Monday, November 19, 2007

Seal stomachs V: How do we know what seals eat?

This set of posts starts here.

Stomach contents are to diet as weather is to climate. At its simplest, the accumulation of records of the weather each day – the temperature, how much rain falls - allows us to build up a picture of the climate for a particular place over time. So it is with animals' diets – if we could watch every meal that every individual of a species or population ate, then, by accumulating records of their meals, we'd have perfect knowledge of their diet. For marine mammals is this obviously impossible – unless our population of interest are all kept in captivity.

So. How do we know what seals eat? The most intuitively obvious answer would be to watch seals feeding. Then we'd know what they ate, because we'd see it happening. Logistically impossible, although technology offers hope. So. We can either find ways of looking at what our animals have ingested, or we can look at where they've gone to feed and work out what's there, or some mix of both of these. And we can do this in ways of varying sophistication.

What do seals ingest?

We can look at the contents of their stomach for recent meals; or in their intestines for slightly older material; or at their faeces, for what was in their intestine without actually handling the seals. Intestinal contents can be obtained without killing seals - they discover the joys of involuntary enemas. For seals that haul out on ice, faecal sampling presents problems that are seen as insurmountable. Enemas on ice are difficult, but not completely impossible.

And what do we do once we've removed the gut contents? There can be whole fish or squid in a stomach, or just remnant hard parts – the ear bones, backbones or ribs of fish; the beaks of squid. By sorting and measuring these back in a lab, and with access to a suitable reference collection - bones and beaks of likely prey, including examples of different sizes of the same species - we can tell what species the seal ate, and what size the prey were. Whole prey items from a stomach are easy: we can simply count, measure and weigh them. But what of hard parts? How do we allow for changes to those that we find – different bits erode in their own ways. And what do we do with hard bits that we can't identify to species, but to some higher taxon (cod-like fishes, for instance)? One way around this is to analyse the DNA of all gut contents, or of problematic goop. These analyses can be reliable and precise, but they're always time-consuming and expensive.

Or we can turn to chemistry for evidence. Chemically, seals are what they eat. Tell-tale fatty acids, absorbed from fish, shellfish, squid - or whatever else that seals might eat - turn up in seals' blubber. The chemical composition of these fatty acids, extracted from blubber samples, can, using sophisticated data-mining algorithms, be compared with the chemical composition of likely prey items. We were collecting blubber samples for Tore's colleague at the Polar Institute who did exactly this. All that's needed is a plug of blubber from the seals, and representative samples of possible prey species for the chemical analysis. And a chemistry lab and a good statistician.

Another chemical option involves comparing the relative composition of isotopes of important elements – primarily carbon and nitrogen – with representative samples from putative prey. Patterns in isotope composition provide information analogous to that from fatty acids, and can be obtained from old bones, as well as from fresh chunks of animal (skin, hair, teeth, internal organs, whatever).

Chemistry smears data over time. It offers panorama, but sacrifices detail. Stable isotopes provide insight into what an animal has been eating for months or longer, and fatty acids for weeks or months, depending on how the species under study lays down fat. But this longer perspective must have a price – it's harder to tell exactly what species has been eaten.

So it would appear that gut sampling is the best option. But detail also has its price. Fresh, just eaten food in a stomach is easy to identify. But what of animals with nothing in their stomach? What are they telling the researcher, other than that a lot of animals must die to provide data? More problematically, what about the half-digested gunk that's usually there as well? Squid beaks are chemically nothing like the hard parts of fish, and seals' stomach acids digest beaks much more slowly. How do we account for this?

Answering these questions requires addressing a far more basic one – why are the data being collected at all? Why do we care what seals eat? After all, tootling around the Arctic in an ice-strengthened research trawler doesn't come cheap. We were supposed to be addressing two questions about the foraging ecology of harp and hooded seals: how did each species' diet vary over the course of a year, and was there dietary overlap between the two species?

The question that goes begging here is – why ask these particular questions?

