2015-09-10

a note on using this presentation

  • press 'p' to see a note with a brief explanation for each slide
  • press 'f' to see the presentation in 'full screen' mode

the ideal world

Representative sample

  • systematic / random
  • for species mapping: planned surveys

the real world

convenience sample

  • geographically unrepresentative
  • environmentally unrepresentative
  • selection bias

Selection bias

any harm done?

  • problem: where to select background/pseudo-absences?
  • risk: to model sampling effort, not ecology
  • questions: does it always matter (when)?

Selection bias

how to fix it

  • bias-correction grid
  • sub-sample (spatial filter)
  • restrict background to part of area
  • select in environmental space
  • stratify domain and fit different models

Concrete case study

Kramer-Schadt et al. (2013)

  • Malay civet (2 species), Borneo
  • real & simulated data
  • no correction, correction grid (1) & sub-sample (2)

Conclusions on bias correction

Kramer-Schadt et al. 2013

  • effects of bias-correction are large

  • spatial filtering works best, followed by correction grid

Another case

Syfert, Smith, and Coomes (2013)

  • Tree ferns, New Zealand
  • real data (2 data sets!)
  • external data for validation
  • correction grid

  • orange - background on basis of tree ferns
  • blue - background on basis of vascular plants
  • black - all NZ locatoins (1 km resolution)

Conclusions on bias correction

Syfert et al. 2013

  • correction grid works well
  • effects are large
  • take care when evaluting 'bias-corrected models' on 'not bias-corrected data'

Last case

Fourcade et al. (2014)

  • turtle & salamander data (2 species) & one virtual species
  • different bias types, varying in degrees
  • 5 bias correction methods
  • AUC and 3 measures of overlap

Conclusions on bias correction

Fourcade et al. 2014

  • sub-sampling worked best for different species & bias type
  • effects are large but differ per error metric

Overall conclusion

  • so … do correct selection bias with:
    • sub-sample
    • correction grid

Discussion points & questions

  • nothing more practical than a good theory, … but can we build it?
  • will more case studies help?
  • how could ZOÖN help?

References

Fourcade, Yoan, Jan O. Engler, Dennis Rödder, and Jean Secondi. 2014. “Mapping Species Distributions with MAXENT Using a Geographically Biased Sample of Presence Data: A Performance Assessment of Methods for Correcting Sampling Bias.” PLoS ONE 9 (5). Public Library of Science: e97122. doi:10.1371/journal.pone.0097122.

Kramer-Schadt, Stephanie, Jürgen Niedballa, John D. Pilgrim, Boris Schröder, Jana Lindenborn, Vanessa Reinfelder, Milena Stillfried, et al. 2013. “The Importance of Correcting for Sampling Bias in MaxEnt Species Distribution Models.” Diversity and Distributions 19 (11): 1366–79. doi:10.1111/ddi.12096.

Syfert, Mindy M., Matthew J. Smith, and David A. Coomes. 2013. “The Effects of Sampling Bias and Model Complexity on the Predictive Performance of MaxEnt Species Distribution Models.” PLoS ONE 8 (2). Public Library of Science: e55158. doi:10.1371/journal.pone.0055158.