Regional soil organic carbon stocks: Enhancing the usability of soil databases

Storage of soil organic carbon (SOC) is an essential function of ecosystems underpinning the delivery of multiple services to society: e.g. the production of food, wood, bio-energy and water and the regulation of the global climate, nutrient availability and the hydrological cycle. Whereas SOC content (%) is part of the routine measurements in general-purpose soil survey projects and in monitoring schemes of agricultural and forest soils, the resulting datasets are not readily fit for regional SOC stock (kg m-2) assessments. This dissertation focused on two bottlenecks: (i) the spatial density of available soil profile collections is mostly too limited and too heterogeneously distributed across the territory to assign at least one observation to each combination of soil type and land cover type, and (ii) the sampling depth is often limited to the topsoil even though a considerable fraction of the total stock is known to be stored in the subsoil. To overcome the lack-of-data-problem related to the limited spatial density of existing soil profile collections, a multi-level generalisation approach was applied. This approach takes full advantage of all available data, by working with detailed land units in regions where sufficient data is available and with generalised units elsewhere. Applied to land units defined by the intersection of the 1990 Corine Land Cover dataset and the soil map of Flanders, Belgium (18,809 non-built-up land units for 929,546 ha), a legacy database (1947 – 1974) of 7,020 profile descriptions with 42,529 horizons and a 3- (coastal region) and 5-level (non-coastal region) generalisation, the approach succeeded in assigning at least one profile to 18,731 land units or 98.71% of the territory’s area. This resulted in an estimated legacy stock of 87.49 Mt OC in the upper 100 cm of soil. This stock is smaller than those revealed by previous assessments even though they are based on the same soil profile collection. Since the proposed method preserves the highest possible spatial and semantic detail, its result can be considered as closer to reality. The applicability of the multi-level spatial generalisation in more data-scarce conditions was confirmed by its application to the soil map of forests in Flanders (153,544 ha) and a collection of 276 forest soil profiles. Approximately 90.45% of the forest area could be characterised and a total stock of 25.31 Mt OC was obtained with a firm contribution of Histosols. However, for the particular case of the forests of Flanders, a digital soil mapping approach using boosted regression trees was more informative. Additionally compared to multiple linear regression, artificial neural networks and least-squares support vector machines, it obtained the best fit (training R² of 0.68 and cross-validated R² of 0.22) and moreover provided insights in the soil system by showing average predictor effects in partial dependence plots. Highest groundwater level, clay fraction, tree genus and soil type were identified as key predictors of the SOC stock in the upper 100 cm. With boosted regression trees, the total SOC stock in forest soil was estimated at 26.99 Mt OC. Also using boosted regression trees and assuming that the potential natural dominant tree genus would occur according to the present soil conditions, the SOC stock in the current forest area was estimated at 30.00 Mt OC, or 21.26 kg OC m-2 on average. When the complete non-built-up territory (1,168,850 ha) would be forested analogously, 255.28 Mt OC would be stored, which is more than double the estimated stock under the actual land cover distribution. The results highlight the importance to conserve and restore carbon hotspots like alluvial forests. New soil inventories should focus on these and other data-scarce land units. Future modelling work can benefit from explicitly taking the soil type and tree genus into account as predictors. The secondly addressed bottleneck was the limited sampling depth, typically the upper 15 to 30 cm of soil. To assess SOC stocks and their changes in the upper 100 cm of the soil profile, vertical extrapolation of topsoil measurements is necessary. The commonly used exponential decline function is not valid, however, for soil types in which subsurface horizons with a larger SOC content, ‘anomalies’, occur. To account for these profile anomalies, an exponential change decline function was conceived and calibrated, assuming that not the SOC content, but rather its change over time declines exponentially with depth. To optimize calibration, more detailed descriptions of subsoil reference profiles, sampled by pedogenetic horizon rather than by fixed depth interval are needed. Applied to 54,041 agricultural land units in Flanders it was possible with this function to model specific profile characteristics such as the presence of spodic horizons, plaggic topsoil and peat substrates, resulting in total arable and grassland SOC stock estimates of 40.44 Mt OC in 388,572 ha and 39.70 Mt OC in 315,074 ha, respectively. For the particular land units characterised by SOC-rich subsoil horizons, the exponential decline function underestimated SOC stocks, which compromised an in-depth assessment of changes in SOC stocks over time. To assess the SOC stock under Low-Input High-Diversity systems (30,556 ha), such as non-forested nature reserves, both bottlenecks related to sampling depth and spatial density needed to be overcome. To this end, depth extrapolation of 139 topsoil (upper 15 cm) measurements using the exponential change decline function was combined with digital soil mapping. Also in this situation, boosted regression trees was the superior modelling approach as it resulted in the lowest cross-validation errors and provided insights in the determining variables of both top- and subsoil stocks. The predictors of the stock in the upper 100 cm were soil type, lowest groundwater level, clay fraction and plant height. These predictors extended with variables describing the site’s slope, specific leaf area, aboveground biomass production, rooting depth, mycorrhizal associations and species diversity also largely explained the topsoil stock variation. The results showed that the topsoil (1.63 Mt OC) stored but 36% of the stock found in the upper 100 cm (4.53 Mt OC). Given the magnitude of subsoil OC and its dependency on typical ecosystem characteristics, it should not be neglected in regional ecosystem service assessments. By combining the abovementioned SOC stock assessments for the agricultural domain, forests and LIHD systems, the SOC stock for 887,745 ha of the non-built-up area in Flanders was estimated to be 111.67 Mt OC or 12.58 kg OC m-2 on average. Soils under more natural land use types including forests (17.58 kg m-2) and LIHD systems (13.72 kg m-2) on average stored more organic carbon than agricultural soils: arable (10.40 kg m-2) and grassland (12.60 kg m-2). Given their large area fraction, however, the latter contain the largest absolute stock. From a final study on a 954 ha study area it became clear that region-wide soil datasets as the ones used in our research have a rather weak performance in the identification of carbon hotspots such as Histosols. This finding shows that local surveys with more dense sampling are necessary when the SOC stock assessments are to serve carbon-aware land management rather than regional reporting.

Jaar van publicatie:2017

Toegankelijkheid:Closed