SIREN

SCAN THIS FOR PROTOTYPE!!

What is SIREN?

SIREN (AI-Based Disaster Response Decision System) is a near-real-time flood disaster decision-support system designed to help emergency responders move from flood detection to actionable response decisions.

During a flood, identifying the affected area is only the first step. Responders also need to understand the level of risk, affected exposure, and whether available emergency resources are sufficient. SIREN brings these components together in a single workflow.

How it works

Area Selection → Satellite Observation → U-Net Flood Detection → Exposure Assessment → Risk Assessment → Resource Assessment → SIREN Response Decision

The system uses satellite observations to analyze a selected area and applies a U-Net deep learning model for flood segmentation. The detected flood information is then combined with exposure, risk, and resource information to identify areas that may require greater attention.

SIREN presents the results through a web-based interactive interface, allowing users to explore affected zones and understand the reasoning behind the response priority.

What inspired us?

Disaster-response teams often have access to satellite maps, flood information, and resource inventories, but these can exist as separate pieces of information. We wanted to build a system that connects them into one decision-support workflow.

Our goal was not just to answer “Where is the flood?”, but also:

“Where should responders focus, and what resources may be insufficient?”

How we built it

We built SIREN using a combination of web technologies, geospatial processing, satellite data, and deep learning.

Sentinel-1 SAR satellite observations for flood analysis
PyTorch-based U-Net for flood segmentation
Python for AI and geospatial processing
Node.js and Express.js for the backend/API
Leaflet.js for interactive map visualization
MongoDB for storing resource and analysis information
Raster/GeoTIFF processing for geospatial flood masks

The prototype supports both current analysis using the latest suitable satellite observation and a historical event demonstration mode using documented flood-event data.

What we learned

Building SIREN helped us understand how AI models can be connected with geospatial data and operational decision-making rather than being used only for prediction.

We also learned the importance of:

aligning satellite imagery and masks spatially,
handling different geographic regions correctly,
separating historical demonstration data from current analysis,
and presenting AI results in a way that is understandable to end users.

Challenges

One of our main challenges was working with large geospatial satellite datasets and ensuring that imagery, flood masks, and geographic zones were correctly aligned.

Another challenge was designing a system that does not simply generate a flood map, but converts that information into explainable risk and resource-gap insights.

Future scope

SIREN can be extended by integrating more satellite observations, larger and more diverse flood datasets, population and infrastructure exposure layers, and authorized disaster-management resource APIs.

The long-term goal is to evolve SIREN into a scalable disaster decision-support platform that can support multiple regions and disaster types.