Forest stand and soil types determine soil organic carbon storage in the Middle Atlas region of Morocco using machine learning models
Mohamed El Mderssa,
Meysara Elmalki,
Joann K. Whalen,
Hicham Ikraoun,
Fatima Zahra Aliyat,
Youssef Dallahi,
Younes Abbas,
Laila Nassiri,
Jamal Ibijbijen
Affiliations
Mohamed El Mderssa
Polyvalent Unit in Research and Development, Polydisciplinary Faculty, Sultan Moulay Sliman University, Beni Mellal, Morocco
Meysara Elmalki
National Water and Forests Agency, Rabat, Morocco
Joann K. Whalen
Department of Natural Resource Sciences, Macdonald Campus of McGill University, Ste-Anne-de-Bellevue, Canada
Hicham Ikraoun
Unit “Environment and Valorization of Microbial and Plant Resources”, Faculty of Sciences, Moulay Ismail University, Meknes, Morocco
Fatima Zahra Aliyat
Unit “Environment and Valorization of Microbial and Plant Resources”, Faculty of Sciences, Moulay Ismail University, Meknes, Morocco
Youssef Dallahi
Plant Physiology and Biotechnology Team, Center of Plant and Microbial Biotechnology, Biodiversity and Environment, Faculty of Sciences, Mohamed V University in Rabat, Morocco
Younes Abbas
Polyvalent Unit in Research and Development, Polydisciplinary Faculty, Sultan Moulay Sliman University, Beni Mellal, Morocco
Laila Nassiri
Unit “Environment and Valorization of Microbial and Plant Resources”, Faculty of Sciences, Moulay Ismail University, Meknes, Morocco
Jamal Ibijbijen
Unit “Environment and Valorization of Microbial and Plant Resources”, Faculty of Sciences, Moulay Ismail University, Meknes, Morocco
Forest soils often contain more carbon (C) than living trees, with significant variation in soil organic carbon (SOC) stocks due to stand type and soil characteristics. This study evaluates SOC stocks in the Moroccan Middle Atlas forests using field measurements and machine learning models. Soil profiles across 16 forest types were analysed, identifying soil typology and measuring SOC stocks. Spatial variation in SOC stocks was influenced by stand type and substratum nature, as determined through supervised extrapolation analysis. SOC stocks ranged from 35 t SOC ha-1 on tree-free land to 252 t SOC ha-1 under mixed cedar (Cedrus atlantica) and zeen oak (Quercus canariensis) stands. To enhance estimation accuracy, Random Forest (RF) and Gradient Boosting Machine (GBM) models were tested. The GBM model outperformed the RF model, with an RMSE of 6.97 t C ha-1 and R2 of 0.99, compared to RF’s RMSE of 10.28 t C ha-1 and R2 of 0.44. For better SOC stock assessment in deeper soil layers, a strategy involving more surface soil samples (0–30 cm) combined with numerical modelling of proximal soil properties is recommended.
