Where Zagreb runs hottest — and who lives there

Main finding

Five districts stand out:Donji Grad, Trešnjevka – sjever,Trnje, Stenjevec, andNovi Zagreb — Istok. Within them,32 neighbourhoods combine extreme heat with extreme population density — twelve of those are in Donji Grad alone, which also has the highest shares of residents aged 65+ in some areas (over 40%).

Open full-size chart

Data table for this chart

District	Mean ΔT (°C)
Donji Grad	+9.6
Trešnjevka – sjever	+9.0
Trnje	+6.9
Stenjevec	+6.2
Novi Zagreb — Istok	+5.7
Podsljeme	−3.5

Complementary air temperature (three reference stations) shows hot afternoons often reaching the mid-30s °C, while seasonal averages sit nearer ~28 °C. Night-time land-surfacesnapshots show a much narrower spread (~17–21 °C) than daytime heat, with the inner city still slightly warmest.

Research questions & answers

The study addressed ten practical questions about where heat concentrates, how strong it is, who is affected, and what complementary satellite and weather data add.

Question 1

Where in the city is it hottest?

The hottest surfaces are in the dense inner city. Donji Grad, Trešnjevka – sjever, and Trnje average about42, 41, and 39 °C land-surface temperature; green Podsljeme averages 29 °C — roughly a13 °C spread across the city in one summer. Built-up areas average 39 °C; tree cover 31 °C; water 29 °C.

Open full-size chart

Question 3

Does vegetation cool the city?

Yes, clearly. Greener pixels are reliably cooler (correlation −0.77). Podsljeme is the greenest and coolest district; Donji Grad the least green and hottest. Tree cover can shift a pixel’s heat anomaly from about +6 °C in built-up areas to −2 °C under trees.

Open full-size chart

Question 5

Who lives where it is hot?

32 neighbourhoods are in both the top quartile for population density and a top-quartile hot district — spread across all five hot districts (12 in Donji Grad, 7 in Trešnjevka – sjever, 5 in Trnje, 4 in Stenjevec, 4 in Novi Zagreb — Istok). The densest include Matko Laginja (~24,000 people/km²) and Nadbiskup Antun Bauer (~20,500 people/km²).

Residents aged 65+ concentrate in Donji Grad (e.g. Hrvatski narodni vladari 46%, Mimara 42%) and Trnje (Cvjetnica 38%) — also the hottest areas. Children (0–14) peak in parts of Novi Zagreb — Istok (Veliko Polje, Jakuševec), itself a hot district.

The people most vulnerable to heat — especially older adults — live in neighbourhoods with the strongest heat-island signal. This is a citywide pattern across dozens of areas, not a handful of outliers.

Air temperature, daily rhythms & elderly exposure

Question 9

People experience air temperature, not roof temperature — and when heat peaks (commute, work, night) matters for health messaging as much as where on the map.

What the weather stations show

Three reference stations (Grič city centre, Maksimir, Pleso) were analysed for June–August 2025. Seasonal averagesunderstate discomfort: afternoon work hours (10:00–16:00) average ~28 °C, but 95th-percentile days reach ~34 °C and individual heatwave hours hit the mid-30s (Pleso peak37.7 °C). Pleso is slightly warmer by day butcoolest at night (~17.7 °C mean); Grič stays warmer overnight than Maksimir.

Modelled air at neighbourhood scale

A European reanalysis product (bias-corrected using the three stations) underestimates daytime summer air by roughly0.7–1.1 °C. After correction, estimated neighbourhood air spans only 24.9–27.8 °C citywide (σ ≈0.45 °C) — the grid is too coarse for block-level differences. Use stations for absolute air and timing; use the map layer for relative ranking combined with census.

Land surface vs air

Where both are available, land surface is on average7.7 °C hotter than air at neighbourhood centroids (range +2.4 to +14.7 °C). Health and comfort narratives need both: satellites show surface hotspots; thermometers show breathed air.

Elderly heat exposure on the map

The map layer Elderly heat exposure score combines share of residents aged 65+ with ranked modelled daytime air. Because neighbourhood air is nearly uniform spatially, the pattern is driven mainly by who lives where, with a modest boost where air sits in the upper tail.

Highest-scoring neighbourhoods:

Night-time land-surface temperature

Question 10

This addresses night-time heat retention at the ground and roof surface — relevant for sleep comfort — not the same as night air temperature (see Question 9).

Three clear night satellite passes were found in early July 2025; two were combined for the map composite. Night surface temperatures were much more uniform across the city (~14–21 °C at neighbourhood scale) than daytime Landsat heat (~33 °C city mean). Typical swath medians were about18–19 °C. Coverage is patchy; treat this as a snapshot, not a full summer night climatology.

Warmest districts: Donji Grad and Trnje (~20.6 °C).Coolest fringe: Brezovica (~16.7 °C). Inner-city neighbourhoods (Petar Krešimir IV., Vrbik, Kralj Zvonimir, etc.) rank highest but only ~0.2–1 °C above the city average. Eastern Sesvete fringe areas are coolest (~14–15 °C).

Station night air (22:00–06:00) runs Grič ~20.5 °C, Maksimir ~19.4 °C, Pleso ~17.7 °C — similar order of magnitude to nightsurface medians, but the two should not be read as the same number for a given neighbourhood.

Pair night LST with the Elderly × warm night LST layer on the dashboard. Because night surface spread is small, high elderly-share areas in Donji Grad and Trnje still dominate the score — similar to the daytime exposure layer.

