Integrating climate projections and machine learning to predict survival of drought resistant trees for climate smart reforestation

Résumé

Climate smart reforestation faces critical uncertainty about tree survival under future drought conditions. Predicting which trees will survive is essential for guiding species selection and management interventions. This study develops an explainable machine learning framework that integrates long term empirical forestry data with climate projections, functional traits, and management practices to predict survival probability for drought resistant trees. We trained and compared multiple models, including Extreme Gradient Boosting (XGBoost), Random Forest, and survival analysis methods, using data from 12,000 individually monitored trees in arid and semi arid Kenya. XGBoost achieved the highest predictive performance (accuracy 98.5%, F1 score 0.99). Explainable AI techniques (SHAP, LIME) revealed that species identity is the dominant predictor of survival, with root depth, genetic drought score, and mulching as key modulators. Critically, survival depends on species specific trait combinations rather than single trait optima. The Cox proportional hazards model identified tree age, elevation, irrigation, and fertilization as significant drivers of mortality risk (p < 0.05). This framework provides an interpretable, forward looking tool for guiding species selection and targeted interventions, thereby enhancing reforestation viability in drought prone landscapes. The approach is generalizable to other regions and supports evidence based climate adaptation in forest ecosystems worldwide.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0