Climate Migration Forecasting With AI: Models, Data Sources, and 2050 Projections

The numbers everyone quotes, and where they come from

Three projections dominate climate-migration discourse: 'up to 216 million internal climate migrants by 2050' from the World Bank Groundswell report; '1.2 billion at risk of displacement by 2050' from the Institute for Economics and Peace Ecological Threat Report; and the IDMC Global Report on Internal Displacement annual disaster-displacement figure, typically 25-35 million per year. These three numbers measure very different things, were produced by different methods, and are not interchangeable. Reporting that conflates them is one of the most common errors in climate-migration coverage.

The model families behind the headline figures

Groundswell uses a scenario-based gravity model that combines downscaled climate projections, agricultural productivity changes, and economic gradients to estimate internal migration flows under three SSP-RCP scenarios. The IEP figure is a composite vulnerability index applied to population, not a flow forecast. The IDMC annual figure is a measured count of disaster-displacement events rather than a projection. AI enters mostly in Groundswell-style flow modelling and in the rapidly maturing field of impact-attribution models that link specific extreme events to specific climate forcings.

What AI adds beyond the World Bank baseline

Three research directions are producing usable forward estimates. Machine-learning emulators of expensive climate-economy models let researchers run thousands of scenario combinations rather than a handful, producing fuller uncertainty distributions. High-resolution agricultural-yield models trained on remote sensing and ground data refine the rural push factor that dominates internal climate migration. Mobility models trained on call-detail records and mobile-money transactions in countries that permit it (Bangladesh, Senegal, Cote d'Ivoire) calibrate the response function from environmental stress to actual movement, which has historically been the weakest link.

Where the models systematically understate the problem

Two structural underestimates recur. First, urban-to-urban and short-distance moves are mostly excluded because the underlying mobility data does not capture them. Second, trapped populations — households too poor to move in response to environmental stress — are invisible in flow models by construction, even though they may be the most vulnerable group. The Foresight: Migration and Global Environmental Change report (UK Government Office for Science) remains the canonical reference on this point.

How to read a climate-migration projection responsibly

Five questions clear most of the fog. Is it a flow or stock estimate? Is it internal or cross-border? What scenario combination (emissions pathway plus socioeconomic pathway)? What horizon? What uncertainty range? The 216 million figure, for example, is internal, flows over the period to 2050, under a high-emissions low-development scenario, with a much lower estimate under a more optimistic combination. The 1.2 billion figure is an at-risk stock, not a projected flow. Treating them as comparable is the error.

The 2026 frontier: attribution and protection

The most consequential methodological progress since 2022 is event-level attribution: linking a specific flood, drought, or cyclone to climate change with calibrated probability. World Weather Attribution and the IPCC AR6 working groups have made this routine for major events. Coupling event attribution to displacement counts produces, for the first time, defensible estimates of climate-attributable displacement at the event level. This matters for the Warsaw International Mechanism on Loss and Damage and for any future protection regime that distinguishes climate-driven movement.

Further reading and primary sources

• World Bank Groundswell: https://openknowledge.worldbank.org/handle/10986/36248
• IDMC Global Report: https://www.internal-displacement.org/global-report
• IEP Ecological Threat Report: https://www.economicsandpeace.org/
• UK Foresight Migration: https://www.gov.uk/government/publications/migration-and-global-environmental-change-future-challenges-and-opportunities
• World Weather Attribution: https://www.worldweatherattribution.org/
• UNFCCC Loss and Damage: https://unfccc.int/topics/adaptation-and-resilience/workstreams/loss-and-damage-ld