Al Safa Park 2 — Site Analysis: Findings & Visualisations

gee/ — Landsat archive, 1972–2025

0 · How the site came to be: the green won't grow by itself

Cloud-filtered median composites from the full Landsat archive show the site's whole life: open desert until the late 1970s, the street grid graded by 1985, villa blocks filling in from the Jumeirah side through the 1990s, and the current urban pattern complete by 2005–2010. The park itself is young — its vegetated fraction was 0 % until 2000, climbed to 7 % (2005) and 20 % (2010), peaked at 36.5 % in 2015, and has slipped to 31.9 % by 2025.

A 10 m Sentinel-2 pair confirms the last decade in detail: vegetation is static (park 22.2 % → 19.9 %, ring 6.7 % → 6.9 % on the S2 scale), with grown street trees and cleared plots cancelling out, and a linear vegetation-loss band along the Sheikh Zayed Road side. Meanwhile the community's population grew +54 % over the same decade (5,127 → 7,913): green space per resident is shrinking. The neighbourhood's greening phase is over — more green now only comes by design, which is the historical case for this redevelopment.

Going further back: declassified reconnaissance frames (Corona 1965, KH-9 1973/1983, down to 0.6 m) are confirmed to cover the site on USGS EarthExplorer — a candidate "hero image" sequence for the submission's opening spread.

gee/ — Landsat & canopy baseline

1 · The park barely cools its neighbourhood

A summer land-surface-temperature composite (Landsat 8/9, Jun–Sep 2024–25, 52 cloud-free scenes, median) puts the park at 48.2 °C against 48.7 °C in the surrounding 250–600 m ring — a park cool island of just +0.4 °C. Across the 2×2 km AOI, surfaces range 46.3–53.5 °C, with the Sheikh Zayed Road corridor to the east running above 52 °C. The hottest single scene was 2025-07-27 (park at 52.9 °C). The method is identical to our earlier Al Karama study, so the two parks are directly comparable.

Benchmarked against 14 Dubai parks: near the bottom

Running the same Park Cool Island metric across 14 Dubai parks (1.3–155 ha) places Al Safa 2 at +0.47 °C — rank 9 of 14, in the weakest-cooling class, and independently reproduces our headline +0.40 °C (a method cross-validation). The comparison carries three design lessons the jury can check against real parks:

• Green is not cooling. Al Safa 2 is 85 % NDVI-green (plenty of lawn) yet cools least; across the 14 parks NDVI does not predict PCI, and across 16 neighbourhoods greenness and summer LST are essentially uncorrelated (R²=0.03). What cools is canopy shade, water and scale — not grass.
• Dense canopy beats size. Nad Al Sheba (5.7 ha, 89 % green) cools +2.24 °C — like a large park. Al Safa 2 can't grow, but it can densify and raise its canopy, which is exactly the intervention-matrix planting strategy — now justified by an external precedent.
• Water cools most. The pond parks (Al Barsha +5.7 °C, Al Quoz +4.7 °C) top the ranking — external backing for the water-feature and water-retentive-paving moves.

And nothing nearby fills the gap: the only walkable cool play is paid

Widening the lens from parks to what a family can actually reach: within a 10-minute walk of Al Safa 2 there are only two other options — Khaja Park (a 0.4 ha pocket green with no real programme) and Splash ’n’ Party, a private kids’ waterpark ~400 m away charging AED 99 per child. The nearest free public park of any size, Umm Suqeim Park (2.6 ha), is ~2.8 km — a drive — and runs a women-&-children weekday-access model. The parks that genuinely cool (Safa +1.95 °C, Al Quoz Pond +4.74 °C) are 45-minute-plus walks. Facility character verified against public sources; Splash ’n’ Party geocoded via Nominatim.

