Overview#
A storm system tracking toward a major metropolitan area will generate a predictable surge in emergency calls starting about six hours before landfall. Without advance warning, dispatch centres scramble for staff when demand peaks. The Predictive domain analyses historical weather-to-call correlations, runs ML demand forecasts, scores risk by zone, and generates pre-positioning orders so resources arrive before the surge, not during it.
Key Features#
- Weather station management with observation recording and alert processing
- Weather-call correlation analysis with historical pattern recognition and surge prediction
- Demand forecasting using multiple ML model types with staffing recommendations
- Special event management with impact radius mapping and historical analysis
- Zone-based risk scoring with real-time weather and event adjustments
- Geographic risk zone management with population and infrastructure data
- Resource staging location management with capacity and coverage tracking
- Pre-positioning scenario planning for weather events, special events, and coverage gaps
- Coverage analysis snapshots with gap identification and recommendations
- Pre-positioning order lifecycle tracking from creation through arrival
Use Cases#
Relevant sectors include public safety communications, critical infrastructure, and national emergency management.
- Predicting call volume surges from weather events and generating staffing recommendations
- Calculating zone-based risk scores to proactively deploy resources before incidents occur
- Optimising resource positioning for major events to minimise response times
- Analysing coverage gaps and generating pre-positioning recommendations
Integration#
The Predictive domain connects with dispatch systems, command operations, dashboards, and alert management. It integrates with National Weather Service APIs and GIS services for external data.
Open Standards#
- GeoJSON (RFC 7946): Geographic zone boundaries, weather alert polygons, risk zone extents, and atmospheric dispersion contours are all stored and exchanged as GeoJSON geometry objects.
- OASIS Common Alerting Protocol (CAP): Weather alerts ingested from NOAA NWS (US territories) and Environment Canada are delivered in CAP format and parsed into a normalised internal representation.
- NOAA National Weather Service API: Live surface observations, active weather alerts, and zone-based severe weather data are fetched from the NWS open REST API (api.weather.gov), covering US territories and Caribbean zones.
- Pasquill-Gifford Atmospheric Stability Classification: The Gaussian plume dispersion model classifies atmospheric stability into classes A through F using Turner's method, and applies Briggs urban dispersion coefficients to model hazardous material plume spread from point sources.
- US EPA Acute Exposure Guideline Levels (AEGL): Hazmat dispersion contours are drawn at AEGL-1, AEGL-2, and AEGL-3 thresholds (60-minute values, mg/m³) as published by the EPA AEGL Programme, alongside NIOSH IDLH values.
- ARIMA / SARIMA (Box-Jenkins time-series methodology): Seasonal ARIMA models, implemented via the statsmodels SARIMAX class, are used for call volume demand forecasting with configurable look-back windows and seasonal periods.
- GraphQL: All predictive analytics capabilities, forecasts, risk scores, weather observations, dispersion calculations, and pre-positioning orders, are exposed through a GraphQL API using the Strawberry schema framework.
- ISO 8601 / RFC 3339 datetime format: All timestamps across forecasts, weather observations, alert effective and expiry times, and pre-positioning order lifecycles are represented in UTC using the ISO 8601 extended format.
Last Reviewed: 2026-02-05 Last Updated: 2026-04-14