MOLUSCCAOpen Analysis
Farm risk experiment · Hosur, Tamil Nadu · 31 years of daily gridded rainfall

Can a five-acre vegetable
farm in Hosur outlast its weather?

Hosur grows tomatoes, beans and marigolds for Bengaluru, on red soil and one borewell, between two monsoons that rarely behave. We took 11,323 days of rainfall, a crop water-balance model and real price history, and asked three questions: how often the weather hurts, which crops survive it, and what a farmer should expect to earn over ten years.

The short answer: yes, but not on tomatoes alone. The binding limit is borewell water, not rain, and the biggest swings in income come from prices, not storms. A six-rotation mix earns about ₹87,600 a year after every cost including land, loses money in roughly one year in four, and is positive over ten years in 98% of simulated futures.
The data

A dry plateau with two wet seasons

Hosur sits at 880 metres on the Mysore plateau. Rain arrives in two bursts, and evaporation outruns it in ten months of the year.

Mean rainfall, P10–P90 whiskersReference evapotranspiration
Monthly rainfall at 12.7365°N 77.8326°E, interpolated from the 0.25° grid, 1995–2025, against FAO-56 reference evapotranspiration from Bengaluru station normals. Only September and October deliver more water than the crop can use. Everything else is irrigation.
Annual rainfall, mm. Dashed line is the 31-year mean. 2003 received 212 mm; 2022 received 1,491 mm. Highlighted years fell below 70% of normal.
What this data is, stated plainly

The rainfall is a gridded areal estimate built from rain gauges, interpolated to the farm point and stored in whole millimetres. It is not a farm gauge, and on a typical wet day the four surrounding grid cells disagree by 9 mm. The temperature grid point reads about 1.9 °C warmer than both Bengaluru and Hosur station records, so we corrected it month by month before applying any crop heat limit.

NASA POWER, ERA5, Agmarknet and TNAU could not be reached from the analysis environment, so none of their data appears here. Humidity, wind and night temperature come from published station averages instead.

Question one

How often does it rain too much?

Rain is a hazard signal, not proof of a flood. On free-draining red soil the damage comes from flower drop, disease and missed harvests during long wet spells, so the triggers below count persistence as well as volume.

TierRuleEvents per year95% intervalYears hitPeak months
WarningSevere
Distinct wet-spell events per year after merging spells less than three days apart. Severe events averaged 0.56 a year from 1995 to 2010 and 1.27 a year from 2011 to 2025.
Why we dropped the original percentile rule

The source report flagged a "severe" spell whenever 5-day rain beat that month's 95th percentile. In January that percentile is 2.7 mm, so a light shower counted as severe. Fixed thresholds tied to return levels and crop damage are easier to defend and do not drift as the record grows.

The extremes

A 5-day total of 176 mm has about a 10% chance in any year (95% range 150–201). The wettest day on record was 117 mm on 10 November 2015. A 20-year event has a 40% chance of happening at least once in a decade.

Question two

How dry does it get, and does it matter with a borewell?

One dry season is an irrigation bill. Two in a row is a groundwater problem, because the hard-rock aquifer under Hosur only refills from rain.

Indicator1995–2025
How drought enters the model

Rainfed ragi and horse gram lose yield through the FAO crop-water response. Irrigated crops lose yield only when the borewell cannot meet demand. After a year with less than 40% of normal rain, borewell output is cut to 60%, rising linearly back to full at 75% of normal. This is an assumption standing in for a pump test. In the historical record it bites in 2002–05, 2007–08 and 2024–25.

The model

Why we simulated the field, not the price chart

A farm's year is decided by what happens to each crop at each growth stage. So the model walks every crop through every day of 930 plausible seasons.

season = Jun–Dec weather of year i + Jan–May weather of year j (30 × 31 = 930 seasons) crop day = soil water − rain + Kc × ET0 ; irrigate when depletion > p × available water yield = attainable × (1 − Ky × water deficit) × (1 − wet-spell loss) × (1 − heat loss) price = base × season × exp(σ·(ρ·regional wet index + √(1−ρ²)·z) − σ²/2) profit = yield × price − pre-harvest − harvest cost − commission − interest − land − fixed

Keeping June to December together preserves the link between the two monsoons (their totals correlate at 0.46). Prices are partly tied to regional wet spells, because a wet September in Kolar is also a wet September for every competing grower. The same random stream, seed 20261010, runs here, in the Excel model and in the report, so the numbers below match those files exactly at default settings.

