Simulation of Nonstationary, Non‐Gaussian Water Levels on Great Lakes

Abstract

A methodology for simulating nonstationary non‐Gaussian water levels on the Great Lakes is presented for use in providing input scenarios for erosion modeling on the Great Lakes. The methodology utilizes a low‐pass filtering technique to reduce the data to a stationary time series after which the data is transformed to a Gaussian series via use of the empirical distribution function coupled with a specialized tail‐fitting procedure for the extreme values in the data. The reduced series can then be resimulated in the frequency or time domain. The present series is resimulated in the frequency domain via a target spectrum, the spectrum of the reduced series. Inverse procedures are then utilized to simulate the original series. Comparisons of statistical and time series properties of the original and simulated series are made for a number of sites on the Great Lakes and it is found that the methodology provides a reasonable simulation for the original data.