Connectivity of green and blue infrastructures: living veins for biodiverse and healthy cities

Methods in ecology: counting things and sampling

You do ecology by counting things. But it's not easy...

There are many ways to generate questions about ecology. Maybe you are looking for a new question no one has asked before, or a new way to ask a question about an old debate, or maybe you just want to ask for yourself a question that has been asked many times, to experience finding the answer. In any case, you might do things like observe, form a mental image, make an abstraction, recognise a pattern, create a pattern, form an analogy, use your kinaesthetic senses (body thinking), empathise with something in the world, make a model, or just play. Now you have a question in the form of an observation, a pattern, an analogy, a model, or a feeling, for example. So far, you would be doing the same thing as an artist, an inventor, a philosopher, a physicist, or an anthropologist. Let's say your observation/ pattern is what Thoreau wrote about how squirrels create forests by forgetting about their buried hickory nuts: "This, then, is the way forests are planted. . .If the squirrel is killed, or neglects its deposit, a hickory springs up." So far this is a nice observation/ image/ speculation, and could be described as natural history. But let's say you want to be an ecologist.

Ecologists count things. Your job as an ecologist is to convert this observation/ image/ speculation into something that can be counted.

There are several ways of counting, here with examples from beetles:
Nominal scale: Categories. Can measure mode (the most frequently occurring number in a sample), frequencies (e.g. percents).
Example: Number of detritivores, herbivores, omnivores, predators.
Ordinal or ranking scale: Ordered but without equal intervals, cannot use arithmetical operations (e.g. the average)
Example: Number of eggs, larvas, adults (the egg comes first in time, followed by the larva, then the adult, so there is an order to the categories)
Interval scale: ordered with equal intervals. Good for most arithmetic operations.
Example: Dates of egg laying
Ratio scale: Interval scale with true zero point. Good for all arithmetic operations.
Example: Beetle body masses

So, if we go to Thoreau's squirrel observation again, we might think of a number of countable things:
-squirrel density
-hickory tree density
-percent hickory trees producing nuts
-distance nut carried by squirrel before deposit
-number of hickory nuts deposited by squirrel
-number of hickory nuts eaten later by squirrel
-number of hickory nuts germinating
-number of hickory nuts surviving n number of years
etc.

However, we can rarely count everything that might be relevant to our question, because (1) the world is BIG, (2) time goes on for a long time, (3) detectability: we cannot find all the squirrels and all the buried hickory nuts and all the germinated hickory nuts (4) the world is very complicated and even figuring out whether one squirrel is really two squirrels or whether five squirrels is actually one squirrel, is more complex than you might think. Consequently, ecologists count a subset of things, and then use statistics to make claims about what would have happened if they had been able to count everything. The subset has to have a couple of key features. Each of the samples that is counted has to be independent: that is, if we say it is one squirrel it really has to be one squirrel and not half of a Siamese squirrel, or five squirrels. In addition, all the samples have to be strictly comparable. This means, in practice, not only that all squirrels are squirrels and not chipmunks, but also that all the observations of squirrels are done in the same way. If I observe all the squirrels at midnight in a blizzard from 100m, some of them may not be squirrels, but they all have a comparable likelihood of being squirrels given the observing conditions. However, if some of my squirrel observations were made at noon in clear weather from 1 m, these are definitely squirrels, with a much higher certainty than in the other sample. According to sampling logic, these two kinds of observations are different and can never be compared.

Of course, we do ultimately end up comparing things that are not comparable, for example when trying to summarise the results of several studies, and this is where interpretation and argument come in.

But going back to counting things, if samples are not independent, this is called "pseudoreplication". Since everything interacts and is connected in ecology, sometimes you could argue that all sampling procedures in ecology lead to pseudoreplication. This can be problematic. In any case, when you measure the same thing multiple times (knowingly), you calculate measurement error; there are no other things you can say about what you counted if everything you counted is the same thing. Measurement error is expressed as the little error bar you see on bar charts or graphs.

One of the goals of sampling is to make a good, non-biased, representation of the "everything" we can't actually count. There are two ways to do this: being very methodical (grids, quadrants or transects (lines), for example), or being very random. You can also carefully combine these. This is also where comparability can be really important, because you have to ask if your methodological or random approach can be carried out in an exactly comparable way in all circumstances. For example, if my methodological approach is to sample every 1 meter along a North-South line, what do I do when I reach a cliff, or a tar pit?

Even detecting the things you are counting can be a big challenge. Hence all the kinds of live traps (that don't kill the trapped animal), sensors, binoculars, camera traps, etc. etc. that we use in ecology. The bat sensors, that record the bats' squeaks and calls, that we use in this project, are an example. Of course, detection problems are almost the entire game in some scientific fields where all the objects of study are microscopic.

Sampling is a big part of BIOVEINS, so I asked Pedro Pinho, from the Lisbon team, about the team's sampling strategy.
Me: How were the cities chosen for the projet?
Pedro: Mostly because these were the cities with which our partners already work with. But we can say that the cities were chosen to represent a large climatic gradient within Europe, from Mediterranean to Hemi-Boreal. They also represent a number of different urbanization types (from the oldest Lisbon and Paris to the more recent types in Poznan and Antwerp. And also city size, from huge Paris to small Tartu. But the main reason is climate, "how does habitat fragmentation influences biodiversity and ecosystem services and how climate influences that relationship"?
Me: If I remember correctly, you chose to have almost equal numbers of parks of different size classes in each city. So, the sampling strategy doesn't represent the true distribution of park sizes in each city; on the other hand it has an equal sample size in each size class for each city. Can you explain why you chose equal sample sizes over representativeness? How does that effect the kinds of conclusions that we can draw from the study?
Pedro: Exactly, we choose not to represent a city, but rather the phenomena under study (the fragmentation). If we wanted to represent the city, this could occur: 80% of the parks chosen would be small, 15% would be medium and 5% large. That would provide a very poor cover of what we are actually interested, fragmentation. Thus this ensure that we would end up with 33% of large, medium and small parks. Representing the environmental gradient of fragmentation ensures that we can make models regarding the effects of fragmentation: rather that representing a city we want to create an "universal" explanation of the effects of fragmentation.
Me: What is the current paper about?
Pedro: We are working in two papers. [The one about sampling] aims at understanding how can we use remote sensing [satellite images] directly to choose where to sample. Normally we decide this using maps (meaning land-cover that was classified a priori for some reason, like Corine or the Urban Atlas). Here we are testing if this could be done directly from remote sensing, with the advantages of being universal and permanently up-to date.

How to ask questions in ecology: Root-Bernstein, M., Root-Bernstein, M., & Root-Bernstein, R. 2014. Tools for thinking applied to nature provide an inclusive pedagogical framework for environmental education. Oryx. 48(4), 584-592.

--Meredith Root-Bernstein