How Machhar reads the weather

A district is compared only with itself

The weather score has no meaning on its own: 0.42 says nothing. So each week's score is ranked against the same week of the year in that district's earlier years, never later ones, and only once there are at least 8 of them. Most districts have 21, from 2005 on. The map shows where this week falls.

• Highest on recordhigher than every past year for this week
• Higher than usualhigher than at least 70% of past years
• About usualbetween 30% and 70%
• Lower than usualbelow 30%
• No history yetfewer past years than the minimum

Machhar says “21 of 21 years”, not “the 100th percentile”. With 21 past years a percentile can take only 22 values, and the count doesn't pretend otherwise. Comparing a district with itself also means a normally wet district isn't flagged every monsoon just for being wet.

Heat opens transmission, then shuts it

Each mosquito trait responds to temperature with its own optimum. The traits are biting rate, the chance a bite passes the virus on, adult lifespan, larval development, eggs laid, larval survival, and how fast the virus matures inside the mosquito. Multiplied together they give a sharply peaked curve. Past about 35 °C it falls to zero: the mosquito dies before the virus inside it can be passed on.

The day, not the week's average

A mosquito lives through the afternoon and the night, not a weekly average. Because the curve above bends so sharply, the average of a hot afternoon and a cool night is not the same as a day at the average temperature. Machhar rebuilds each day's temperature cycle from ERA5's daily minimum and maximum, works out every trait hour by hour, and then combines them. This is rate summation (Lambrechts et al. 2011).

Below about 19 °C the warm afternoon hours add transmission. Above that, the hot hours past the limit take it away. That is why districts with dry, hot pre-monsoon weather look so different from a model that only reads the monthly mean.

Monthly means of the live model's temperature term for dengue, from each district's weekly record. A value under 0.05 counts as closed. Shimla stays closed all year on temperature alone, with no fitting involved.

Dry air shortens a mosquito's life

Adult Aedes lose water to dry air, and what matters is how much the air can still absorb (the saturation deficit), not relative humidity alone. The same humidity dries a mosquito out far faster at 32 °C than at 24 °C (Brady et al. 2013). Machhar lowers survival as the deficit grows, down to a floor, because Aedes aegypti rests indoors where air stays moister.

More rain is not more mosquitoes

Dengue and chikungunya mosquitoes breed in containers: buckets, tyres, roof tanks. Three things act on them. Rain fills containers, until they can hold no more. Very heavy rain washes larvae out. And in dry spells, households on unreliable supply store water, which is close to ideal breeding habitat. Together the response rises, peaks and falls.

Malaria's Anopheles breeds differently, in pools and puddles that build up over weeks of rain and dry out slowly. Machhar gives malaria its own habitat term: rain accumulates and drains over about four weeks, with no washing out.

The delay shortens as it warms

Rain today matters weeks later: larvae develop, adults emerge and bite, the virus matures inside them, and people fall ill. Development and incubation both speed up with heat, so the same rain counts sooner in a hot week than a cool one. Machhar spreads each week's habitat over the following weeks using a delay set by that week's own temperature, rather than one fixed lag.

The delay is set from Aedes aegypti's development and incubation rates for all three diseases. For malaria that is an approximation (see Limits).

Weeks ahead: forecast, then outlook

Past weeks use observed weather. The next two weeks use ECMWF's 15-day forecast. After that come 4 outlook weeks, where heat, humidity and rain are set to the district's normal for that week. An outlook week still means something, because the delay above reaches back to rain that has already fallen. Machhar measures how much of each week rests on rain already on the ground, and says so on the map.

Median across districts, from this week's national run. Past the fourth outlook week a reading would say more about the calendar than about this year, so Machhar stops there.

What goes in

Population and housing density are the same in every week of a district's record, so they don't change where a week ranks against that district's past. They affect nothing on the map. Goa also runs on a 5 km grid built for its own pilot. Official dengue counts (NCVBDC, yearly, by state) appear on the Goa page for context only. They are not an input.

Has it been checked against real cases? Not yet.

India does not publish weekly dengue, malaria or chikungunya counts by district, so these readings can't yet be tested against real outbreaks. Until they can, a reading describes how favourable the weather is, not how many people are ill. Read it alongside surveillance, not instead of it.

The test is written and waiting for data. It asks whether Machhar would have flagged the districts that later surged, how many weeks early, and how many false alarms it raised with each state limited to 5 flags at a time. It must also beat three cheap baselines: the usual level for that week of the year, last week's value, and lagged rainfall alone. A model that can't beat lagged rain on false alarms hasn't earned its complexity. The results will be published here, whatever they show.

What Machhar can't do

• It can't see a breeding site. Aedes breeds in buckets and roof tanks, far below what any satellite or weather model resolves. Weather gives the conditions, not the sites.
• No immunity. A district that had a big outbreak last season carries partial immunity and is less likely to surge again. Machhar doesn't know this yet, so it may over-flag last year's hotspots.
• Water supply is assumed, not measured. Every district gets the same reliability, so the stored-water effect is the same everywhere.
• One delay for three diseases. The delay uses Aedes rates, so malaria's timing is approximate.
• Districts and 25 km weather are coarse. Transmission happens street by street. A district reading says where to look, not where to spray.
• One district is missing. Old Delhi has no published boundary yet, so it has no weather series. It's shown as missing, never filled in.
• Favourable weather is not an outbreak. The virus has to be there, and people have to be exposed. Machhar describes the conditions and stops there.