National-scale mapping of soil moisture and nutrient regimes in Norwegian forests

Recent drought events in the Nordic have led to an uptick in drought damage and forest mortality particularly in Norway spruce, either as consequence of the drought itself or caused by subsequent bark beetle attacks. In Norway, this has sparked a debate about existing spruce plantations on unsuitable sites, i.e. sites with insufficient moisture supply, and the need for alternative management strategies. However, identifying those sites at highest risk necessitates a high-resolution, national-scale map of soil water retention characteristics which does not exist. In order to overcome the paucity of mapped soil information relevant for forest management decisions and particularly species selection, we combine registrations from the national forestry inventory (NFI), machine learning and various landscape covariates to qualitatively map soil moisture and nutrient regimes of forest soils at national scale.

In detail, we used registrations of vegetation type from the NFI to classify all plots along seven soil moisture classes (wet to dry) and five soil nutrient classes (poor to rich) placing each plot on an edaphic grid showing relative moisture and nutrient regimes. We employed machine learning, i.e.  boosted regression tress, to develop models that predict the probability of belonging to a certain class based on an extensive set of potential predictor variables. These include high-resolution maps and data products covering climate, land cover, terrain characteristics and soil parent material as well as remotely sensed information on forest structure (airborne laser scanning) and spectral vegetation properties (Sentinel-2).  

Results showed on overall good agreement between field registrations and predicted soil moisture and nutrient class. We found that models utilizing remotely sensed information on vegetation structure and spectral properties performed significantly better than those that solely relied on climatic and physiographic information.