Wyniki wyszukiwania - BazTech

Ograniczanie wyników

2 Acta Geophysica Polonica

Powiadomienia systemowe

Sesja wygasła!

Znaleziono wyników: 2

Liczba wyników na stronie

Wyniki wyszukiwania

Wyszukiwano:
w słowach kluczowych: asymptotic bias

Sortuj według:

Ogranicz wyniki do:

Physics of flood frequency analysis. II. Convective diffusion model versus lognormal model

Strupczewski W., Singh V., Węglarczyk S., Mitosek H.

Acta Geophysica Polonica

2003

Vol. 51, nr 1

85-106

The similarity between the convective diffusion (CD) model and the lognormal (LN) distribution is shown by comparison of their moment estimates. Both models are tested using annual peak discharges observed at 39 gauging-sections of Polish rivers. The average value of the ration of the coefficient of skew ness to the coefficient of variation equals about 2.52, a value closer to the ration of the CD model than to the gamma or the lognormal model. The likelihood ratio indicates the preference of the CD over the LN model for 27 out of 39 cases. Applying the maximum likelihood (ML) method, one should take into account the consequence of the wrong distributional assumption on the estimate of moments. In the case of CD, the ML-estimate of the means is unbiased for any true distribution, which is not the case with the LN model, moreover the ML-estimate of the two fist moments of CD remains asymptotically unbiased if LN is true, while there is small bias in the opposite case. To check the objectivity of our inferences from empirical findings, a simulation experiment was carried out, which comprised generated CD- and LN- distributed samples and both the moment and likelihood criteria for the distribution choice. Its results clearly show that normal hydrological sample sizes are far too small for selecting the true distribution.

Model error in flood frequency estimation

Strupczewski W., Węglarczyk S., Singh V.

Acta Geophysica Polonica

2002

Vol. 50, nr 2

279-319

Asymptotic bias in large quantiles and moments for three parameter estimation methods, including the maximum likelihood method (MLM), moments method (MOM) and linear moments method (LMM), is derived when a probability distribution function (PDF) is falsely assumed. It is illustrated using an alternative set of PDFs consisting of five two-parameter PDFs that are lower-bounded at zero, i.e., Log-Gumbel (LG), Log-logistic (LL), Log-normal (LN), Linear Diffusion (LD) and Gamma (Ga) distribution functions. The stress is put on applicability of LG and LL in the real conditions, where the hypothetical distribution (H) differs from the true one (T). Therefore, the following cases are considered: H=LG; T=LL, LN, LD and Ga, and H=LL, LN, LD and Ga, T=LG. It is shown that for every pair (H; T) and for every method, the relative bias (RB) of moments and quantiles corresponding to the upper tail is an increasing function of the true value of the coefficient of variation (cv), except that RB of moments for MOM is zero. The value of RB is smallest for MOM and the largest for MLM. The bias of LMM occupies an intermediate position. Since MLM used as the approximation method is irreversible, the asymptotic bias of the MLM-estimate of any statistical characteristic is not asymmetric as is for the MOM and LMM. MLM turns out to be the worst method if the assumed LG or LL distribution is not the true one. It produces a huge bias of upper quantiles, which is at least one order higher than that of the other two methods. However, the reverse case, i.e., acceptance of LN, LD or Ga as a hypothetical distribution while LG or LL as the true one, gives the MLM-bias of reasonable magnitude in upper quantiles. Therefore, one should be highly reluctant in choosing the LG and LL in flood frequency analysis, especially if MLM is to be applied.