Ecology-informed symbolic machine learning: a methodological framework for classification of forest succession

Area of Science:

• Ecology
• Machine Learning
• Ecological Modeling

Background:

• Classifying forest successional stages is challenging due to ecological complexity and the limited transparency of standard machine learning (ML) models.
• Black-box ML algorithms, while accurate, often lack the interpretability needed for practical ecological applications like restoration and regulation.

Purpose of the Study:

• To introduce and evaluate an ecology-informed symbolic machine learning (EISy-ML) framework for transparent and interpretable forest successional stage classification.
• To integrate symbolic regression with ecological constraints, such as monotonic biomass trajectories and structural complexity proxies, into ML models.

Main Methods:

• Developed an EISy-ML framework combining symbolic regression with ecological constraints derived from allometric functions.
• Applied the framework to field data from 467 plots in Brazil's Subtropical Atlantic Forest.
• Benchmarked EISy-ML performance against eight standard ML classifiers using metrics like balanced accuracy, macro F1, Cohen's kappa, and Matthews correlation coefficient.

Main Results:

• EISy-ML generated interpretable and biologically plausible equations for classifying forest successional stages.
• Achieved the highest performance metrics: test accuracy (0.899), F1 (0.905), Kappa (0.829), and MCC (0.803).
• Demonstrated no statistically significant performance difference compared to top-performing standard ML models, while offering superior transparency.

Conclusions:

• The EISy-ML framework significantly enhances transparency, reproducibility, and ecological coherence in successional classification models.
• This approach enables direct application in ecological restoration monitoring and environmental auditing.
• Validated the hypothesis that integrating ecological constraints with symbolic ML yields robust and interpretable models for ecological applications.