I am currently working with activity data on European badger (Meles meles). They are nocturnal, and begin their winter sleep (not true hibernation) approx. early November lasting to early March. Their activity in the winter is limited indirectly by low temperature (frost) and snow cover due to less access to food (e.g. digging after earthworms).
This circannual rhythm of activity might be endogenously generated, although external cues such as photoperiod, temperature or snow cover can function to entrain a circannual rhythm.
I working with data from wild badgers and I will not be able to control for the different variables experimentally.
I am planning to test if the decrease in activity in the fall and increased activity in spring could be explained by the time of the year, photoperiod, temperature and snow cover.
However, I will not be able to separate photoperiod from time of the year, but temperature and snow cover do varies. I am seeking the most parsimonious model to explain what causes the badgers to go to winter sleep.
I wish to discuss and get advice on how I should treat the data.
Response variable:
Activity index (numbers of observations adjusted for monitoring effort).
Explanatory variables:
Temperature
Snow (depth)
Time of the year
Time of the year is the variable I find challenging and I want to hear your opinions. I am in particular interested in the fall vs winter and winter vs. spring. Could one solution be to split the data set into summer and winter solstice, i.e. June and December solstice (Northern Hemisphere)? Then I could use days until winter solstice and days since winter solstice as an explanatory variables respectively in addition to temperature and snow.
Hi Ronny, can you not use the inverse of photoperiod, i.e. night length, as your proxy for time of year? Admittedly, matched days either side of the winter solstice have similar night lengths, but if you co-factor days since last emergence I think it might capture some of the drive for activity (e.g. increasingly longer nights and longer gaps between emergences contribute to lower activity).
Hi Ronny, it sounds as though you are not only interested in factors that induce winter sleep, but also spring emergence. Since each of these falls, respectively into the fall equinox-winter solstice and winter solstice-spring equinox periods, it would make sense that you would split the analytical timeline (as you suggest) at the winter solstice to avoid the issue of similar daylengths on either side of the solstice boundary, but also because in the former case, you are analyzing co-variates potentially related to the induction of winter sleep, whereas in the latter, your interest is co-variates related to spring emergence. This becomes especially true if you are including day length or a John suggests, night length, as a focal co-variate. My suggestion for a boundary using the equinoxes is to allow enough of a timeline period before the engagement of the winter sleep or spring emergence, respectively, to enable a reasonably robust analysis.
Also, on using temperature as an explanatory variable, you should consider grabbing soil temperature as well as air temperature if the former is available. Soil temperature tends to be uniformly highly predictable in its slow seasonal change and have a much more direct influence on the overwintering site of the animal than rapidly changing air temperature that is not only much less predictable, but has a much less direct influence on your animal.
Lastly, to confidently answer this question, it would seem that you would need temporal or spatial replication of some kind to develop a robust model. Since between-year variability is frequently high, a number of years of repeat data would be ideal. You might be able to do this with spatial replication if your spatial replicates are different enough in context of the explanatory variables that you are measuring. Some combination of both may also be possible. I would expect that replication via repeat years would likely give the best analytical results.
Hi Ronny,
Since badgers are following an annual routine where they seek to optimize different aims at different season and because food availability varies a lot between spring and autumn, I would recommend you to code the seasons too and force the seasonal variables into the model: weather effects should be analysed as additional variation to seasons/life history variables (if you have any of these). To make the model as statistical simple as possible, I would personally divide seasons into categories (e.g. months or quartiles depending on the size of the dataset) at least to a start (post-hoc , you can consider using polynomial or sinus functions of Julian date if these functions gives better fit).
In addition to night length (and time of night if you have data of that resolution) and temperature, precipitation (e.g. dry vs. wet/humid weather) may also play a role in terms of the type of food available. I would probably test all weather effects as quadratic effects since any influence on behaviour may be non-linear.
Maybe Julian date (the days from a specific date) could be a good option?
Good luck!
If you have populations separated by several degrees of latitude, you may be able to assess whether there is an effect of light or dark accumulation going into and out of the winter solstice. Some species show dose-dependent responses to light or dark. For example horse mares exposed to long days abruptly in the winter will begin estrous cycles quickly. But if they are exposed gradually, they will not start their breeding cycles until they have crossed a threshold such as 14 hours of light. One could look at specific thresholds and see if these are more important than say the upward or downward slope of the increasing or decreasing light pattern or dark pattern.
Thanks for reply. I will as suggested divide the year into seasons (fall, winter, spring and summer) in comparison with Julian date (days since summer and winter solstice). Earthworms constitute a major part of their diet and I agree that wet and humid might be favorable conditions for earthworm availability. Unfortunately I haven't measured climatic variables locally (soil temperature), but data on rainfall and ambient temperature are available from local weather stations. There are spatial variation between locations of observations, 200 km at most, however it might not be enough for significant difference in day-length period.
Thanks for replies.
Response to Marc P. Hayes suggestion "Since each of these falls, respectively into the fall equinox-winter solstice and winter solstice-spring equinox periods, it would make sense that you would split the analytical timeline (as you suggest) at the winter solstice to avoid the issue of similar daylengths on either side of the solstice boundary"
Rather than split the analytical timeline to avoid the issue of similar day lengths on either side of the solstice boundary, could it be better to just include an interaction variable between 'day length' and the categorical variable ‘fall equinox-winter solstice and winter solstice-spring equinox’ in the model?
Ronny, no question that you could take the interaction variable route. I thought that you are dealing with two different questions, entry into overwintering versus emergence that might have different underlying issues, it made more sense to split the baby in context of easy interpretation.