# Modeling soil organic carbon with Quantile Regression: Dissecting predictors' effects on carbon stocks

**Authors:** Luigi Lombardo, Sergio Saia, Calogero Schillaci, P. Martin Mai, Raphaël Huser  
**arXiv:** [1708.03859](https://arxiv.org/abs/1708.03859)  
**Date:** 2017-08-13  
**Summary page:** https://www.summarizepaper.com/en/paper/1708.03859v1/

## Key points

- Soil Organic Carbon (SOC) estimation is important for managing natural and anthropic ecosystems
- SOC has a relationship with greenhouse gas emissions
- Quantile Regression (QR) can be used to spatially predict SOC distribution in agricultural areas
- QR models were tested in a highly variable semi-arid region in Sicily, Italy
- Topographic and remotely sensed predictors were used to estimate SOC stock (0-30 cm depth)
- QR models produced robust performances and allowed for the recognition of dominant effects among predictors with respect to the considered quantile
- The predictive map generated at the median showed lower errors than those from Joint Research Centre and International Soil Reference and Information Centre benchmarks
- QR is an effective method for mapping SOC using legacy data in agro-ecosystems
- R code scripted for QR is included in this study providing a valuable resource for future research efforts in this field.

## Summary

The estimation of Soil Organic Carbon (SOC) is crucial for managing both natural and anthropic ecosystems, particularly in light of its relationship with greenhouse gas emissions. While statistical applications have dominated SOC stock mapping at a regional scale, the use of Quantile Regression (QR) to spatially predict SOC distribution has been largely unexplored by the scientific community. In this study, QR was tested to estimate SOC stock (0-30 cm depth) in agricultural areas within a highly variable semi-arid region (Sicily, Italy), covering approximately 25,000 km2. Topographic and remotely sensed predictors were used, and results were compared with available SOC stock measurements. The QR models produced robust performances and allowed for the recognition of dominant effects among predictors with respect to the considered quantile. This information suggests that QR can discern predictor influences on SOC stock at specific sub-domains of each predictor. The predictive map generated at the median showed lower errors than those from Joint Research Centre and International Soil Reference and Information Centre benchmarks. These results suggest that QR is a comprehensive and effective method for mapping SOC using legacy data in agro-ecosystems. Additionally, R code scripted for QR is included in this study providing a valuable resource for future research efforts in this field. Overall, this study highlights the potential benefits of utilizing QR as an alternative approach to traditional statistical methods when predicting SOC distribution in agricultural areas.