Probabilistic regional and seasonal predictions of twenty-first century temperature and precipitation

Working Paper 23


The rationale for international agreements on climate change mitigation comes from the global scope of impacts, irrespective of the location of greenhouse gas (GHG) emissions.

By contrast, one of the motivations for national commitments to such agreements, and for national adaptation planning, is concern about national scale impacts.

Climate predictions on regional scales are therefore highly sought after by policy- and decision-makers. Yet robust, relevant predictions on these scales raise practical and philosophical challenges for climate science.

Existing methods underestimate uncertainty through limited exploration of model error and ad hoc choices regarding the relationship between model diversity and real world probabilities.

We present a new method for extracting model-based probabilistic information on regional and seasonal scales, utilising the world’s largest climate ensemble exploring the consequences of model uncertainty. For the first time, ensemble filtering is implemented to counter problems of in-sample bias in future analyses.

A probabilistic interpretation is presented of the regional scale consequences of targets to halve global GHG emissions by 2050, using a scenario with an estimated 32 per cent probability of exceeding 2oC global warming (relative to pre-industrial levels).

Meeting such a target leads to the model’s winter climate for Northern Europe being between 0.5 and 5.9oC warmer and -5 and 34 per cent wetter in the 2090s. A business-as-usual scenario provides ranges of 6.8 to 14.5oC and 22 to 71 per cent. Higher precipitation increases are found for North Asia.

That these ranges are large illustrates the need for adaptation strategies which minimise vulnerability rather than optimise for the future.

The method is potentially useful for making probabilistic statements about future seasonal mean model temperatures in many of the 22 predominantly land regions studied, as well as for model precipitation in a small number of high latitude regions.

David Stainforth