Diep Ngoc NGUYEN (1, 2), Jacopo FURLANETTO (1, 2, 3), Silvia TORRESAN (1, 2,3), Andrea CRITTO (1, 2)
1. Ca' Foscari University of Venice - Italy
2. Euro-Mediterranean Center on Climate Change - Italy
3. National Biodiversity Future Center – Italy
The research aims to better understand current impacts and future multi-risks of single and compound hazards to disturbances in water quality and riverine ecosystems, supporting the decision-making process for environmental quality protection and adaptation planning under future changes and uncertainties.
Veneto region (NE Italy) with river water bodies of WFD interests, with diverse landscape from mountains to low plains.
1. EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR DRIVER ATTRIBUTION
• Using ML and explainable AI as a robust framework for multi-(hazard-)risk assessment, revealing the complex interplay between land use and climate extremes and their compound impacts on water quality.
• Implying the needs to address the amplifying role of human activity and prioritize natural landscape conservation to safeguard water quality under increasing climate extremes.
2. REGIME-AWARE HYBRID GRAPH NEURAL NETWORK FOR WQ PREDICTION
• Spatio-temporal GNN with embedded regime learning from unsupervised ML effectively learns group-specific response patterns instead of over-generalizing across the network, helps the GNN resolve behavirors that network topology alone cannot explain
• Address real-world challenges in water monitoring: data gaps/spatially sparse sensors and hydrologic connectivity across heterogeneous basins
• Contributes a generalizable modeling framework: supports cross-basin management decisions, land and climate scenario assessments, and targeted interventions in riverine systems