Non-stationary analysis of water level extremes in Latvian waters, Baltic Sea, during 1961–2018

Abstract

Analysis and prediction of water level extremes in the eastern Baltic Sea are difficult tasks because of the contribution of various drivers to the water level, the presence of outliers in time series, and possibly non-stationarity of the extremes. Non-stationary modeling of extremes was performed to the block maxima of water level derived from the time series at six locations in the Gulf of Riga and one location in the Baltic proper, Baltic Sea, during 1961–2018. Several parameters of the generalized-extreme-value (GEV) distribution of the measured water level maxima both in the Baltic proper and in the interior of the Gulf of Riga exhibit statistically significant changes over these years. The most considerable changes occur to the shape parameter ξ. All stations in the interior of the Gulf of Riga experienced a regime shift: a drastic abrupt drop in the shape parameter from ξ≈0.03±0.02 to ξ≈-0.36±0.04 around 1986 followed by an increase of a similar magnitude around 1990. This means a sudden switch from a Fréchet distribution to a three-parameter Weibull distribution and back. The period of an abrupt shift (1986–1990) in the shape parameters of GEV distribution in the interior of the Gulf of Riga coincides with the significant weakening of correlation between the water level extremes and the North Atlantic Oscillation (NAO). The water level extremes at Kolka at the entrance to the Gulf of Riga reveal a significant linear trend in shape parameter following the ξ≈-0.44+0.01(t-1961) relation. There is evidence of a different course of the water level extremes in the Baltic proper and the interior of the Gulf of Riga. The described changes may lead to greatly different projections for long-term behavior of water level extremes and their return periods based on data from different intervals. Highlights. Water level extremes in the eastern Baltic Sea and the Gulf of Riga are analyzed for 1961–2018. Significant changes in parameters of generalized-extreme-value distribution are identified. Significant linear trend in shape parameter is established at Kolka. The shape parameter changes in a step-like manner. The shape parameter of GEV has regime shifts around 1986 and 1990 in the gulf.