in a soil map of greater accuracy than a map obtained using a non-spatial model, i.e. a model that ignores spatial dependence in the soil type variable. The GLGM is central to the framework of model-based geostatistics. We adopted a pragmatic approach in which the five soil types in a cultivated peatland were separately modelled with a binomial logit-linear GLGM. Prediction with soil type-specific GLGMs resulted in five binomial probabilities at each prediction location, which were standardised to multinomial probabilities by selecting the soil type with maximal probability. A soil map was created from the predicted probabilities. In addition, two non-spatial models were used to map soil type. These were the multinomial logit model and the generalised linear model for Bernoulli-distributed data. Validation with independent probability sample data showed that use of a spatial model for digital soil type mapping did not result in more accurate predictions than those with the non-spatial models.

► The generalised linear geostatistical model is presented for soil type mapping. ► The model is applied to map the soil in a peatland area using real-world data. ► Predictive soil map was validated with independent probability sample data. ► Map was not more accurate than soil maps obtained by non-spatial models.