The Rabi Season Now Arrives Weekly. A Model Has to Earn Each Number.

The rabi season used to arrive as a single estimate, argued over for months. This year it is arriving in instalments. On Monday the Agriculture Ministry's weekly sowing update put the area covered as of 30 October at 37.56 lakh hectares, against 69.02 lakh hectares a year earlier — down 46 per cent. A hard headline. But look at its shape: 19.54 lakh hectares of that total, more than half, was sown in the single week ending 30 October. The season is not missing. It is late, compressed, and moving fast — and only a weekly series can show you that.

Three clocks, not one season estimate

A crop model fed on public data now runs on three clocks at once. The sowing update lands weekly. The Central Water Commission publishes reservoir storage weekly. The India Meteorological Department's extended-range release looks two weeks ahead at a time — the current one, dated 29 October, runs to 11 November. Each clock measures something different, each fails in a different way, and treating any one of them as "the season" is where models go wrong. Start with what the sowing clock actually said this week:

Rabi sowing as of 30 October (lakh hectares)	2026	2025	Change
All rabi crops	37.56	69.02	−46%
Mustard	20.50	38.07	−46%
Pulses	9.45	19.56	−52%
Gram	5.90	14.35	−59%
Wheat	0.19	3.34	−94%
Coarse cereals	3.78	4.22	−10%
Rabi paddy	3.16	3.29	−4%

Source: Union Agriculture Ministry weekly rabi sowing update as of 30 October 2026, reported by BusinessLine (2 November) and Rural Voice via the UPAg portal (3 November). Farmioc.

The wheat line is a calendar, not a collapse

Wheat at 19,000 hectares, down 94 per cent, is the number a naive model will choke on. Same-date, year-on-year comparison is the easiest feature to build, and at the start of November it is close to useless: the main wheat sowing window only begins in the first week of November and runs for about two months, varying from State to State. Last year that window produced a record 336 lakh hectares against a five-year average of 315.23 lakh hectares, and an all-time high production of 120.66 million tonnes. A model that flags a supply collapse every late-start year, and relaxes every early-start year, has learned the calendar, not the crop. What carries signal is the weekly increment against the window still remaining — and this week's increment, 19.54 lakh hectares across all rabi crops, says the catch-up has started.

The national figure hides the split that matters. In Rajasthan, the largest mustard State, sowing had reached only 891,000 hectares against 2.206 million hectares a year earlier, down nearly 60 per cent, while Uttar Pradesh was down 22 per cent. Officials attribute the slow start to insufficient moisture and daytime temperatures too high for sowing, and expect the pace to pick up as temperatures cool. That is a State-level, week-level story. Any exposure model still running on a national rabi average is averaging away the only part of the data that prices risk.

Storage is not soil moisture

The second clock tells a different, older story. In its weekly bulletin as of 15 October, the Central Water Commission put storage in 178 key reservoirs at 126.72 billion cubic metres — 68.58 per cent of live capacity, a quarter below the 167.91 billion cubic metres held a year earlier, and about 16 per cent below the ten-year average. The southern region held just 44.3 per cent of capacity, against 89.7 per cent a year ago, while the central region sat at 86.4 per cent. At basin level, the Pennar and Cauvery basins were running storage deficits of 72.32 and 60.50 per cent against normal.

Here is the limit a modelling team has to respect. Reservoir storage is a stock of water in monitored dams. It is not field moisture, and this season proves the gap: the sowing delay was blamed on dry fields and hot days, in a year when central-region reservoirs are 86 per cent full. A full dam does not wet a rainfed gram plot two districts away. Storage belongs in a model joined at basin and command-area level to the crops that actually draw on it — as a national covariate it flatters irrigated and rainfed districts alike, and it flatters them wrong. Note, too, the publication lag: the sowing update reached the market on 2 November for a 30 October position, while the fullest reservoir breakdown in public circulation is still the 15 October one. Weekly cadence is not real time. A system that timestamps each series honestly, and refuses to blend a three-week-old water stock with a three-day-old sowing count, will beat a faster system that blends them silently.

The rain deficit is regional, and the forecast is a forecast

The third clock explains why fields were dry. Cumulatively, from 1 October to 28 October, all-India rainfall ran 48 per cent below the Long Period Average — 37.4 mm against a normal 71.3 mm. Yet the final week of that period came in 5 per cent above average nationally, because the South Peninsula recorded 54.1 mm against a normal 32.1 mm as Cyclone Arnab crossed the Andhra Pradesh coast on 28 October. Both statements are true; they describe different windows. The seasonal split is stark: Central India ran 85 per cent below normal for 1–28 October, East and Northeast India 68 per cent below, the South Peninsula 19 per cent below and the Northwest just 3 per cent below. A single all-India departure figure would have hidden every one of those positions.

What comes next is a forecast, and it should enter a model labelled as one. The IMD's extended-range forecast of 29 October expects rainfall for the week of 5–11 November to be normal to below normal for the country as a whole, with minimum temperatures above normal across much of the east, northeast, centre and peninsula, and no cold wave in either forecast week. Warm nights matter here: if sowing is waiting on cooler days, as officials say it is in the mustard and wheat belts, the temperature forecast is as much a sowing forecast as the rain map is. When the next release lands, that forecast gets replaced — it should never harden into a planning assumption in between.

Working a weekly season

Three working rules follow for lenders, insurers and procurement desks. First, judge sowing against the window, not the date: track the weekly addition and the share of the normal area it still has to cover, crop by crop and State by State. Second, keep the three series separate in the model — sown area is a ground report, storage is a basin stock, rainfall departure is a gridded weather measure — and let each carry its own timestamp and its own error. Third, re-run exposure weekly while the catch-up is live: the government's rabi foodgrain target of 177.72 million tonnes, including 121 million tonnes of wheat, is the denominator, and the next three sowing updates will say more about whether it holds than any season estimate published this month. Farmioc's DataStore organises sowing, water, weather, yield and price series by crop and district on that weekly rhythm, and PriceHub tracks what the mandis do with the same numbers — because in a season moving this fast, the age of a figure is part of the figure.