Heat Map

The urban heat island is neither a metaphor nor a turn of phrase. It is a measurable physical effect: dense construction, dark roofs and sealed ground absorb solar radiation during the day and release it at night, while the evaporation from vegetation that cools open landscapes is all but absent in the city. The result is a persistent temperature difference between neighbourhoods of one and the same city — in our measurements, up to thirteen degrees.

The instrument of observation, meanwhile, is out of scale with its subject. An urban weather network offers one or two points for a city of a million, and a bulletin reading «Volgograd, +36» describes the whole city with a single number — even though the Krasnoarmeysky district and the reservoir shore are two different climates. Someone choosing where to live, a physician assessing heat-stroke risk and a planner deciding where to plant trees all work from a picture with no spatial resolution at all.

A satellite provides exactly what the weather network cannot: a regular thirty-metre grid across the entire city, a decade of retrospective data, and one identical method for Makhachkala and for Moscow. The institute framed the question as an engineering one: without a satellite of our own, without closed data and without a budget for permanently running infrastructure, can we build a tool that tells a city administration, a physician, a journalist and a resident the same thing — where in this city it is hot, by how much, and why.

Heat Map is our answer. The working formula is short: open data, open models, monitoring of fact instead of forecast, and a cheap static map. And one principle that sets the tone — the physics is to blame, not the people. The map does not accuse the city; it shows where physics works against the resident, and what exactly can be changed.

The project measures LST — land surface temperature, the temperature of the ground cover itself: asphalt, roofing, tree canopy, water. This is not what a thermometer on a balcony shows, and the difference between the two is not a correction factor but a physically different quantity. At midday a surface can be twenty degrees hotter than the air above it; at night, colder.

We deliberately refuse to «convert» one into the other for the reader's convenience. The quantity we can stand behind is the measured one; the quantity we would have to model would look better in a headline and worse in science. So the map repeats this caveat in the interface, in the layer captions and in every export — and modelled quantities, such as the daily peak, are labelled as estimates.

The practical meaning is not lost by this, only shifted. Surface temperature describes directly how the urban fabric handles solar energy: where a roof acts as a heater, where a courtyard has no shade, where a patch of greenery pulls a whole block down. That, rather than the reading of one screen shelter in a park, is what answers the question of what to do.

A public tool lives exactly as long as somebody pays for its infrastructure. Heat Map is therefore built deliberately to have no moving parts: the pipeline's output is a set of static files, and the map in the browser reads them directly over ordinary HTTP range requests. No database, no tile server, no background process, no access keys on the client. The cost of keeping such a map alive tends towards the cost of serving files, and its resilience equals the resilience of static hosting.

The computations that do need to be interactive were moved into the browser: the per-pixel trend for an arbitrary period is calculated on the client from annual value grids while the user drags the slider. This is the rare case where pushing computation onto the user is not a saving but an architectural decision — a server would have nothing to compute faster.

Reproducibility is delivered by the shape of the pipeline rather than by a declaration: model versions and checksums are pinned, steps are idempotent, each city's state is machine-readable, the pure logic is covered by tests, and scenes and intermediate products are cached to disk so that a repeat run does not depend on external services being up. Any step reproduces with one command from source — including a full run on synthetic data, if the reviewer has no access to satellite catalogues.

Ten cities are currently published on real satellite data: Moscow, Rostov-on-Don, Volgograd, Astrakhan, Samara, Saratov, Krasnodar, Stavropol, Makhachkala and Simferopol. The sample leans south on purpose: there heat is an annual norm rather than an anomaly, and the cost of a planning mistake is higher. Each city has all seven layers, real administrative boundaries, a district ranking, an eleven-year annual record and a shareable card. The pipeline runs end to end and the interface builds into a static export.

Immediate tasks: publication at a permanent address on the institute's domain and a load test of tile delivery; refining the trend to the median of per-pixel slopes; a common valid-pixel mask for annual medians, to remove the noise of comparing composites with different coverage; narrowing the vegetation class in the heat island computation so that dry summer cropland stops masking the effect of built-up ground; and checking reference regional indicators against primary sources. After that, more cities.

← All projects of the institute