Regime-Aware AI Post-Processing of Monsoon Rainfall Forecasts

The receipts 🧾

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

1. Exhibit A

   Heavy rainfall events are rare, so the categorical scores that matter most are computed on small samples and will be noisy — report confidence intervals or a sceptical judge will rightly discount a single season's improvement

2. It gets worse

   Splitting the data by regime shrinks the sample available to fit each correction, which is exactly how regime-aware methods overfit and end up worse than a single global correction

3. Still reading?

   Transition days belong to no regime cleanly and are often when forecasts are worst, so a hard classifier will mishandle precisely the cases the system exists for — consider soft weighting across regimes

4. And the finisher

   The product is a verification table, which is dry, so plan how to make a regime-stratified skill improvement legible to a panel that is not made entirely of meteorologists

The damage report 📉

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

• Feasibility

  4/5

  Actually buildable, which on this slate is rarer than it sounds. Do not squander it on scope.

  Everything needed is free and real — global forecast archives provide the raw rainfall, gridded observed rainfall over India provides verification, and the monsoon regime definitions have published objective criteria you can implement rather than invent.

• Innovation scope

  4/5

  There is something genuinely new here. Do not bury it under another dashboard.

  The statement identifies the idea — that correction should be conditioned on regime — and prescribes nothing about how you classify regimes, how you correct within them, or how you handle the transition days that belong to neither, which is where the interesting problem lives.

• Clarity

  4/5

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

  Concise but properly specified: it names the regimes, the five deliverables and, unusually, the exact verification metrics including equitable threat score, critical success index, hit and false alarm rates and the fractions skill score, so success is defined externally rather than by you.

• Acceptance potential

  4/5

  Strong footing before you have written a line. Try not to waste it.

  Free real data, published regime criteria, named verification metrics and a measurable improvement over an honest baseline make this one of the most rigorously assessable statements in the block, and rainfall post-processing is unglamorous enough that the field will be thin.

• Effort

  Heavy

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

  Assembling matched forecast and observation archives across seasons, implementing objective regime classification, fitting and validating per-regime corrections and building a proper verification harness across six metrics is four workstreams with the archive assembly the slowest.

• Demo-ability

  Medium

  Demoable, if you rehearse it. Nobody rehearses it.

  The verification table is the demo and it is genuinely convincing to a meteorologist because it uses their own metrics, but it is numbers rather than a visual and a general judge will need the regime breakdown explained before the result lands.

• 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.

• The statement names the verification metrics itself, which means your evaluation protocol is fixed by the sponsor and your improvement claim is directly comparable to operational practice
• Active and break monsoon phases have published objective definitions, so your regime classifier can be validated against an accepted criterion rather than defended as a design choice
• Gridded observed rainfall over India is freely available and is exactly the verification data an operational centre uses, so your skill numbers mean the same thing to the panel as their own