Title: Combining depth extrapolation and digital soil mapping to assess soil organic carbon stocks: what you get is more than what you measure (ECOPLAN)
Authors: Ottoy, Sam
Van Meerbeek, Koenraad
Sindayihebura, Anicet
Hermy, Martin
Van Orshoven, Jos
Issue Date: 19-Sep-2016
Conference: European Ecosystem Services Conference: Helping nature to help us location:Antwerp date:19-23 September 2016
Abstract: The soil organic carbon (SOC) stock is an important indicator in ecosystem service
assessments. Whereas routine soil sampling is often limited to the topsoil, a non-negligible
fraction of OC is known to be stored in deeper horizons. To assess SOC-stocks including
the subsoil, vertical extrapolation of topsoil SOC-measurements is necessary. However, the
commonly used exponential decline functions underestimate the stock in soil types
characterised by carbon-rich subsurface horizons like podzols and peat soils. To take these
‘anomalies’ into account, we have developed an exponential change decline function,
assuming that not the %OC but rather the change in %OC between the target (2009-2011)
and a historical (1947-1974) reference topsoil measurement declines exponentially with
Another shortcoming of routine sampling is its limited and heterogeneous spatial density
which is a weak basis for regional SOC-assessments. To cope with this lack of data, digital
soil mapping approaches have been proposed which exploit the covariance of a soil variable
like %OC with predictors related to one or more of Jenny’s soil forming factors: climate,
organisms, topography, parent material, age and geographic position.
In this study, we combined the vertical extrapolation of 139 topsoil (upper 15 cm) %OCmeasurements
with digital soil mapping techniques to assess the SOC-stock (t ha-1) in the
upper 100 cm of the soil of Low-Input High-Diversity (LIHD)-systems in the region of
Flanders, Belgium. Managed nature reserves and road verges are typical LIHD-systems
characterized by low levels of inputs (e.g. manure application) and a high species diversity.
We assessed the performance of four modeling techniques: multiple linear regression,
boosted regression trees, artificial neural networks and support vector machines. Soil
texture fractions, depth of the ground water table, plant functional traits and soil type were
identified as key predictors of the SOC-stock. Overall, machine-learning methods
performed best through their capability to model non-linear relationships.
Publication status: published
KU Leuven publication type: IMa
Appears in Collections:Division Forest, Nature and Landscape Research

Files in This Item:
File Description Status SizeFormat
AbstractESS.pdfAbstract Published 30KbAdobe PDFView/Open
ESSFinal.pdfPoster Published 1500KbAdobe PDFView/Open

These files are only available to some KU Leuven Association staff members


All items in Lirias are protected by copyright, with all rights reserved.