Use of Maxent for predictive habitat mapping of
Use of Maxent for predictive habitat mapping of CWC in the Bari canyon Bargain Annaëlle Foglini Federica, Bonaldo Davide, Pairaud Ivane & Fabri Marie-Claire Ifremer Mediterranée & CNR Bologna (ISMAR)
BARI CANYON 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 2
Modelisation Processings Relate species occurrence data (distribution = biological data) with environmental predictor variables (EGVs = Ecogeographic variables) Explain the contribution of each environmental variable to the species distribution Produce continuous maps of potential species or habitat
Steps of the CWC predictive habitat modelisation 1. Data acquisition • Méthode générale Species records Choose of statistical Methods Environmental data 2. Model settings Evaluate Variable contribution Model assessment Is the model better than random model ? 3. Model applications Predictive habitat maps Sites comparisons Marine planification tool
Maxent software • Software http: //www. cs. princeton. edu/~schapire/maxe nt/ • by Steven Phillips, Miro Dudik and Rob Schapire, with support from AT&T Labs. Research, Princeton University, and the Center for Biodiversity and Conservation, American Museum of Natural History 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 5
Opening the software 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 6
Opening the software To enter the species data To enter the Ecogeographic variables Parameters of the model 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 7
Data acquisition : CWC presence points Use of ADELIE-SIG (Ifremer) on video transects and digital images CWC presence points datapoints have been reduced to one presence points per cell of EGVs resolution (BARI = 20*20 m)
Species files in Maxent software • Files need to be in. csv • Export the table from arcgis into. txt, open in Excel, and change the file in this form : Species/long/lat • The file must be then save in. csv • Open the. csv file with wordpad and change « ; » with « , » • Open the file in maxent 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 9
Data acquisition : EGVs 1. Use of the high resolution bathymetry data With arcgis, and Benthic Terrain Modeler module (http: //www. Arc. GIS. com/home/item. html? id=b 0 d 0 be 66 fd 33440 d 97 e 8 c 83 d 220 e 7926) • limit the bathymetric data to the canyon (bathymetry under -180 m) • compute benthic indices at different resolution Terrain roughness Topography Rugeness (3, 5, 11 m resolution) Surface Area to Planar Area Slope Orientation Eastness Northness Bathymetric Position Index (3, 9, 17, 25, 33, 65) Curvature Profile Plan
Data acquisition : EGVs 1. Use of hydrodynamic data at 1 km resolution § Data from 1 rst of november to 28 th june § Climatic event from 25 th Jan. To 14 th Feb. • interpolate the measure points to the bathymetric resolution (20 m) with natural neighbor. Temperature Max Mean SD Water Density Max Mean SD Water Salinity Current speed Max Mean SD
EGVs study • Variables need to be non-correlated in a model • A statistical analysis is thus necessary to choose the best non-correlated ones before using the model (PCA, dendrogram…) • The model is also better if the sample points incompass all the variable value range -> Need to compare the EVGs values at each points to the global EVGs values 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 12
EVGs finally selected for CWC in the Bari canyon • Slope • Ruggeness at 11 pixels resolution • Bathymetric Position Index at 25 pixels resolution • Bathymetric Position Index at 65 pixels resolution • Northness • Eastness • Profile curvature • Transversal curvature 25/11/2015 • Mean water current speed • Mean water salinity • Mean water density • Maximum water temperature Bargain Annaelle – Modélisation des coraux d’eau froide 13
EGV files in Maxent software • Files need to be in. asc • Export the table from arcgis into. asc • Open the directory layer containing all the files with Maxent • Then, choose the EVGs that will be use in the model, selecting the good type of variable (continuous or categorical) 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 14
Software settings Feature types to be use during training. « Hinge » only has been selected, which is according to Philipps & Dudik (2008) a good approximation of the global distribution of the species Settings has many options for Maxent model In « basic option » , the regularization multiplier has been changed to « 3 » , to avoid over-fitting 10 replicates, using crossvalidate has been selected to test the model The other parameters has been left 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide In « Advanced option » , the default prevalence has been changed to « 0. 7 » , as the dive to study CWC were not totaly random The other parameters has been left 15
Software settings This options have been selected to do jackniffe tests and study each variable importance for CWC distribution The logistic format has been left, to have a probability of suitability for each pixel, from 0 to 100 %. Define here the output directory And then RUN the model 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 16
MAXENT outputs • One html file, that summaries all results • All replicates html files • Many tables and graphs 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 17
MAXENT outputs • ROC curves (and AUC values) • Maps • CWC distribution response to variable variations • Variable importance in the final model table • Jackniffe tests 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 18
MAXENT outputs for CWC distribution in the Bari Canyon The ROC curve is very high, with an AUC value close to 1 (0. 99) The results of the habitat mapping show two main areas for CWC distribution in the BARI canyon (left map), with no big differences between replicates (SD very low, map on the right) 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 19
MAXENT outputs for CWC distribution in the Bari Canyon The importance of each variable in the final model shows that Slope, rugosity and water current speed are the main variable to explain CWC distribution Jackniffe tests on training gain had the same conclusions, including also the mean water density in the main variables The absence of the salinity in the model also decrease the most the final gain, showing that this variable is essential in CWC distribution 25/11/2015 Bargain Annaelle – Modélisation des coraux d’eau froide 20
Conclusions 25/11/2015 Model High probability area Threshold Maxent 3. 9 km² 0. 6 Bargain Annaelle – Modélisation des coraux d’eau froide 21
- Slides: 21