Estimating return values for precipitation in Norway by combining a large ensemble data set with gridded observations to re-parameterize the extreme value distribution

Extreme precipitation events that lead to excess surface water and flood are becoming an amplifying societal cost as a result of both the increasing precipitation amounts in recent years and urbanization. Knowledge about extreme precipitation events is important for the ability to predict them, but also to know how often they occur with various intensities in order to estimate design values for constructions and critical infrastructure. 

To study extreme precipitation events by applying statistical analysis requires long timesteries, which often is a challenge when using conventional or new observational data records. 

In the present study, a data set constructed from the numerical seasonal prediction system at ECMWF, SEAS5, has been applied in order to increase the event sample size compared to conventional observational or re-analysis data sets. The data are analyzed by fitting them to a GEV-distribution. This distribution is compared to an equivalent GEV-distribution for the gridded observational data set SeNorge. While this data set has a smaller sample size, the fine scale horizontal resolution allows for more spatial heterogeneity in the data set. In this study we propose a method to estimate return values by combining the two datasets, and in that way exploiting the advantages of both data sources: sample size from SEAS5 and spatial distribution from SeNorge. The combination is done by using a normalized “growth curve” from both data sets. The large sample size is important for the shape and scale parameters of the fit to the GEV and is thus used from the SEAS5 data set. These parameters will define the shape of the curve. Location from SeNorge is then used to get the correct level of the curve.