How best to account for seasonality when analysing bird monitoring data

[Ally30]

Hello All,
I have some monitoring data from seabird surveys carried out in an offshore construction site. Surveys were undertaken before, during and after construction and I am trying to evaluate whether there were significant effects on the observed seabird densities as a result of disturbance or displacement cause by the construction works.
My issue is that surveys were carried out through the year and there are likely to be seasonal effects on the bird densities. I am trying to work out the best way to account for these, so that I can answer the question:
"Was there a significant effect of the development on the density of seabirds in the area, after having accounted for variability in counts introduced by seasonality".
I have tried a GLM (with negative binomial distribution as the data are extremely overdispersed). The response is bird density and the covariates are period (as a factor, "before", "during" and "after" construction) and season (as a factor, with 4 levels).
What I am unsure of is interpretation of the results. Both factor covariates are significant in the model. But I can't work out if this means that there are significant independent effects of period of construction and season (as may be expected) or whether it means that period of construction has a significant effect on bird density AFTER having accounted for the effect of season on the bird numbers.......
I hope that this is making sense!
I am considering whether introducing interactions between 'period' and 'season' may be sensible. I am also considering a mixed model, with season as a random variable.
Any comments or advice would be very welcome!

 

[chiro]

Hey Ally30.

The interaction term with regards to period and season seems like a good place to start to establish whether some kind of bias is introduced with regard to the season.

One other suggestion might also be that if you have each season as an individual group (i.e. a random variable), then doing a group by group comparison in various ways could establish whether there is evidence of differences of various parameters between the groups. The simplest test would be something like an ANOVA relevant to the assumptions of the data. For this, you would re-classify densities according to seasons.

If you are looking at variability within the seasons, the next logical simple test would be to look at variance and different spread measures between seasonal data (including separate variation, and relative variation between the groups).

In terms of a regression, you can use the above to get an idea of how you can incorporate the relative variation in terms of a regression model: if the variation (and relative variation) is significant then this will be reflected in the various regression models.

You can do F-tests to compare various fitted models, but getting an idea of these kinds of variation can be seen if you divide things into groups.