A spatial bootstrap technique for parameter estimation of rainfall annual maxima distribution

Abstract

Estimation of extreme event distributions and depth-duration-frequency (DDF) curves is achieved at any target site by repeated sampling among all available raingauge data in the surrounding area. The estimate is computed over a gridded domain in Northern Italy, using precipitation time series from 1929 to 2011, including data from historical analog stations and from the present-day automatic observational network. The presented local regionalisation naturally overcomes traditional station-point methods, with their demand of long historical series and their sensitivity to very rare events occurring at very few stations, possibly causing unrealistic spatial gradients in DDF relations. At the same time, the presented approach allows for spatial dependence, necessary in a geographical domain such as Lombardy, complex for both its topography and its climatology. The bootstrap technique enables evaluating uncertainty maps for all estimated parameters and for rainfall depths at assigned return periods

Observed changes in precipitation extremes across 11 basins in China during 1961–2013

“…The GEV distribution often makes assumptions of stationarity and homogeneity within the time period of the fit (Ghosh et al , ). Up to now, precipitation series have been successfully fitted by GEV distributions in many studies (Kharin and Zwiers, ; Ghosh et al , ; Min et al , ; Uboldi et al , ). These studies imply that the population distribution of precipitation can be approximately fitted by the GEV distribution and it is appropriate to estimate precipitation return levels from GEV so long as non‐stationarity issue is not prominent during the process.…”

Evidence of non‐stationarity in a local climatology of rainfall extremes in northern Italy

“…The present study focuses on areas, rather than particular points. Regional averaging is appropriate for evaluating climate changes (Klein Tank et al , ) because it enables to filter out local systematic errors, reduce sensitivity to outliers (Uboldi et al , ) and compensate for station density changes. Moreover, the study of climate and its variations generally concern areas, rather than individual locations, where a long series of climatic observations may or may not be present.…”

“…Besides the regionalization discussed above, quality control checks on individual values and series (object of a presentation at EMS 2013: Uboldi et al , ) were performed to find and eliminate gross errors. Each time series underwent simple quality control tests (such as climatological range) and internal consistency tests (such as monotony among maxima of different durations for the same year).…”

“…In the recent past, we collaborated to updating the hydrological database of ARPA Lombardia (the environmental agency of Lombardy, Italy) and its rainfall annual maxima statistics. A new estimate of Depth–Duration–Frequency curves was obtained by introducing a new method, the ‘spatial bootstrap’ (Uboldi et al ., ). That work also lead to some preliminary indications on the stationarity of rainfall statistics (Uboldi et al ., ).…”

“…A new estimate of Depth–Duration–Frequency curves was obtained by introducing a new method, the ‘spatial bootstrap’ (Uboldi et al ., ). That work also lead to some preliminary indications on the stationarity of rainfall statistics (Uboldi et al ., ). Here, we specifically investigate stationarity of the rainfall annual maxima statistics in Lombardy for several different event durations, from 1 to 24 h.…”