Data management of extreme marine and coastal hydro-meteorological events

Abstract

In statistical extreme value analysis and forecast modeling, data screening and management are necessary steps before fitting a probability distribution to represent adequately the observed data. These methods include trend analysis, steadiness tests, seasonality analysis, and long-memory studies; are critically reviewed and applied to coastal datasets. It was shown that the smaller the timescale of the coastal process, the more likely it tends to be non-stationary. The seasonal variations in the autocorrelation structures are present for all the deseasonalized daily, 1/3-monthly and monthly coastal processes. The investigation of the long-memory phenomenon of coastal processes at different timescales shows that, with the increase of timescale, the intensity of long-memory decreases. Only the daily water level series exhibit a strong long-memory. Comparing the stationary test results and the long-memory test results, these two types of tests are more or less linked, not only in that the test results have similar timescale patterns, but also in that there is a general tendency that the stronger the nonstationarity, the more intense the long-memory.

Keywords:

• Data screening
• extreme value analysis
• long-memory study
• seasonality analysis
• steadiness test
• trend analysis