Characterizing the Size of Atmospheric Rivers: an Approach Independent from the Detection Algorithm.

Atmospheric rivers (ARs) have important consequences in the hydrological cycle of midlatitude regions. They bring both positive and negative effects in continental regions. It has been shown that landfalling ARs are associated with about 40-75% of extreme wind and precipitation events over 40% of the world's coastlines. Their hazardous or beneficial effects have been quantified in terms of their intensity (Integrated Water Vapor) and duration. Assuming ergodicity, AR size and duration are equivalent; therefore, AR size is directly related to the benefits and hazards associated with ARs. The question "how large are atmospheric rivers?" is largely unconstrained and is heavily influenced by how researchers have quantitatively identified ARs. Results from the Atmospheric River Tracking Method Intercomparison Project (ARTMIP) indicate that the size of ARs depends strongly on the tracking algorithm employed, with a range that varies over several orders of magnitude among different methods. We use the ERA5 reanalysis (2017) from the European Centre for Medium-Range Weather Forecasts, to randomly sample AR events and create a composite and characterize the size of ARs. We utilize three independent approaches to estimate AR size: a method based on IVT-weighted principal components, a method based on the probability distribution within ARs versus that of the background field, and one based on a lagrangaian perspective. We apply principal components (PC) analysis to define AR's center of mass, and sample integrated vapor transport (IVT) along and across the AR. We calculate the probability density function (PDF) of AR lengths and widths, and the bivariate PDF of the IVT with the distance to the center of mass. Finally, we estimate the area of ARs from a Lagrangian perspective. Preliminary results suggest that the surface of AR from the composite is between 10¹² and 10¹³ m² for 75% of the sampled ARs. The mean length and width are approximately 2500 and 850 km respectively. This is consistent with sizes indicated by ARTMIP algorithms and provides a more constrained estimate. There is evidence that this method might provide a robust and objective answer to this question, independent of the AR detection algorithm.