Future changes in multidecadal mean water availability, represented as the difference between precipitation and evapotranspiration, remain highly uncertain in ensemble simulations of climate models. Here we identify a physically meaningful relationship between present-day mean precipitation and projected changes in water availability. This suggests that the uncertainty can be reduced by conditioning the ensemble on observed precipitation, which is achieved through a novel probabilistic approach that uses Approximate Bayesian Computation. Comparing the constrained with the full ensemble shows that projected extreme changes in water availability, denoted by the 5th and 95th percentile of the full ensemble, are less likely over 73% and 63% of land, respectively. There is also an overall shift toward wetter conditions over Europe, Southern Africa, and Western North America, whereas the opposite occurs over the Amazon. Finally, the constrained projections support adaptation to shifts in regional water availability as imposed by different global warming levels.
Plain Language Summary Human-induced emissions of greenhouse gases are warming our planet and if unabated would continue to do so. Projected climate change can affect rainfall, evapotranspiration, and streamflow and hence water availability over land. Climate model projections of future average water availability still vary substantially across different models. We find that part of this disagreement can be explained by the amount of rainfall that each model simulates for the present. In a drying region, for example, if a model has more water in the present, then the simulated loss is likely higher; whereas if it is already dry in the present, then there is only a small amount of water that can be lost. Therefore, we improve the projections by placing more confidence on the accuracy of those climate models whose average rainfall from the past decades is more similar to observed rainfall amounts. We show that previous projections of very extreme future changes in water availability are less likely to occur on 73% of the land's surface. We also provide improved projections of regional changes that are useful for the management of water resources, indicating more drying in the Amazon and less drying in Europe, Southern Africa, and Western North America.
Padron, Ryan S.,Gudmundsson, Lukas,Seneviratne, Sonia I.. Observational Constraints Reduce Likelihood of Extreme Changes in Multidecadal Land Water Availability[J]. GEOPHYSICAL RESEARCH LETTERS,2019-01-01,46(2):736-744