“The main goals of this proposal are to: 1) examine the potential value and applicability of five global circulation models (GCMs) part of the North American Multi-Model Ensemble (NMME) project in forecasting monthly and seasonal precipitation and temperature over the continental United States and Europe; 2) Develop a multi-model averaging procedure to increase the forecast skill of these models.
We will focus on the continental United States and Europe and evaluate the forecast quality of five GCMs part of the NMME project in forecasting seasonal precipitation and temperature. The GCMs we consider are two GFDL models (CM2.1, and FLORb01), NASA-GMAO-GEOS-5, COLA-RSMAS-CCSM3, and CCCma-CanCM4. The outputs from these models are available from the early 1980s to the present. They have a resolution of 1-degree and monthly, with a forecast lead time up to one year. We will also focus on extreme precipitation and temperature. More specifically, we will examine the skill of the models in forecasting extended periods with high temperature and/or low precipitation (leading to drought conditions), and periods characterized by extreme precipitation (leading to flooding). The forecasts from these models are compared against gridded monthly temperature and precipitation data created by the PRISM Climate Group for the United States and the E-OBS data for Europe.
Aspects of forecast quality are quantified using a diagnostic skill score decomposition that allows the evaluation of the potential skill and conditional and unconditional biases associated with these forecasts. This skill score will allow for a better understanding of the utility of these models for flood and drought predictions. It also represents a diagnostic tool that could provide model developers feedback about strengths and weaknesses of their models.
Quantification of the forecast quality represents the necessary step towards improving on the models’ skill. We will apply a simple new Bayesian model averaging procedure that leverages the strength of each model at different lead times for different months. Using the relationship between hindcast model forecasts and observations from the verification, the procedure assigns weights to each historical observation given the multi-model forecasts; the weights represent the likelihood of each historical outcome given the multi-model forecasts. The weighted samples of observations not only define an optimal bias-corrected multi-model ensemble forecast, they can also be used to selectively weight historical forcings as an atmospheric ensemble pre-processor method for hydrologic forecasting. This Bayesian multi-model weighting procedure will be performed at the pixel scale for all the months/seasons and lead times.
This proposal is relevant to NOAA’s Next Generation Strategic Plan (NGSP) because it provides information valuable to “assessments of current and future states of the climate system that identify potential impacts and inform science, service, and stewardship decisions,” to improve “freshwater resource management,” and to reduce “loss of life, property, and disruption from high-impact events,” which are some of the NGSP’s objectives. In evaluating the temperature and precipitation forecasts from five GCMs part of the NMME project, our goals lie squarely with those of the Modeling, Analysis, Predictions, and Projections (MAPP) Program and solicitation. In particular the proposed work will result in the evaluation of “the prediction of large-scale, extended lead time conditions conducive to extremes such as heat waves, extreme precipitation.” Moreover, the development of weighted ensemble predictions aligns with the “application of the NMME for the development of new prediction products.””