Convective scale nowcasting for Thunderstorms, Hail & Cloudbursts (06 hr)

The receipts 🧾

Every red flag on this statement, in full. These are the four places it bites.

1. Exhibit A

   Hail probability, downburst velocity and lightning strike density each require labelled observation records that are not publicly available for India, so three of the four stated outputs cannot be trained or verified properly

2. It gets worse

   Two of the three named input streams are not openly distributed, so the multi-source fusion architecture the statement is built around may reduce to satellite-only in practice — plan for that from the start

3. Still reading?

   This statement substantially overlaps two others in the same cycle covering thunderstorm and cloudburst nowcasting, so you will be compared directly against near-identical submissions and need a clear reason yours differs

4. And the finisher

   Advection of the current field is a strong short-range baseline, and a system that does not beat it at two hours has demonstrated nothing regardless of how sophisticated the fusion is

The damage report 📉

Every score this statement earned, and what each one actually costs you.

• Feasibility

  3/5

  Buildable. Not comfortably. There is a week in here you have not planned for yet.

  Satellite imagery is free and sufficient to nowcast convective development on its own, but the statement's other two named inputs are not openly distributed in India — radar products are only selectively public and bulk lightning network data is not — and each of the four severity parameters needs its own labelled event record that does not exist for this region.

• Innovation scope

  3/5

  Mildly interesting. The novelty will not carry the room; the build has to.

  The lead time, resolution, data sources, fusion approach and the four forecast parameters are all specified, so what remains open is the architecture and how you handle detection of initiation, which is genuinely the interesting part.

• Clarity

  4/5

  The ask is unambiguous, which quietly removes your favourite excuse.

  Precise about the operational envelope — zero to six hours, one to three kilometres, the three input streams and the four severity outputs are all named — though it says nothing about verification or what accuracy would make the system useful.

• Acceptance potential

  2/5

  The numbers do not like you. Bring something the numbers cannot see.

  Four distinct severity parameters each need labelled observations that are not available for India, two of the three named input streams are not openly distributed, and this statement substantially duplicates two others in the same cycle — so you would be competing against near-identical submissions on a problem where the data does not support the stated outputs.

• Effort

  Heavy

  Heavy. Somebody on this team is not sleeping in week three. Pick who, on purpose.

  A high-cadence multi-source ingestion engine, initiation detection, four separate severity forecasts and a live GIS dashboard with arrival timing is five components, and aligning streams arriving at different cadences and resolutions is the persistent difficulty.

• Demo-ability

  Medium

  Demoable, if you rehearse it. Nobody rehearses it.

  A replay showing initiation caught before maturity with a correct arrival countdown is compelling, but it depends on you having found an event with all three data streams available, and probability fields need interpretation.

• Data

  None supplied

  No dataset comes with this one, so every accuracy figure you quote is a number about labels you invented.

  Nothing is provided with the statement. You are sourcing, cleaning and labelling it yourself, and that work is invisible in the demo but very visible in the questions.

The demo they will have already seen 🪞

Somewhere around 140–330 teams are heading here, and the description is doing the choosing for most of them. They will read the same brief, reach the same architecture, and build a version of the same demo you are planning. Being correct is the floor. If your five minutes could be swapped with the team before you and nobody in the room would notice, you have not picked badly — you have built predictably, which costs exactly the same and hurts more.

What survives 🛡️

The ground worth standing on when the questions start.

• Detecting convective initiation before a storm is mature is the genuinely valuable capability here and is the one thing advection-based nowcasting fundamentally cannot do, so it is both the hard part and the differentiator
• Geostationary satellite imagery alone is free and sufficient to build a working convective nowcast, so the project is viable even without radar and lightning access
• The arrival countdown per location is an unusually concrete and legible output for a meteorological product and gives you a demo moment that needs no explanation