Deriving Precipitation Frequency Estimates from High-Resolution Weather Radar Rainfall Data

Climate change and the expected resulting changes in precipitation patterns call for robust and resilient climate adaptation solutions, including urban drainage systems that can handle the precipitation of the future.  One of the most commonly used engineering tools when designing urban drainage systems is precipitation frequency estimates (PFEs) that allow for estimation of extreme precipitation rates, also called return levels, and the associated return periods. Oftentimes PFEs are based on point rain gauge measurements, either directly computed from a rain gauge precipitation time series or from a regionalized model that is based on a larger network of rain gauges, when representing extreme precipitation in ungauged areas. Weather radar precipitation measurements pose an alternative data source for computing PFEs as longer precipitation time series become available. Here, PFEs can be computed directly at the weather radar pixel scale (corresponding to the spatial resolution of the radar data) without the need for interpolation or other models of ungauged areas.

In this study, we aim to investigate how weather radar derived PFEs compare to rain gauge derived PFEs, especially at the short timescales that are necessary in urban drainage design. In addition to rain gauge radar pixel PFE comparisons, we aim to utilize the fully spatially distributed weather radar derived PFEs to analyze the spatial structure of model parameters over a study area in Denmark. Utilizing a 18-year long C-band weather radar record, PFEs are derived in the form of IDF curves at the pixel scale, along with the corresponding PFEs of rain gauges located within the study area. Timescales ranging from 1 minute to 2 days are considered. The weather radar and rain gauge data sets are analyzed using the median plotting formula for empirical return levels and extrapolated to longer return periods by constructing a partial duration series (PDS). The PDS is then modelled by the Generalized Pareto distribution, where model parameters are determined via maximum likelihood estimation. The resulting PFEs display clear scale differences, where weather radar derived PFEs are underestimated at short timescales. However, IDF curves converge at timescales around 200-300 minutes. The spatially distributed model parameters reveal novel insights with regards to spatial variation of extreme precipitation in the study area. Clear gradients are found in the number of yearly exceedances, the mean exceedance, and the shape parameter controlling the PFEs. Moreover, these parameters are also clearly dependent on the timescale considered, where higher timescales equal smoother parameter surfaces with higher spatial correlation. These results highlight the advantages of supplementing rain gauge data with weather radar data for supplementary information about spatial variation of extreme precipitation over a given area. They also underline methods for determining the specific timescales where users should be aware of scale differences, given the inherent different measurement techniques of rain gauges and weather radar.