Utilizing InVEST ecosystem services model combined with deep learning and fallback bargaining for effective sediment...

Area of Science:

• Environmental modeling
• Ecosystem services assessment
• Computational hydrology

Background:

• Sediment retention (SR) is crucial for watershed health and water quality.
• Traditional models may not fully capture complex spatial-temporal dynamics of SR.
• Integrating advanced computational techniques can enhance ecosystem service modeling accuracy.

Purpose of the Study:

• To integrate game theory and deep learning with the InVEST Ecosystem Services Model (IESM) for Sediment Retention (SR) modeling.
• To compare the performance of game theory and deep learning algorithms in predicting SR in the Kasilian watershed.
• To identify optimal sub-watersheds for SR and map spatial SR distribution.

Main Methods:

• Sediment Retention (SR) mapping using the InVEST Ecosystem Services Model (IESM).
• Implementation of the Fallback bargaining algorithm (game theory) for sub-watershed prioritization.
• Application of deep learning algorithms: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Network (RNN) for SR modeling.
• Statistical analysis including error metrics (MAE, MSE, R², RMSE) and Area Under the Curve (AUC) for model evaluation.
• Comparison of model outputs using similarity percentages and Alpha Diversity Indices (ADI).

Main Results:

• Geo-environmental factors like rain erosivity, soil erodibility, LS, elevation, and land use significantly influence SR.
• The RNN deep learning model achieved optimal performance with high accuracy (R²: 0.79, AUC: 0.97).
• Deep learning models (CNN, LSTM, RNN) showed higher similarity (79-84%) to InVEST model results compared to the game theory algorithm (47%).
• Spatial SR mapping indicated higher potential in northern sub-watersheds, with sub-watershed 5 identified as having the highest potential.

Conclusions:

• Deep learning models significantly outperform game theory algorithms in sediment retention modeling accuracy and spatial prediction.
• The RNN model is identified as the most effective for SR distribution modeling in the Kasilian watershed.
• The study provides valuable insights for watershed management and conservation strategies through accurate SR zonation.