On the use of radar-based quantitative precipitation estimates for precipitation frequency analysis

- We examine key challenges in using radar-based information for frequency analysis.
- Radar-based frequency estimates show underestimation biases compared to gauges.
- Estimation methods did not show significant differences in mean frequency estimates.
- Short radar records had a clear impact on the variance of frequency estimates.
- Underestimation in frequency estimates are explained by radar conditional biases.

SummaryThe high spatio-temporal resolutions of radar-based multi-sensor Quantitative Precipitation Estimates (QPEs) makes them a potential complement to the gauge records for engineering design purposes, such as precipitation frequency analysis. The current study investigates three fundamental issues that arise when radar-based QPE products are used in frequency analysis: (a) Effect of sample size due to the typically short records of radar products; (b) Effect of uncertainties present in radar-rainfall estimation algorithms; and (c) Effect of the frequency estimation approach adopted. The study uses a 13-year dataset of hourly, 4 × 4 km2 radar-based over a domain that covers Louisiana, USA. Data-based investigations, as well as synthetic simulations, are performed to quantify the uncertainties associated with the radar-based derived frequencies, and to gain insight into the relative contributions of short record lengths and those from conditional biases in the radar product. Three regional estimation procedures were tested and the results indicate the sensitivity of the radar frequency estimates to the selection of the estimation approach and the impact on the uncertainties of the derived extreme quantiles. The simulation experiments revealed that the relatively short radar records explained the majority of the uncertainty associated with the radar-based quantiles; however, they did not account for any tangible contribution to the systematic underestimation observed between radar- and gauge-based frequency estimates. This underestimation was mostly attributable to the conditional bias inherent in the radar product. Addressing such key outstanding problems in radar-rainfall products is necessary before they can be fully and reliably used for frequency analysis applications.