UNDERTAKING THE LARGEST-EVER ASSESSMENT OF THE UK’S WILDLIFE

Tom August, Nick Isaac, Gary Powney & Charlie Outhwaite

Wild bird populations in the UK, 1970-2015 . 2017. DEFRA

How do we know these things?

• Long term surveys of animal and plant populations
• Undertaken by professional scientists
• Undertaken by volunteer naturalists

Countryside survey:

Surveys from 2007, 1998, 1990, 1984 and 1978.

How do we know these things?

Professional surveys

Carey, P.D.; Wallis, S.; Chamberlain, P.M.; Cooper, A.; Emmett, B.A.; Maskell, L.C.; McCann, T.; Murphy, J.; Norton, L.R.; Reynolds, B.; Scott, W.A.; Simpson, I.C.; Smart, S.M.; Ullyett, J.M.. 2008 Countryside Survey: UK Results from 2007. NERC/Centre for Ecology & Hydrology

UK Ladybird survey receives 20,000 records a year from unpaid members of the public

How do we know these things?

Volunteer surveys

Who is the BRC?�Biological Records Centre

The national focus for terrestrial and freshwater recording

​

Biases in ad-hoc data

Photo: Richard Comont

Time

Space

Detectability

Effort

Accounting for bias

Isaac et al (2014) Meth Ecol Evol 5: 1052-1060

​

Occupancy models

• Account for variation in recording effort and detectability over time
• Account for bias in space and time

What is the computational issue

Analyses were previously run on our in-house cluster for ~4,000 species.

We parallelised across species

1 CPU

Input

Output

X 4000

What is the computational issue

Analyses were expanded to 12,000 species some with ~2 week runtimes

1 CPU

Input

Output

X 12,000

What is the computational issue

Analyses were expanded to 12,000 species some with ~2 week runtimes

1 CPU

20,000 iterations

1 chains

Input

Output

X 36,000

1 CPU

20,000 iterations

1 chains

Input

Output

1 CPU

20,000 iterations

1 chains

Input

Output

What is the computational issue

Analyses were expanded to 12,000 species some with ~2 week runtimes

X 720,000

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

Jasmin solution

X 720,000

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

1 CPU

1,000 its

1 chain

1 CPU

1,000 its

1 chain

x20

Input

Output

X 12,000

Challenges?

• Creating thousands of jobs to run required extra scripting
• Managing thousands of simultaneous jobs was a headache.
• Tracking jobs that failed and restarting them required extra code
• Handling many more output files added to overheads

Rewards

• Trends for thousands of species including groups which could not have been run in-house (e.g. moths)
• Access, at times, to over 2000 cores simultaneously
• A workflow that will allow re-analysis in the future
• An opportunity to learn, and train others within the organisation
• An analysis unprecedented in complexity and scope within the UK

Results redacted, sorry!

Next steps

• The UK Biodiversity indicators will be published this summer
• What is driving the changes we see?
• How do trends vary across the UK and between taxonomic and functional groups?
• How do we best share the results for both scientists and recorders?

​

Conclusions

• Wildlife is changing both globally and nationally
• Citizen science data is a valuable source of information
• Cutting edge models require large HPC facilities
• Results give evidence to support policy decisions and report on successes and failures

With thanks to…

Charlie Outhwaite

@CharlieLouO

PhD Student

Gary Powney

@GaryPowney

Quantitative Ecologist

Nick Isaac

@drNickIsaac

Macroecologist

Thousands of volunteer records who contribute there time for free

​

Recording schemes and societies who support recorders and generously give their data and expertise

​

The JASMIN team, especially Fatima Chami

Biases in ad-hoc data

Biases in ad-hoc data

Top Grasshopper recorders

Isaac & Pocock (2015) Biol J Linn Soc 115: 522-531

Contribution to citizen science projects often follows the 80:20 rule where approximately 80% of the data comes from only 20% of the recorders