Amount, determining factors and spatial distribution of soil organic carbon storage in the Dano catchment (Southwest Burkina-Faso)

Abstract:

The ability to project and to mitigate the impacts of climate change is closely related to the evaluation of soil organic carbon (SOC) content and stock across different types of land use and soil groups. Therefore, this study aimed at estimating the surface and subsoil organic carbon stocks in different land use systems and across various soil groups. A further aim was to assess the spatial variability of SOC content and stocks and how this is controlled by climate and site properties. The Random Forest (RF) modelling was used and compared to Ordinary Kriging interpolation (OK) for the topsoil SOC and stock. About 70 soil profiles were described along 16 transects with 197 samples collected from different horizons up to 1 m depth where possible. In addition, 1205 samples were collected within an intensive auger grid mapping. Mid-infrared spectroscopy and partial least-squares analysis were used as a fast and low-cost technique to handle the large amount of samples for the SOC estimation. The natural/semi natural vegetation recorded the highest SOC stock in the topsoil (28.6 t C ha^-1) as compared to the cropland (25.5 t C ha^-1). Over 1 m depth, Gleysols (87.4 t C ha^-1) stored the highest amount of SOC stock followed by the Cambisols (76. t C ha^-1) and the Plinthosols (73.1 t C ha^-1) while the lowest were found in the Lixisols (57.8 t C ha^-1). For the topsoil, the RF model revealed soil properties such as cation exchange capacity (CEC) and stone content as main factors affecting SOC content variability while CEC and bulk density were the major drivers for the subsoil. The carbon stock variability was mainly affected by the CEC and the reference soil group in the topsoil while horizon thickness and bulk density constituted the main factors for the subsoil. The geostatistical evaluation proved that the SOC content in the Dano catchment has a moderate spatial autocorrelation while the carbon stock was strongly spatially dependent. The RF gave a better prediction for the topsoil SOC content with the lowest root mean square error of 4.98 g kg^-1 while the OK performed better for the topsoil carbon stock with a root mean square error of 14.98 t C ha^-1. About 52 % of the carbon stock over 1 m depth was held in the upper 30 cm and is proned to release upon natural vegetation clearance and non-sustainable farming practices.