Decisions regarding water resource management, agricultural practice, and energy allocation often require information about future climate conditions weeks to months in advance. Skillful and reliable seasonal climate prediction can significantly facilitate and benefit the decision making process. However, it is a challenge to make skillful predictions about the climate system at these time scales because the processes that contribute to the seasonal predictability are not fully understood thus not adequately represented in climate models. Research is needed to assess the current prediction skills from the state-of-the-art climate forecast systems, and to develop “best practice” procedures for optimal use of climate forecasts in applications that are directly relevant to decision making. This proposed project addresses those very issues.
In this project, we will carry out research activities to 1) evaluate the state-of-theart climate forecast systems, quantify their seasonal prediction quality, and assess factors that contribute to the prediction skill; 2) assess the optimal choice of ensemble members and scales, and to develop best practice procedures for combining and post-processing multiple forecasts to achieve better forecast quality; and 3) demonstrate the usefulness of seasonal climate prediction and evaluate the new post-processing procedures with seasonal drought prediction. The innovation of the proposed research is mainly reflected in the second activity. We will develop two innovative methods (multiscale Bayesian merging and structured output regression) in parallel to combine forecast information across multiple characteristic spatial and temporal scales. These methods will address outstanding issues like spatial and temporal dependence (or correlation structure) that is practically ignored when combining forecast members in an ensemble or multimodel ensemble system currently. It is our intention to make comparative evaluation of these two methods that grew out of two research communities. These statistical methods have the potential to significant advance the seasonal climate forecast skills. We will demonstrate the improvement in prediction skills and usefulness of climate prediction in regional hydrological applications by performing seasonal drought forecast for selected drought events in the US using these new methods.
We have assessed the feasibility of the project and have a clear understanding of possible difficulties. The proposed methods are new to seasonal climate prediction, but have been used in research fields of data assimilation, data mining and machine learning. The research team has experience in dealing with these methods, so we expect this project will progress smoothly. The project is relevant to MAPP program because it directly responds to the first priority area solicited by MAPP for FY2010, i.e., advance intro-seasonal to decadal climate prediction. In particular, this proposed research focuses on objective assessment of climate prediction skill from state-of-the-art climate forecast systems, and development of “best-practice” procedures for post-processing the predictions for hydrological applications. The outcome of this research will contribute to NOAA’s operation in seasonal climate forecasting.