• The market has already priced the gap. Families pay for cool water-play 400 m away because the public realm offers none — demand for a free public cool refuge is proven, not assumed.
• Don’t compete on scale. Safa and Al Quoz own the “day-out”; Splash ’n’ Party owns the paid indoor party. The empty niche is a free, public, all-ages, daily, walkable cool refuge.
• SWOT in one line each. S — monopoly on the walkable catchment plus existing bones (2 entrances, sports courts, playgrounds); W — barely cools, canopy-poor, no water/F&B draw; O — fill the walkable cool-refuge gap, water is the strongest lever, Nad Al Sheba proves a small park can cool; T — a paid competitor already owns kids’ water-play, regional pull, the green-≠-cool maintenance trap.

So what can be done: reclaim the adjacent car park

If the park is too small and too canopy-poor to cool, the one adjacent piece of land it can actually annex is the 2,768 m² asphalt car park on the north-west/west edge. Converting it to shaded green space adds 18.8 % to the park's area, 18 new 8 m trees (903 m² of canopy) and +22,187 m²·h/day of shade — a 28 % increase on the park's entire current shade asset — plus ~20 t CO2 stored and 662 kg/yr sequestered.

Honest premise: the car park's summer LST is essentially the same as the park's (+0.1 °C), so this is not about removing a hot surface — it is about adding the three things the benchmark showed the park lacks: canopy shade, tree ecosystem services and usable green area. Because the parcel sits on the hot afternoon (west/north-west) side, right beside the must-keep #113 tree cluster, the new canopy can form a continuous afternoon shade belt rather than an isolated patch. It is the single highest-leverage move in the whole study for a park that cannot otherwise grow.

Canopy structure explains why. Meta/WRI 1 m canopy-height data shows 25.9 % of the actual park polygon under canopy taller than 2 m (13.0 % above 5 m) — note this corrects an earlier “~10 %” figure that was measured over a 150 m circle including roads and villas, not the park itself. Trees are modest: median 5 m, 90th percentile 10 m, tallest 13 m.

gee/ — ERA5-Land + thermofeel, 10 years

2 · Extreme heat is the design condition, not the exception

We built a 10-year hourly thermal-comfort baseline (87,567 hours, Jul 2016 – Jun 2026) computing UTCI in sun and in shade with the ECMWF thermofeel library (Di Napoli et al. 2020 MRT, Erbs direct/diffuse split).

• The 10-year worst day, 2019-08-09, peaked at UTCI 54.7 °C in sun — and still 49.8 °C in shade, both in the “extreme heat stress” category (>46 °C). Peaks in the 52 °C class occur nearly every year (the top-10 days spread across 2016–2023).
• Summer (Jun–Sep) averages 3.5–4.9 extreme hours per day, rising to 5.4–5.5 h/day in July–August. The 10-year trend is flat (−0.8 h/day per decade); year-to-year variability dominates.
• Usable time (median UTCI < 38 °C): July offers 13 h/day in sun vs 15 h in shade. Shade’s biggest wins are the shoulder months — +3 h/day in June and September.
• In deep summer the park is only usable after dark. Median evening UTCI (18–23 h) stays above the “strong” ceiling (32 °C) every hour in Jul–Sep, yet sits below “very strong” (38 °C) for 4–6 of those hours — so the evening is the one window that is merely uncomfortable rather than dangerous, against afternoons at 46–49 °C.
• The day/night optimum flips. The pre-dawn low still holds at 29–30 °C in Jul–Aug (a true tropical night), and after sunset the open sky runs ~1.5 °C cooler than under canopy — open ground radiates its heat to the sky while canopy and built mass trap it. Daytime shade is essential; evening comfort wants open sky plus low-thermal-mass (high-albedo) paving.

shade/ — vector shadow model, hourly

3 · The park must supply its own shade — and the north-east gets none

An hourly shadow sweep for the design day 2025-08-01 (worst recent UTCI day; solar geometry also valid for the 10-year worst day) projects shadows from 1,336 OSM building footprints, 214 tree-crown patches, and the 68 as-built parasols.

