The eight hazard classes, and what each one actually measures

HazardNet classifies every district-horizon unit into one of eight classes and scores it. These pages state what each class means, how its independent physics cross-check is computed, what it cannot tell you, and what the current run says.

Pick a hazard

Each page below is generated from the same artifacts this deployment ships, so the numbers in them match the map.

Current run at a glance

Prediction date 2026-09-16; 74 district-horizon units classified across 0 districts.

  • Riverine and monsoon flooding, scored from how much rain falls and how hard it falls. Current run: 0 district-horizon units classified as Flood.
  • Short-duration, high-intensity flooding — the burst case of the rainfall proxy. Current run: 71 district-horizon units classified as Flash Flood.
  • Wind-driven coastal hazard — sustained wind above 50 km/h, with the rain it carries. Current run: 3 district-horizon units classified as Tropical Cyclone.
  • Heat load against the rainfall the horizon is expected to bring. Current run: 0 district-horizon units classified as Drought.
  • Exceedance above 30 °C, plus persistence. Current run: 0 district-horizon units classified as Heat Wave.
  • How far the minimum falls below 16 °C, plus persistence. Current run: 0 district-horizon units classified as Cold Wave.
  • Vegetation and burn risk from heat, wind and drying. Current run: 0 district-horizon units classified as Fire.
  • Squall wind plus a convective downpour. Current run: 0 district-horizon units classified as Severe Local Storm.

Shared rules across every class

  • The eight classes above are the model's entire output vocabulary (`Models/labels.json`), enforced on ingest. A hazard outside the list is not mapped onto a nearest neighbour — an unrecognised label is rejected and reported.
  • Confidence bins (Certain ≥ 0.85, Probable 0.70–0.85, Uncertain < 0.70) describe the model's own certainty in the class it chose. They are not accuracy, and no accuracy figure is published because no held-out evaluation has been run on real labels.
  • Nothing above the WATCH level can be published automatically while no calibration map exists, and nothing at all can be published while the pipeline stamps no model version — the alert engine records that as `publication_blocked` rather than issuing the alert.

Questions and answers

Why only eight classes?

Eight is the model's output vocabulary (Models/labels.json), fixed by the training data and enforced on ingest. A hazard outside that list is not mapped onto a nearest neighbour: the event loader rejects and reports the label instead, and the forecast ingest refuses the row.

Which class is most likely in the current run?

The current-run counts are listed above, and each hazard page repeats them. What the counts do not tell you is which is likely — they describe one model run, not a climatology.

Content reviewed 2026-09-16. HazardNet is decision support, not an official warning service — see the methodology for scope and limitations.

Loading the interactive HazardNet application…