UrbanPulse · Singapore

Greenery that proves who it cools.

One score for every HDB block. A loop that lets residents verify what greenery delivers. A pass that turns mandatory greenery into returns developers can bank.

See the five findings Explore the atlas

UrbanPulse/ Atlas / Singapore, all HDB townsneed-weighted2021–25

Blocks scored

10,749

Block score

less resilientmore

Pillars · weight

Heat · 25

Shade · 20

Green-blue · 15

Community · 15

Access · 15

Flood · 10

10,749HDB blocks scored (99.4%)

27towns benchmarked

89Landsat scenes, 1988–2025

16,087 kmwalkways mapped for shade

00Highlights

Five things 10,749 blocks told us.

01 / 05

01 · Inclusion

24points

hidden inside one town's average.

In Bedok, the typical block outscores its worst-served fifth by 24 points. A town average cannot show that, so every UrbanPulse plan reports the gap.

02 · Cooling

−0.5°C

for every +10 points of tree canopy within 200 m.

Tree canopy is the strongest cooling lever a planner controls. The effect holds with town, paving, block height and age fixed, and across 27 models that each predict a town they never saw.

03 · Markets

≤0% premium

Buyers don't pay for cooling yet. Lenders can.

Across 73,304 HDB resales, block-level canopy earns no price premium. UrbanPulse makes cooling bankable instead: verified outcomes unlock sustainability-linked loan step-downs.

04 · Developers

−41%

30-year greenery cost, same URA rules.

On a 2023 Woodlands GLS site, the index-guided design adds a median S$2.1M per site after fees, positive in all 20,000 simulations. It also puts 898 seniors and rental households within 200 m of a public cool refuge.

05 · The moat

+1.4S$M

bankable value per site after 8 verified sites.

Unverified, a lender's cautious (P10) case is −S$0.6M per site. Each site residents verify narrows the forecast for the next. That evidence library is what a competitor cannot copy.

01The problem

Planners can model wind and sun, but not who a plan cools.

Singapore is getting hotter, and seniors, low-income households and rental-flat tenants are the least protected. Evidence on heat, greenery, flooding and vulnerability sits in separate datasets, often at planning-area level. That makes it hard to see which past design choices worked for residents, or which vulnerable groups a new plan might leave out. Residents, who could help cool their neighbourhoods, have no role and no incentive.

New HDB blocks under construction in Tengah, with bare ground and little shade

~½

of 400+ HDB flats studied were hotter inside than outdoors.

SUTD home-heat study, via Her World, May 2026

>90%

of rental flats surveyed lack air-conditioning.

Chew, Chung & Ang, WSBE 2026 abstract (preliminary)

heat targets in the Green Plan 2030, which counts parks, trees and buildings.

Singapore Green Plan 2030 targets

Who	Pain point observed	What we resolve
Vulnerable residents	Most exposed, least protected: about half of 400+ HDB flats studied were hotter inside than outdoors.	Plans scored on who they cool, with an inclusion check for seniors and rental households.
Planners	Tools model wind, sun and temperature, not people; evidence is scattered and at planning-area level.	One map, block-level scores, benchmarks from towns that worked.
Developers	Greenery is mandatory (30–40% of site area; Green Mark Platinum Super Low Energy on government-sold land) but seen as a cost.	A project pass that turns required greenery into returns: sustainability-linked loans, lower upkeep, subsidies, rent.
Developers & lenders	Green claims can't be verified after completion.	Verified post-occupancy data from community credits, e.g. kg harvested on rooftop farms.
Everyone	Rules count leaves, not beneficiaries; the Green Plan sets no heat target.	An index measuring cooling, access and use by people, kept current by residents' reports.

02Gaps in today's toolkit

The physics is modelled well. The people layer is missing.

Existing tools model the physical environment. What is missing is the people and equity layer, and a link between past design choices and outcomes.

Gap 01

Design tools model physics, not people

HDB/A*STAR's Integrated Environmental Modeller (used for Tengah) simulates wind, sun and temperature; URA's cooling page does not mention elderly or vulnerable residents.

URA Master Plan 2025: A Cool City

Gap 02

Greenery rules count leaves, not who benefits

Green Plot Ratio 3–4, tiered by plot ratio and location, not by heat or vulnerability.

URA greenery requirements

Gap 03

Liveability and heat are measured separately

A 203-subzone liveability index has no heat; heat-vulnerability maps focus on exposure and population.