• Park shade fraction bottoms out at 32.6 % at noon (parasols and in-park buildings included), recovering to 65 % at 07:00 and 70 % at 18:00. From 10:00 to 16:00, more than two-thirds of the park is in direct sun.
• The hours-in-shade map shows all-day shade only beneath the south-west tree cluster; the north-east quadrant — the playground side — gets roughly 0–2 shaded hours out of 13.
• Surrounding villa shadows reach at most ~45 m and never enter the park: neighbouring buildings will not help; shade must be created inside the park.

The trees worth protecting: five patches carry 93 % of the shade

An asset ledger now covers all 214 canopy patches — height, crown area, trunk count, daily shade contribution and a preservation tier. The park's ten in-park patches deliver 80,328 m²·h of shade per day, and the distribution is extremely top-heavy: the south-west cluster #113 (2,989 m² crown, 19 trunks, 10 m) provides 59 % of in-park shade by itself, and Tier A — just five patches — carries 93 %. Preserving #113 is worth roughly the mature shade of 15–20 of the intervention matrix's new trees. The "trees to protect" are a short, explicit list, not a vague green blob.

Field observations from iNaturalist (19 in the AOI, fetched via the public API and spatially joined to the patches) put real species names on the top asset: #113 is Neem (Azadirachta indica), Cordia and Plumeria obtusa. The field records also fill a satellite gap: some observed date palms and acacia stand outside the Tier A patches — fine-leaved trees the 1 m satellite canopy does not resolve, so ground observation complements the remote sensing. Honest limits: 10 of 19 observations remain at genus level or higher pending community ID, and the CO2 figure (~4.9 t/yr for the park's patches) is an order-of-magnitude estimate until species are confirmed.

Applying urban-forestry (i-Tree) coefficients to the 4,911 m² of in-park canopy turns the earlier single CO2 estimate into a four-service ledger: 108 t CO₂ stored, 3.6 t/yr sequestered, 39 kg/yr of air pollutants removed and 19.6 m³/yr of stormwater intercepted (each with a literature range; the old ~4.9 t/yr sits at the sequestration upper bound, a consistency check). Because Tier A carries 96 % of the canopy, it carries ~96 % of every service — preserving those five patches is the cheapest green infrastructure the park has. Honest limit: the coefficients are temperate/US-derived, so for a sparse arid canopy these are upper-bound figures, and cooling is reported separately (measured shade + the SOLWEIG −4.6 °C) rather than folded in.

The Tamtam cafe sits on the park’s hottest, least-shaded ground

The field visit flagged the east-side Tamtam cafe as too hot and awkward to reach from the north front entrance. Sampling the shade and comfort layers at its location (field/cafe_relocation_check.py) confirms the heat half squarely and reframes the access half. The cafe gets just 1 shaded hour of the 12-hour design day — the 0th percentile, the single least-shaded spot in the park (park mean 6.4 h). Its afternoon UTCI is 53.1 °C (85th percentile — hotter than 85 % of the park) and its mean radiant temperature 74.6 °C, +28 °C above the SW ficus canopy. The handheld FLIR independently read p98 46.7 °C at the cafe plaza on the evening visit. A cafe is where people sit and linger, so it currently occupies the worst-possible ground for dwelling.

• Access: near, but not usable. The cafe is only ~30 m from the north entrance, so the objection is not distance — it is that it sits off the shaded central palm spine, on the hot east edge, so the approach crosses the park’s least-shaded ground (the access model already found no cool route at 15:00).
• The move works only if the new spot is shaded. The entrance gate itself is barely better (3 h, 53.0 °C); but a position ~25 m in, within the date-palm plaza, already gets 7 h shade / 52.7 °C (park median). So the data backs relocating to a shaded front position, paired with shade provision — not simply “to the entrance”.

solweig/ — 1 m microclimate simulation

4 · What tree shade is actually worth: one full UTCI category

A SOLWEIG simulation (solweig-gpu, 1 m resolution, hourly) maps mean radiant temperature and UTCI across the park for the design day. At the 15:00 peak:

