evidoria

← Back to browse

Good practice Imported

A Community-Built Machine-Learning Flood Forecasting Prototype for Kasese District, Uganda

Uganda · Kasese · See the Uganda profile

Evidence: Descriptive / self-reported Top 88% 33/100 · Ask Evidence Copilot about this practice

Researchers and Kasese District stakeholders built a low-cost, sensor-fed machine-learning prototype (linear regression and k-nearest neighbours) that forecasts Nyamwamba River flood risk from rainfall, temperature and humidity, sending SMS alerts, but the published study frames

Details

Maturity
Pilot
Promoter
Kasese District Local Government (with local university researchers and community stakeholders)
Period
2023–2025 (proof-of-concept)
Keywords
disaster risk management, flood early warning, local government

Context

Kasese District in western Uganda is repeatedly flooded by the Nyamwamba River, which flows from the Rwenzori Mountains into the district's main town, prompting researchers and local stakeholders to look for an affordable early-warning option suited to a resource-constrained rural area.

Objectives

The project set out to build a low-cost, sensor-based machine-learning system that could forecast flood risk on the Nyamwamba River and warn residents in time to act.

Activities

Using a participatory design-science approach involving district stakeholders and students from the affected area, the team combined inexpensive sensors and Arduino boards with the ThingSpeak platform for real-time data collection, then trained linear-regression and k-nearest-neighbours models on rainfall, temperature and humidity readings to predict flood levels, with alerts sent by SMS and social media.

Results

A 2025 peer-reviewed evaluation in the Journal of Flood Risk Management reports that the prototype successfully collected and analysed real-time sensor data and produced flood-level predictions, with its low cost and reliance on existing mobile-phone coverage highlighted as strengths for rural deployment.

Conclusions

The study's own authors describe the system as an unvalidated proof-of-concept built on a narrow set of climate indicators, requiring further validation before any wider rollout.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years)
Staffing & skills
Kasese District Local Government, local university researchers, community stakeholders and students from the affected area

Conditions for success

  • Participatory design-science process involving district officials and affected-community students
  • Low-cost, off-the-shelf hardware (sensors, Arduino boards) suited to a resource-constrained district
  • Reliance on existing mobile-phone penetration for SMS-based alerts rather than new infrastructure

Common failure modes

  • Built on a narrow set of climate indicators (rainfall, temperature, humidity only)
  • Requires further validation before any wider rollout, per the study authors
  • No formal government agency ownership, funding line or sustainability plan documented

Commonly funded by

Own resources / municipal budget Philanthropic / foundation funding

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful