Estimating megadrought risk from the CMIP5 archive using model emulators

Abstract:

Prolonged droughts lasting decades – megadroughts – have occurred in the past throughout western North America as well as other parts of the world. In the US southwest, recent work suggests the risk of these events during climate change may be as high as 30% to 50% this century due to decreasing precipitation alone. However, this view is incomplete because it does not factor in changes in water availability from higher temperatures, nor does it address the risk of continent-wide megadrought. Here the current risk estimates for megadrought in the Western US, Australia, Southeast Asia, Sahel region of Africa, and the Amazon region of South America are revised by factoring in temperature demands as well as spatial covariance structures. It is found that previous estimates of 30-50% for most arid regions may be quite low, especially by the end of the 21^st century. Further, there is a moderate risk of widespread megadrought encompassing most of western North America. Such an event would be similar in its spatial structure to the drought currently underway, but could last for decades. Finally, using the probabilistic methods employed here to estimate future megadrought risks from CMIP5 models, the statistics of past megadroughts are reassessed; it is found that the prolonged epochs of aridity that occurred during medieval climate anomaly (850-1300 CE) are extremely unlikely to have occurred simply by chance, supporting the claim that external or internal forcing mechanisms are needed to explain large scale hydrologic fluctuations of that period. Taken together, these findings argue that megadrought, like other kinds of emergent climate risks, need to be factored into regional adaptation strategies.