• Cross-validation: SOLWEIG’s open-ground UTCI of 52.6 °C at 15:00 matches the independent ERA5 + thermofeel point calculation (52.6 °C) exactly — two methods, same answer.
• Canopy stays 4–5 °C UTCI (and >20 °C T_mrt) below open ground throughout 11:00–17:00 — yet 48 °C is still “extreme”. Tree shade alone cannot make a midsummer afternoon comfortable, but one category’s difference is the difference in actionable hours.
• Spatially, the only cool refuges are the south-west tree cluster and thin building-shadow strips; the north-east records the worst values — consistent with the shadow model.
• Field check (4 Jul): a FLIR thermography session measured sunlit walkway at 45.9 °C vs tree-canopy zone 35.4 °C — a 10.5 °C surface contrast, in the rank order the models predicted (the two locations the model marks least-shaded measured hottest of all 18 frames). Note these are pre-sunset values, not daily maxima; the 14–16h peak window is planned for visit 2.

Testing the fixes in silico: the intervention matrix

The proposed moves were re-simulated in SOLWEIG (now with a DXF-derived landcover grid). The combined package — 31 new trees on the shade-poor zones, a water-retentive plaza, the NW mist patch and cool coating on paths — delivers UTCI −1.3 °C mean / −2.2 °C at the 95th percentile at 15:00, improving 45 % of the park and cutting extreme hours from 7.05 to 6.61 h/day.

• The model derived a design rule: high-albedo paving alone worsens pedestrian UTCI (+0.4 °C mean, +1.2 °C p95) — shortwave bounces onto people. Reflective surfaces belong on roofs; walkable ground gets water-retentive paving.
• Mist reads ~0 by construction — SOLWEIG does not model evaporative air cooling; the mist case rests on the wet-bulb-depression analysis in Finding 6, not this table.
• Night (21:00) deltas are ~0 in the model while FLIR measured 39–46 °C paving at sunset — the open validation point that visit-2's 21:30 loop will test.
• Scale note: trees read −0.8 °C here as a park-wide walkable mean; the −4.6 °C above is the local under-canopy effect — different denominators.

What the camera adds: materials matter after sunset

The thermal pairs also carry a finding the models don't: surface material governs evening heat. Paving and the boundary wall were still at 39–46 °C with the sun nearly down — stored daytime heat re-radiating exactly when night programming wants people in the park. Meanwhile playground sand under the shade structure sat at 35.4 °C with a tight spread (34.4–36.0 °C) — effectively air temperature — and grass read ~33 °C, cooler than adjacent paving even in evening shade. Shade plus a low-mass surface works; hardscape stores what vegetation sheds.

access/ — OSMnx walk network

5 · A porous edge puts 62 more households within a 5-minute walk

Who is this for: the official community, Al Safa Second (no. 357), holds 7,913 residents (end-2024, DDSE Population Bulletin 2024, Table 5) at ~4,180/km² on the official 1.893 km² boundary. The series runs 5,127 (2010) → 7,217 (2015) → 6,823 (2020) → 7,443 (2023) → 7,913 (2024) — growth resumed after a 2020 dip. The park's catchment also reaches the neighbouring communities: Al Wasl 13,661 · Al Safa First 10,237 · Umm Suqeim First 13,902 · Al Manara 9,838 · Umm Al Sheif 5,158.

Finer attributes exist only emirate-wide (nothing community-level since Census 2005): 56.7 % of residents are aged 25–44, ~13.1 % are children (0–14) and only ~1.7 % are 65+; the emirate's 68.5 % male skew is driven by labour-camp communities and is balanced at household level — the defensible assumption for a villa district; ~92 % expatriate; average household ≈ 4.0 persons. People of Determination have no published count at any level (the only official figure is UAE-wide: 15,782 card holders, 2016) — inclusive-design sizing uses the WHO ~15 % benchmark with the gap stated. Full citations: demography/README.md.

