Method & data

How HEATSYNC works, and where it stops.

Problem statement

During a heatwave, district control rooms in Maharashtra must decide — every day, for 36 districts — which areas are worst, which neighbours will be next, where relief can be staged from, and what to tell citizens, farmers and hospitals. The inputs exist but arrive scattered: hourly weather streams, decades of records, IMD’s criteria. HEATSYNC is a response engine that turns them into a ranked, clustered, explained and audience-ready picture in milliseconds, using one classic data structure per step.

Experiment → module

Every lab experiment, mapped.

ExpData structureModuleJob in the heat responseKey operations
01Array of structuresStation RegistryFixed-capacity registry of the 36 district automated weather stations; insert/delete at the end, linear search by code or name.insertLast · O(1)deleteLast · O(1)search · O(n)display · O(n)
02Singly linked listAlert ChainNewest-first chain of heat bulletins; follow-ups are inserted after the bulletin they update.insertAtBegin · O(1)insertAfter · O(n)deleteBefore · O(n)deleteFirst · O(1)
03Stack using SLL and infix to postfixRule CompilerCompiles IMD heatwave criteria written as infix rules into postfix and evaluates them for every district.push · O(1)pop · O(1)peek · O(1)tokenize · O(n)
04Circular queue (counter method)Sensor Ring24-slot ring buffer of hourly station readings giving a rolling 24-hour max without shifting memory.enqueue · O(1)dequeue · O(1)push (overwrite oldest) · O(1)peek · O(1)
05Binary search treeHotspot IndexDistricts keyed by heat score; reverse in-order gives the live hotspot ranking and range queries find every district in a severity band.insert · O(h): avg O(log n), worst O(n)search · O(h): avg O(log n), worst O(n)delete · O(h): avg O(log n), worst O(n)min / max · O(h)
06Graph BFS (adjacency matrix)Heat Spread GraphDistricts sharing a border are edges; BFS finds contiguous heatwave clusters, alert rings and the nearest cooler district for relief staging.build matrix · O(V²)hasEdge · O(1)enqueue / dequeue · O(1)bfs · O(V²) with an adjacency matrix
07Sorting and binary searchClimate ArchiveSorts 11 seasons of ERA5 daily maxima so binary search can rank today's heat against history in O(log n).insertionSort · O(n²) worst, O(n) bestquickSort · O(n log n) avg, O(n²) worstmergeSort · O(n log n)binarySearch · O(log n)
08Hash table (linear probing)Instant LookupO(1) lookup of a district by name, former name or HQ PIN code.insert · O(1) average, O(n) worstsearch · O(1) average, O(n) worstdelete · O(1) average, O(n) worstloadFactor · O(1)

The C implementations of all eight experiments, adapted to this use case, are in the repository’s c/ folder with their run transcripts.

Heatwave criteria

IMD’s rules, as code.

The rule compiler tokenises each rule, converts it to postfix with a linked stack and evaluates it for every district. A reference classifier written directly in TypeScript checks the compiled rules on every snapshot.

Plains stations

severe
tmax >= 47 || (tmax >= 40 && dep >= 6.5)
heatwave
tmax >= 45 || (tmax >= 40 && dep >= 4.5)
watch
tmax >= 40

Coastal stations (Konkan)

severe
tmax >= 37 && dep >= 6.5
heatwave
tmax >= 37 && dep >= 4.5
watch
tmax >= 37

tmax = daily maximum (°C) · dep = departure from normal (°C). “Watch” is HEATSYNC’s early flag for a hot day that does not yet meet IMD’s heatwave definition.

Open-Meteo historical weather API · CC BY 4.0

ERA5 reanalysis

Daily maximum temperature for all 36 district points, 1 March – 30 June, 2015–2025 (48,312 values); May 2024 daily and hourly fields for the replay.

Open-Meteo forecast API · CC BY 4.0

Open-Meteo forecast

Live mode: 7-day maximum, feels-like, humidity and wind, plus the last 24 hourly readings, fetched in one request and cached for 30 minutes.

geoBoundaries IND ADM1 / ADM2 · ODbL / CC BY

District boundaries

36 Maharashtra districts, simplified with shared borders preserved; 77 border-sharing pairs become the edges of the adjacency matrix.

India Meteorological Department

Heatwave criteria

Plains, coastal and absolute-temperature thresholds, written as rules and compiled by the stack module.

Limitations

What this prototype is not.

  1. 01

    ERA5 is a model reanalysis on a ~25 km grid, sampled at one point per district. It runs a few degrees cooler than station maxima in hot, dry interiors, so fewer days meet the heatwave criteria than IMD station records would show.

  2. 02

    “Normal” is the ERA5 2015–2024 mean for the same calendar window (±7 days), a documented stand-in for IMD’s 30-year station normals. Outside March–June no normal is defined and only absolute thresholds apply.

  3. 03

    Live mode uses numerical forecast guidance, not observations. HEATSYNC never presents either as an official warning.

  4. 04

    IoT weather stations (the use case’s Phase III) are represented by the sensor-ring module fed with reanalysis or forecast hourly data; physical stations would plug into the same ring buffer.

  5. 05

    Advisories come from a deterministic template engine following national heat-action guidance. They are drafts until a person approves them.