Seasonal predictions are increasingly acknowledged by the adaptation community as a valuable resource to inform management of risks and opportunities. There is also growing awareness of the role that decadal variability plays in our experience of climate change. Given that decadal prediction experiments will be part of CMIP5, it is almost certain that the adaptation community will want to add these predictions to their portfolio of climate information. The problem is that predictions across these timescales carry various model biases, including probabilistic unreliability, unless they are recalibrated.
Our proposal offers a systematic framework to recalibrate seasonal-to-decadal predictions to yield estimates of forecast uncertainty that can be used to inform decisions such as planning and risk management across these timescales. A comprehensive sensitivity study will also provide guidance on the optimal design of prediction systems in the face of limited resources often faced by many modeling and forecast centers. This guidance will allow for informed decisions on trade-offs between, amongst other factors, the frequency and number of historical hindcasts, ensemble sizes, and the complexity of the recalibration scheme. Through this study we will assess the appropriate level of complexity of recalibration scheme for a given prediction problem, quality of ensemble predictions, and design of hindcasts: in the face of small hindcast samples or small ensemble sizes, more complex schemes may actually degrade the predictions through the addition of noise. This work will involve the development and use of mathematical models in order to test the full range of design choices associated with seasonal-todecadal hindcasts and forecasts. The mathematical models synthetically represent the ensemble prediction and observation time series, and the relation between the two. Existing dynamical model data sets of seasonal-to-interannual simulations and hindcasts and seasonal-to-decadal initialized and uninitialized hindcasts will be used both to parameterize the mathematical models and to demonstrate the impact of recalibration on forecast performance.
This work is highly relevant to NOAA’s long-term goal of climate adaptation. This work will assist in the effort to create and sustain enhanced resilience in communities and economies by creating climate prediction and projection information, and associated methodologies, that enable society to better anticipate and respond to climate and its impacts. Our work directly addresses two of NOAA’s five-year climate objectives: “assessments of current and future states of the climate system that identify potential impacts and inform science, service and stewardship decisions”; and, “adaptation choices supported by sustained, reliable, and timely climate services”. Of particular relevance to the MAPP program and Priority Area 1 to advance intraseasonal to decadal climate prediction, the proposed work will “assess the optimal choice of ensemble members, forecast times, and model diversity in order to characterize the impact of initial condition and model uncertainties in climate prediction”, and also “improve understanding of the impact of climate model biases and their evolution in forecast time on prediction skill, and the ‘best practice’ for post-processing predictions”.