Leveraging explainable machine learning models to assess forest health: A case study in Hainan, China.

Abstract

Global forest area has declined over the past few years, forest quality has declined, and ecological and environmental events have increased with climate change and human activity. In the context of ecological civilization, forest health issues have received unprecedented attention. By improving forest health, forests can better perform their ecosystem service functions and promote green development. This study was carried out in the WuZhi Shan area of Hainan Tropical Rainforest National Park. We employed a decision tree algorithm, a machine learning technique, for our modeling due to its high accuracy and interpretability. The objective weighted method using criteria of importance through intercriteria correlation (CRITIC) was used to determine forest health classes based on survey and experimental data from 132 forest samples. The results showed that species diversity is the most important metric to measure forest health. An interpretable decision tree machine learning model was proposed to incorporate forest health indicators, providing up to 90% accuracy in the classification of forest health conditions. The model demonstrated a high degree of effectiveness, achieving an average precision of 90%, a recall of 67%, and an F1 score of 70.2% in predicting forest health. The interpretable decision tree classification results showed that breast height diameter is the most important variable in classifying the health status of both primary and secondary forests. This study highlights the importance of using interpretable machine learning methods for the decision-making process. Our work contributes to the scientific underpinnings of sustainable forest development and effective conservation planning.