How district risk is scored
Every district page carries three Risk cards. Heat and rain ask how unusual this place is right now, for this time of year, by its own record; hazards ask what has struck it over the long record. Every figure is read from the index's own files; "How exactly" opens the working.
Heat and rain against their record
the same rule for both
Its own normal
30 yearsof the same weeks, in the same district, 1991–2020
The score is where today falls in that record: 100 is the most adverse it has been.
How exactly
Heat: the 3-day mean of the day's peak wet-bulb against every day within ±15 days of the date, 1991–2020, kept as 101 quantiles per district. Rain: the 60-day and 30-day totals against the same windows ending within ±3 five-day steps of the date. Linear between quantiles; ties take the middle rank.
Five bands
1.0%of 1991–2020 district-days were severe or extreme for heat; 2.6% since 2021
A band starts at a physical line: the body's limits for heat, IMD's departure classes for rain; as many districts can be severe as the weather makes so.
How exactly
Heat: the level the body reaches on two of the last three days: within 3 °C of the resting adult's limit is watch; past the limit of a woman over 65 (27.08 °C wet-bulb in humid heat, PSU HEAT 2024) high; within 1 °C of the resting limit severe; past it extreme. Each band above normal needs the 3-day wet-bulb over the district's 1991–2020 median, severe and extreme 1 °C or more over it. Rain: the departure from the 1991–2020 mean. The 1 °C bar and the −40% and −80% rain steps are Maplore's choice, between IMD's lines. The score beside it is still the percentile of the district's own record: how unusual, not how dangerous.
Two sides of rain
60 · 30days for a deficit · for an excess
The higher band of the two sides is the headline. In the dry season neither goes past watch: there is little rain to miss.
How exactly
Bands: departure from the 1991–2020 mean: deficit −20/−40/−60/−80%, excess +20%, then at +60% by rain over normal of 64.5/115.6/204.5 mm (IMD's heavy-rain lines). The −40% and −80% steps are Maplore's, halfway between IMD's −20% (deficient) and −60% (large deficient). Score: deficit = 100 − 2 × the percentile of the 60-day total when below the median; excess = 2 × the percentile of the 30-day total − 100 when above. Guards: a deficit is held at watch where the normal 60 days bring under 20 mm; an excess where the 30 days brought under 25 mm.
What each reads
heat for people, rain for crops
Heat
783districts, daily from 1990, to 2 November 2026
Banded on Garmi's limits: within 3 °C, past an older woman's limit (27.08 °C), within 1 °C, past the resting body's limit; severe and extreme also need 1 °C over its normal.
How exactly
Garmi's long record: per district and India-time day, the peak wet-bulb, the night's low and the smallest margin to the limit (PSU HEAT chamber studies), computed from the hourly ERA5 archive on our server, topped up nightly. ERA5 is Copernicus data, used with attribution.
Rain
5 daysa step, at 0.05°, from 1981 to 5 November 2026
CHIRPS v3 rain, weighted by area into each district. The latest five days are preliminary until the final arrives.
How exactly
CHIRPS v3 pentads from the Climate Hazards Center: finals from 1981, preliminary for the last weeks; India's window read from each global file; districts rasterised at 0.01° into area shares of each 0.05° cell. Public domain (CC BY 4.0, rights waived); Funk et al. 2026, Scientific Data 13:718.
Checked three ways
against its definition, against IMD, against the harvests of the monsoon crops
Heat bands, 1991–2020
76.2%of 1991–2020 district-days read normal; 65.5% since 2021
Filled: 1991–2020. Dashed: since 2021. No share is fixed; each band holds what the weather put in it.
Rain bands, 1991–2020
32.5%of 1991–2020 five-day steps read normal; 31.4% since 2021
Filled: 1991–2020. Dashed: since 2021. Rain strays from its normal often: watch is the commonest band.
Rain against IMD, 2026
95%of 779 districts within one of IMD's season classes
Outlined: IMD. Filled: ours. Ours runs milder at the extremes; IMD is a check here, never an input.
Against rice harvests, 1991–2019
1.68×as many bad rice years after a rain deficit at high or above at the end of September
A bad year: yield 10% or more under the district's own trend. The rarer the band, the likelier it. Heat as built for the body shows no crop skill (1.01×).
