Soil organic carbon predictions across Mexico at 1 m of soil depth and 90 m of spatial resolution (1999-2009)

Abstract

Studies of spatial variation of soil organic carbon (SOC) are essential to improve knowledge about the global carbon cycle. This work documents the development of a digital SOC map for Mexico at 1 m of soil depth and at 90 m of spatial resolution representative of the period 1991-2009. A model ensemble of regression trees with a recursive elimination of variables explains 54% of the total variability using a cross-validation technique of independent samples. The predictive model produces an average error of 0.54 kg m2 of SOC at 1 m depth. The limitations of the proposed map and the research opportunities to improve the accuracy in future work are discussed. A total of 16.03±4.24 Pg of SOC is estimated in the f irst meter of mineral soil for the Mexican territory. This result is conservative compared to previous works (global and national). In this study we provide a reference framework on digital soil mapping useful for enabling state and municipal SOC monitoring programs with low computational cost.