Controls on soil carbon storage from topography and vegetation in a rocky, semi-arid landscapes

Soil properties can exhibit strong spatial variation, even at the small catchment scale. Especially soil carbon pools in semi-arid, mountainous areas are highly uncertain because bulk density and stoniness are very heterogeneous and rarely measured explicitly. The effect of topographic and vegetation variables, on stoniness, bulk density and soil carbon has been explored in a 2.7 km(2) watershed of Sierra Morena in south Spain. Soil core samples were collected from 67 locations at 6 depths up to 0,3 m. Stoniness and bulk density were measured with standard methods, total organic carbon through elemental analysis. These soil properties were then used to calculate carbon stock and related to solar insolation, elevation, slope, curvature, TWI, TPI, SPI and NDVI. Stone content depends on slope, indicating the importance of water erosion on long-term soil development. Spatial distribution of bulk density was found to be highly random. By means of conventional statistical methods, with the help of a random forest method, solar radiation and NDVI proved to be the key variable controlling soil carbon distribution. Total soil organic carbon stocks were 4.38 kg m(-2) on average, with stocks about double as high on north versus south-facing slopes. These results confirm the importance of the coupled soil moisture and vegetation dynamics on the carbon balance in semi-arid ecosystems. However, validation of the random forest model showed that the different covariates only explained 18% of the variation in the dataset. Apparently, present-day landscape and vegetation properties are not sufficient to fully explain the full variability in the soil carbon stocks in this complex terrain under natural vegetation. This is attributed to a high spatial variability in bulk density and stoniness, key variables controlling carbon stocks. Future improvement of mechanistic soil formation models could help estimating these soil properties better. (C) 2016 Elsevier B.V. All rights reserved.