Predicting tipping points
Nobody can say reliably when ice sheets, ocean circulation or rainforest will tip; warning signs and models are not yet good enough.
Open in the interactive tree →Tipping elements such as the Greenland and West Antarctic ice sheets, the AMOC, the Amazon and permafrost have estimated thresholds spanning wide ranges (Greenland 0.8-3.0 °C). 'Critical slowing down' is a candidate early-warning signal but is noisy.
As of October 2026
The Global Tipping Points Report 2025 judged coral reefs as already tipped and the AMOC, Amazon and ice sheets as at risk below 2 °C of warming, without dating a crossing. A statistical analysis by Ditlevsen (2023, updated August 2025) puts AMOC collapse around 2065 with a 95% range of 2037-2109, but critics note it rests on proxy data and intermediate-complexity models. Direct monitoring of the AMOC (the RAPID array) began in 2004 and covers only about two decades.
What is missing
- Longer continuous observations of ocean circulation and ice sheets
- Models that resolve ice dynamics, ocean eddies and clouds at kilometre scale
- Validated statistical early-warning methods
- Better paleoclimate records of past abrupt shifts
- Compute for large ensembles of coupled simulations
Becomes possible once solved
- A tipping-point early-warning system
- Targeted adaptation (coastal planning, water, farming) with known time frames
- Clearer limits for emission and removal targets
Open steps
- Validated early-warning indicators Medium AI leverageDevelop statistical warning signals that rise before a tipping point, and test them on real ocean, ice and forest records.
- Longer AMOC record from satellites Medium AI leverageExtend the roughly 20-year direct AMOC record backwards and fill gaps by reconstructing circulation from satellite and ocean data.
- Fast ice-sheet model emulators High AI leverageMake ice-sheet and surface-melt models fast enough to run thousands of members, so odds of crossing ice-sheet thresholds can be estimated.
- Kilometre-scale coupled ensembles Medium AI leverageRun coupled ocean, ice and atmosphere models at kilometre scale in large ensembles to see abrupt shifts that coarse models smooth over.
- Past abrupt shifts from proxy records Low AI leverageReconstruct how fast and how early past abrupt climate shifts happened from ice, sediment and cave records, to test warning methods.
Where AI could help
Medium AI leverage. AI can sharpen early-warning statistics and cheapen model ensembles, but tipping points rest on short observations and untested physics.
- Train early-warning detectors on large model ensembles and test them on the roughly 20-year AMOC record
- Emulate ocean and ice-sheet models so thousands of runs fit the compute budget
- Fuse sparse observations, proxies and models into one estimate of the distance to a threshold
- Search paleoclimate records for past abrupt shifts and their precursors
Shown so far
- In September 2021, a deep-learning early-warning algorithm trained on simulated bifurcations beat generic indicators in sensitivity and specificity on 268 model and empirical time series (PNAS). source
- In April 2026, a not yet peer-reviewed preprint trained a CNN to estimate the AMOC's distance to its tipping point in one climate model and tested it on another, using idealised freshwater forcing. source
Prerequisites
- Climate models1967
- Ice cores1987
- GRACE Gravity Satellites2002GRACE tracks mass loss of Greenland and Antarctica, ice sheets near tipping
- Coral reef tipping point2025