Speaker: Chris Chan
Chris will be presenting the attached paper: Using climate models to estimate the quality of global observational data sets.
Here are his thoughts on the paper:
Observational datasets are often used to evaluate the performance of climate models. However, these observational references are uncertain due to a variety of error sources. In this paper, Massonnet et al managed to answer the question that can the climate models be used to assess the quality of observational references? By assuming that both observations and model outputs are noisy versions of the unknown true climate signal, and employing metrics that respect underlying hypothesis of symmetry (correlation), they show in a toy model that observational error can also contributes to poor correlations. When applied to the sea surface temperature dataset, they show that over the Niño 3.4 region, the correlation between models and the most advanced ESA-CCI dataset are systematically higher than that with the ERSST4 dataset by 0.07. But there is not a universally best SST dataset. ERSST4 over powers ESA-CCI where ship measurements dominate. The authors concluded that "considering climate model evaluation as a bidirectional exercise is essential to remember that observations, no matter how good they appear, are also intrinsically uncertain".