True network-distance analysis on the pedestrian graph (1.2 km radius, 1,853 nodes) against 1,323 villa households. Today the park has effectively two entrances, both on the south-west. Adding openings at ~40 m spacing along the feasible edges only (the east side abuts Umm Suqeim Model School and the north a rehabilitation facility — neither can open) yields eight entrances.

• Households within a 5-minute walk (400 m network distance): 261 → 323 (+62, +24 %). Full-perimeter opening would be the theoretical ceiling at +96 (+37 %).
• Median detour factor (homes within 700 m straight-line): 1.94 → 1.85 — many residents currently live “behind” the park yet must walk around to its front.

Round 2: which gates matter — and the walk itself is the heat barrier

Attaching the shadow model to every edge of the walk network shows the approach streets are 13 % shaded at 08:00, just 5 % at 15:00, 19 % at 17:00 (length-weighted). At the peak there is no cool route to the park — the coolest route and the shortest route are identical everywhere. The walk to the park is the biggest heat barrier, which is the evidence for street trees and shading beyond the park fence.

• Heat-dose gate ranking (walking minutes × sun/shade UTCI): G4-west serves 375 households, existing G1 323, G2-south 239 — the other candidates capture zero. Eight candidate openings collapse to three priority gates covering 937 of 942 villas — roughly 3,750 residents.
• Flow corridors: origin–destination routing puts up to 293 households on a single western street — the three main corridors are the priority segments for continuous street-tree shade.
• Families: 159 households have both the park and a nursery within a 5-minute walk — a child-mobility "golden zone" concentrated north-east/east of the park.

A shade-seeking pedestrian model (an agent-based method from the technology scan, run over the SOLWEIG 1 m UTCI grid as a cost surface) tests how much a visitor can escape the heat by choosing shaded routes. A 12 % longer walk cuts time-in-direct-sun by 19 % (−1.7 °C per metre walked) — but the honest finding is that cumulative heat dose barely changes: the whole park sits at 47–54 °C, so a detour cannot outrun it. The design implication is sharp — shade belongs on the direct desire-lines people actually walk, not on scenic detours, and the model’s desire-path map is the priority list for where to put it.

gee/ — intervention scenarios

6 · Water is an afternoon tool; the porous edge also buys usable hours

Evaporative cooling potential (wet-bulb depression, 10-year record): summer afternoons 14:00–17:00 hold a median of 13.5 °C of depression (>10 °C in 92 % of hours) — misting and water features work hardest exactly at the peak. But evenings 19:00–23:00 drop to 6.7 °C (below 8 °C in 70 % of hours): water for the afternoon; shade and airflow for the night.

Ventilation scenarios (walled edge = 40 % of free wind, porous = 70 %, open = 100 %, same radiation): a porous edge recovers +1 usable hour/day in Jul–Sep and +2 h in May versus a walled park, while peak UTCI barely moves (41.4 → 41.0 °C). The porous edge doesn’t cool the peak — it wins back usable time at the edges of the hot season. An OpenFOAM CFD upgrade of this scenario model is in progress (watertight geometry done: as-built wall vs 8-opening porous variant).

siting/ — weighted suitability grid

7 · Where the splash zone should go: the north-west, upwind and tree-shaded

A 1 m weighted overlay scores every cell for water-play siting: afternoon shade accumulation (×0.3) + playground proximity (×0.3) + upwind position (×0.2) + open sightlines (×0.2). Summer daytime wind is remarkably consistent: 2.1 m/s from 302° (NW) — Dubai’s sea breeze — so cooled air from a NW water feature drifts across the park.

