Effects of the number of presences on reliability and stability of MARS species distribution models: the importance of regional niche variation and ecological heterogeneity

摘要

Question: What are the effects of the number of\udpresences on models generated with multivariate\udadaptive regression splines (MARS)? Do these effects\udvary with data quality and quantity and species\udecology?\udLocation: Spain and Ecuador.\udMethods: We used two data sets: (1) two trees from\udSpain, representing high-occurrence number data\udsets with real absences and unbalanced prevalence;\ud(2) two herbs from Ecuador, representing lowoccurrence\udnumber data sets without real absences\udand balanced prevalence. For model quality, we\udused two different measures: reliability and stability.\udFor each sample size, different replicates were\udgenerated at random and then used to generate a\udconsensus model.\udResults: Model reliability and stability decrease with\udsample size. Optimal minimum sample size varies\uddepending on many factors, many of which are\udunknown. Regional niche variation and ecological\udheterogeneity are critical.\udConclusions: (1) Model predictive power improves\udgreatly with more than 18-20 presences. (2) Model\udreliability depends on data quantity and quality as\udwell as species ecological characteristics. (3) Depending\udon the number of presences in the data set,\udinvestigators must carefully distinguish between\udmodels that should be treated with skepticism and\udthose whose predictions can be applied with reasonable\udconfidence. (4) For species combining few\udinitial presences and wide environmental range variation,\udit is advisable to generate several replicate\udmodels that partition the initial data and generate a consensus model. (5) Models of species with a\udnarrow environmental range variation can be highly\udstable and reliable, even when generated with few\udpresences.