Speaker
Description
- Forests, Biodiversity, Costal and Marine Areas: The role of natural ecosystems in mitigation and adaptation
Forests play a strategic role in climate change mitigation and in achieving the country's Nationally Determined Contributions (NDCs) through carbon sequestration. Accurate estimation of aboveground biomass (AGB) is therefore essential for forest carbon accounting and sustainable ecosystem management. This study aimed to evaluate and improve the methodological framework for quantifying carbon stored in forest aboveground biomass through the development and comparison of predictive modelling approaches (conventional allometric equations and machine learning (ML) algorithms) for Argania spinosa of central-western Morocco. A destructive sampling approach was applied to 124 harvested argan trees in the Mesguina forest (Agadir region). Stem volume was determined using Smalian's formula, while circumference at breast height (CBH), tree height (H), and stand regime were considered as predictor variables. The representativeness of the sampled trees was verified by comparison with a reference stand of 186 trees using descriptive statistics, the Kolmogorov-Smirnov test, Weibull distribution fitting, and graphical analyses. Conventional approaches, including Box-Cox transformed linear models, nonlinear models (NLM), and generalized additive models (GAM), were compared with three machine learning algorithms: Random Forest Regression (RFR), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR). Model performance was evaluated using R², RMSE, MAE, MAPE, and repeated cross-validation procedures. The harvested sample accurately represented the reference stand (Kolmogorov-Smirnov test, p = 0.939). Among all tested approaches, Random Forest provided the highest predictive performance (R² = 0.969; RMSE = 0.070 m³; MAPE = 12.09%), clearly outperforming conventional statistical models. Circumference at breast height was identified as the most influential predictor, followed by tree height, whereas stand regime contributed little to prediction accuracy. The best nonlinear allometric model also produced satisfactory results (R² = 0.862), although its predictive accuracy remained lower than that of Random Forest. Cross-validation analyses confirmed the robustness and stability of the ML models across different resampling strategies. These findings demonstrate that machine learning, particularly Random Forest, offers a powerful and reliable alternative to conventional allometric equations for estimating argan tree volume and aboveground biomass. The proposed approach provides valuable support for carbon stock assessment, climate change mitigation initiatives, and sustainable management of Moroccan argan ecosystems. Future studies should validate these models across a wider range of ecological conditions to improve their transferability and operational applicability.