Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Physics-Aware Downsampling with Deep Learning for Scalable Flood Modeling
Niv Giladi, Zvika Ben-Haim, Sella Nevo, Yossi Matias, Daniel Soudry
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace average pooling with learned DNN downsampling | coarse-grid flood simulation · Elevation-map generalization test | Huber loss: 6.04\cdot10^{-3} -> 3.07\cdot10^{-3} | Improved |
| Evaluate DNN on hardest samples (90th percentile) | DNN downsampling · Elevation-map generalization test | Huber loss 90 percentile: 2.80\cdot10^{-2} -> 1.10\cdot10^{-2} | Improved |
| Replace average pooling with DNN downsampling | coarse-grid flood simulation · Elevation-map generalization test | Inundation error 90 percentile: 5.64\cdot10^{-2} -> 1.77\cdot10^{-2} | Improved |
| Replace average pooling with DNN downsampling | coarse-grid flood simulation · Elevation-map generalization test | MSE: 1.37\cdot10^{-2} -> 6.64\cdot10^{-3} | Improved |
| Replace average pooling with DNN downsampling | coarse-grid flood simulation · Elevation-map generalization test | Inundation error: 9.77\cdot10^{-3} -> 4.44\cdot10^{-3} | Improved |
| Train DNN on a single map with multiple boundary conditions vs average pooling | DNN downsampling · Single-map multi-boundary-condition test | Huber loss vs fine-grid solution | Improved |
| Train DNN on one hour of flow, evaluate 20 hours on Ganges region | DNN downsampling · Ganges river 175x175 coarse grid | flood extent accuracy | Improved |
| Bilateral filter then average pooling instead of plain average pooling (ablation) | average pooling baseline · Edge-preserving downsampling comparison | per-sample loss | No clear effect |
| Bilateral filter then max pooling instead of average pooling (ablation) | average pooling baseline · Edge-preserving downsampling comparison | per-sample loss | Hurt |
| Train with Huber loss rather than MSE (ablation) | DNN training · loss function choice | convergence | Improved |
| Use plain ResNet-18 with normalization layers (ablation) | DNN architecture · architecture comparison | elevation scale preservation | Hurt |
| Remove normalization only in the residual connection (ablation) | ResNet-18 · architecture comparison | convergence and generalization | Improved |
Closest papers by meaning
- Pluvial Flood Emulation with Hydraulics-informed Message Passing (ICML 2024 · similarity 0.90)
- RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting (NeurIPS 2025 · similarity 0.86)
- FIDLAR: Forecast-Informed Deep Learning Architecture for Flood Mitigation (AAAI 2025 · similarity 0.86)
- The Merit of River Network Topology for Neural Flood Forecasting (ICML 2024 · similarity 0.84)
- Quantile-Regression-Ensemble: A Deep Learning Algorithm for Downscaling Extreme Precipitation (AAAI 2024 · similarity 0.83)
- Spatial-Logic-Aware Weakly Supervised Learning for Flood Mapping on Earth Imagery (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
