Flood flow frequency analysis in the presence of outliers

Abstract

Flood flow frequency analysis is performed on the series of hydrological observations. The observations introduce uncertainties in the analysis results, inherited from limited data sets, operational difficulties in high flow measurements, flood flow estimates in the rating curve extrapolation zone, as well as different mechanisms of runoff generation. Difficulties in defining flood flow probabilistic model arise in the presence of outliers (significantly larger or smaller value compared to the remaining data set values). The paper provides an overview of methods for the detection of outliers from traditional practice to the latest research results, as well as procedures for outlier treatment in the further statistical analysis, focusing especially on the recommendations from the recently published Bulletin 17C in the USA. The assumptions about the population, procedures of outlier detection, the treatment of transformed data sets, and potential difficulties in the application are highlighted. Statistical tests and procedures are applied to flood data at 70 hydrological stations in Serbia. Analysis of flood flow frequency in the presence of outliers is performed by the methods of the annual maxima and peaks over threshold (POT). The examples include both upper and lower outliers. Comments are given on the differences in the results from the applied outlier detection and treatment procedures. Recommendations to the practitioners are given in the conclusions in order to reduce uncertainty in the flood flow frequency analysis in the presence of outliers.

Statistička analiza velikih voda se radi na nizovima hidroloških osmatranja, koja u rezultate ove analize unose neizvesnosti u pogledu ograničene dužine osmatranja, teškoća u merenju vrlo velikih protoka i njihove procene iz zone ekstrapolovanih krivih protoka, kao i zbog različitih mehanizama generisanja oticaja. Definisanje probabilističkog modela velikih voda je otežano i zbog postojanja izuzetaka u nizovima, tj. značajno velikih ili malih vrednosti u odnosu na većinu. U radu se daje pregled postupaka i statističkih testova za otkrivanje izuzetaka od standardnih do najnovijih, kao i načini postupanja sa izuzetnim vrednostima. Najveća pažnja posvećena je otkrivanju i tretmanu donjih izuzetaka, koji u velikoj meri utiču na raspodelu a time i na usvojene računske velike vode. Poseban osvrt je dat na najnovije preporuke iz američkog Biltena 17C. Komentarišu se pretpostavke o raspodeli populacije, procedure testiranja, postupanje sa transformisanim nizovima i ukazuje na moguće poteškoće u primeni. Sama primena testova i procedura prikazana je u praktičnom delu, gde se statistička analiza velikih voda u prisustvu izuzetaka sprovodi za karakteristične primere sa hidroloških stanica u Srbiji na nizovima formiranim metodama godišnjih maksimuma i prekoračenja preko praga. Primeri obuhvataju i gornje i donje izuzetke. Komentarišu se razlike u rezultatima prilikom primene različitih procedura u prisustvu izuzetaka i predlaže niz iskustvenih mera kako bi ocena velikih voda bila što pouzdanija.