Benita, Kalashnikov & Tunçer 2021

Gap 04

Heat risk mapped too coarsely

The Singapore-ETH 2020 vulnerability map is by planning area, far coarser than the block where people live.

Mothership, Feb 2024

Gap 05

No town-level link from design to outcome

City-level heat is tracked 1990–2024; no town-by-town study linked to design choices was found.

npj Urban Sustainability 2025

Gap 06

Indoor heat is unmapped

About half of 400+ flats hotter inside than outdoors; more than 90% of rental flats lack air-con.

Her World 2026 · WSBE 2026

Gap 07

No heat target

The Green Plan 2030 targets parks within a 10-minute walk, trees and buildings, but sets no heat target.

Green Plan 2030

Gap 08

Averages hide inequity

Green-space cooling is inequitably distributed; lower-income groups drive the gap.

Xu et al. 2025, Ecosystem Services

03The proposal

Data flows in, scores flow out, and revenue funds the residents who report.

An interactive platform that helps planners and developers design towns that are cooler, greener, more flood-resilient and inclusive, before anything is built. It gets smarter with every town Singapore builds.

1 · The index

Open data on one map scores every block 0–100 on six pillars. Planners and developers upload a plan and get a score, an inclusion check and nature-based tweaks showing the gain each would bring.

2 · Community data

Residents and Active Ageing Centres earn credits for verified actions and reports, led by rooftop farms (kg harvested). Reports validate the model and re-learn its weights, so each town improves the next. This is the piece open data cannot provide.

3 · How people pay

Developers, starting with Government Land Sales sites, buy a project pass that turns mandatory greenery into returns, backed by verified post-occupancy data for ESG reporting. Planners buy seats or a licence. Part of the revenue funds residents' credits.

Solid lines: data and tools. Dashed: payment. Green: the community-data loop, the element no existing tool has.

What is new

	Today	UrbanPulse
Unit of analysis	Planning area, or one site at a time	Every HDB block, and who lives in it
What is measured	Wind, sun and temperature (physics simulators); leaf area (Green Plot Ratio)	Cooling, shade, access and use by people, with a confidence flag on every indicator
Equity	Averages	Need weights for seniors and rental households, plus an inclusion gap on every plan
After handover	No data; green claims cannot be verified	Verified community data: kg harvested, use, upkeep, comfort, species
Money	Greenery is a compliance cost	Verified KPIs unlock sustainability-linked loan step-downs and lower upkeep
Learning	Each new town starts from assumptions	Each verified site re-learns the weights for the next

03bThe whole flow

Seven steps, one loop.

01 / 07

Step 01

Integrate

One live map of heat and felt heat, satellite heat history, greenery, flooding, buildings, mobility, seniors and access, all from open data.

Here: 16 data.gov.sg datasets + the PUB flood list, 89 Landsat + 7 Sentinel-2 scenes, an OSM walk network.

Step 02

Learn & benchmark

Score existing HDB towns down to block level, track how their heat changed, and turn the best performers into design targets.

Here: 27 towns ranked; targets from the top-5 towns per indicator.

Step 03

Score the scenario

A planner or developer draws or uploads a scenario; the index scores it on all six pillars and runs the inclusion check.

Fixed national reference, so a gain is a real move up the distribution.

Step 04

Suggest tweaks

Specific nature-based changes with their score gain. The user accepts or rejects each; whole generated alternatives come later, once community data has validated the model.

Here: green pockets, rooftop farms and shaded walkways for Woodlands.

Step 05

Check inclusion

A plan cannot score well on average if its worst-served 20% of blocks, or its senior flats, fall far behind.

Here: an average-first package widens the gap; an inclusion-aware one closes it.

NEW

Step 06

Keep it alive with the community

After move-in, residents and AAC seniors earn credits for verified shade, green pockets and neighbour check-ins; actions and indoor readings feed back into step 2.

The loop that turns a static index into a learning one.

Step 07

Show the value

A separate value lens shows resale effects of cooling and greening, kept outside the score, so seniors' homes are never marked down.

Here: 73,304 resale transactions, 2024–2026.

Each new town feeds the next through community data. "Here" notes what this repository implements.

04Part 1 · The index

Six pillars, every block, 0–100.

Higher means more resilient. Each indicator becomes a percentile across all HDB blocks, and a pillar is the mean of its indicators. Vulnerability weights the result rather than adding to it. Added as a pillar, a block would score higher simply by housing fewer seniors.

How the pillars fit together

indicator → percentile 0–100 pillar = mean(indicators) block = Σ w·pillar / Σ w need = (seniors + 1.5·rental) × 1.3 if built before 1990 town = Σ need·block / Σ need inclusion = median − mean(worst 20%)

Default weights w = 25/20/15/15/15/10. Seniors per block = non-rental units × 3.0 persons × the Census 2020 share aged 65+ in that subzone. Shade enters access only through heat-adjusted walking routes, where an unshaded metre counts as 1.5, so it is not counted twice.

Why these inputs: five tests

An input is in only if it passes all five; anything that fails is parked for phase 2. One candidate failed during the build:

Input	Evidence	Plannable	Block-level	Not redundant	Predictive

05Atlas

One map, every block, your weights.

Switch layers, re-weight the pillars, and watch the town ranking and inclusion gaps respond. Click any block for its pillar breakdown.

Town ranking (need-weighted)

Click a town to zoom the map.

Inclusion gap

Median block minus the mean of the worst-served 20%. A town with a large gap leaves part of its residents behind even if its average looks fine.

06Satellite heat history

Three decades of surface heat, block by block.

Drag the slider to compare 1988–97 (Landsat 5) with 2021–25 (Landsat 8/9). Each scene is cloud-masked and expressed relative to that day's island median. The maps therefore show where heat moved, not absolute warming.

Estates cool as their trees mature. Blocks built in the 1980s are now relatively cooler than in the 1990s. Blocks built 1998–2009, on land that was green in the 1990s, warmed most (median +1.1 °C relative).

Canopy is the strongest plannable lever. Holding town, imperviousness, height and age fixed, +10 percentage points of tree canopy within 200 m goes with about −0.5 °C surface heat.

2021–251988–97

−6 °C coolerisland median+6 °C hotter

Figure 1: surface heat maps, change by build era, heat drivers and canopy partial dependence — **Figure 1.** a Median surface-heat anomaly 2021–25. b Change since 1988–97. c Change by block build era (bars: 5–95% and 25–75% of blocks). d Standardised OLS coefficients with town fixed effects; bars are town-clustered 95% and 50% confidence intervals. e Partial dependence of surface heat on canopy across 27 gradient-boosted models, each trained with one town held out. The official ST band has gaps in north-east Singapore, so LST is computed with a single-channel method and NDVI-based emissivity. Where both exist, it matches the official product with RMSE 0.99 °C.

07Learn & benchmark

Turn the best performers into design targets.

For each plannable indicator, the target is what the median block in the five best-performing towns already achieves. These targets are observed in Singapore, not assumed.

Naturalised Kallang River at Bishan-Ang Mo Kio Park

Indicator	Island median	Target	Best towns

Precedents the index can learn from

Singapore already builds blue-green infrastructure. What is missing is evidence of who it serves; each of these can become a scored, community-verified design pattern.

Aerial before and after of Bishan-Ang Mo Kio Park — **Concrete canal → river park.** Bishan-Ang Mo Kio Park turned a 2.7 km concrete canal into a naturalised river (PUB ABC Waters). Pagodashophouse, CC BY-SA 3.0

Punggol Waterway at dusk — **Blue-green spine for a new town.** Punggol Waterway; Punggol's blocks are, on average, the coolest of any town relative to the island median. Erwin Soo, CC BY 2.0

HDB blocks beside a planted canal near Segar LRT — **Everyday water edges.** A planted drain beside blocks near Segar LRT in Bukit Panjang, the top-ranked town. Teojincheng, CC0

Covered walkway leading to a bus stop near Block 670A — **Shade to the bus stop.** Covered linkways like this one, expanded under LTA's Walk2Ride programme, count as fully shaded in heat-adjusted walks. LN9267, CC BY-SA 4.0

Lush community garden in Jalan Senang estate — **Gardens residents keep.** Community in Bloom gardens are today's proxy for the Community pillar. Wzhkevin, CC BY-SA 4.0

New HDB blocks and construction in Tengah — **The next test.** Tengah, the "forest town", scores lowest today because its trees, gardens and amenities are still being built. ZKang123, CC BY-SA 4.0

08Suggest tweaks · check inclusion

Woodlands: which six changes, for whom?

The engine screens three kinds of change: green pockets on URA Master Plan 2025 reserve and open-space parcels, community farms on HDB car-park roofs, and shading for the walkways residents use most. Each shows the score it adds and who it reaches. Tick to build your own package.

Figure 3: Woodlands tweak package, comparison of two rankings, and who gains — **Figure 3.** Six tweaks chosen greedily, at most three per type and at least 300 m apart. Re-ranking the same candidate pool by gains for the worst-served 20% gives up 0.5 points of town average, and the inclusion gap closes instead of widening.

09Part 2 · Community data

Open data predicts. Residents verify.

Two parks with the same score may draw 40 versus 5 seniors an hour because only one has shaded seats and good upkeep. Community reports tell the model how green spaces are actually used, and credits keep the reports coming.

credits = Δ indicator × block need weight (more seniors and rental households → more credits) report-only actions → fixed amount under-sampled spaces → bounty

Covered walkway with benches at Ghim Moh

Community credits

Action	Indicator it feeds	Verified by	Output measure
Shade added along a route (sail, trellis, planter)	% of route shaded: recalculate that segment	Photo + town council	Shaded metres added; route shade (pp)
Tree planted	Canopy within 200 m, once established	NParks or town council	Crown area added (m²)
Green pocket created or maintained	Seniors within 200 m of a cool refuge	Photo + town council	Seniors newly within 200 m
Neighbour check-in on hot days	% of at-risk seniors visited on high-WBGT days	AAC	% visited per 30 days
Indoor temperature reading	Indoor heat (measured, not modelled)	AAC or volunteer	Indoor °C per block (mean, peak)
Join or host a garden session; bring a new neighbour	Social connection: attendance, new ties	QR at site + AAC	Attendees by age; first-timers
Quarterly 2-minute survey (3-item loneliness scale)	Loneliness trend	Anonymous	Mean score per block (3–9)
Report a garden problem; adopt a bed; watering shift	Condition and upkeep	Geotagged photo + town council	Open issues; days to fix; stewardship hours
Rate a garden after a visit	Restorative quality	QR at site	Mean rating by time of day
Plant natives; log butterflies and birds	Species seen per month	Photo + garden lead	Species per site per month

Hero feature: the rooftop farm

Its output, kilograms harvested, is easy to verify and links to several returns. HDB car-park rooftops have already been tendered to urban farms (2020 round: 9 sites, 1,808–3,311 m², 3-year terms). Urban farming is already a target in a S$300M sustainability-linked loan to a Singapore developer.

Metric	Community source	Return it proves
Harvest (kg / month, by crop)	Logs weighed at collection	Loan target met; produce or rent income
Volunteer hours, residents involved	QR check-ins	Social connection; ESG reporting
Produce shared with seniors	AAC confirms receipt	Inclusion; ESG reporting
Roof surface temperature, farmed vs bare	Volunteer readings + satellite	Top-floor cooling (hypothesis to test)
Crop failures, pests, time to fix	Upkeep reports	Lower maintenance cost

Rules

Verify before paying; only the first valid report of a problem counts.
Pay for taking part, never for the answer.
Weekly cap per person, so a few users don't dominate the data.
Physical actions decay unless re-verified (the period is adjustable per action type).
Maps separate measured, claimed and human-reported data; scores recalculate monthly.
Seniors can log through AAC staff on paper; no smartphone needed.
Consent, and aggregates by block or site only, in line with the PDPA.

How community data improves the model

Validate: test whether high scores match real use and comfort.
Learn weights: set each indicator's weight by how well it predicts use and comfort, instead of guessing.
Set targets: features linked to high use (e.g. shaded seats every 100 m) become benchmarks.
Ground the simulator: replace assumptions (seniors use a cool spot within 200 m) with Singapore observations.

Status: no community data exists yet, so the Community pillar uses NParks Community in Bloom gardens within 400 m as a low-confidence proxy.

10Part 3 · How people pay and benefit

Making mandatory greenery pay.

Greenery is a compliance cost every developer already carries. The project pass turns it into measurable returns. The verified data behind those returns is what open data cannot give, so it is what customers pay for.

Who pays, for what

Customer	Product	Indicative price*	What they get
Developers, starting with Government Land Sales sites	Project Pass, per site	S$80k one-off	Score the design against its 400 m neighbourhood; URA greenery compliance check; tweaks that place public green space where residents need it most; return calculator
Developers and lenders	Monitoring, per development	S$36k a year, 5 years	Verified post-occupancy KPIs (harvest, microforest survival, use, comfort) for sustainability-linked loans, marketing claims and ESG reporting
Owners of existing buildings	Retrofit monitoring	S$15k a year	Evidence for Skyrise Greenery Incentive retrofits, which co-fund existing buildings
Planners and agencies	Professional seats or an annual licence	by tender	Scenario scoring for new towns, inclusion check, benchmarks from towns that worked
Residents and AACs	Credits, funded by part of revenue	—	Vouchers or health-app points, garden steward recognition

* Illustrative assumptions used in the business model below, not quotes.

