Case study · Network & telemetry AI
Flood depth prediction
A fast, data-driven surrogate for urban flood simulation: statistical features from multi-resolution SWMM outputs and maximum-depth (H_max) rasters, compressed with PCA and modelled with CatBoost, predict flood depth with very low error.
- 0.0018RMSEOn held-out test scenarios
- 0.0002MAE80/20 train–test split
- 30–120 mmulti-resolution inputSWMM outputs and H_max rasters
The challenge
Physics-based urban flood simulations are accurate but slow, which limits how many rainfall scenarios planners can explore. A data-driven surrogate has to reproduce both the depth values and their spatial pattern.
Our approach
We summarized rainfall, duration and location features from multi-resolution (30–120 m) simulation data, reduced them to 100 principal components and trained CatBoost (1,000 iterations, depth 6). Validation checked both error metrics and whether the predicted spatial distribution matches the simulated H_max.
Workflow
How it works, step by step.
The pipeline in 5 stages, from ingest to validate. Each stage has a clear input, a clear output and a reason to exist.
- 01
Ingest
Multi-resolution (30–120 m) SWMM outputs and H_max rasters.
- 02
Engineer
Statistical summaries plus rainfall, duration and location features.
- 03
Compress
PCA reduces the features to 100 components.
- 04
Model
CatBoost regressor — 1,000 iterations, depth 6.
- 05
Validate
80/20 split; error metrics and spatial agreement with H_max.
Results
What it produced.
Key findings
- Strong linear correlation between actual and predicted depth values.
- Predicted spatial distribution matches the simulated Hmax with high fidelity.
- Consistent performance across all test scenarios.
- Validated on an 80/20 train–test split.
Stack & techniques
- CatBoost
- PCA
- SWMM
- Geospatial rasters
- Python
Delivered by members of the Saerosense founding team. Client names are withheld unless public — references are available on request.
Related service
Start a project
Have a similar problem?
Send a few lines about your data and the outcome you want. We reply within two business days with a feasibility read and a proposed scoping plan.