Estimation of Great Lakes Water Level Statistics: Conditioning Via “The Bootstrap”

Abstract

Reliable lake level frequency distributions are a critical component of any comprehensive strategy for coping with Great Lakes water level fluctuations. However, statistical techniques commonly used on riverine systems are inappropriate for large lake systems, due to the levels’ long-term persistence and dependence on the prevailing climatic regime. To illustrate an alternative methodology, we present a series of resampling analyses modeled after well-known bootstrap techniques applied to 130 years of monthly Lake Erie water level records. The analyses show that lake level exceedance probabilities should be conditioned on 1) length of planning horizon, 2) starting month of planning horizon, 3) initial lake level, and 4) climatic regime. Our methodology can be extended to additionally consider storm and wind effects on levels, to incorporate levels data available for discontinuous periods prior to 1860, and to develop other types of lake level statistics useful to decision makers, such as duration and time-to-exceedance probabilities.