What is already required

Requirement	Demands	Confidence
URA landscape replacement + Green Plot Ratio	Greenery of 30–40% of site area (70–100% in strategic areas); GnPR 3–4; publicly accessible	checked
BCA Green Mark (mandatory)	New buildings of 2,000 m²+ GFA: at least Green Mark certification	search summary
Government-sold land, sold on/after 30 Jun 2022	Green Mark Platinum Super Low Energy (≥60% energy improvement over 2005 codes)	search summary

How required greenery can earn

Return	Evidence
Cheaper loans	Developer sustainability-linked loans now carry nature targets, e.g. the CDL–DBS S$300M SLL (2026) with microforest and urban-farming targets. Margins step down when verified targets are met.	lead return
Lower upkeep	In the case study, a different greenery mix under the same rules cuts upkeep from S$35 to S$16 per unit per month.	model
Higher sale price	Condos within 1 km of Jurong Lake Gardens averaged $1,617 psf vs $1,383 within 2 km (EdgeProp, Jul 2025). These are simple averages, not causal, and our HDB value lens finds no canopy premium.	weak
Subsidies	Skyrise Greenery Incentive: up to 50% co-funding. It covers existing buildings, so it is a retrofit line, not a return on new GLS sites.	scope
Greenery that earns rent	HDB car-park roofs rented to urban farms; tendered rents about S$11–24/m²/yr.	tender data

The value lens: the market does not price cooling yet

We fitted a hedonic model to 73,304 HDB resale transactions (2024–2026), within town and month, controlling for flat type, model, size, storey and remaining lease. It finds no premium for block-level canopy, while distance to MRT is priced strongly. These are associations, not causal effects.

Why this shapes the business model: if buyers do not reward cooling, its value has to be made visible another way, through verified data that unlocks cheaper loans and lower upkeep. That is the community-data loop, and it is why the lead return is finance, not sale premium.

10bDeveloper case study

A 2023 GLS condominium site in Woodlands

A mixed-integer programme (Gurobi; 129 variables, of which 64 binary, and 183 constraints) is solved to proven optimality in milliseconds. It chooses the greenery mix and where public cool-refuge pockets go, subject to URA's greenery rules, site capacities, a maintenance-fee cap and two loan KPIs. Pocket siting uses the index: each nearby block is weighted by need × resilience shortfall. Site figures come from public GLS records, and the site is anonymised.

−41%

30-year greenery cost vs a typical showcase design, same GnPR 3.5

898

seniors + rental households within 200 m of a public cool refuge (best single-edge placement: 712)

+S$2.1M

median 30-year value per site vs showcase, after fees; positive in 100% of 20,000 Monte Carlo draws

S$35→16

greenery upkeep per unit per month

Figure 5: developer case results — **Figure 5.** a Three designs under identical URA rules. b 30-year net value over 20,000 draws of uncertain inputs (premium, upkeep, capex, cooling, loan step-down of 2.5–10 bps, farm rent). c The verification flywheel: as verified sites accumulate, the forecast narrows and the value a lender can underwrite (P10) turns positive. Assumptions are tagged in `case_gls_site/src/params.py`.

Why it can win

Evidence, not claims. "Lush landscaping" becomes "shaded seating every 100 m doubles senior use". That evidence backs amenity choices, marketing and ESG reporting.
A data moat. Verified post-occupancy data compounds. After 8 verified sites, the bankable value per site moves from −S$0.6M to +S$1.4M in the case model. A competitor without the library starts from zero.
Aligned incentives. Developers save on upkeep and finance, residents are paid for taking part, and planners get evidence. Part of revenue funds the credits that keep data flowing.
A ready first market. URA's 2026 Confirmed Lists carry 9,320 GLS units, all under the strictest green standards.

Business model (illustrative)

Assumes 3 → 30 new sites a year plus retrofits of existing buildings, with monitoring renewing for 5 years. In the case study, client value ≈ 9.7× the present value of fees. Proposed pilot: one GLS site plus one existing microforest, with the first verified nature-KPI report within 12 months.

