Abstract
Natural disasters such as tropical cyclones, severe floods, and strong winds continue to devastate Malawi on a regular basis. Devastating events like Cyclone Idai (2019), Cyclone Ana and Gombe (2022), and the record-breaking Cyclone Freddy (2023) claimed hundreds of lives, displaced over a million people, and caused economic losses running into hundreds of millions of dollars. With more than 85% of the population living in rural areas characterized by poor road networks, limited electricity, and low mobile internet penetration, timely and accurate early warnings rarely reach the people who need them most. Existing government and humanitarian alert systems often suffer from delayed dissemination, lack of village-level precision, and dependence on radio or SMS services that fail in remote and low-connectivity zones.
This project proposes a scalable, low-cost Real-Time Disaster Alert System (RDAS) specifically designed to overcome these challenges through the intelligent use of machine learning, real-time data processing, and hyper-local risk assessment. The system integrates multiple data sources—including global weather APIs (OpenWeatherMap, IBM Weather, ECMWF), satellite rainfall estimates (NASA GPM, EUMETSAT), historical disaster records, topographic data, and crowd-sourced ground reports—to detect emerging threats and predict their likely impact hours or even days in advance.
At its core, the system employs proven machine learning models such as Random Forest and Long Short-Term Memory (LSTM) networks trained on more than three decades of regional weather and disaster data. These models generate continuously updated, village-level risk scores (Low/Medium/High/Extreme) for floods, flash floods, and destructive winds, with special focus on high-risk areas along the Shire River basin, Lake Malawi shoreline, and lowland districts such as Nsanje, Chikwawa, Phalombe, and Karonga, and Mangochi.
To ensure alerts reach everyone regardless of connectivity, the system uses multiple low-bandwidth channels: automated voice messages and SMS in Chichewa and English, USSD push notifications that work on the most basic feature phones, integration with community radio stations, and targeted WhatsApp/Facebook broadcasts. An offline-capable Android application is also provided to agricultural extension workers, disaster committees, and local leaders. A built-in feedback mechanism allows communities to report real-time conditions (e.g., rising river levels or damaged infrastructure) via a toll-free USSD code, creating a valuable feedback loop that further improves prediction accuracy.