Sensitivity of future climate change and uncertainty over India to performance-based model weighting

Multi-model ensembles are used to understand present and future climate change. Differences between individual model projections of future climates result in uncertainty in what we can expect in the future. Typically, models from such ensembles are treated equally but two reasons have been used to move away from this model democracy. The first reason is that models may not be equally skillful and therefore, their projections would need to be weighted differently based on model performance over the observed period. A second reason has to do with whether models are truly independent, that is they do not share components that can result in simulations that are closer, giving the impression of a lower uncertainty. This study examines the sensitivity of mean change and uncertainty in projected future surface air temperature and precipitation over India to weighting strategies based only on model performance but not their independence. It is found that some model performance metrics do not provide any value, as most models tend to have uniformly high or uniformly low weights. Considering multiple dimensions of model performance that add value in distinguishing between models is important for model weighting to be useful. Using multiple observational datasets for the two variables analysed, it is found that observational uncertainty can affect the weighted future change as much as the choice of weighting schemes over some regions. Careful selection of model performance metrics and reliable observations are necessary for a more robust estimate of future change and associated uncertainty.