论文信息 - Species Recommendation using Intensity Models and Sampling Bias Correction (GeoLifeCLEF 2019: Lof_of_Lof team)

Species Recommendation using Intensity Models and Sampling Bias Correction (GeoLifeCLEF 2019: Lof_of_Lof team)

This paper presents three algorithms for species spatial rec-ommendation in the context of the GeoLifeCLEF 2019 challenge. Wesubmitted three runs to this task, all based on the estimation of speciesenvironmental intensities through Poisson processes models: The first isdirectly derived from MAXENT method used for species distributionmodels. The second method is a modification that uses sites were speciesobserved as background points in MAXENT to correct for spatial sam-pling bias due to heterogeneous sampling in the training occurrences. Thelast method jointly estimates species and sampling intensities to correctfor sampling bias. The best run was the MAXENT method which wasranked 14 over 44 runs with a top30 accuracy of 0.111 on the test setwhile the worst performing method was LOF with an accuracy of 0.086(ranked 19)

Christophe Botella | Pascal Monestiez

[1] Pierre Bonnet,et al. Overview of GeoLifeCLEF 2019: Plant Species Prediction using Environment and Animal Occurrences , 2019, CLEF.

[2] Antoine Guisan,et al. What we use is not what we know: environmental predictors in plant distribution models , 2016 .

[3] Steven J. Phillips,et al. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. , 2009, Ecological applications : a publication of the Ecological Society of America.

[4] Jürgen Symanzik,et al. Statistical Analysis of Spatial Point Patterns , 2005, Technometrics.

[5] Steven J. Phillips,et al. Point process models for presence‐only analysis , 2015 .

[6] Robert P. Anderson,et al. Opening the black box: an open-source release of Maxent , 2017 .

[7] J. Elith,et al. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time , 2009 .