Research note and disclaimer. UrbanPulse is an independent academic research prototype built entirely from public information. Companies, agencies, loans and projects are cited only as public examples of market practice. No affiliation, endorsement or access to private data is implied, and the developer case uses an anonymised site. Prices, returns and revenues are model estimates under stated assumptions. They are not quotes, valuations, or investment or planning advice.

11Method, validation and limits

What we can stand behind.

Coverage. 10,749 of 10,817 residential HDB blocks (99.4%) are joined to official footprints without geocoding, by matching block-number sets between street names and HDB street codes. A check against 929 OneMap-geocoded blocks found 899 correct and 2 wrong.
Satellite heat. 52 + 37 cloud-masked Landsat scenes. Our single-channel LST reproduces the official product with RMSE 0.99 °C; the bias of −0.47 °C is removed by per-scene anomalies.
Heat drivers. R² 0.68 with town fixed effects; gradient-boosted trees predict an unseen town with R² 0.59.
Robustness. Town rankings hold under stakeholder weight lenses (Spearman 0.80–0.97) and 2,000 random weightings. Changing the shade penalty from 0 to 1.0 moves no town by more than 0.6 points.
Redundancy. No pair of the 13 scored indicators exceeds |ρ| = 0.8.

Limits, stated plainly

NDVI > 0.6 at 10 m approximates tree canopy; it cannot separate trees from dense shrubs.
Walkway shade depends on how completely OpenStreetMap records covered linkways. Building shade is not yet modelled.
Seniors per block are estimated from Census 2020 subzone shares. Tengah had no residents then and uses the island median.
Flood resilience uses distance to 36 named PUB flood-prone locations, each geocoded to one representative point.
Surface temperature is not air temperature or felt heat; WBGT and indoor readings are phase 2.
The value lens and heat drivers are associations. The developer case uses benchmarked assumptions, tagged in the code.
Community data does not exist yet; that is the point of Part 2.

SDG 11.7Universal access to safe, inclusive and accessible green and public spaces, in particular for older persons. The inclusion check and need weights target this.

SDG 11.bIntegrated policies for climate adaptation and resilience. UrbanPulse supplies a heat target the Green Plan does not yet set.

SDG 11.3Inclusive, participatory planning. Residents and AACs are data partners, paid for taking part.

12Data & references

Open data used

HDB Property Information; HDB Existing Building footprints; HDB resale prices (2024–2026) · data.gov.sg
URA Master Plan 2019 subzones; Master Plan 2025 land use; waterbodies · data.gov.sg
SingStat Census 2020 residents by subzone and age · data.gov.sg
MOH eldercare services and CHAS clinics; NEA hawker centres; LTA MRT exits and bus stops · data.gov.sg
NParks parks and nature reserves, Park Connector Loop, Community in Bloom gardens · data.gov.sg
PUB List of Flood Prone Areas (Nov 2025), geocoded with OneMap (SLA)
NEA WBGT, 1 Mar–31 May 2026, 27 stations · data.gov.sg real-time API
Landsat 5/8/9 Collection 2 Level-2 (USGS) and Copernicus Sentinel-2 L2A, via Microsoft Planetary Computer
OpenStreetMap contributors (ODbL), BBBike Singapore extract of 9 Oct 2026
Basemap: OneMap (Singapore Land Authority)

Contains information from data.gov.sg accessed October 2026, made available under the terms of the Singapore Open Data Licence version 1.0.

References

URA Master Plan 2025: A Cool City in a Warming World
URA greenery requirements for flats and condominiums
Singapore Green Plan 2030 targets
Her World (2026): SUTD home-heat study
Chew, Chung & Ang (2026): rental-flat thermal inequality, WSBE abstract
Benita, Kalashnikov & Tunçer (2021): Spatial Livability Index
Xu et al. (2025): equity in heat mitigation by green space
Mothership (2024): Singapore-ETH heat vulnerability hotspots
npj Urban Sustainability (2025): SE Asian capitals heat 1990–2024
Science Centre: which street shade keeps you cool (CARES, one-north)
World Bank (2018): Walk2Ride covered walkways
CDL and DBS (2026): S$300M sustainability-linked loan for nature-based urban development
EdgeProp (2025): green premium near Jurong Lake Gardens
AgFunderNews (2020): HDB car-park rooftops for urban farms
Sobrino, Jiménez-Muñoz & Paolini (2004): land surface temperature retrieval from Landsat TM 5, Remote Sensing of Environment 90, 434–440.

Photographs

Skyline illustration adapted from the team's slide template.