The crops

Twelve crop cycles, one bed at a time

Each crop has its own planting date from TNAU guidance. Shift any date by up to 45 days and its weather exposure is recomputed across all 930 seasons.

Crop cycleShift (days)PlantingHarvest windowIrrigation mmWeather yieldP(loss ≥30%)Profit /haScore

Score out of 100: 30% weather resilience, 30% profit, 20% chance of not losing money, 20% water economy. Weather yield is the share of attainable yield left after simulated rain, heat and water losses.

The one crop to drop

Tomato transplanted on 1 February flowers in April, when adjusted daytime highs pass 34 °C on most days, and harvests into the May price trough. It loses about a fifth of its yield to heat and money in an average year. Kharif and rabi tomato stay in the plan, capped.

The portfolio

Six rotations instead of one bet

Land goes to year-round rotations so two crops never share a bed. A mixed-integer optimiser chose the areas under limits on water, labour, capital, rotation and plot size.

StrategyMixExpected profitBad year (P5)Loss yearsWorst 5% avg

Profit is per year for the whole farm after land rent of ₹50,000/ha. Optimised for a 5-acre farm and scaled to the area set below.

Your allocation

RotationThrough the yearAcresGap between crops
Irrigation demand, dry-year (P80)Borewell supply
Monthly irrigation demand of the current allocation against the borewell. July is the tight month, when every kharif crop is being established at once. Bars turn red where demand exceeds supply.
The forecast

Ten years, a hundred thousand seasons

Each simulated decade draws ten seasons from the 930, with fresh price, cost and yield shocks every year. The range matters more than the middle.

Annual farm profit across all simulated years, ₹ thousand. Red bars are loss-making years. The long right tail is tomato and marigold in shortage years.
Cumulative profit, June 2027 to May 2037, median with 10th–90th percentile band. Single loss years are near certain within a decade; a loss over the whole decade is rare. These are scenarios, not a forecast of particular years.
Crop cycleMean profit /haBad year (P5)Loss chanceYield loss ≥30%
Try it yourself

Move the assumptions

Change the farm, the water, the market or the climate. The full simulation reruns: the same weather engine, the same random stream.

Expected profit a year
–
Loss-making years
–
Worst 5% of years
–
Ten-year total, median
–

Crop prices, yields and costs

CropYield t/ha (cons · base · opt)Price ₹/kgVolatilityPre-harvest ₹/haHarvest ₹/tCommission
Crop cycleYield t/haPrice ₹/kgRevenue /haCosts /haNet /haNet /acreBreak-even yieldBreak-even price
The honesty test

What moves the answer most

Each bar changes one assumption and reruns 30,000 simulated years. Prices and yields swing the result far more than any climate scenario.

Adverse settingFavourable setting
Expected annual profit, ₹ thousand, for the recommended mix on 5 acres. A 20% price fall costs more than doubling every weather-loss assumption.
Where this model is weak, stated plainly

The losses we assign to wet spells and heat at each crop stage are reasoned estimates. No farm in Hosur has published damage records we could fit them to. Doubling them takes about ₹84,000 off the expected annual profit, which is why the first season's records matter more than any refinement of the climate data.

The model also says years with a severe storm earn slightly more on average, because they refill the borewell and lift regional prices. If prices do not respond that way, the sign flips. Treat that result as unconfirmed.

The calendar

What to do, month by month

The operating year runs June to May for the rotations in the current allocation. N nursery, P planting, G growing, H harvest and sale.

TriggerRuleAction within 72 hours
Conclusion

The question behind the question

"Is Hosur too wet or too dry?" is the wrong question. It is both, often in the same year.

The farm is limited by water it can pump, not rain it receives. At two litres a second for eight hours a day the borewell carries about 1.2 hectares of dry-season vegetables. Every extra quarter of supply is worth roughly ₹27,000 a year, so a pump test comes before a planting plan.

The income risk is a market risk. Tomato prices at Kolar have swung from ₹1.5 to ₹74 a kilo in four years. Spreading harvests across eleven months, contracting gherkin and timing marigold to the festival season do more for a bad year than any drainage design.

Storms still need a plan. A warning-level wet spell arrives in nineteen years out of twenty, mostly in September and October. Raised beds, open drains and picking ahead of forecast rain are cheap, and they are what the triggers above are for.

Theme