Inspiration
Freshwater events rarely happen in isolation. Heavy rainfall can amplify runoff and sewage inputs, environmental signals can propagate downstream, and the same locations can repeatedly become vulnerable. Yet environmental teams often have to piece these signals together from different sources and investigate after an event has already happened.
AquaPulse was inspired by a simple question:
What if a river could have a memory?
Instead of treating every anomaly as a completely new event, AquaPulse connects current environmental signals with historical patterns, spatial relationships, rainfall conditions, citizen observations, and previous events.
What AquaPulse Does
AquaPulse is an AI-assisted freshwater event-intelligence system designed to help identify potential environmental events and prioritize where human verification should happen.
Its intelligence loop is:
Observe → Detect → Correlate → Remember → Predict → Prioritize → Verify → Learn
The system can:
- Detect environmental anomalies from multiple signals.
- Compare current conditions with historical event patterns through River Memory.
- Identify vulnerable Hot Moments, such as periods following heavy rainfall.
- Estimate likely downstream propagation using river connectivity and historical patterns.
- Generate an explainable event confidence rather than a black-box prediction.
- Prioritize locations for field verification.
- Incorporate citizen observations as an additional evidence layer.
- Provide contextual explanations through the AquaPulse Guide.
AquaPulse does not claim that an anomaly is automatically pollution or that water is unsafe. Instead, it identifies potential event patterns and communicates uncertainty so that human experts can verify them.
How We Built It
The prototype uses a React + Vite frontend with React Leaflet for geospatial visualization and a Python FastAPI backend for the intelligence layer.
The system architecture separates:
- Environmental and rainfall observations
- Citizen observations
- Spatial river and drainage relationships
- Historical events
- Event features and anomaly signals
- Explainable event scoring
- Verification prioritization
The prototype combines real external data and derived contextual baselines with clearly labelled simulated scenarios for demonstration. Real data sources include CPCB water-quality/drain records and NASA precipitation data, while the architecture is designed to incorporate satellite observations and future IoT or laboratory measurements.
The current intelligence engine is intentionally explainable and rule-based. It combines rainfall, water-quality anomalies, citizen observations, historical similarity, spatial coherence, and temporal coherence rather than presenting an opaque AI prediction.
What We Learned
Building AquaPulse showed us that freshwater intelligence is not only about collecting more data. The difficult part is connecting different types of evidence while preserving context, uncertainty, and human oversight.
We learned how environmental signals can be interpreted differently depending on rainfall, location, historical behaviour, and downstream connectivity. We also learned the importance of distinguishing between a potential environmental event, evidence quality, impact, and verification status.
Challenges
One of the biggest challenges was avoiding false conclusions from individual signals. For example, heavy rainfall can naturally change river conditions without indicating a pollution event.
To address this, AquaPulse includes negative evidence and multi-signal correlation rather than treating rainfall alone as proof of an event.
Another challenge was working with sparse historical environmental observations. Instead of presenting limited historical measurements as a continuous live baseline, AquaPulse labels them as contextual evidence and keeps the prototype's simulated demonstration data clearly separated.
Impact and Vision
AquaPulse is designed around a human-in-the-loop approach: the system helps answer "Where should we look first, and why?" rather than replacing environmental experts.
Its long-term vision is to connect citizen science, environmental monitoring, satellite observations, IoT sensors, laboratory verification, and historical event intelligence into a shared freshwater resilience layer.
By remembering what happened before, AquaPulse aims to help communities move from simply observing freshwater problems to anticipating where and when attention may be needed next.
Built With
- api
- data
- environmental
- fastapi
- geospatial
- javascript
- leaflet.js
- learning
- machine
- numpy
- pandas
- python
- react
- rest
- scikit-learn
- visualization
- vite
Log in or sign up for Devpost to join the conversation.