Autonomous Monitoring of Vulnerable Habitats And other tales

Autonomous Monitoring of Vulnerable Habitats And other tales. . Robin Freeman, CEES, Microsoft Research 13 July 2007

Overview • Introduction • Previous Work – Analysing Avian Navigation • Habitat Monitoring • Brief Results • Future Work

Introduction – About Me • BSc CS-AI, MSc Evolutionary and Adaptive Systems, • D. Phil (Engineering and Zoology) – Part of the Life Sciences Interface Doctoral Training Centre, Oxford – Trains physical and computation sciences graduates in biology before starting Ph. D in life sciences. • Now a Post-Doc at Microsoft Research – Computational Ecology and Biodiversity Science Group – European Science Initiative, External Research Office. 3

~9 hrs ~15 min

Introduction • Analysing Avian Navigation • GPS Tracking of Pigeons, Oxford • GPS Tracking of Manx Shearwaters, Skomer • Habitat Monitoring • Manx Shearwater – Skomer Island, Wales 6

Introduction – Zoological Interest • Specific questions (Sensory basis of navigation), • Conservation (home range, behavioural anomalies), • Other general questions. – Technical Interest • Novel algorithms/methods – Analysis of positional information – Feedback to bio-robotics, Complex Systems, Artificial Life, etc 7

Pigeons? - Why Pigeons? • Model Navigational Species – Much easier to study than wild birds, • Birds return to a maintained loft (Wytham). – Allows attachment of GPS device – Large body of research to draw on. • Pigeon navigation has been studied for over 100 years.

How Do They Navigate? • Two hypotheses for the sensory basis of navigation in the familiar area – ‘Map and Compass’ • Compass controlled navigation (as it is at unfamiliar locations). – Series of decision points using compass. – ‘Pilotage’ • Independent of a compass, relying directly on visual cues – Oh look, there’s that house!

Clock Shift • Experiment – Train the birds to ‘recapitulate’ routes to home, – Then ‘clock-shift’ the birds by 90° • Sets up a direct competition between visual landmarks (the recapitulated route) and erroneous compass instructions With D Biro, J Meade, T Guilford & S J Roberts

• Nearest Neighbour Analysis • Shows offset and variance between controls and familiar clock-shift.

Tracks ranked by Mahalonobis distance from recapping distribution Delayed Clock shift response (landmark related)

• Demonstrates that both mechanisms must be involved. – The birds must be able to home using visual information alone (they recapitulate) – Consistent deviation from recapitulated path • Offset? Zigzag? Biro D, Freeman R, Meade J, S. Roberts, Guilford T. (2007) PNAS. 104(18)

Behavioural Segmentation - Hidden-Markov Models - Positional Entropy 14

Landscape Analysis • More likely to fly over edge ‘rich’ areas • Flight pattern becomes less predictable over edge rich areas. Lau KK, Roberts S, Biro D, Freeman R, Meade J, Guilford T. (2006) J. Theo. Bio. 239(1) pp 71 -78

Paired Homing Pigeon Flight n GPS data for 48 Pigeons from 4 diff. sites n All possible pairs considered n Any real interaction between the birds should be seen as higher coupling between real pairs n Other pairs may show n High coupling due to same landscape/other unknown variables Actual pair Bird paired with self Bird & random bird from different site

Birds which flew together show significantly (p < 0. 05) higher coupling than other possible pairings. Implies some form of information transfer.

Manx Shearwater (Puffinus puffinus) • Highly pelagic, migratory seabird. • Burrow dwelling, central place forager. • UK summer breeding • Winters in South America • 250, 000 – 300, 000 breeding pairs. • 45% on three Pembrokeshire islands, Skomer, Skokholm and Middleholm; • 36% on Rum. 18

Motivation • Ecology and Behaviour very similar to other Procellariiformes – Albatrosses, Petrels and Shearwaters. • 19 of 21 Albatross Species now globally threatened; • Devastating impact of long-line fishing • Understanding their behaviour, habitat and ecology may allow us to reduce this decline. 19

Motivation UK Seabird decline over recent years Source: JNCC, UK Seabirds 2005 20

21

Skomer Island • Small Island (~2 km long) off coast of Wales • Home to large populations of Guillemots, Razorbills, Kittiwakes, Puffins, Fulmars • Worlds largest population of Manx Shearwaters • Well established research centre and study programmes 22

Skomer Island 23

Previous Work • GPS Tracking of Manx Shearwater – Distribution of foraging was largely unknown; • South to Spain; – Interaction • With fisheries? • Environmental variables? – Establishment of Marine protection zones. 24

– Foraging largely confined to Irish Sea; – Birds did not fly far south. . • Even when they had the opportunity to do so. • Climate effect? – Clustered areas; – Rafting. Right: Distribution of individual over trips of 1 to 7 days. Red shows incubating birds, blue chick rearing 25

– Each 2 -hourly fix gives a small burst of 1 Hz data. – Bursts can be segmented into different behaviours. – Speed Vs Directionality

Sitting & Erratic Movement Directional Movement 27

– Speed has no obvious effect on depth – Time of day appears to (right) 28

Autonomous Habitat Monitoring • Working closely with Academic Partners – University of Oxford • Prof. Tim Guilford, Animal Behaviour • Prof. Chris Perrins, Edward Grey Ornithology Institute – University of Freie Berlin • Tomasz Naumowicz, PHD, Free University Berlin • Prof Torben Weis, U Duisburg-Essen 29

Autonomous Habitat Monitoring • Create and deploy a wireless sensor network that can: – Monitor the visitations of individual birds; – Monitor environmental conditions inside and outside the burrow; – Provide a pilot system for eventual integration with GPS tracking; – Do this all night, every night… 30

Methods • Approx. 10 Burrow monitored – Ringed and RFID tagged pair of birds in each burrow; – Sensors & wireless sensor node to each burrow; 31

Methods • Network – Scatter. Web platform from Freie Universitat Berlin; • Nodes – 2 x Passive Infrared – 2 x Temp/Humidity – RFID Detector 32

Initial Results • No observable impact on birds’ behaviour – No evidence of digging, distress or abandonment. • Of 10 monitored burrows – 7 hatched (last week) – Remainder still on eggs 33

Initial Results – Obvious nocturnal distribution of activity • Bimodal? – Resolution and density of data already significantly higher that achievable using traditional methods. All recorded events 34

2007/05/14 12: 00 2007/05/15 00: 00 2007/05/15 12: 00 35

Initial Results 36

00: 00 Temperature Variation over 4 days (20 -23 June) • 06: 00 18: 00 • Red: Temp Outside Green: Temp Inside 12: 00 37

Future Questions… – Do individuals return at specific times? – How do pairs alternate feeding strategies? – How does activity/environment vary across space and time? – How do the results vary with weather? 38

Future Directions • Deploy second network – Pilot has allowed us to iron out most problems; – Hope to set up additional network this winter. • Create a toolkit that any ecologist can deploy and use. • Integrate GPS tracking with network – Continual monitoring of foraging behaviour. 39

~9 hrs ~15 min

An Aside (1) 41

An Aside (2) 42