NOVEL METHODS IMPROVE PREDICTION OF SPECIES’ DISTRIBUTIONS FROM OCCURRENCE DATA

Número de documento:

WS-VME-09/P01

Autor(es):

Anderson, R.P., Ferrier, S., Moritz, C., Dudík, M., Loiselle, B.A., McC. Overton, J., Nakamura, M., Lohmann, L.G., Hijmans, R.J., Manion, G., Scachetti-Pereira, R., Elith, J., Lehmann, A., Guisan, A., Williams, S., Peterson, A.T., Nakazawa, Y., Wisz, M.S.

Punto(s) de la agenda

WS-VME-09 - - undefined

Resumen

Predictionofspecies’distributionsiscentraltodiverseapplicationsinecology,evolutionandconservationscience.Thereisincreasingelectronicaccesstovastsetsofoccurrencerecordsinmuseumsandherbaria,yetlittleeffectiveguidanceonhowbesttousethisinformationinthecontextofnumerousapproachesformodellingdistributions.Tomeetthisneed,wecompared16modellingmethodsover226speciesfrom6regionsoftheworld,creatingthemostcomprehensivesetofmodelcomparisonstodate.Weusedpresence-onlydatatofitmodels,andindependentpresence-absencedatatoevaluatethepredictions.Alongwithwell-establishedmodellingmethodssuchasgeneralisedadditivemodelsandGARPandBIOCLIM,weexploredmethodsthateitherhavebeendevelopedrecentlyorhaverarelybeenappliedtomodellingspecies’distributions.Theseincludemachine-learningmethodsandcommunitymodels,bothofwhichhavefeaturesthatmaymakethemparticularlywellsuitedtonoisyorsparseinformation,asistypicalofspecies’occurrencedata.Presence-onlydatawereeffectiveformodellingspecies’distributionsformanyspeciesandregions.Thenovelmethodsconsistentlyoutperformedmoreestablishedmethods.Theresultsofouranalysisarepromisingfortheuseofdatafrommuseumsandherbaria,especiallyasmethodssuitedtothenoiseinherentinsuchdataimprove.