Multi-level statistical soil profiles for assessing regional soil organic carbon stocks

• Like for many regions, the spatial density of legacy soil data in Flanders is heterogeneous
• Multi-level generalisation approaches are able to deal with this heterogeneity
• For Flanders we obtain a lower but more realistic estimate of the baseline SOC-stock
• Imposing more observations for selecting the generalisation level leads to lower estimates and higher uncertainties

To support the assessment of soil-related ecosystem service delivery throughout an area of interest, the information content of soil maps needs to be extended with quantitative data about e.g. the soil organic carbon content. Such data are typically provided by soil profile descriptions with associated horizon data. However, the magnitude and spatial density of available profile collections are mostly too limited to allow the coupling of at least one profile description with each soil–land cover combination, further termed land unit (LU). To overcome this lack-of-data-problem, spatial generalisation and modelling approaches have been developed. However, these approaches do not take full advantage of all available data when their density is heterogeneous across the territory. We present a multi-level, both spatial and semantic, generalisation approach to work with detailed LUs in regions where sufficient data is available and with generalised units elsewhere.Applied to the soil map of Flanders, Belgium (18,809 LUs for 9295 km2), a database of 7020 profile descriptions with 42,529 horizons and a 3- (coastal region) and 5-level (non-coastal region) generalisation approach, the method succeeded in assigning at least one profile to 18,731 LUs or 98.7% of the territory. Increasing the minimum required number of matching profiles increased the number of LUs characterised at higher levels of generalisation. Using the horizon characteristic ‘organic carbon content’, we estimated the total historical SOC-stock in Flanders at 87.49 Mt C and found that increasing the minimum required number of profiles decreased this estimate and increased its uncertainty. The assessed stock is lower than the ones found in previous studies even though they are based on the same collection of soil profiles. We consider our results as more realistic since the proposed method preserves the highest possible spatial and semantic detail.