Last year at the BES Annual Meeting in Ghent, I took part in an R package workshop where I learned the rudiments of scripting R packages. The ‘traitfindr’ package that our group came up with and started is, well,….still in progress. Nevertheless, the writing and introduction of new R packages is somewhat of a cottage industry these days. And, a valuable service is provided by authors of papers that demonstrate the utility of a package by way of a worked example addressing important ecological concepts.

Methods in Ecology and Evolution Senior Editor, Aaron Ellison recently noted in his ‘Also of Interest‘ post that all the BES journals publish methodological papers. But, we all know that MEE, more than any of the other journals, provides the primary home for these papers. Scanning recent issues of MEE, I came across three that struck me as being particularly important and that I expect will interest Journal of Ecology readers.

Twitter as a source of citizen science ecological data

Hart, A.G. et al. (2018). Testing the potential of Twitter mining methods for data acquisition: Evaluating novel opportunities for ecological research in multiple taxa.

OK, so this paper doesn’t present a new R package, but as a regular Twitter user (@davidjohngibson), I couldn’t resist highlighting this paper. The very idea that my tweets or anyone else’s could actually be of some ecological value is fascinating. The authors show that twitter-mined data can be used to answer certain types of ecological questions. They searched Twitter for terms (hashtags, e.g., #flyingants, #housespiders, #murmuration) related to winged ant emergence, autumnal house spider sightings, and starling murmurations, respectively. They then compared these data with published citizen science data collected over the same time period. The Twitter-mined data were most successful for quantifying temporal ecological patterns, but were less successful in determining microhabitat influences on winged ants or weather patterns on starling murmuration behaviour.

Flying Ant Image by Charles Lam, Flickr CC 2.0

There are limitations in using Twitter-mined data including determining the relevancy of certain hashtags to the topic of interest, and biases from social media trends. However, there are advantages and opportunities that other sources of data don’t offer such as the immediacy of the data for memorable phenomena (tweets are generally made the same day that an observation is made). As with other citizen science data, the data needs validating by comparison to other sources reporting the same phenomena. Previous suggestions have been made to use Twitter-data in ecological studies (e.g., Daume, 2016), but Hart et al. provide the first extensive testing of this approach.

The potential for using social media to collect citizen science data is just starting to be realized. Other ecological applications that have been proposed include the provision of bio-diversity data, mapping species distributions, studying phenology and range shifts (see citations of Daume’s paper and references in Hart et al.). Social media platforms that are being mined include Facebook, Flickr, and Google Images (see Leighton et al., 2016) in addition to Twitter. So far, I doubt that any of my tweets will be used as Twitter-mined data, but here’s hoping #redwine #cappuccino #invasivespecies.

Identifying functional trait-environmental associations

Miller, J.E.D. et al. (2018) Functional traits and community composition: a comparison among community-weighted means, weighted correlations, and multilevel models.

Understanding functional trait-environmental relationships is of tremendous importance for determining patterns and mechanism of community composition and structure. We now recognize that simple species-environment relationships only scratch the surface and that we need life-history and evolutionary-based analyses. The most popular of the available approaches is community-weighted mean regressions (CWMr) where species trait values are averaged across a site and regressed against an environmental variable. The problem with this approach is that many species will be represented in many sites meaning that sites are not independent, leading to high Type I errors. In other words, we may be misled into interpreting spurious relationships. The authors used a test dataset and simulations to demonstrate this problem. They also compared the propensity to obtain Type I errors and the Statistical Power of CWMr with four weighted correlation metrics, and multilevel models (MLM).

The authors concluded that there was no ‘best’ method, but CWMr should definitely not be used and that Jamil et al.’s (2013) MLM approach offered the highest statistical power along with routines to help control for inflated Type I errors. The authors provide R code to allow readers to make use of the MLM approach. At this point it remains to be seen whether ecologists will take up and use the methods tested here, but the very least we need to be more cognizant of the risk of making Type I errors whatever approach is used.

Modelling microclimate

Maclean, I.M.D. et al. (2018) Microclima: An R package for modelling meso- microclimate.

Down-scaling climate data from the coarse-gridded 10-100 km scales at which measurements are taken to the smaller scale of organisms that ecologists are interested in is a time-consuming and difficult problem. Maclean et al. present an R package; Microlima, that provides a flexible and accurate hybrid approach to modelling fine-scale variation in temperature at meso- and microscales. They illustrate their approach by modelling hourly air temperatures for 12 months at a 100m resolution across ~280 km² of the Lizard Peninsula, Cornwall, UK, and also for 1 month (May) at 1m resolution across 1 km²  (as shown in the figure).

In testing their approach the authors focus upon producing meso- and microscale temperature estimates of their study area but without relating the patterns to organismal distributions or performance. However, impressive finescale resolution of near-ground temperatures were modelled with slope and aspect being the principal determinants of the spatial variation. It would be fascinating to see how such fine-scale hourly temperature variation relates to fitness and performance of, say, annual plants such as Arabis alpina that shows extremes of phenotypic plasticity and local adaptation across elevational gradients (de Villemereuil et al., 2018).

Methodological Advances

The three papers highlighted here share a common theme that they propose and test novel quantitative methods of interest and value to ecologists; as they should do to be published in MEE or any other BES journal. In talking about the use of quantitative methods, my former PhD advisor and Journal of Ecology Editor, the late Peter Greig-Smith, said in the preface of the first edition of his 1957 Quantitative Plant Ecology book, that there was “a growing awareness among ecologists of the need to place their science on a more exact basis”. This sentiment remains true today and while I am unsure what he would have thought about social media let alone data-mining, I know that he would approve of the advances in quantitative methods that the papers highlighted here provide.

You can find out more about the Methods in Ecology and Evolution on their blog. And you can read more of their articles on the Wiley Online Library.

David Gibson, Executive Editor, Journal of Ecology