Building Forecast-X: a WMO-aligned multi-hazard warning system for 600 upazilas
Most early-warning systems in Bangladesh are built around a single hazard — flood, cyclone, or heat — tracked by a single agency. Forecast-X started from a different question: what would it take to give every upazila a consistent, multi-hazard risk picture, built on the same methodology from Teknaf to Tetulia?
The ensemble problem
Any single NWP model has systematic biases in a country as climatically varied as Bangladesh. Forecast-X blends four ensembles — ICON-EPS, NCEP-GEFS, ECMWF-IFS, and UKMO — across 13 variables, sourced through the Open-Meteo Professional API. Blending rather than picking a "best" model matters most at the threshold boundary, where a single model's bias can flip a forecast from "watch" to "warning."
From variables to a risk matrix
Thirteen raw variables don't mean much to a decision-maker on their own. The core of the system is a multi-hazard threshold framework spanning 11 hazard types with sub-types, mapped against the WMO 4×4 likelihood/impact risk matrix. This is what turns "rainfall probability 68%" into a status a district disaster management committee can act on.
What's operational vs. roadmap
Rolling multi-window rainfall computation and WBGT (heat stress) calculation are live in the current dashboard. Some hazard sub-types are still being validated against historical events before being marked operational — a distinction the system tracks explicitly rather than presenting everything as equally mature.
What's next
The next phase focuses on tightening validation for the newer hazard modules and extending the choropleth dashboard's usability for non-technical district-level users.