We will hopefully have a paper coming out soon that lays out the 2012 Gulf of Maine story--expect a link and more info when it's ready. As you recall, 2012 was really warm. Lobsters started moving inshore and began molting very early. This led to an increase in lobster landings but a collapse in prices.

One of the statements we make in the paper is that "Many biological responses to the warmer 2012 conditions could have been predicted using data that are routinely collected by ocean observing systems." This statement is meant to be a challenge to the community to start developing ecosystem forecasts using the observing systems. After all, many of these systems were sold based on their ability to provide information useful to resource management.

To test our assertion, I put together some very simple statistical forecasts for weekly surface temperatures at Buoy E in the Gulf of Maine based on the temperature conditions during the first week of May. This is basically an extension of my prediction from a few weeks ago. I also developed a simple statistical model that predicts the change in the start of the high-catch period of the lobster fishery. A description of the models is at the end of the post. Here's what we get for 2013:

The blue region is the mean temperature cycle surrounded by 95% confidence bounds. The gray region is the 95% confidence intervals surrounding the prediction (black line). The red crosses are the actual observations. At the start of May, we were just under a degree warmer than average, and the model predicts a slightly above normal summer. Note that the most recent observation was actually a bit below normal. It will be interesting to see how this progresses, given that NOAA is still predicting an above average summer using their more complex models. I predict a slightly earlier start to the lobster season.

To check the validity of this approach, I went through the full forecasting process for each year from 2002-2012. For each year, I didn't include any data from that year when I fit the temperature and lobster models. Thus, the correspondence between the red crosses (the observations) and the forecast (black line) should be a good test of the approach. Here's how we would've done in 2012:

Given how anomalous 2012 was, it's a bit surprising that this is one of the best forecasts. However, this might actually make some sense, especially for the lobsters. The lobster model only includes temperature effects--it doesn't include other processes that might influence landings like the price of fuel or bait, the number of lobstermen, or biological factors. Since the 2012 temperature signal was so strong, it likely overwhelmed all of these other factors. Here is the complete set of hypothetical forecasts:

So, it looks like there is some potential to forecast aspects of the lobster fishery, and potentially summer water temperatures as well. The accuracy of the forecasts could be improved within the year by assimilating information (temperature and lobsters) as it comes in or by incorporating other factors into the models.

*The Models*The temperature model predicts the sea surface temperature in week j (SST_j) using the previous week's temperature (SST_j-1) and the temperature at 20m in week 16 (T16, first week in May). For each weekly period (after week 16), I fit the model:

SST_j=a_j*SST_j-1 + b_j*T16 + c_j

using all data from that weekly period. My very quick assessment is that the fits are pretty good and that including T16 is an improvement over the SST model, but this will need to be more carefully analyzed. To make the forecasts, I use SST_16 and T16 to compute SST_17. I then feed SST_17 into the model for week 18 and keep rolling forward. It would be easy to substitute the observed temperatures.

The lobster model is even simpler. It is just a vanilla linear regression between the start date and T16. The trick is how to define the start date. We grabbed monthly landings data from Maine DMR and NOAA. For each year, we divide the monthly values by the total landings for the year. I found the day (using linear interpolation) when landings should first reach 8% of the total. Of course, the fishery runs year-round, but this captures the start of the really intense summer period. I have not looked at the end date. For the figures, I just applied the shift in the start date to the end date (assuming an early year is also a late year). This is more to show how we might visualize the duration of the season.

## Leave a comment