Post Provided by Charlie Outhwaite & Nick Isaac
Nick and Charlie are giving a presentation on ‘Biodiversity Indicators from Occurrence Records’ at the BES Annual Meeting on Wednesday 16 December at 13:30 in Moorfoot Hall. Charlie will also be presenting a poster on Tuesday 15 December between 17:00 and 18:30 on ‘Monitoring the UK’s less well-studied species using biological records‘ in the Lennox Suite.
Biodiversity Indicators are some of the most important tools linking ecological data with government policy. Indicators need to summarise large amounts of information in a format that is accessible to politicians and the general public. The primary use of indicators is to monitor progress towards environmental targets. For the UK, a suite of indicators are produced annually which are used to monitor progress towards the Aichi targets of the Convention on Biological Diversity as well as for European Union based commitments. However, this is complicated by the fact that biodiversity policy within the UK is devolved to each of the four nations, so additional indicators have been developed to monitor the commitments of each country.
A range of biodiversity indicators exist within this suite covering the five strategic goals of the Convention; which include addressing the causes of biodiversity loss, reducing pressures on biodiversity and improving status of biodiversity within the UK. Within strategic goal C (improve status of biodiversity by safeguarding ecosystems, species and genetic diversity) there are currently 11 “State” indicators that use species data to monitor progress towards the targets underlying this goal. Most existing species based indicators use abundance data from large scale monitoring schemes with systematic protocols. However, there are other sources of data, such as occurrence records, that can offer an alternative if they are analysed using the appropriate methods. This post will discuss the development of species indicators for occurrence records to complement the current UK species based indicators, specifically relating to the C4b priority species indicator and the D1c pollinators indicator.
Recording schemes are one source of occurrence records. There are many recording schemes within the UK, from the Bees, Wasps and Ants Recording Scheme and the British Lichen Society to the National Bat Monitoring Programme; data are being collected all year round on species that are less well studied within a scientific context.
These data have an unstructured collection process. Volunteers can pick the time and location of their recording, as well as what they record, so there are a number of inherent biases. Because of this not many people have taken advantage of this wealth of data even though it is easier and cheaper to collect than the standardised monitoring equivalent.
Advances in methodology
Recent applications of hierarchical Bayesian occupancy models have made it possible to estimate species status and trends from this unstructured occurrence data, by employing statistical descriptions of the data collection process (see Isaac et al. (2014)). Our model differs from the classical occupancy model in a number of crucial ways which increase its suitability for working with unstructured occurrence data. Differences include, the addition of an observer model, the inference of non-detections from the observation of closely-related species, a random effect for site in order to deal with the unbalanced nature of the data and the use of list-length as an estimate of sampling effort.
These advances have made it possible to more easily estimate status and trends for many species in taxonomic groups for which no standardised monitoring data are available. For example Powney et al. (2015) use occurrence records to look at dragonfly occurrence trends in Britain and Ireland. The output from each species’ model is a time-series of occupancy estimates. Superficially, these data are similar to abundance data used to generate existing species indicators.
Development work was undertaken to improve and increase coverage of the UK priority species and pollinators indicators through the use of occurrence records. Working with these kinds of data offers additional challenges for indicator development. Due to the fact that many of the species being modelled are rare, or simply have fewer records, we are really pushing the limits of what is capable when using this methodology. Although a high number of species were modelled, for the 2015 indicator around 80% of these are lost as there are insufficient data to produce a reliable result. Fortunately, Bayesian models tell you when there’s not enough information in the data and so a set of rules were developed to determine which index values were suitable.
Rule number 1: Have your chains converged?
Check the Rhat statistic to ensure the Markov chains have converged over the set chain length. The usual rule is that any value greater than 1.1 means the chains have not converged.
Rule number 2: Is your occupancy estimate meaningful?
Within a Bayesian model, prior information is given which represents what we already know about the parameter values before running the model. For the prior on the occupancy estimate we use uninformative priors with a median value of 0.5. If estimates only including information from the prior were included in the final indicator, this could lead to misleading results. A threshold standard deviation value of 0.2 was chosen as it marks the point at which a unimodal (informative) distribution can be distinguished from a uniform distribution (indicative of little to no information within the data).
Rule number 3: Which species should be included?
After checking the first two rules, the number of reliable occupancy estimates each species was able to contribute to an indicator varied a great deal. A threshold value of at least 20 reliable years of data was set as this was considered to be sufficiently robust to take forward into a final indicator that was being constructed from 1970 to 2012.
The implementation of these rules results in species time series of occupancy which have missing values for those years where thresholds were not met. These missing values were interpolated from reliable values in adjacent years.
The final indicator!
Finally, we are ready to put a composite indicator together. All of these rules and indicator generation processes can be carried out using an R package, BRCindicators, to generate the appropriate form of indicator.
Although this method has helped us to learn more about the status of UK biodiversity, there are still opportunities for further development. Due to the nature of the data and the requirements of the models many species are too rare to use this method, with around 80% of model outputs being unreliable. To improve this we can try broadening the time window, sacrificing temporal precision to increase species coverage.
Categorising species is also a challenge. Currently species trends within the UK suite of indicators are grouped according to five categories of change. These were first developed for the wild bird indicators which are based on change in abundance and therefore may not be suitable for categorising species in terms of change in occurrence. This is something that needs to be considered for future updates of the indicators.
Communicating uncertainty to policy-makers is tricky, or rather… different from communicating it to fellow scientists. You can say something is uncertain, but they want to know “can I still use it?” Explaining what is meant by expressing the 90% credible intervals used in the Bayesian framework was one point that was contested in the final write up. The use of the term “occupancy” was also discouraged within the technical document and it was decided that this term should be replaced with “distribution” to aid communication with a wider audience. They want a ‘narrative’ for the indicator trajectory and they ask questions that they think politicians will ask them. Although this can be frustrating it does make you think more about how accessible your work is to a wider audience.
Although there are still challenges to be faced, the use of this type of data has enabled the expansion of the UK biodiversity indicators set and increased the taxonomic breadth of what we are now able to monitor. Both the C4b priority species indicator and the D1c pollinator indicator use unstructured occurrence records to monitor progress towards Aichi targets 12, 14 and 15. The next update of these indicators will be released with the 2015 UK biodiversity indicators set at 09:30 on the 17th December 2015 on the JNCC website.
DEFRA. (2014). UK Biodiversity Indicators 2014: Measuring progress towards halting biodiversity loss. Retrieved from http://jncc.defra.gov.uk/page-4229
Isaac, N.J.B., van Strien, A.J., August, T. a, de Zeeuw, M.P. & Roy, D.B. (2014). Statistics for citizen science: extracting signals of change from noisy ecological data (B. Anderson, Ed.). Methods in Ecology and Evolution, 5, 1052–1060. Retrieved August 27, 2014, from http://doi.wiley.com/10.1111/2041-210X.12254
Powney, G.D., Cham, S.S.A., Smallshire, D. & Isaac, N.J.B. (2015). Trait correlates of distribution trends in the Odonata of Britain and Ireland. PeerJ, 3, e1410. Retrieved November 19, 2015, from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=4655099&tool=pmcentrez&rendertype=abstract