To know and understand the spatial distribution of soil organic carbon (SOC) is the first step for management of this important pool in the global carbon cycle, and to develop actions to address climate change. The objective of this work was to generate maps of percent SOC across Andalucía through the use of consistent models, quantifying the associated uncertainty and identify controls of the spatial variability. We used a legacy soil profile collection with 1500 soil profiles and 20 environmental covariates as prediction factors for climate, topography, and ecosystem functional attributes related to the dynamic of primary production. A combination of linear models and an ensemble of regression trees coupled with geostatistics was used to estimate the spatial distribution (horizontal and vertical) of SOC, maps of SOC distribution across six soil depths (0-5, 5-15, 15-30, 30-60, 60-100 y 100-200 cm). Explained variance of our models varied between 63 to 57 %, with high uncertainty at sites with the highest values of SOC (up to 8%). The variability of SOC corresponded to a complex interaction of factors, whereas precipitation is an outstanding predictive factor across all depths, annual primary production (EVI) at the superficial horizons, and topography across deep horizons. Generated maps result in a useful tool for environmental policy, because they facilitate the periodical update of SOC and provide information for the management of this pool.
Keywords: carbon vertical distribution; uncertainty mapping; lineal models; random forests; quantile regression forests (QRF)