Regional frequency analysis of extreme precipitation with consideration of uncertainties to update IDF curves for the city of Trondheim

• Published in: Journal of hydrology (Amsterdam) 2013-08, Vol.498, p.305-318
• Main Authors: Hailegeorgis, Teklu T., Thorolfsson, Sveinn T., Alfredsen, Knut
• Format: Article
• Language: English
• Subjects: Balanced bootstrap resampling; Earth sciences; Earth, ocean, space; Exact sciences and technology; Extreme precipitation events; Extreme values; Hydrology; Hydrology. Hydrogeology; Intensity–Duration-Frequency curves; L-moments; Precipitation; Quantiles; Regional; Regional frequency analysis; Sampling; Trends; Trondheim; Uncertainties; Uncertainty
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2013.06.019

•We applied regional frequency analysis (L-moments) for extreme precipitation events.•We consider uncertainties (trends, homogeneity, distributions and sampling).•Significant sampling uncertainty quantified from balanced bootstrap resampling.•The IDF curves from this study are more reliable than the existing one.•This study contributes to endeavors for estimating reliable IDF curves. Regional frequency analysis based on the method of L-moments is performed from annual maximum series of extreme precipitation intensity to update Intensity–Duration-Frequency (IDF) curves for the city of Trondheim. The main problems addressed are (1) reduction of uncertainties of different sources for reliable estimation of quantiles: (i) testing of trend patterns and stationarity of the data series from the target site and demonstrating the dependency of results on the data used; (ii) testing regional homogeneity of extreme precipitation events for the climate regime in the study area and “pooling” of regional data for data augmentation and reduction of uncertainty due to short length of data series; and (iii) selection of distributions for extreme precipitation events of different durations to reduce the uncertainty due to choice of distributions; and (2) assessment and quantification of sampling uncertainty in terms of interval estimates (confidence bounds) of quantiles. Trend patterns and check for stationarity were demonstrated for a data from a target site based on both non-parametric Mann–Kendall and parametric regression tests. Selection of distributions was performed based on Z-statistics and L-moment ratio diagrams. Non-parametric balanced bootstrap resampling was used to quantify the sampling uncertainty. For extreme precipitation events of shorter durations (5–30min) there are statistically significant increasing trend patterns for the data series with start years of 1992–1998 while there are no significant trend patterns for recent extremes and there are no statistically significant trend patterns for longer durations (45–180min). The results of the analyses indicate that: (1) significance tests for trend patterns and stationarity are dependent on the data series used but the stationarity assumption is valid for the data series used from the target site. (2) the extreme precipitation events from four sites in Trondheim are homogeneous and can be “pooled” for regional analysis; (3) different types of distributions fit to extreme precipitation events of different du