Observed change 2005–2025 anchored in published records, and a projection to 2045 under the IPCC's intermediate-emissions pathway (SSP2-4.5), for 30 world cities and 50 major Chinese cities. Climate change here means more than warming: shifting rain, harsher extremes, rising seas, drought, and storms. Every city card separates verified observations from model-based estimates.
Pick a metric — temperature change, extreme-heat days, precipitation change, or dominant hazard — and drag the year slider: 2005→2025 follows each city's observed trend (a trend fit, not raw measurements), 2030→2045 is projection. The 50 Chinese cities appear as smaller dots here and in full on the China map below. Click any dot to open its card. Terrain: NASA Blue Marble (topography + bathymetry).
A city that warms 1 °C does not get 1 °C-worse heatwaves — it gets several times more very-hot days. Hangzhou averaged ~33 days ≥35 °C in 1991–2020 and reached 41.9 °C in 2024; Beijing logged its first-ever three consecutive 40 °C days in June 2023; London hit 40 °C for the first time in 2022, Paris 42.6 °C in 2019. By 2045 the 'furnace' cities of the Yangtze (Chongqing, Changsha, Nanchang, Wuhan, Hangzhou) are projected to spend 7–9 weeks a year above 35 °C. Heat is the clearest, best-verified signal in this dataset.
Across most of the 80 cities, annual precipitation changed modestly (±10 %), but its delivery changed sharply: Zhengzhou received 201.9 mm in a single hour in July 2021 (China's hourly record, ~380 deaths in Henan); Seoul recorded 141.5 mm/hour in 2022; Dubai got ~250 mm in one day in April 2024 — over two years of normal rain. Planners should design drainage for hourly intensity, not annual totals.
The 2020 record Poyang Lake flood was followed in 2022 by its record low; São Paulo went from near 'Day Zero' (2014–15) to record storms; Los Angeles from megadrought to atmospheric-river winters. This 'whiplash' — wet and dry extremes amplifying together — is a defining feature of the observed record since 2005.
Global mean sea level rose ~111 mm since 1993, with the rate doubling from 2.1 to ~4.5 mm/yr. China's coast is rising faster than the global mean (+4.0 mm/yr; 2024 was the highest on record). Where cities are also sinking — Jakarta (up to 10–25 cm/yr in the north, enough to force the capital's relocation), Bangkok, Tianjin, Shanghai — relative sea-level rise is several times the global figure. By 2045 another ~10 cm of global rise is locked in under SSP2-4.5.
China warmed 0.31 °C/decade over 1961–2025, with the north and west fastest: the Tibetan Plateau (~0.35+ °C/decade) and Inner Mongolia/Ningxia (~0.4+) outpace the Pearl River Delta (~0.22). The northwest is getting wetter ('warming-wetting'), while the southwest has trended drier. One consequence in this dataset: by 2045, Harbin's projected annual mean (6.7 °C, +1.5 °C vs 2005) exceeds Changchun's 1991–2020 normal, 300 km to its south — climate zones are migrating north.
The strongest China landfalls since 1949 all occurred in this period near our study cities — Saomai 2006 (Wenzhou), Rammasun 2014 (Haikou), Meranti 2016 (Xiamen), Yagi 2024 (Hainan, strongest autumn landfall on record). Mangkhut (2018) and Hato (2017) battered the Pearl River Delta; Lekima (2019) flooded four provinces. Warmer seas load storms with more rain: typhoon rainfall extremes, not counts, are the rising threat.
Mexico City, Cape-Town-style 'Day Zero' scares (São Paulo 2015, Chennai-like stress in Istanbul 2021), North China Plain groundwater depletion, Yunnan's multi-year droughts, and glacier-fed supplies (Urumqi's Glacier No. 1 set melt records in 2022; Lhasa's plateau is warming at twice the global rate) all point one way: for many of these 80 cities the binding constraint by 2045 is water, not temperature.
Under SSP2-4.5 the world passes ~1.9 °C above pre-industrial by ~2045; near-term differences between emission scenarios are small (±0.2 °C) because of climate-system inertia. Adaptation (drainage, cooling, water storage, coastal defense) is governed by the numbers on this page; how much worse 2050–2100 gets is governed by choices made now.
2005–2025 (trend fit to observations). Each city starts from its published 1991–2020 climate normal (annual mean temperature, precipitation, days ≥35 °C) — the normal is centered on 2005.5, so it anchors the 2005 value directly. The city's own observed warming trend (from national meteorological services, the CMA Blue Book, Berkeley Earth and peer-reviewed studies; urban heat island included) carries it to 2025. This segment is a linear reconstruction anchored in observations, not raw measurements — real years vary around it (the event lists on each card show the actual extremes).
Projection (2030–2045). Pattern scaling under SSP2-4.5 (intermediate pathway, adjusted upward to the observed 2015–2025 rate): each city's observed amplification relative to global mean warming over the same historical window (clamped to 1.0–2.0× inland, 0.85× floor for marine-moderated coasts — land does not warm slower than the global mean) is applied to the trajectory below. Hot days follow a saturating logistic model — near-exponential growth for cities where 35 °C is rare, leveling toward a climate-class ceiling for hot deserts (winters stay below 35 °C even in Riyadh). Precipitation continues its observed trend.
| 2005 | 2010 | 2015 | 2020 | 2025 | 2030 | 2035 | 2040 | 2045 | |
|---|---|---|---|---|---|---|---|---|---|
| GMST vs 1850–1900 (°C) | 0.87 | 0.98 | 1.10 | 1.24 | 1.44 | 1.53 | 1.66 | 1.79 | 1.92 |
Verification. City data was gathered by six parallel research agents against published sources (each card lists its sources and which fields were verified vs estimated), then the full dataset and this methodology were independently audited by a second, stronger model (Claude Opus 5) checking baselines against 1991–2020 normals, trends against national assessments, and every landmark event for factual accuracy. Two Opus 5 audit agents returned 11 methodology findings (2 critical, both fixed: hot-day saturation and an amplification-window mismatch) and 20 city-level corrections, all applied; see AUDIT.md in the repository.
Limits. These are estimates, not forecasts: 5-year smoothed values hide single-year extremes (a 2045 heatwave will be far worse than the 2045 mean); city-scale precipitation trends have low signal-to-noise; scenario spread by 2045 is ≈ ±0.2 °C on the global mean; urban heat island growth beyond the observed rate is not added. Sea-level exposure is described per city rather than modeled. Full method: data/METHODOLOGY.md in the project repository.