Improved Prediction of Eurasian Beaver Gnawing Preferences in Riparian Habitats: A Machine Learning Approach

Area of Science:

• Ecology
• Conservation Biology
• Computational Biology

Background:

• Eurasian beavers (Castor fiber) are ecosystem engineers modifying riparian ecosystems through damming and foraging.
• Beaver gnawing significantly impacts riparian forest composition and structure.
• Traditional statistical models have limitations in analyzing complex beaver-woodland interactions.

Purpose of the Study:

• To investigate the impact of Eurasian beavers on riparian woodlands in Central Italy.
• To explore the potential of Machine Learning (ML) algorithms for analyzing beaver activity and impacts.
• To identify key factors influencing beaver gnawing behavior.

Main Methods:

• Implemented three ML algorithms: Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Random Forests (RF).
• Collected in-situ tree measurements (diameter, distance from riverbank, species) and data on beaver damage (gnawing signs, impact severity).
• Utilized a two-step ML approach to predict tree damage and its severity.

Main Results:

• ML algorithms achieved high accuracy in classifying damaged/undamaged trees (up to 93%) and predicting damage severity (85%).
• Models maintained high accuracy even with reduced training data (85% with 20% of data).
• Random Forests (RF) emerged as the most suitable ML method due to accuracy and computational efficiency.
• Tree diameter and distance from the riverbank were identified as the most significant predictors of beaver gnawing activity.

Conclusions:

• ML techniques offer a more effective and cost-efficient method for assessing beaver impacts on riparian woodlands.
• ML can significantly reduce the required field data collection efforts.
• Understanding key influencing factors like tree diameter and proximity to water is crucial for managing beaver-woodland interactions.