A review of multimodel superensemble forecasting for weather, seasonal climate, and hurricanes. (11th May 2016)

• Record Type: Journal Article
• Authors: Krishnamurti, T. N.; Kumar, V.; Simon, A.; Bhardwaj, A.; Ghosh, T.; Ross, R.
• Is Part Of: Reviews of geophysics. Volume 54:Number 2(2016:Jun.)
• Page Start: 336
• Page End: 377
• Publication Date: 2016-05-11
• Subjects: multimodel -- superensemble -- ensemble mean; Geophysics -- Periodicals; 550.5
• DOI: 10.1002/2015RG000513
• Languages: English
• ISSNs: 8755-1209

Abstract

Abstract: This review provides a summary of work in the area of ensemble forecasts for weather, climate, oceans, and hurricanes. This includes a combination of multiple forecast model results that does not dwell on the ensemble mean but uses a unique collective bias reduction procedure. A theoretical framework for this procedure is provided, utilizing a suite of models that is constructed from the well‐known Lorenz low‐order nonlinear system. A tutorial that includes a walk‐through table and illustrates the inner workings of the multimodel superensemble's principle is provided. Systematic errors in a single deterministic model arise from a host of features that range from the model's initial state (data assimilation), resolution, representation of physics, dynamics, and ocean processes, local aspects of orography, water bodies, and details of the land surface. Models, in their diversity of representation of such features, end up leaving unique signatures of systematic errors. The multimodel superensemble utilizes as many as 10 million weights to take into account the bias errors arising from these diverse features of multimodels. The design of a single deterministic forecast models that utilizes multiple features from the use of the large volume of weights is provided here. This has led to a better understanding of the error growths and the collective bias reductions for several of the physical parameterizations within diverse models, such as cumulus convection, planetary boundary layer physics, and radiative transfer. A number of examples for weather, seasonal climate, hurricanes and sub surface oceanic forecast skills of member models, the ensemble mean, and the superensemble are provided. Key Points: Improved weather forecasts are possible from a combination of many models to produce a consensus forecast Model consensus is achieved through using a weighted mean of models; the weights (as many as 10 million ) vary in space and time Model ensembles utilize a suite of models and the methodology entails the reduction of systematic bias errors of each member model