New Ideas in Statistical Seasonal Forecasting: Application to North American Temperature and Precipitation

点击翻译按钮获取中文摘要

The project will develop and compare, alone and in combination, several recently suggested ideas for further improvement in statistical seasonal forecasts, with specific application to northern hemisphere surface temperature and precipitation. These ideas are: 1) use of sea surface temperature (SST) training data from the 19th century rather than from the mid-20th century only (which is conventional ), in order to improve statistical stability of the fitted models; 2) use of additional low-frequency surface predictors in addition to SSTs, specifically predictors derived from North American snow cover; 3) exploration and comparison of Canonical Correlation Analysis, Maximum Covariance Analysis, and Redundancy Analysis as statistical prediction frameworks; 4) modeling and accounting for nonstationarity in predictand means due to ongoing climate change and potentially other low-frequency variations through a time-dependent "hinge" mean function; and 5) exploration of a novel approach to filtering apparently unpredictable intraseasonal variations from predictand seasonal means, through computation of potentially more predictable "slow" patterns onto which gridded predictand values can be projected. Fully out-of-sample retrospective forecasts constructed using various combinations of these five elements will be evaluated and compared in an experimental setting that simulates real-world constraints in the operational production of seasonal forecasts.Broader impacts of this project include the potential to 1) produce practical results leading to improved seasonal forecasts based on a consensus of dynamical and statistical forecast tools, in order to better support long-range decision making in a variety of enterprises sensitive to seasonal climate variations; 2) have a significant impact on weather and climate risk management, potentially benefiting businesses, consumers and public policy makers. The project will contribute to the training of a Ph.D. student in the area of statistical climate diagnostics and prediction.