Methods Flashcards
> Tell me about your project (~ 3-4 minutes)
Problem
We know that mean shifts in climate are having a profound influence on the distribution, abundance, and conservation status of species. But climate change is also associated with an increase in extreme climate events.
I want to focus on those extremes — and in particular, the possible resulting extremes in marine population dynamics.
I want to know,
- How can we forecast when and where catastrophic population dynamics will occur?
- Is there a climate change signal to the prevalence and magnitude of population dynamic extremes?
- And, can we use this knowledge to make better decisions about how we conserve depleted species in our oceans?
Solutions?
To do that, I will work with Eric Ward and his fisheries stock assessment colleages at the Northwest Fisheries Science Centre / NOAA, Michelle McClure (the director of fisheries science at Northwest Centre), Trevor Branch an assitant professor at the University of Washington, and Jodie Toft and Paul Dye at the Nature Conservancy in Seattle.
We want to apply cutting edge space-time species distribution models to the extensive trawl survey datasets that are collected along both coasts of the United States.
These kinds of models have only been possible in the last year or so with advances in statistical software and approaches, and have been largely developed by the people I plan to work with at NOAA.
These models are super powerful, in that we can now estimate ecological processes (like productivity and density dependence) that vary across both space and time.
We want to extend these models by first incorporating a mechanistic population dynamics model and, second, allowing for catastrophic recruitment events.
Unfortunately, at this point these models are inaccessible to the majority of researchers — these models involve learning new statistical software, a fairly extensive knowledge of statistics, and extensive coding in a low-level programming language.
One output of our project will be a package for the statistical software R that creates a user-friendly interface to applying these statistical models. This will be useful far beyond fisheries science, to any ecological abundance data collected through space and time.
We will use these models to investigate the frequency and predictability of localized catastrophes in fisheries recruitment (the number of fish entering fisheries each year) and assess whether there’s a link between these population extremes and mean shifts and extremes in climate variables.
This research is of fundamental long-term conservation importance, and assessment scientists at the Northwest Center are really excited to incorporate these models into how they model fish populations in stock assessment.
And then, there’s a second more basic output that the Nature Conservancy is really interested in.
We can also use these models to come up with abundance estimates at a fine scale through space and time (say down to the month) of the abundance of fish species that they’re particularly interested in. Jodie Toft and Paul Dye at TNC work fairly closely with fishers, fisher groups, coastal tribes, and council members as part of the process of designating rockfish conservation areas and essential fish habitat zones. These areas are re-evaluated at regular intervals but decisions about the boundaries of these zones have been rather coarse, mostly based on depth contours, and some would say a bit arbitrary.
A second product of our project will be an interactive website (using the new R website backend, Shiny) that explores the output from our space-time models that Jodie at the TNC can use one-on-one and at meetings with fishers, fisher groups, coastal tribes, and at council meetings to explore how population projections might look under different climate and fishing scenarios.
We think these can go a long way towards bringing more science to bear on the decision making about how to best draw lines in the ocean to protect threatened and endangered fish species and essential fish habitat.
So what?
As a whole, this work directly targets clauses in the Magnuson-Stevens Act that call for measures to avoid catastrophic fisheries disasters.
On a global scale, the Intergovernmenal Panel on Climate Change is focussed on managing the risks of extreme events - on social, economic, and biological level — which our work addresses.
Locally, our work will tie directly into stock assessment practices at the Northwest Fisheries Science Center, and likely eventually at other assessment agencies as well.
Our work also ties into the Washington State TNC’s goals to help fishers fish in more stustainable ways while making a stable income, and having less impact on marine habitat.
Ultimately, our project has implications for everyone who eats fish, depends on fisheries for a living, and for the structure and function of our oceans.
> How would you explain your project to a grade ~6 student?
- Climate change means the world is mostly getting warmer, but it’s also getting weirder; temperature is varying up and down a lot more and we’re getting more frequent really really hot months and years and more frequent big storms like hurricanes and flooding
- this is happening on land, like where we are today or at your home, but it’s also happening in the oceans, where all the fish that we eat live, and all the sharks, and whales, and other animals you might have seen in a movie, on a boat, or in a zoo
- in the oceans, not only does the temperature change, but there are also changes in the properties of the water itself; for example, there can suddenly be large areas with little oxygen in them that the fish need to survive
- I want to figure out how all these changes in climate are affecting fish in the ocean
- I’m especially interested in how all these weird climate changes might be causing sudden and surprising changes in how many fish are born in certain places in the ocean
- ultimately, I want to help us make better decisions about when and where we can catch some kinds of fish and when we should protect them so that we can make sure we have lots of fish to eat now but also wonderful animals in the oceans and and lots of fish to eat for our kids, and your kids, and for 100s and 1000s of years.
> What species do you want to work with? What areas?
- we want to work across a wide variety of life histories and ultimately a variety of regions too (PNW, down coast to California, up to Alaska, and probably groundfish data on the East Coast too)
- e.g. forage fish, which are lower in the food chain, may be more susceptible to sudden shifts in the distribution of plankton (and thus temperature);
- Northern anchovy have gone through some dramatic shifts in the past
- eulachon are listed as threatened under ESA and in Canada under SARA, have had some dramatic swings in the past and are an important part of the NE Pacific food web, as food to marine mammals, birds, salmon, etc.
- rockfish conservation zones
- groundfish (e.g. petrole… TNC especially interested in sablefish, dover, petrole, and hake)
- possibly Dungeness crab too; TNC quite interested in and in the same trawl data, but not sized, recorded as total weight only
> So explain what these models might look like.
I imagine we’d be fitting a stock-recruit function as a process model (such as a Ricker or Beverton Holt), then you can let one or both stock-recruit parameters (e.g. productivity and density dependence parameters) vary with space and/or time and be influenced by environmental variables, possibly in some non-linear function.
(Jim Thorson and Eric Ward at the NWFSC just published a paper that came out last month working with a similar model in which density dependence is allowed to vary by space. They find that ignoring spatial variability can result in strongly biased estimates of density dependence and model comparison statistics, like AIC, strongly favour models with spatial variance in density dependence.)
Things to think about: which distributions to use, whether catastrophic failures symmetric or not. If we can integrate size class data.
> What contingency plans do you have if technical or implementation isn’t working?
In the near-term TNC is most interested in simply having easily accessible interactive tools for applying rigorous science to the ongoing designation of essential fish habitat and rockfish conservation zones and engaging fishers and fisher groups in those decisions — that should be doable regardless of the extremes element.
Essentially, what I’m proposing has a risky but potentially very important element to it (a focus on extremes and local catastrophes in recruitment and a possible link to the environment). But it also has a less risky component. Even without the extremes element, we can build more rigorous models that can inform and communicate our best knowledge about species distributions through time and inform conservation area designations.
> Can you explain what a random effect is?
- ‘random effects’ or ‘latent states’, or ‘varying parameters’, or ‘hierarchical parameters’
- refer to parameters that are assumed to arise from some stochastic process, and we can estimate the distribution of likely values
- most commonly we assume the parameters arise from a normal distribution; so we assume these ‘random effect’ parameters are constrained by some normal distribution; we then let the data tell us how similar the parameters are to each other — in other words, how big the variance is across these random effects
- if the variance is really small then they’re very similar and the model will pool a lot of inference across ‘random effect’ levels, if the variance is very large (relative to the effect size) then they are nearly independent, and what we see at one ‘random effect’ level doesn’t tell us much about what’s happening at another
> What’s a parameter?
- something we estimate from data
- at a basic level, estimating the mean or average value of a dataset is estimating a parameter
- often we combine a model of how we think a certain process works with data; our model contains values that we can estimate from the data — say how quickly an average fish grows — and by fitting our model to data we can estimate the size of these values and how certain we are in their values
> How is your project risky? Where could things go wrong?
- part of my project involves asking whether there’s a climate signature to patterns of population extremes in the oceans; it’s possible that effect will turn out to just not be a big deal, or that the data are not yet informative enough to be able to say one way or the other
- this involves cutting edge method development, and there will be some challenging problems we’re going to have to think carefully about; e.g. when you allow for movement through space and time, it can be hard to say whether a population has experienced a local catastrophe at a given area in time or has simple moved to a new location in space
> Have you got started at all at this point?
- we’ve been discussing it as a group
- we have the data… (some of it now public from Malin Pinksy’s Smith work)
- Eric, Jim Thorson, and others at NWFSC have published a number of papers in the last year describing the basic methods behind what we’re proposing
- (briefly, these involve what are called spatial random fields — they are essentially allowing spatial processes to be drawn from a random effect distribution… in fact Jim and Eric just had a paper in Ecology this month on a spatial Gompertz model using random fields to allow density dependence…, AIC…)
- in some capacity we plan to carry out this work, but I think it would be a great opportunity for the Smith program to be involved
- recently, I’ve been looking at time series of population abundance for all sort of species from around the world and looking at how often jumps in abundance from year to year are far more than we’d expect under typical population dynamics
- we call these ‘black swan’ population dynamics
- are you familiar with the term ‘black swan’?
- black swans are a term coined by Nassim Taleb in his book ‘Black Swan’
- he uses the term to refer to unexpected and extreme events that have undue influence and are often rationalized in retrospect; for example the onset of world wars, stock market crashes…
- we find that across species somewhat around 5% of population exhibit jumps in abundance (~90% downward jumps) that are nearly impossible given typical population models over an average span of about 30 years; they are usually associated with big swings in climate, hunting, or predation and often a combination of those at once
- this may have profound influence on projections of extinction risk and I think plays well into our proposed project, where we want to look at this more mechanistically with high quality data and on a geographic scale that is relevant to policy making
> How is this different from tipping points, regime shifts, and PVA?
- my work complements many other fields
- tipping points and regime shifts refer to permanent or semi-permanent changes
- PVA could/can incorporate extreme deviations and spatio-temporal random effects…
> How is this project cutting edge?
- cutting edge models…
- large scale data synthesis, continuing on from Malin’s work
- working with recruitment, so mechanistic
- using brand new software
- brand new interactive web graphics capabilities
- big increase in focus on risk and extremes in ecology in recent years
> What is essential fish habitat?
From :
“Fish require healthy surroundings to survive and reproduce. Essential fish habitat includes all types of aquatic habitat—wetlands, coral reefs, seagrasses, rivers—where fish spawn, breed, feed, or grow to maturity.”
> What are forage fish?
“‘Forage fish’ species are small or intermediate-sized pelagic species (e.g. sardine, anchovy, sprat, herring, capelin, krill) that are the primary food source for many marine predators” Pikitch et al. 2014
> Expected papers?
- method development paper with simulated and real data (integrating catastrophic distributions into spatiotemporal models)
- large-scale data analysis of drivers affecting recruitment
- ss3sim paper with recruitment catastrophes; account for them or don’t
- perspectives pieces precautionary recommendation about rolling closures to protect against — more of a thought piece; integrate with climate from GCMs over say 50 years; how might we design reserves to buffer against those (longterm projections)
What’s the big advantage of these recent space-time models in layman’s terms?
- big advantage in layman’s terms is: ecologists have worked with spatial in past, but we can now let spatial random effects vary from one year to the next, and these models are often heavily favoured over spatially constant population models
- we can make the easily make the case that by including complexity we’re increasing the precision of predictions
- in previous models, we’ve linked to fixed habitat, e.g. depth, bottom type, distance to rocks, but a lot are static in model
- our proposal includes variables like climate that are varying spatially and temporally
