A soil nutrient regime index for forest practitioner decisions in Hesse, Germany: spatial explicit modelling of soil chemistry and integration by fuzzy-logic

Forests face considerable pressure from climate change, while demand of provided ecosystem services is high. Managing and planting forests need well informed decisions by practitioners, to fulfill the goal of sustainability. In Germany, informed decisions are derived from forest site evaluation maps, integrating biogeoecolocigal conditions (climate, soil water, nutrients). Here, we focus on mapping of nutrients in the federal state Hesse, Germany. For Hesse, a forest site map exists, which indicates a soil nutrient regime (SNR) index (classes very poor, poor, medium, rich, very rich). Site mapping was done in the field by experts, considering ground vegetation and soil morphology. Guidelines exist for choosing management options (i.e. suitable species composition, harvest restrictions, etc.), but if spatial information is not accurate, management decisions will be misguided.

Three major challenges regarding the currently available site information exist: (1) the spatial proportion of “medium” sites is exceptionally high (65% of mapped forest area) and while there is differentiation between parent materials, topography is neglected. (2) Whereas 80% of Hesse’s forests were mapped, there is need to fill the gaps. (3) The existing SNR index does not take analytical measurements of soil nutrients into account. Objectives were (1) to refine and expand the existing map of SNR by (2) including soil chemical properties from the second National Forest Soil Inventory (NFSI), (3) which have to be regionalised beforehand.

Stocks of Ca, Mg and K, base saturation, effective cation exchange capacity (90cm depth and organic layer), and C/N ratio (organic layer or 0-5 cm) of 380 profiles from the NFSI were chosen to characterise the SNR. Regionalisation was performed with generalised additive models (GAM) by using environmental relationships of the target variables with variables of climate, vegetation, parent material and soil properties (soil map 1:50,000). Ten-fold cross validation revealed R² values from 0.54 to 0.79, with low relative root mean square deviation (5 to 17%) and slopes not significantly different from 1. From the six successfully modelled target variables, we inferred a single SNR for each soil map polygon. This was challenging, because variables provided contrasting information regarding the SNR. We addressed this by using the Soil Inference Engine (SIE), which bases on fuzzy logic. Each variable received an optimality value for each SNR class. Using an expert-driven weighting system a SNR membership was inferred, whereas highest membership defined the SNR class. The result was highly sensitive towards parent material and topography. For instance, acidic parent material had lower SNR classes compared to base rich parent material. Within a given parent material, ridges where judged less nutrient rich compared to planes and topographic positions, where material is accumulated.

The results provide a much more differentiated and complete map for SNR, which mirror actual expectations of nutrient distribution across Hesse’s landscape units. The approach is transparent and inter-subjectively reproducible. The new map will be used to guide reforestation activities in Hesse after the severe forest disturbances by recent climatic extremes (e.g. drought, storms) and the approach can be transferred to other regions.