Ground height
DeltaDTM v1.1 (Pronk et al. 2024), a 30 m bare-earth model of the world's coastal lowlands. It starts from Copernicus GLO-30, corrects its heights with ICESat-2 satellite laser measurements, and uses ICESat-2 and GEDI laser points to separate bare ground from buildings and trees. Heights are relative to the EGM2008 geoid, which the authors treat as mean sea level. No shift is applied.
DeltaDTM only covers ground up to about 10 m; land above that counts as high ground. Shares are of each planning area's land within URA boundaries. Nationally it puts 23% of land below 5 m, against the roughly 30% in Singapore's National Climate Change Secretariat materials, so shares may be modestly low overall.
Sea-level rise
Singapore's Third National Climate Change Study (V3), median relative sea-level rise averaged over six Singapore locations, from a 1995–2014 baseline: 0.45 / 0.57 / 0.79 m by 2100 and 0.72 / 0.95 / 1.37 m by 2150 under SSP1-2.6 / SSP2-4.5 / SSP5-8.5. Years in between follow a smooth curve through those points.
Likely ranges are wide: 0.23–1.15 m by 2100 across all three.
The safety line and the coastal clock
Before 2011 the minimum level for newly reclaimed land, and the minimum platform level for new developments, was 3 m above mean sea level; in 2011 both were raised to at least 4 m to allow for sea-level rise. The "safety line" is this site's own measure: 3 m plus projected rise. It is not an official standard; today's standard for new reclaimed land is at least 4 m.
The coastal clock counts years until a further 10 percentage points of an area's land falls below the line. The 10-point threshold is a prototype setting and has not yet been tested against insurer or lender behaviour.
Flash floods
PUB's 36 flood-prone areas (list as of Nov 2025) and 22 hotspots (list as of end 2024), placed on the map with OneMap and OpenStreetMap road junctions. PUB publishes names, not coordinates, so points are approximate.
Calibration: 39 of the 58 PUB sites (67%) sit lower than the ground within 300 m, against 48% of random land (p ≈ 0.002).
Downpour clock: years until Singapore's days with 50 mm or more of rain reach 1.5 times the 1995–2014 level, from V3 daily rainfall averaged over the island. It is one date for every area: V3's 2 km cells differ by up to 20 years between neighbouring areas, which is model noise, not a real local difference. Whether a street floods depends on drains, canals and dips, which is why each area is described by PUB's own list. Daily totals also miss the short bursts that cause most flash floods.
Heat clock
V3 daily maximum air temperature at 2 km. Heat clock: years until an area averages 30 or more days a year at or above 35°C. Between 1995 and 2014 Singapore averaged under 2 such days.
Check against 16 NEA weather stations, 2014–2024: the average is close (about 5 hot days a year in both) and the ranking broadly agrees (rank correlation 0.58, p = 0.02). V3 runs hot in the north (Woodlands about 1.7×, Mandai about 3× the station record) and cool at Paya Lebar and on offshore islands. Treat northern heat clocks as early.
Humidity: muggy days use V3 daily mean temperature and humidity (dew point ≥ 25°C), rare in 1995–2014 (about 4 a year) and common by 2100. Warm nights (lows ≥ 26°C) were tested but already occur on over 40% of nights, so they do not separate areas.
Surface heat ranking
Ground heat by day: Landsat 8/9 surface temperature from 108 clear-sky passes (about 11 am), 2016–2025, with cloud, shadow and water removed. Each area is compared with the island median on the same day, then averaged. Hottest: Pioneer, Boon Lay, Serangoon. Coolest: Central Water Catchment.
Ground heat at night: MODIS Aqua surface temperature (about 1:30 am), 2016–2025, good-quality cells only, compared with the island mean. Cells are 1 km, so read it as a ranking. Warmest: Geylang, Kallang, River Valley.
Tree cover: ESA WorldCover 2021 (10 m), share of land under trees and mangroves, water excluded. Tall trees: Meta and WRI canopy height 2020. Both run in Google Earth Engine. Satellites measure the ground surface, not the air, and see rooftops and shade from above, which is why tall, shaded centres such as Downtown Core read cooler than their streets feel.
Check against NEA: areas Landsat ranks hotter do tend to have warmer air at the same time (r = 0.45 across 14 stations), but that is not yet statistically solid (p = 0.10). No relationship with WBGT, which has only a year of history.
V3 projections
CCRS/NEA's V3 regional model, bias-corrected at 2 km over Singapore, five global models, daily data for 1995–2014, 2040–59 and 2080–99. Each area's value is the area-weighted average of the 2 km cells it covers; figures are the median of the five models, with the range in the side panel.
Years between those slices follow the IPCC AR6 global-warming path for each scenario (best estimates for 2021–40, 2041–60 and 2081–2100); scenarios share one path until 2030, so "today" is the same in all three. "Today" values are modelled, not observed. "2100" shows the 2080–99 average. Hot and muggy days use one rising curve per area through all seven slices ordered by warming, interpolated on a log scale because days above a threshold grow roughly exponentially with warming. Rainfall measures (heavy-rain days, wettest day, dry spells) are noisy between 20-year slices, so for each area one line is fitted against global warming across all scenarios, and each scenario moves along it at its own pace. Thresholds of 30 hot days and 1.5 times heavy-rain days are prototype settings.
Haze
NEA hourly 24-hour PSI, April 2014 onwards, worst of five regions each day. An unhealthy day is PSI above 100. History only: no public model projects haze. Dry spells (15 or more days with island-average rain under 1 mm) come from V3 and are shown as a risk condition, not a haze forecast.
What this does not model
- Seawalls, Marina Barrage, pumping and drains. Land below the line today is defended now; the clock measures how fast the defended share grows.
- Storm surge, high tides and waves. The safety line uses mean sea level plus projected rise only; Singapore's National Climate Change Secretariat notes that surges and high tides on top of rise could push water levels up to about 5 m.
- Building platform levels and basement entrance heights, which PUB sets for new developments.
- Water flow. This is ground height, not a flood simulation.
Known limits of the data
- A 30 m model has errors of around half a metre: DeltaDTM's authors report an average error of 0.45 m overall and 0.57 m in built-up areas. Older low-lying areas (Marine Parade, Kallang, Geylang, Rochor) come out lowest; modern reclaimed platforms (Tuas, Changi, Marina South) come out at their built 3–5 m.
- In dense central areas the ground under buildings is estimated by DeltaDTM, not measured.
- Shares between the 3 m and 5 m bands are interpolated from area statistics.
- Elevations are relative to mean sea level as published by DeltaDTM; V3 uses a 1995–2014 sea level. The gap is small next to the elevation layer's own uncertainty.
- V3 temperature is bias-adjusted against ERA5-driven model runs, because Singapore's temperature stations are too sparse for a gridded reference (rainfall uses 28 stations), so differences between neighbouring areas may be understated. Coastal and offshore areas run cooler partly because their cells include sea.
- Small central areas cover only two or three 2 km cells, so their heat and rain values are close to their neighbours'.