# Machine Learning Flood Prediction: HEC-RAS vs Random Forest
> Explore a master's thesis on using Random Forest as a surrogate for HEC-RAS 1D and 2D flood modelling in Bavaria, improving real-time forecasting speed.

Tags: machine-learning, flood-prediction, hec-ras, random-forest, environmental-engineering, hydrology, data-science, bavaria
## Master's Thesis: Machine Learning–Based Flood Prediction
* **Author:** Dhwisha Jatinkumar Saraiya (Technical University of Munich)
* **Focus:** Comparing 1D and 2D flood modeling in Kulmbach and Isar River areas.

## Introduction & Problem Statement
* Hydrodynamic models (HEC-RAS) are accurate but computationally expensive and slow.
* Real-time forecasting requires faster surrogate models like Machine Learning (ML).

## Methodology & Data Features
* **Study Areas:** Kulmbach (1D) and Isar River (2D) in Bavaria, Germany.
* **Input Features:** Digital Elevation Model (DEM), Manning's Roughness (n), Discharge (Q), Hydraulic Radius, and Water Surface Elevation.
* **Algorithm:** Random Forest Regressor (scikit-learn) using an 80/20 train-test split.

## HEC-RAS vs Random Forest Results
### 1D Model (Kulmbach)
* **RF Performance:** R²: 0.94, RMSE: 0.12m, MAE: 0.08m.
* **Time:** HEC-RAS takes hours; RF takes seconds.

### 2D Model (Isar River)
* **RF Performance:** R²: 0.91, RMSE: 0.18m, MAE: 0.13m.
* **Extent Accuracy:** 87% F1-score for flood inundation patterns.

## Summary of Findings
* Random Forest trained on HEC-RAS outputs provides near-instant predictions with high accuracy (R² > 0.88).
* Manning's n and Water Surface Elevation are the most critical features for model accuracy.
* ML offers a viable, efficient complement to traditional hydrodynamic modeling for emergency flood forecasting.
---
This presentation was created with [Bobr AI](https://bobr.ai) — an AI presentation generator.