Representation Uncertainty in the Earth Sciences

The first Joint Workshop on Representation Uncertainty in the Earth Sciences was held in March 2021. This brought together the Earth observation, data assimilation and forecast verification and post-processing communities, alongside metrologists to discuss the definition and quantification of representation uncertainty within the Earth sciences. The aim of the workshop was to facilitate cross-disciplinary discussion, establishing where existing methodologies could be shared and to foster future collaboration. A key outcome of the workshop was a working definition of representation uncertainty applicable across all the Earth science communities, which is presented in this white paper. The cross-disciplinary discussions at the workshop highlighted the need for scientists to work with metrologists to establish a common vocabulary for uncertainties, accessible to Earth science applications. Further recommendations included regular workshops to discuss progress in defining and quantifying representation uncertainty and awareness of cross-disciplinary funding opportunities to further address representation uncertainty issues. Plain Language Summary This paper describes a workshop which brought together experts from different Earth science disciplines to discuss and attempt to define the term "representation uncertainty". We make observations of the Earth using satellites, ground based instruments (such as weather stations), air and sea-borne sensors. These observations are used in their own right, and also by computer models to generate weather forecasts. The observations themselves are imperfect and we quantify these imperfections using the term "uncertainty". In this paper we discuss the uncertainty that occurs when we compare two different sets of observations, two different models, observations and models, or where there are differences in underlying assumptions. As well as the uncertainties inherent in models and observations, there is also an uncertainty due to the fact that the two things being compared are not representing a phenomenon in exactly the same way. For example, a satellite observation may represent an average value over a few hundred square meters, while an instrument on the surface measures only at a single point (typically one-square-meter or less), and the model represents an area of several square kilometres. Understanding those differences is essential to be able to properly combine different sets of observations, and observations with models.