Skillful Decadal Prediction of Droughts Over Large‐Scale River Basins Across the Globe.

Long‐term droughts, closely linked to oceanic climate variability, are a great threat to the global economy, as well as food and water security. Skillful predictions will provide enormous benefit to the human societies. By taking advantages of the newly available decadal hindcasts and target experiments from the Coupled Model Intercomparison Project phase 6 (CMIP6), this paper evaluates potential predictability and decadal prediction skill of droughts over major river basins across the globe. The results show that the CMIP6 drought prediction is skillful within 3–6 years, and the skill score is 11% higher than CMIP5 over 47% river basins. Prediction skill is enhanced by 15% (16%) over 47% (49%) basins, forced by the actual sea surface temperature over tropical eastern Pacific (North Atlantic). Our findings imply the possibility of drought adaptation based on skillful decadal prediction with sufficient lead time. Plain Language Summary: A reliable drought prediction is helpful to reduce adverse impact and economic loss by increasing the preparedness. Although operational seasonal drought prediction has been applied over the world, decadal prediction is still a grand challenge due to limited knowledge about the predictability and mechanism of long‐term drought. Here we apply the decadal hindcasts from CMIP6 to evaluate the probabilistic decadal prediction skill of drought and analyze the impacts of sea surface temperature on the prediction through CMIP6 pacemaker experiments. Our results show the skillful prediction can be found over half of the major river basins at 1–4 years lead. In addition, incorporating the actual sea surface temperature in tropical eastern Pacific or North Atlantic can efficiently enhance the drought prediction skill over 48% basins as well. This study suggests that it is possible to provide useful long‐term drought information for water and agricultural management through using coupled models, but the skill can be improved further by optimization of initialization and ensemble methods. Key Points: Probabilistic predictions of basin‐scale droughts by dynamical models are skillful within 3–6 years across the globeThere is an 11% improvement from CMIP6 models compared to CMIP5 over 47% of the largest 177 global river basinsSSTs in the tropical eastern Pacific and the North Atlantic have critical importance for the success over about 48% basins [ABSTRACT FROM AUTHOR]