AI explainability framework for environmental management research

Area of Science:

• Environmental Science
• Artificial Intelligence
• Computer Science

Background:

• Deep learning (AI) models are crucial for environmental management but often lack transparency.
• Explainable AI (XAI) has received limited attention in environmental applications.
• Current AI models may not be suitable for edge devices or provide interpretable decisions.

Purpose of the Study:

• To develop a triadic explainability framework for AI in environmental management.
• To enhance the generalizability and reduce overfitting of AI models.
• To enable the deployment of efficient AI models on edge devices and improve decision interpretability.

Main Methods:

• Context-based augmentation of input data for improved generalizability.
• Direct monitoring of AI model layers and parameters for leaner network design.
• Output explanation procedures focusing on interpretability and robustness of AI predictions.

Main Results:

• The framework successfully augments input data, minimizing overfitting.
• Leaner AI networks suitable for edge deployment were developed.
• The output explanation procedure enhances the interpretability and robustness of AI decisions.

Conclusions:

• The developed XAI framework significantly advances the state of the art in environmental management AI.
• The framework offers practical implications for better understanding and utilizing AI in environmental research.
• This work paves the way for more transparent and reliable AI applications in environmental decision-making.