6 m
Flood-analysis grid resolution
Products · Analytics
AI Flood Risk Analysis
Summary
Combines field rainfall with terrain and drainage data to estimate flood risk by zone. The results are delivered to the Flood Watch monitoring screen, where staff can check them.
From analysis to monitoring
- 01
Build terrain & environment database
Integrates GIS data such as DEM, river networks, drainage pipes and building heights.
- 02
Real-time input
Feeds in quality-controlled one-minute rainfall and public observation data.
- 03
Risk estimation
Calculates expected flood depth and risk level by zone, aiming to look up to three hours ahead.
- 04
Display & alert decision
Shows risk levels on the map and notifies staff when alert thresholds are reached.
- 05
Post-event validation
Compares against actual outcomes to analyze error and retrain the model.
Verified scope and limits
- Prediction accuracy and lead time are continuously validated and improved using pilot operating data.
- Field variables such as drainage blockages, construction work and pump operation may not be reflected in the model in real time.
- Analysis results support staff judgment and do not substitute for an official alert.
Technical details · Why two models are used together
Physics-model scenarios
SWMM analyzes sewer-network flow, and a two-dimensional flood-analysis model builds flood maps by rainfall frequency and duration. This makes the physical basis of results easy to explain.
Deep learning model
LSTM, CNN-LSTM and Attention architectures were reviewed, and CRU-Net — a U-Net-based structure with Attention and Residual Blocks added — is applied to quickly estimate spatiotemporal data.
Technical details · Model-build scope
The values below describe the conditions under which the model was built, not performance figures.
1,322 × 1,097
Gwanak-gu flood-grid size
10–180 min · 5–250 mm
Rainfall scenario range (duration · total rainfall)
76 / 1,296
Number of maximum-flood-depth maps / time-series flood maps
Read the Rain. Protect What Matters.