How exactly
ICRISAT District Level Database (CC BY 4.0), rice, 433 districts that are each one district of today, 11,996 district-years; trend by a Theil–Sen line per district. Bands fixed before the test. Over the whole of July to September the lift is 1.28×. ICRISAT's rice is all seasons together, which dilutes the monsoon signal; irrigation hides it; about 2 in 3 alarms are not followed by a shortfall, so this is a risk score, not a harvest forecast.
Where rain matters
1.75×Odisha, the strongest response
Rain-fed states answer to the index most; Haryana and Andhra Pradesh least.
Every monsoon crop
11 of 12kharif crops answer a September rain deficit with a lift of 1.25× or more
The same test for every crop in the district records. Rain carries the signal for millets, pulses, oilseeds and cotton too; heat built for the body carries none.
Hazards on the long record
earthquakes since 1900, cyclones since 1981, lightning 1998–2023; one score, weighted by Maplore
Each on its own measure
5 hazardseach measured where people live, then placed on the same five bands
A band starts at a physical threshold, never at a fixed share of districts: as many can be severe as the record makes so.
How exactly
Each measure is scored linearly between its edges (watch 50, high 75, severe 90, extreme 98) and reaches 100 at one and a half times the extreme edge. Floods: share of the district's people inside JRC's 1-in-100-year river flood extent. Cyclones: storms of gale force (34 knots) a decade passing within 100 km of where people live, IBTrACS seasons 1981–2025, and of cyclone force (64 knots) on its own edges of 0.2, 0.4, 0.7 and 1.0, the higher of the two scored; counts fade inland by one e-fold every 170 km (Kaplan and DeMaria 1995). Earthquakes: the Modified Mercalli shaking reached by the most exposed quarter of the district's people over every event of magnitude 4.5 or more since 1900, from USGS ShakeMaps where recorded and otherwise modelled from the USGS catalogue (Allen, Wald and Worden 2012, rupture length from Wells and Coppersmith 1994, ground corrected by the USGS Vs30 mosaic); shaking of IV and below counts as none. Landslides: share of people on IIT Delhi's high or very-high landslide ground, counted only where the land within about 3 km rises 100 m or more (the low-relief line of IS 14496 Part 2; heights from Copernicus GLO-90). Lightning: flashes a km² a year where people live, NASA's LIS record 1998 to November 2023.
Weighted by Maplore
30 · 25 · 20floods, cyclones, earthquakes; landslides 15, lightning 10
The weights are Maplore's own, stated here, and the same for every district.
Two hazards add
96.6a district at 90 for floods and for cyclones; 90 for either alone
Losses add, so a place hit two ways stands above a place hit one way; one hazard alone keeps its own band.
How exactly
Each hazard's score s is turned into a loss of w × 10^((s − 50)/25), every 25 points ten times the loss, w its weight; the losses are summed and read back on the same scale against the heaviest weight: 50 + 25 log₁₀(sum ÷ 30). Above 98 the scores are spread to 100 over everything up to all five at 100, so the most exposed stay apart.
India today
197 of 783districts severe or extreme for hazards
188 of them are severe on one hazard by itself; adding hazards lifts the other 9.
Why the normal stays 1991–2020
a fixed yardstick shows the change; a moving one hides it
Heat, every year
2.5×days at high or above, 2021–2025 against the 1990s (7.1% to 17.6%)
Share of India's district-days at high or above, each year.
Rain, every year
1.0×steps at high or above, 2021–2025 against the 1990s (23.3% to 22.5%)
Share of India's district five-day steps at high or above, each year since 1981.
The decision
Every score is read against the WMO standard normal, 1991–2020, and it will not move. A lender's or insurer's trigger must mean the same next year as this year; a rolling window, as Garmi uses for comfort, would lower the bar each decade and hide exactly the change they carry. WMO's next normal goes beside this one, never in its place.
What it can't tell you
- What comes next. Scores describe weather that has happened: heat to yesterday, rain to the last five days.
- Crop heat. The heat index reads the body's margin; a crop needs daytime heat over its flowering or grain-filling weeks. Being built.
- Anything finer than a district. Blocks and panchayats share the registry's codes; their values come later.
- IMD's view. No IMD radar, station, lightning or gridded data is in any score; IMD's classes are only a check.
- Irrigated farms. Where canals and tube-wells carry the crop, rain says little about the harvest.
- Hazards where open records stop. Lightning ends in November 2023, when NASA's sensor did; earthquakes before 1900, and old ones USGS has no recorded shaking for, are modelled, not observed.
- Next year's hazard. The hazard score is a standing record of what has struck a place, not a forecast for a season.