Because an animal's diet is the food its eats over time, the chemistry-based approaches (fatty acids and isotopes) offer the advantage of having time resolved, sacrificing detail to do so. But what of gut contents? At best, intestinal or faecal sampling provide data on meals over the past few days. So how can we make inferences about feeding over time - diet - from meals - snapshots in time?

Drawing inference from field data preoccupies ecologists. Sampling theory – how can we infer population-level conclusions from a set of field samples – drives the design of our protocols for data collection. How can we ensure, to the best of our ability, that our samples will allow us to come to unbiased conclusions about the populations of interest? It's here that a couple of key concepts come into play.

Randomization is vital. To a scientist, random means something special, something different from haphazard. In normal usage, the two are interchangeable. In scientific sampling, they are worlds apart. Random sampling is like a lotto draw – each sample has the same probability of selection. Haphazard sampling is when no thought is given to the probability of selecting a sample. Bias – the bane of being able to make inference from data – will almost certainly result from haphazard sampling, and what's worse, even if there is no bias, no-one can ever know.

Pseudoreplication is another key concept. Now that computer programs handle the grunt work of statistical analyses, there's a presumption that ecologists will collect a lot of samples of their data. These samples are supposed to be independent, otherwise they provide another source of bias when it's time to make inferences. Pseudoreplication involves collecting samples that aren't truly independent for the inferences made in the study. Like beauty or pornography, pseudoreplication can be in the eye of the beholder. Are the inferences that we make from our data appropriate? An example from the world of marine mammal stomach contents offers clarification.

A Norwegian government scientist goes into the North Sea on a whaling ship. The whalers kill a dozen or so minke whales, and the scientist checks the whales' stomachs to see what they've been eating. The whalers are interested in killing their quota of whales as quickly as possible (they're out there earning a living, after all), and find many whales in one discrete area. Most of their quota of a dozen are killed in this one spot, a few square miles across. The whales' stomachs are full of sand eels, a small fish that occurs in huge, discrete aggregations.

What inference can the scientist draw from these data? That working from a commercial vessel provides data that can be biased, because commercial considerations override a scientist's desire for randomization? Sure, and rightly so. That a large aggregation of sand eels can attract minke whales, that will feed on them? Not entirely unreasonable. What about – most minke whales in the North Sea eat sand eels? Absolutely not.

That final inference must be based on a few assumptions. The most important is that the whales killed were a representative sample of all minke whales in the North Sea. But we know - from surveys in the North Sea, and from other information on the biology of minke whales – that minkes occur elsewhere, not just in that one specific area where there happened to be a lot of sand eels. The assumption that the stomachs were a representative sample is what's known as a strong assumption – an assumption which, if it's wrong, causes everything about the study to fall apart and leaves the scientist looking like a bit of a goose.

This example – with the inference drawn that sand eels are the main food of minke whales in the North Sea – was published in a scientific journal a few years ago. The author treated each stomach as an independent sample, in order to make his inference about the whole of the North Sea. But most of the whales that were killed had aggregated to feed on the same thing (sand eels), which is why there were where they were, and so available for killing and having their stomachs investigated. So his samples weren't independent samples. Hence the term, pseudoreplicates.

Place matters, particularly when studying marine mammals that can move through entire ocean basins. So what about looking at where marine mammals feed? These days, SLTDRs are tool of choice for seals that probably can't be handled twice (like harps and hoods). There are other options for seals whose behaviour makes it likely that they can be found again – ones that return regularly to a haulout site accessible to scientists. The coolest toys for the well-heeled scientist in this situation are tiny video cameras that store digital imagery, instead of storing depth data. So finally, we can watch how seals feed. But as we can't ever be sure of re-catching a harp or hood, this isn't an option with them. SLTDRs (Satellite-Linked Time-Depth Recorders, remember?) are used, but they're kissed goodbye once they've been attached.

After our day ashore, Tore had his permission to hunt in Icelandic waters. So we all made our way back to the Jan Mayen, the ship cast off, and we steamed into Denmark Strait. Thanks to the accident while at Jan Mayen (the island), we were about to hunt seals in an area well south of where we intended to originally.

Continued here

No comments: