Digital Mapping of Soil Organic Carbon Stocks at Regional Scale: An Application in the Peruvian Central Andes.

The Andean Region is among the most threatened of ecosystems under current predicted global warming scenarios. Reliable soil organic carbon (SOC) stock measurements are essential for predicting the influence of future climate on global soil carbon pools. However, there is insufficient information of the spatial variability of SOC stocks in this understudied ecosystem.

The aim of the present study was to characterize and model the spatial distribution of SOC stocks in the Central Andean region using soil-landscape modeling techniques. For that purpose, three study areas were identified across the Peruvian Central Andes (Jauja, Concepcion and Huancayo). A total of 400 topsoil samples (0-30 cm) were collected and analyzed for SOC. among other soil properties.

Six modeling approaches including random forest (RF), boosted regression trees (BoRT), bagged regression trees (BaRT), classification and regression trees (CART), support vector machine (SVM) and regression kriging (RK) were used to predict and map the spatial pattern of SOC stocks in the three study areas. The models were assessed using a 10-fold cross validation.

Overall, the measured SOC stocks ranged from 0.6 kg m^-2 to 87.6 kg m^-2, corresponding to the minimum and maximum value reported at Jauja and Huancayo, respectively.

Random Forest outperformed the other models in predicting and mapping the spatial distribution of SOC stocks, accounting for 83%, 87% and 82% of the total variation in Jauja, Concepcion and Huancayo, respectively. The hierarchy in prediction performance was as follows: RF>SVM>BaRT>BoRT>CART>RK, for Jauja; RF>BaRT>CART>SVM>BoRT>RK for Concepcion and RF> BoRT>BaRT>CART>SVM>RK for Huancayo.

These results suggest that RF is a promising approach in predicting the spatial distribution of SOC stocks at regional scale in the Andes.