• Optimal cell: 25.155971, 55.221620 (suitability 0.90) — near the west/north-west edge, east of the large canopy cluster (afternoon tree shade), close to the central playground.
• The Tamtam-adjacent plaza scores 0.48 — below the park average (0.54): downwind and short on afternoon shade. A café-splash pairing there would pay a 0.42 suitability penalty.
• Framing: splash at the NW optimum, plus a shaded route linking it to the Tamtam kiosk — the café keeps its draw, the water sits where the grid says it works. The grid arbitrates the design decision.

water/ — FAO-56 ET₀ + WUCOLS hydrozones

8 · More shade, less water: a native palette halves irrigation

The cooling proposal adds canopy — which invites the obvious objection in a water-scarce city: won’t more greenery cost more irrigation? We built a full landscape water budget to answer it. Reference evapotranspiration (FAO-56 Penman–Monteith) from the same 10-year ERA5-Land series is 1,814 mm/yr (summer 6.3 mm/day) — the rate at which Dubai’s heat pulls water out of any planting. Applying WUCOLS landscape coefficients to each hydrozone (park 14,740 m²; 9,115 m² hardscape is not irrigated; 5,625 m² is plantable):

• Today the plantable area is almost entirely cool-season amenity turf — a high-water-use grass — on spray irrigation. Baseline demand ≈ 8,500 m³/yr.
• Proposed: keep a 1,500 m² activity lawn, convert the rest to a native / drought-tolerant palette (shrub + groundcover + mulch) on drip. Demand drops to ≈ 4,200 m³/yr — a 50% cut, saving ~4,300 m³/yr, about 21 Dubai households’ annual water.
• Greenery goes up, not down. The 4,911 m² of in-park canopy is kept and densified, plus +18 trees / +903 m² crown / +22,187 m²·h/day of shade (expansion/). Native trees also cool by transpiration while drinking far less than turf.
• Even the full expansion (the NW car park re-planted as xeric park, +18% green area) uses ≈ 5,700 m³/yr — still 33% below the as-built turf park.

viz3d/ + reference/ — interactive model & as-built

9 · Interactive 3D explorer, grounded in the as-built drawings

A deck.gl web viewer combines all layers for design discussion: extruded OSM buildings and tree crowns, an hourly shadow slider (06:00–18:00, same engine as the shadow analysis so the pictures match the numbers), the hours-in-shade overlay, and the SOLWEIG UTCI overlay. It is now live for everyone at lamma-park.pages.dev/viz3d/alsafa2_3d.html — password-gated — any username, password from Makoto (the project landing page at lamma-park.pages.dev also links the POI and photo maps).

The competition’s as-built DWG has been georeferenced (similarity fit, 0.65 m residual) and mined: 66 hexagonal parasols + 2 large shade structures and the round Tamtam kiosk are now real geometry in the shade, siting and 3D analyses — so “existing shade” is no longer just trees.

A second, geometry-faithful model assembles the same as-built data into a portable 3D scene (trimesh → glTF/GLB) shown in a browser viewer lit by the real design-day sun with cast shadows (NOAA solar position, 06:00–18:00 slider) at lamma-park.pages.dev/proposal_1/view_proposal.html (the consolidated viewer — its “Before” model). It carries the buildings, both walls (with their real gate gaps), the octagonal Tamtam kiosk and its pergolas, the three recessed sand playgrounds with their equipment, the sports goals/hoop/net, and path furniture — with trees typed palm vs broadleaf. Its elements were built and corrected against the site photos, the 360 panoramas and the user’s Mapillary imagery; what is surveyed vs. representative is logged in anim/twinmotion/MODEL_PROVENANCE.md. It is the base geometry for the Twinmotion photoreal render and the concept film.

field/ + pano/ — 360° site survey (Insta360 X4), 2026-07-09 evening

10 · Field visit: a 360° survey ground-truths the model

An evening walk-through captured ten equirectangular panoramas and a dusk timelapse. The files carry no GPS (auto-geotagging was off), so the shooting points were fixed on-site by the photographer on a satellite map and then georeferenced (affine fit on the park corners + kiosk; the park outline and all three playgrounds reproject to within ~10–15 m). The panoramas themselves — read against the mosque to the south, the floodlight mast, the central date-palm allée and the perimeter buildings — supply what the frames actually contain at each point.

