Monitoring strategies for phytoplankton in the Baltic Sea coastal waters

Phytoplankton monitoring in the Baltic Sea is to a large extent harmonised through the HELCOM COMBINE protocol. This ensures that the methods of sampling and analysis are quite similar and that data should be relatively comparable. There are differences in the spatial and temporal coverage of samples taken within the different monitoring programs. Moreover, within the national monitoring programs there can be large variations in the number samples taken at different stations, between years and during the year. Most monitoring stations are sampled more frequently during summer. Although the chlorophyll a and species-specific phytoplankton biomass has been measured routinely and by standard methods since the early 1970s, most national monitoring programs have had a reasonable monitoring effort after about 1990 only. New methods for collecting data, such as ships-of-opportunity and remote sensing, provide additional information to the traditional shipboard sampling and other new emerging technologies may provide alternative means for monitoring phytoplankton. We investigated the variation in phytoplankton biomass on the basis of the CHARM phytoplankton database and proposed a statistical method to improve the precision of biomass indicators. The precision of the annual phytoplankton biomass can be greatly improved by taking the seasonal variation into account, but describing the correlation structure in data contributes to improved precision as well. This latter method attempts to separate variations in phytoplankton biomass into systematic and random variations, thereby obtaining more correct estimates of the residual variance. Consequently, the number of observations required to obtain a given precision could almost be reduced by 50%, simply by interpreting data from another perspective. Nevertheless, variations in the phytoplankton biomass are still substantial and it may not be realistic to expect precisions below 30% from biweekly to monthly sampling. However, it is possible that improved modelling of the variations by including covariables may reduce the residual variance even further, improve the precision and thereby reduce the monitoring requirements, but this will require more detailed analysis that are outside the scope of the present work.