The view of a data scientist - BONUS BIO-C3 & BONUS INSPIRE
The changing effect of climate on the productivity of coastal sea
Best place to study what drives the zooplankton variability over long time is Pärnu Bay. This is because the Estonian Marine Institute, and its predecessors, have sampled the zooplankton in Pärnu Bay since 1957 (to predict the herring recruitment). The resulting time series is probably the longest, and most detailed zooplankton time-series from the Baltic Sea that I am aware of.
In Pärnu Bay, zooplankton is dominated by small copepods (look like tiny shrimps under the microscope), especially two taxa: Eurytemora affinis, and Acartia spp. What happens with E. affinis is also of economic importance, since it is the primary prey item for both, the larval herring in Pärnu bay in May-June, as well as for the adult herring in open Gulf of Riga year round.
We "re-found" that effect of winters was strongest in the beginning of the growth season - in May, and in June. When the winters were milder, there were more copepods in the water. Exactly as all the previous papers I found said it would be (see the list in the end).
We also noticed that warmer water in spring and summer was speeding up the growth and increasing the average age of the population. It takes some time for the copepods to grow up, from the baby-copepods ("nauplii") to adults that can start reproduction, about two months. When the water is warmer, this time becomes shorter, and we saw it in the increasing share of adults. So far still not so much a news.
Things became more interesting, when I started to look how consistent have these effects been over the time.
Non-stationary and non-linear links
A hot topic in the marine ecology right now is how to detect causal links between different organisms and environment, when those links are not simple and linear (e.g. recent papers by Liu et al, or Klein et al. - warning! - the methods described in those papers are numerically quite demanding!)
In our paper, I (half-accidentally) realised, that non-linear and non-stationary links can be detected also with quite simple (linear) methods. For example with a sliding window correlation analysis.
In short, "sliding window correlation" means that you slice up your time series into shorter bits (preferably at least 15 years long), and see, if you find the similar correlations between variables in all of the short subsets. The slicing up is not done randomly, but with a moving window: first set is the first 15 years (1957-1971), next one is the window moved forward by 1 year (1958-1972), and so forth. When the whole time series is 57 years long, you can make 42 such steps. Looking at the time series of model results can tell you already, whether the links you found using the whole dataset are robust, or not.
And they were not, of course (they almost never are in nature). In the periods, when winters were on average colder, warming had a stronger effect on the spring production than the similar increase in temperature during periods that were already quite warm. So, during colder winters, copepods were more stressed, and more relieved by even small warming. And if the winters are mild enough, it does not make anymore much of a difference, when they get even warmer. That means that the link between winter conditions and copepods was non-linear.
But we also found that with the sea surface becoming warmer in May and June, the Acartia spp. was no longer affected by winter conditions so much, and become rather dependent on sea surface temperature. This shift happened around 1980s, and was an example of a non-stationary effect, which usually happens when the effect of one variable depends on the mean level of another variable.
So, the bad news is - it's not over yet, we are not there yet. We only figured out that we need to switch to methods that can tell us what affects who also when the effect itself changes over time. But for one, results like this help to understand, why the different researchers that use only some parts of the whole time series can sometimes reach very different conclusions. And demonstrated again the importance of very long time series, in our quest of unravelling the whole web of links that make up the marine ecosystems.
We are deeply grateful to all the researchers and technical staff in the Estonian Marine Institute, and its predecessor institutes, whose dedication and hard work has kept the zooplankton monitoring alive over the many decades.
These papers all report positive effect of milder winters on small-sized copepods in different parts of the Baltic Sea: