Reading group: Index this

Sumatran tiger on camera trap. Photo via wikimedia commons (which states that “this animal didn’t like camera traps and proceeded to destroy three over a weekend”)

In our latest QAECO reading group, Kate Giljohann selected an editorial published in the Journal of Applied Ecology titled “Management by proxy? The use of indices in applied ecology”. The article, led by Philip Stephens and editors of the journal, discussed the role of proxies in applied ecology and ways to improve their use in environmental management.

Most of us use proxies or indices in our research. But what are they? Well, scientists and managers often want to measure and keep track of key attributes of ecological systems to inform decision-making. For example, estimating population size and the rate at which it is changing is fundamental for determining the status of species, whether they should be managed, or how they have responded to management intervention. But, often these quantities can be difficult or time consuming to measure directly, so instead we measure something else that’s closely related. The assumption is that if we keep track of what our proxy is doing, we can also keep track of what our real quantity of interest is doing as well.

Continue reading

Posted in Uncategorized | Leave a comment

Poll open! Vote for your #favRpackage

Thanks for all the great nominations for #favRpackage!  The poll is open – vote now for your favourite R package for ecology or conservation-related data analysis. Poll closes 10 October.

Posted in Polls, Uncategorized | Tagged | Leave a comment

Reading Group: Species-fragmented area relationship

leal1Species-area relationships (SARs) are perhaps the closest thing we have to a law of community ecology. Accordingly, they’ve received copious study over the last two decades, from theoretical development and formalisation, to parameterisation for natural systems and use in conservation planning.

Species-area relationships are often modelled with a power law function relating species richness to habitat area. One underlying theoretical idea is that SARs represent a relationship between habitat area and extinction risk for individual species: more habitat = lower extinction risk per species = more species. But what if habitat is fragmented? In human-dominated landscapes, for example, loss of habitat almost always entails habitat fragmentation, which should have important consequences for species-area relationships. Metapopulation and metacommunity theory says that it must, because habitat fragmentation increases extinction risk for species over-and-above the loss of habitat extent. As such, traditional species-area relationships might significantly overestimate species richness in fragmented landscapes.

In our most recent reading group, we looked at a potential solution to the overestimation problem published by Ilkka Hanski and colleagues in 2013. In one of his last major projects, Hanski produced a simple but elegant adjustment to the power law relationship between habitat area and species richness to account for the degree of habitat fragmentation. The adjustment is a multiplication of the traditional species-area relationship by the fraction of species that would persist for a given amount of habitat fragmentation, denoted P(λ), where λ is the ‘metapopulation capacity’ of the fragmented landscape (itself a function of patch areas, inter-patch distances and species’ dispersal capacities). Assuming a generic λ, Hanski et al. used the following form for estimating P(λ):

P(λ) = exp(-b/λ),

where b is a parameter to be estimated from species richness data among fragmented landscapes. This fraction can simply be multiplied by a standard SAR formulation to adjust it for fragmentation. In the example in the paper, the commonly used power law formulation: cAz (where A is the area, and c and z are parameters) is adjusted to yield an expected number of species: S = cAz exp(-b/λ). The fragmentation component could in theory be bolted on to other SAR formulations too.

Hanski et al. demonstrated that their simple formulation worked quite well for both simulated and real data, and we felt that its derivation from first principles is a major strength. Nevertheless, we wondered if some slight extensions could make the approach more flexible and powerful, and improve its application to real-world datasets.

Of course we hear you cry, λ requires knowledge of extinction and migration rates, how are you going to derive those extra parameters? Well, in the PNAS paper Hanski et al. argue and demonstrate that the assumption of similarity among species, meaning a single value for each of b and λ, has little effect on the overall model fit when the focus is taxonomically or ecologically related species (e.g., forest birds). We felt that further work in other empirical cases would be beneficial to validate this. And indeed, as another way of dealing with the parameterisation, we talked about how one might use a hierarchical model to estimate b from likely sparse observations across species in real-world datasets.

One of the drawbacks of the simple power law SAR is that the parameters don’t have a direct biological interpretation and must be estimated from species-area data for each context. This model can be used by ecologists to fit curves to their data, but not to predict the number of species in other ecosystems (though there are other SAR functional forms that are more interpretable). Whilst λ in the SFAR is deeply rooted in ecological theory, the paper doesn’t provide an ecological interpretation for the parameter b or for the functional form of P(λ). If we could interpret (or reformulate) P(λ), it could be bolted on to a more interpretable SAR model, and enable us to make and test real predictions about species counts in fragmented habitats.

Overall, we felt this study provides a solid first step for adjusting SARs for the near-universal issue of habitat fragmentation. Further empirical testing of the approach is fertile ground for future research, as is refinement of the approach to relax assumptions required in the estimation of b and P(λ), inject clarity as to their ecological interpretation and allow greater flexibility to tackle messy real world datasets.

Posted in Reading Group | Leave a comment

These are QAEco’s favourite R packages, what’s yours? #favRpackage

Are you an ecologist?

What stats program did you use to analyse your data ten years ago?

How about now?

A recent paper by Justin Touchon and colleagues reviewed changes in the statistics and statistical programs used by ecologists over the past 24 years. It seems that, not only are we publishing more, but classic analysis methods are being replaced by ‘sophisticated modelling frameworks’. Good old ANOVAs and Man-Whitney U tests are dropping out of fashion, while AIC is being used more than 100 times more frequently in ecological papers now than back in 1990.

Furthermore, the open source program R has gone from nowhere in the early 2000s to being the most widely-cited statistical program – mentioned in a third of all articles published in 2013!

So, if you’re an ecologist aboard the R boat, what’s your favourite R package? QAEco is planning a poll to find out. Currently, our favourites include:

Spatial data – analysis, mapping and modelling

raster – nominated by Nick Golding, raster allows you to analyse spatial data and ‘do all the GIS you want (and more)’ in R. raster also interfaces with a host of other GIS software including sp for vector GIS, GDAL for raster file I/O, and rgeos for everything in between. Most importantly, being able to do all your GIS work in R (rather than point-and click software) makes it easy to create automated and reproducible analyses – Nick reckons it has saved him more hours than he could possibly count. To get started with raster and the rest of the R GIS ecosystem, check out @frodsan’s tutorial.

dismo – this is Jane Elith’s favourite because it makes it much easier to run species distribution models efficiently. dismo relies on other modelling packages for fitting models, but allows the typical steps necessary for distribution modelling, and efficient prediction to large rasters. The main author is Robert Hijmans.

ggmap Saras Windecker’s nomination, ggmap combines the spatial information of mapping programs with the layered grammar of ggplot2. This allows for the production of modular spatial graphics that are easily tweaked to your specifications.

adehabitat  (particularly adehabitatLT and adehabitatHR) – Bronwyn Hradsky finds this suite of packages super-useful. Written by Clement Calenge, adehabitat facilitates the analysis of animal movement data, such as relocation data from GPS or VHF collars. The packages make it easy to convert movement data in to trajectories, visualise, error-check and manipulate these data, and analyse home ranges and habitat selection. They also come with a set of very readable vignettes, which provide a great introduction to this field.

Data manipulation

dplyr – Elise Gould says that dplyr makes wrangling your data frames a breeze. dplyr is a metaphorical set of ‘pliers’ for wrangling your data frames, to do things like row- or column-wise subsetting, conduct group-wise operations on multiple subsets of data, or merge data frame and matching rows by value rather than position. Using dplyr (rather than base R) means that common data manipulation problems take less code and less mental effort to write. Moreover, much of dplyr’s work is implemented behind the scenes in C++ code, making wrangling larger data frames lightning-fast! To get started with dplyr, have a look at the data wrangling cheatsheet. For more detailed explanations, see the ‘wrangle’ section of Hadley’s forthcoming book, R for Data Science, and read this great explanation of ‘tidy data’.

reshapeEsti Palma finds this package very helpful for data management. It allows the user to summarize, re-configure and re-dimension datasets, using only two functions; melt() and cast(). There are heaps of online tutorials about how to use reshape (and its faster reboot reshape2). Quick-R and Sean Anderson provide two simple options to get anyone started. Both reshape and reshape2 have been developed by Hadley Wickham.

For interfacing between R and externally-compiled code

RcppJian Yen is a fan, because Rcpp makes easy to run C++ code from R – all you need to do is write a C++ function and run one line of R code. Better still, Rcpp provides extensions to standard C++ syntax, which means that you can write C++ code that looks a lot like R code. This is awesome because many of us will have spent hours staring at a screen waiting for R scripts to finish running. Sometimes we can get around this by writing better R code, but there are times when that won’t be enough. That’s where Rcpp comes in, letting you write and compile C++ functions that target bottlenecks in your code. Hadley Wickham provides an overview and short guide, and the Rcpp website also has a lot of useful information.

jagstoolsGerry Ryan’s favourite. jagstools allows to you work with the Bayesian hierarchical modelling engine JAGS because, as Gerry says, who doesn’t love to Gibbs sample? jagstools takes results object from JAGS via the package R2jags — which are basically complex lists — and returns a simple matrix of the parameters. This makes your results much easier to work with and plug into graphs or other models. jagstools was made by QAEco-logist (and R-extraordinaire!) John Baumgartner. Here JB runs through simple worked examples of how to use jagstools, and what you might use it for.

How about you?

QAEco wants to find out which R packages are most popular among ecologists – but first we need some nominations. Tell us about your favourite R package for ecology or conservation-related data analysis in the comments below, or #favRpackage and tag @Qaecology.

Nominations close 16 September. Back in 2014, we found that Australia’s favourite eucalypt was the Mountain Ash – this new poll is a wee bit geekier!

Posted in Polls, Statistics | 6 Comments

PhD opportunity: threatened native mammals, invasive predators and fire at Glenelg Ark

QAEco is seeking an enthusiastic and highly qualified candidate for a PhD project, researching the influence of fire and invasive predator management on threatened native mammals at Glenelg Ark, south-western Victoria.


Potential research topics include:

  • Threatened native mammal responses to fox baiting and fire management. Potential focal species include southern brown bandicoot, long-nosed potoroo, dayang (heath mouse) and swamp antechinus
  • Interactions between fire and introduced predators:
    – Red fox and feral cat responses to fire (behaviour, habitat selection, diet etc.)
    – Effects of habitat structure on invasive predator hunting success
  • Unintended consequences of fox control:
    – Changes in macropod, possum or feral cat behaviour and abundance in response to fox control, and implications of these changes for vegetation and threatened native fauna

Applications close 30 September 2016.  See  GlenelgArk_PhDopportunity_2017 for more details.

Posted in Events, Jobs | Leave a comment

Reading group: Environmental DNA sampling — hope or hype?

In our latest QAEco reading group, Emily McColl-Gausden and Reid Tingley led a discussion on ‘Environmental DNA for wildlife biology and biodiversity monitoring’ by Kristine Bohmann, Alice Evans and colleagues.

Bohmann et al. examined eDNA from a monitoring/ecology perspective, and therefore didn’t delve too deeply into the genetic details (which some of our less genetically-inclined members appreciated!). The paper did, however, highlight some very interesting, and perhaps less well-known eDNA applications from both aquatic and terrestrial environments. Surveying for endangered deer using leech blood and collecting eDNA from Arctic fox paw prints were two intriguing examples. Bohmann et al. also highlighted some important limitations of eDNA sampling, such as the occurrence of false positive detections both in the field and in the lab. The paper looked to the future in its longer term aspirations for eDNA, in addition to more realistic shorter-term goals.


Former QAECO student, Adam Smart, takes a water sample for eDNA analysis from a roadside drain (Image: Reid Tingley)

We all agreed that there were many potential applications of eDNA sampling, and that researchers were only beginning to tap into its potential. There are also certainly benefits from an animal ethics point of view. However, one issue that caused much discussion was the ability to estimate abundance from eDNA. Some members of the group were rather dubious about the strength of this correlation, and how it might change in different environments. We came to the conclusion that eDNA-based abundance data could potentially be used as an index to rank the abundance of species, rather than an absolute number (e.g., for differentiating rare vs. common species).

For some in the group, the paper was too optimistic in its predictions of global networks of eDNA monitoring stations. Others, however, thought that technological advances could enable widespread data collection in the not-too-distant future. Handheld devices for sampling and analysing eDNA data were mentioned as an example of such an advance.

More broadly, this paper catalysed an interesting discussion regarding how intensely we scrutinise new technologies in ecology. The point was raised that we don’t seem to question the more established sampling techniques as much as we probably should. For example, electrofishing is not 100% effective for detecting fish, yet imperfect detection is rarely discussed or considered. And, of course, as a group of quantitative ecologists who are interested in conservation decision-making, we wondered about the value of resolving many of the uncertainties regarding eDNA sampling. Given a fixed budget, should we invest in resolving such uncertainties, or are we better off taking additional environmental samples? Following on from that point, we agreed that we would like to see more papers on the cost-effectiveness of eDNA sampling, as well as more focus on imperfect detection. Did we mention that we’re quantitative ecologists?

Posted in Reading Group | Tagged | 1 Comment

Reading group: towards unification

This is the second instalment of our fortnightly reading group blog series. This time, our group was led by Skip Woolley and we discussed the paper Towards a unification of unified theories of biodiversity by McGill 2010.

This is an interesting paper that identifies different community ecology theories that are generated by inherently different mechanisms, but in turn, generate very similar community ecology patterns. These patterns are: species-area relationships, a hollow shaped abundance curve (RAD) and distance-decay relationships (Fig. 1).


Figure 1. Intrinsic patterns in community ecology. a) Species-area relationship; b) Rank abundance distributions; and c) Distance-decay relationships.

A case is put forward by McGill (and others) that this generality can be used to unify these theories and establish a set of rules that are applicable to all these approaches and thus, can be taken as generality.

McGill settles on the rules:

  1. Individuals within a species are clumped together
  2. Abundance between species follows a hollow shaped curve
  3. Species are treated as independent and are placed without regard to other species.

As a group, we liked the summation of these approaches and the respective patterns they produce. Lending us to accept certain mathematical forms, the rules are shared in common across these approaches and thus can be taken as generality. Similar concepts have been applied to other fields of science and the concept has previously been discussed amongst our reading group (The common patterns of nature by Frank 2009).

We did feel that the unification of these theories was a slight stretch, and that a way forward would be to test under which conditions these rules stand up? For example, a number of our reading group participants are involved in modelling joint species distributions, so we felt that the assumption of species independence was not a requirement to generate the patterns. Given this, what would be the bare number of rules required to generate these patterns? One way to do this would be to start with a set of necessary starting conditions (for example, species interactions from a food web) and assess which patterns arise from these interactions.

We also felt a more interesting question was: what drives these patterns? Along these lines of reasoning, we discussed the idea that these patterns could be viewed as null models and further investigation via theoretical or empirical studies could help highlight deviations from these expected patterns. As a whole, we felt that understanding the cases where these rules breakdown would help us understand their underlying mechanisms. In this vein, McGill touches on a need to understand processes that drive richness and abundance patterns at the end of the paper. A better understanding on the processes that shape these patterns will help theoretical ecologists more accurately investigate the relationships between these theories, and have the added benefit of assisting applied ecologists in their efforts to manage biodiversity.

Until next time.

Posted in Reading Group | Tagged , , | Leave a comment