The survey corrected things the as-built drawings and OSM missed, now fixed in the 3D explorer: the walled sports enclosure is two courts — a sand beach-volleyball court plus a green futsal pitch, not one; a ~16 m floodlight mast stands on their shared edge (absent from every source layer); and a perimeter building was re-set to its observed 2-storey height. The walk also fixes the park’s front: the main entrance, the front car park and the clinic sit on the north side, with the sports enclosure reached through a single gated wall to the south. These are the same panoramas that put field-verified species names on the canopy patches, so the model now answers to the ground. The 3D explorer now carries a clickable 360 photo at each of the ten confirmed spots — a “Field 360 photos” layer whose pins open an in-page pan/zoom panorama (with an auto-rotate button), turning the model into a walk-through of the real site.

What the numbers add up to

• The one land lever is the adjacent car park. For a park the benchmark ranks 9/14 and calls too small to cool, converting the 2,768 m² NW car park adds 18.8 % area and a 28 % jump in shade — the highest-leverage single move, and the parcel is not even a hot surface (its value is shade and usable green, not removing heat).
• Shade is deficit №1, and it must be home-grown. The park cools its surroundings by only +0.4 °C, sits two-thirds in direct sun through the extreme hours, and no neighbouring building can shade it. The near-shadeless north-east (playground) quadrant is the priority zone.
• Sell shade honestly: it buys one UTCI category (−4.6 °C) and 20+ °C of radiant relief — not midsummer comfort. Its real yield is +3 usable hours/day in June and September: shade extends the season.
• Split the day: evaporative water features carry the afternoon peak (13.5 °C wet-bulb depression); nights favour open sky and airflow, not canopies — so night programming belongs in the open areas.
• Choose materials like they're infrastructure: FLIR measured paving at 39–46 °C at sunset against grass at ~33 °C — low-mass, vegetated surfaces and less continuous paving are what make the night programme physically comfortable.
• The porous edge is a twofer — and it is three gates, not eight: G4-west, existing G1 and G2-south serve 937 of 942 villas — roughly 3,750 residents at the DDSE average household of 4.0 (+24 % within 5 minutes), and ventilation recovers +1–2 usable hours/day at the season’s shoulders.
• Site water NW-upwind: the park has no existing fountain, so the splash zone is a free choice — and the suitability grid puts it at the north-west edge (0.90) over the Tamtam-adjacent plaza (0.48), riding the 302° sea breeze across the park.
• Methods are cross-validated and comparable: SOLWEIG and the ERA5 point model agree at 52.6 °C, first FLIR field measurements match the predicted rank order, and LST/shade methods mirror the Al Karama study — a strong story for the AI Methodology Report (20 % of the score).

Data sources

Everything below is free and openly accessible except the competition-supplied CAD (used under the challenge's IP terms, converted locally, never redistributed). Satellite and reanalysis data come through Google Earth Engine (project uobdubai; free for research use).

Models & tools (all open source): Google Earth Engine · thermofeel 2.2.0 (ECMWF) · SOLWEIG (solweig-gpu 1.2.22, Lindberg et al.) · OSMnx · deck.gl · OpenFOAM (in progress) · GeoPandas/Rasterio stack. Method choices mirror our earlier Al Karama study so results are directly comparable.

Brief coverage tracker — 32 tasks, 8 categories

The competition brief expects AI use across eight areas. Mapping this report against the team task tracker: 16 covered · 10 partial · 6 not started. Environmental analysis is complete; the user-side and design-side items are the remaining work.

Caveats to carry into any deliverable

• ERA5-Land is one ~9 km grid point — a neighbourhood-scale time baseline, blind to in-park microclimate (that is what the SOLWEIG maps and the FLIR field measurements are for).
• Shade UTCI assumes MRT ≈ air temperature — real tree shade transmits and reflects some radiation, so shaded values are slightly optimistic.
• Tree crowns are projected as opaque solids — dappled shade is over-counted; crown depth is ignored; the FLIR check shows shade is also over-predicted at grazing sun angles (harmless for the 10–16h decision window).
• 97.5 % of building heights are the 7.5 m villa default (Al Safa is a homogeneous two-storey villa district, so impact is limited, but local 3-storey buildings are under-shaded).
• LST is surface temperature, not felt heat — we use LST for spatial pattern and UTCI for the human story; keep them distinct in the report.
• SOLWEIG ran one design day; the baseline maps used uniform ground cover, while the intervention matrix uses the DXF-derived landcover grid. Other days need only a new met file.
• Entrance detection is OSM-based; the as-built CAD remains authoritative for boundary and gate positions.
• No confirmed hot-arid Gulf precedent exists for AI-in-park-design (from a verified technology scan, research/tech_scan.md) — a genuine novelty claim for the entry, but every cooling figure borrowed from milder-climate precedents is a method that transfers, not a result that will replicate at Al Safa 2.
• Demography is community-level for population only — age, gender and nationality exist only emirate-wide (nothing local since Census 2005); People of Determination has no published count at any level, so inclusive-design sizing uses the WHO ~15 % benchmark with the data gap stated. Population figures for 2022–24 were revised under the register-based "Dubai Population Now" method — do not mix them with older bulletins. (And 61,353 seen in search snippets is Al Qouz Third, not Al Safa 2.)

Visualisation inventory

Summer LST composite, park vs ring (PCI +0.4 °C)

gee/lst_summer_map.png

Park Cool Island across 14 Dubai parks — Al Safa 2 ranks 9/14

benchmark/park_comparison.png

16-neighbourhood LST/greenness benchmark

benchmark/neighborhood_benchmark.png

Car-park-to-park extension — +19% area, +28% shade

expansion/parking_to_park.png

1 m canopy height, AOI

gee/canopy_height_map.png

Median UTCI, hour × month, 10 yr

gee/utci_heatmap.png

Worst day 2019-08-09, sun vs shade

gee/utci_worst_day.png

Usable hours by month, sun vs shade

gee/utci_usable_hours.png

10-year extreme-hours trend

gee/utci_trend.png

Evaporative-cooling potential, hour × month

gee/evap_potential_heatmap.png

Walled / porous / open edge scenarios

gee/wind_scenarios_usable_hours.png

Development history 1972–2025 (Landsat archive)

gee/historical_development.png

2015 vs 2025 Sentinel-2 pair + ΔNDVI change

gee/change_2015_2025.png

Hours-in-shade map, 1 m

shade/shade_hours_map.png

Hourly shadow snapshots, design day

shade/shadow_snapshots.png

Tree asset ledger — Tier A 5 patches = 93 % of shade

trees/tree_inventory_map.png

iNaturalist species over the canopy inventory

trees/inat_overlay_map.png

Tree ecosystem-services ledger (i-Tree) — 108 t CO₂ stored

trees/tree_benefits.png

T_mrt & UTCI maps at 15:00 peak

solweig/solweig_peak_maps.png

Intervention matrix — combined package −1.3 °C, 45 % improved

solweig/intervention_matrix.png

5-min walkshed, 2 vs 8 entrances

access/walkshed_before_after.png

Approach-street shade — 5 % at 15:00

access/street_shade_map.png

Heat-dose gate assignment (3 priority gates)

access/heat_dose_map.png

Shade-seeking pedestrian ABM — shade goes on desire-lines

access/shade_abm_map.png

Walking flow corridors — street-tree priority

access/betweenness_map.png

Nursery × park 5-min overlap (159 households)

access/isochrones_kids_map.png

Water-play suitability grid (Tamtam kiosk marked)

siting/fountain_siting.png

FLIR session 1: planned targets vs geolocated shots

flir/plan_vs_actual.png

Five annotated visible/thermal pairs (materials evidence)

flir/report_pairs/

3D explorer — shadow slider + overlaysinteractive

lamma-park.pages.dev/viz3d/…

3D analysis view with as-built parasols

viz3d_parasols.png

Site panorama + 46 geotagged field photos; POI map (181 POIs)interactive

pano/ · photo map · POI map