Heatwave detection & spatial disparity

Question 11

Using the DHMZ-style definition (≥3 consecutive days with Tmax ≥30 °C and Tmin ≥20 °C — a tropical night), the analysis detected 3 city-wide heat waves in summer 2025, consistent with the official DHMZ assessment.

When detection runs independently for each of the 218 MO centroids, a10× spatial disparity emerges: eastern lowland MOs (Sesvete) experienced up to 5 events / 23 heatwave days, while elevated Medvednica MOs (Podsljeme) had zero.

The strongest predictor of heatwave days per MO is the summer mean nighttime minimum temperature (Pearson r =0.91), far stronger than daytime LST (r = 0.36). MOs that cool below 20 °C at night cannot meet the tropical-night criterion on most days.

TALEA-inspired analytical layers

Four additional raster layers on the interactive map are inspired by theTALEA platform (Municipality of Bologna, European Urban Initiative — Innovative Actions). They appear on the dashboard under the TALEA Layers group.

Heat × population, markets & cycling

Hot districts (recap)

A district counts as “hot” if its mean heat anomaly is at or above the75th percentile of all 17 districts (+5.7 °C above the green/rural reference): Donji Grad, Trešnjevka – sjever, Trnje, Stenjevec, Novi Zagreb — Istok.

Population × heat

Headline: Vulnerable populations and the urban heat island geographically coincide — the strongest equity finding in the study.

Outdoor markets in hot areas

8 of 31 city outdoor markets sit in hot districts, including all three Donji Grad markets: Kvaternikov trg, Branimirova, Zagreb UPRAVA; plus Trešnjevka, Trnje, Savica, Špansko, and Utrina. These are places where people — often elderly — are outdoors in the hottest hours. Shading, drinking water, and cooler surfaces at these markets would have visible public-health benefit.

Cycling infrastructure in hot areas

The city reports 274 km of cycling-related network when shared paths and road markings are included; about61 km is strictly dedicated bike path or lane. In hot districts:

• Trešnjevka – sjever (second-hottest district) hasno strictly dedicated cycling infrastructure — only shared paths.
• Donji Grad has only ~1.7 km dedicated of ~20 km total — narrow historic streets limit options.
• The longest dedicated networks are in cooler or mid-heat districts (e.g. Novi Zagreb – zapad, Trešnjevka – jug).

In hot areas, feasible improvements are oftenshade and microclimate along existing shared paths rather than new segregated lanes.

Priority neighbourhoods & interventions

Ten neighbourhoods score on multiple dimensions at once: hot district, high density, and above-average elderly share. They are a defensible starting point for targeted heat action.

Intervention types (by impact)

1. Street and courtyard tree planting — often the most cost-effective; mature trees can lower local heat anomaly by 2–5 °C.
2. Cool roofs and façades — especially on large post-war estates.
3. Shade at outdoor markets — eight named hot-district markets.
4. Drinking-water points along bike corridors and at markets.
5. Light, permeable surfaces in dense pedestrian zones.
6. Green façades and pocket parks where street trees are not possible.

Workshop case studies

What this means for the team

1. Prioritise neighbourhoods, not only districts. The 32 dense-and-hot areas — 12 in Donji Grad — are the actionable units.
2. Tie heat to people. Elderly concentration in Donji Grad and Trnje is the most policy-relevant pattern.
3. Use markets and bike corridors as concrete project lists. Shade, water, and trees at named places.
4. Invest in vegetation. Cooling Zagreb means adding green, not only limiting building.
5. Combine surface heat, air timing, and population. The map shows where; station profiles show when; census shows who.
6. Target nighttime heat retention. Heatwave exposure is driven by tropical nights (Tmin ≥20 °C), not daytime peaks. Permeable surfaces, tree canopy, and green roofs in eastern lowland MOs would cut heatwave days disproportionately.

Future research directions

• More night-time satellite coverage for a stable summer night picture.
• Multi-year trend (2015–2025) — implemented: seeRQ-E and NDVI trend.
• Heatwave spatial analysis — implemented: DHMZ-style detection per MO; 3 city-wide events, 10× spatial disparity (seeheatwaves).
• TALEA-inspired analytical layers — implemented: seeTALEA.
• Heat combined with air quality at monitoring stations.
• Schools, kindergartens, and care homes overlaid on hot neighbourhoods —implemented: see RQ-B.
• Tree-cover gap analysis per hot neighbourhood —implemented: see RQ-C anddistrict greenery.
• Walking access to parks and cool public buildings —implemented: see RQ-D.
• A single heat-vulnerability index per neighbourhood —implemented: see RQ-A.

Limitations

• The main heat metric is land-surface temperature, not air. Air adds three reference stations and a coarse citywide model — not a dense monitoring network.
• Neighbourhood air differences on the map are small; the elderly exposure layer ranks air and combines it with census age structure.
• Not every neighbourhood had a paired land-surface sample in summer 2025; some statistics use subsets.
• Station averages are not the same as peak heatwave hours — use hot-day percentiles for “how hot it felt” messaging.
• Night surface results use only two clear July satellite passes; neighbourhood values average whole areas (parks, water, and buildings together).
• Census is from 2021; satellite and weather from2025.
• Small districts with few sample points are less statistically stable — though the Donji Grad signal is strong enough that conclusions hold.
• Cycling data does not cover every district equally (e.g. gaps for Podsljeme and Brezovica).

See sources and methods for every layer