A First Look at Decadal Hydrological Predictability by Land Surface Ensemble Simulations

The prediction of terrestrial hydrology at the decadal scale is critical for managing water resources in the face of climate change. Here we conducted an assessment by global land model simulations following the design of the fifth Coupled Model Intercomparison Project (CMIP5) decadal hindcast experiments, specifically testing for the sensitivity to perfect initial or boundary conditions. The memory for terrestrial water storage (TWS) is longer than 6 years over 11% of global land areas where the deep soil moisture and aquifer water have a long memory and a nonnegligible variability. Ensemble decadal predictions based on realistic initial conditions are skillful over 31%, 43%, and 59% of global land areas for TWS, deep soil moisture, and aquifer water, respectively. The fraction of skillful predictions for TWS increases by 10%–16% when conditioned on Pacific Decadal Oscillation and Atlantic Multidecadal Oscillation indices. This study provides a first look at decadal hydrological predictability, with an improved skill when incorporating low‐frequency climate information. Plain Language Summary Decadal prediction, which was initially proposed to more accurately project near‐term (e.g., 10–30 years) climate change by using the experiences in weather and seasonal climate forecasting, has raised a wide concern since the IPCC CMIP5 project. The climate community is now trying to transition the decadal prediction from a pure research to a quasi‐real‐time operational effort. However, very limited information is known about the decadal hydrological predictability over land, which is more relevant to the livelihood and stakeholders. Here we combine an ensemble simulation method that is widely used to assess seasonal hydrological predictability, with the experimental design of the CMIP5 decadal climate hindcasts, to provide a first look at decadal hydrological predictability and skill by carrying out over 2,000 years global land model simulations. We found skillful decadal prediction for terrestrial water storage over one third land areas where deep soil moisture and aquifer have a nonnegligible variability. And the skill can be further enhanced by incorporating low‐frequency teleconnection information from Atlantic and Pacific Oceans. This study suggests that it is possible to provide water resources managers useful hydrological forecast information over arid and semiarid regions a few years or even a decade in advance. Key Points Decadal hydrological predictabilit