Coherency detection of multiscale abrupt changes in historic Nile flood levels

Abstract

A scanning algorithm based on the classical Student t ‐test is extended to detecting the presence and coherency of multiscale abrupt changes in two time series. It functions similarly to the Haar wavelet transform, but also provides thresholds of the statistical significance for testing change‐points. The algorithm is applied to the historic series (AD 622–1469) of the maximum and minimum flood levels of the Nile River using the “Table‐Look‐Up Test” for correction of dependence in the series. The results reveal several abrupt changes, and separate wet/dry episodes on 30–160 year time‐scales. Most of the episodes coincide with historic records of catastrophe in Egypt. Some changes correspond to changes in China and Europe, and to reconstructions of fish stocks off the west coast of North America, indicating that they are driven and linked by global teleconnection processes.