How to map species distribution using citizen science data when there is observer bias.

1. Dealing with observer bias when mapping species
distribution using citizen science data; an example
on the distribution of brown bears in Greece.
Anne-Sophie Bonnet-Lebrun, Alexandros A. Karamanlidis,
Miguel de Gabriel Hernando, Olivier Gimenez

2. INTRODUCTION – Citizen science
New technologies
Importance of mapping species distributions:
 Citizen science!
- Define priority areas for conservation
- Map problematic interactions
- Spatial information
- Increasingly connected world

3. INTRODUCTION – Citizen science
- Time and money
Citizen-science: pros and cons
BUT
- Quality
- Quantity
- Presence-only
- Observer bias
Sampling effort not evenly distributed
Impossible to evaluate detectability
- Large spatial cover
Greece

4. Threats:
- Habitat loss and fragmentation
- Human-bear conflicts
Map its distribution in Greece
Inform conservation strategies
INTRODUCTION – Monitoring brown bears
Conservation status:
- Globally: Least Concern (IUCN Red List status)
- Locally in Europe: small and isolated populations

5. Brown bears, like other large
carnivores, are difficult to monitor:
 Citizen science!
INTRODUCTION – Using citizen science to monitor brown bears
- Cryptic and solitary
- Low density in very large areas

6. INTRODUCTION – Species Distribution Models
+
Partial information on the
species’ presence
Environmental variables
Probabilities of presence in
the whole area of interest
Traditional methods to infer species distributions:

7. METHODS – Dealing with citizen-science data
-presence-only data
 inhomogeneous Poisson point process
(Warton & Shepherd 2010)
Homogeneous
Intensity = constant
- Intensity: average number of points per unit area
- Poisson point process: random process to generate
points scattered in space
Inhomogeneous
Intensity = f(spatial variable)

8. -opportunistic data
 Model observer bias (Warton et al. 2013)
METHODS – Dealing with citizen-science data
Make the difference between:
• ecological (forest cover, altitude, …) variables
• observer bias (distance to the roads, …) variables
- Affect the species’ presence
- Used for building AND projecting the model
- Affect the probability to detect the species
- Used only for building the model
(projection with a common level of bias)

9. Maps of estimated intensity (in presence points per square kilometre)
of Eucalyptus apiculata from three different models.
Ecological
variables only
Ecological +
observer bias
variables
Ecological + observer
bias variables,
conditioning on a
common level of bias
METHODS – Dealing with citizen-science data
Warton et al. 2013

10. METHODS – Environmental variables
Variables used in the model:
- Mean slope
- Altitude
- Density of rivers
- % Agricultural land
- % Forests
- Human population density
- Distance to roads
Ecological
Observer bias

11. RESULTS
Average expected number
of observations
Model based on
opportunistic data
Model based on
presence-absence data
Probabilities of presence
The results of the two
models seem coherent

12. RESULTS – Modelling observer bias

13. DISCUSSION
Distance to the roads

14. DISCUSSION
Ecological vs. observer bias variables:
the example of human population density
- : ecological variable
+ : observer bias variable
The more people, the
more likely they are to
detect a bear
Bears are likely to
avoid areas with a
lot of people

15. DISCUSSION
- Model opportunistic data with Poisson Point Processes?
- Deal with presence-only data
- Possibility to combine different sources of data
(Dorazio 2012, O’Hara 2014)
- Model observer bias?
- Ecological vs. observer bias variables
- Difficulty to find a relevant observer bias variable
- Really reflecting the spatial observer bias process
Model the citizen’s behaviour

16. Thank you for your attention!