Application-specific model selection and model weighting of global climate models with application to regional environmental management of red tide
Abstract
In contrast to generic evaluation of global climate models irrespective of the application, in this research we illustrate the importance of using decision-relevant metrics for climate service applications. We present a case study in the West Florida Shelf in the Gulf of Mexico about regional environmental management of red tides, which are harmful algal blooms that occurs worldwide with severe environmental and socioeconomic impacts. We consider the two approaches of prescreening-based subset selection and optimal model weighting to improve ensemble predictions of global climate models. Independent ensemble members are categorized, selected, and weighted based on their ability to reproduce physically-interpretable features of interest that are problem-specific. Application specific prescreening-based subset selection can be viewed as an extreme form of weighting such that models that are not suitable for this application or variable is discarded. Our results show the advantages of using prescreening-based subset selection with decision relevant metrics to identify non-representative models, understand their impact on ensemble prediction, and provide insights about the validity of the model weights. In addition, we show that while optimal model weighting can potentially improve predictive performance, it can result in including non-representative models with both over and underestimation due to error cancellation. Including non-representative models can underplay the weights of robust models. This emphasizes that importance of prescreening-based subset selection, as optimal model weighting can underplay robust ensemble members by optimizing error cancellation. By illustrating advantages of prescreening based subset selection and the caveats optimal model weighting, our findings are pertinent to many climate services.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.A42M..05E