Oxygen saturation trends in the Baltic Sea : Problems with statistical inference of marine environmental data.

Marine monitoring programmes are seldom planned and realized to fulfil the requirements of the statistical methods used. The shortcomings of these methods are rarely discussed in the literature. In this work two different methods for trend analysis are applied to a real case and attention is given to the possible violation of the underlying assumptions. Long-term changes in oxygen supersaturation in the Baltic Sea are the object of the study. Linear regression and non-parametric methods are those chosen. The linear regression gave significantly increasing trends of oxygen saturation in Arkona, Eastern Gotland basin and in the Bothnian Sea, but the tested assumptions turned out to be violated. The corresponding non-parametric trend test only confirmed the increasing trend in Arkona. It is suggested that non-parametric methods, like the one used in this work, should be favoured in the future for similar types of marine environmental data.