On the interpretation of constrained climate model ensembles

摘要: [1] An ensemble of models can be interpreted in two ways. The first treats each model as an approximation the true system with some random error. Alternatively, a sample drawn from distribution models, such that and truth are statistically indistinguishable. Both interpretations ubiquitous have different consequences for uncertainty projections, but rarely defended. Here we argue seemingly conflicting views fact complementary, interpretation may evolve seamlessly former to latter. We show ‘truth plus error’ like properties exist historical present day climate simulations CMIP archive, they explained by design tuning observations, although both imperfect. For future structural differences response arise which independent state thus ‘indistinguishable’ is increasingly favored. Our inability define performance metrics identify ‘good’ ‘bad’ having largely exploited available observations. remaining error observations often uninformative further reduce biases or range projections covered ensemble. discussion here motivated use multi ensembles arguments generic any situation where multiple constrained used describe same system.