5/18/2015  |   10:30 - 10:45   |  101B

IMPROVED METHODS FOR WEIGHTING SPECIES DISTRIBUTION MODELS TO IMPROVE ENSEMBLE MODEL PREDICTIONS It has become increasingly common to make species distribution predictions and forecasts from ensembles of multiple models. Methods for weighting competing models have lagged, however, and the prevailing approaches are either to weight models equally or to use simple decay functions. We present a general approach to model weighting that more accurately preserves the relative differences in performance of alternative models. It involves (1) creating numerous bootstrapped datasets from the original dataset; (2) running each model on each dataset; and (3) recording the proportion of times each model is selected as “best” for a dataset using a given performance criterion. This proportion is the model weight. We illustrate the approach with a set of species distribution models built from large trout dataset. The R functions to implement the method are freely available and will soon be adapted into a formal package.

Seth Wenger (Primary Presenter/Author), University of Georgia, sethwenger@fastmail.fm;


Nicholas Som (Co-Presenter/Co-Author), US Fish and Wildlife Service, nicholas_som@fws.gov;

5/18/2015  |   10:45 - 11:00   |  101B

DISTRIBUTION PREDICTIONS IN THE GERMAN LTER-SITE RHINE-MAIN-OBSERVATORY: LONG-TERM MONITORING DATA MEET HIGH-RESOLUTION, CATCHMENT-BASED SDMS A long term monitoring dataset from the German long-term ecological research (LTER) site Rhine-Main-Observatory (RMO) was used to set up a species distribution model (SDM) in the Kinzig catchment. 175 taxa of stream macroinvertebrates were modeled and projected on the stream network at high resolution using bioclimatic, topographical, hydrological, land use and geological predictors. On average model performance was good, with a TSS of 0.83 (±0.09 SD) and. a ROC of 0.95 (± 0.03). The extensive knowledge on the monitoring data provided by the LTER-site framework delivers valuable insights on three possible sources of bias affecting SDMs in general: (a) the level of taxonomic identification of the modeled organisms, (b) the spatial arrangement of sampling sites, and (c) the sampling intensity at each sampling site. Results based on distribution predictions indicate that, for the RMO-LTER, occurrence data shows both spatial and temporal bias, while taxonomic identification does not affect model performance.

Mathias Kuemmerlen (Primary Presenter/Author), Senckenberg Research Institute, mkuemmerlen@senckenberg.de;


Stefan Stoll (Co-Presenter/Co-Author), Senckenberg Research Institute, sstoll@senckenberg.de;


Andrea Sundermann (Co-Presenter/Co-Author), Senckenberg Research Institute and Natural History Museum, asundermann@senckenberg.de;


Peter Haase (Co-Presenter/Co-Author), Senckenberg Research Institute and Natural History Museum, peter.haase@senckenberg.de;

5/18/2015  |   11:00 - 11:15   |  101B

NEW NEAR-GLOBAL 1 KM SPATIALLY CONTINUOUS FRESHWATER ENVIRONMENTAL VARIABLES FOR BIODIVERSITY ANALYSES AND SPECIES DISTRIBUTION MODELING The lack of spatially continuous freshwater-specific environmental variables hamper comparative biogeographical analyses across large spatial gradients and on a fine spatial grain. We developed a near-global 1 km spatially continuous data set for freshwater environmental variables based on the HydroSHEDS hydrography and by delineating the sub-catchment for each 1 km grid cell along the stream network. We then related continuous global data sets on climate, topography, river topology, land cover and surface geology to each sub-catchment, and summarized each environmental variable using several metrics (average, minimum, maximum, range, sum, inverse distance-weighted average and sum). Further, we extended the variables of the river network to lakes and reservoirs of the Global Lakes and Wetlands Database. Finally, we summarized the monthly climatic variables to 19 long-term hydro-climatic variables following the ‘bioclim’ framework to provide input data for species distribution models. This newly developed set of continuous river network variables provides an improved basis for analyzing and mapping freshwater biodiversity on a near-global extent yet on a fine spatial grain.

Sami Domisch (Primary Presenter/Author), Leibniz Institute of Freshwater Ecology and Inland Fisheries, domisch@igb-berlin.de;


Giuseppe Amatulli (Co-Presenter/Co-Author), Yale University, giuseppe.amatulli@yale.edu;


Walter Jetz (Co-Presenter/Co-Author), Yale University, walter.jetz@yale.edu;

5/18/2015  |   11:15 - 11:30   |  101B

ASSEMBLAGE PREDICTABILITY AND BETA DIVERSITY IN STREAM ECOSYSTEMS: MACRO-SCALE EFFECTS OF ENVIRONMENTAL HETEROGENEITY, ISOLATION, AND PRODUCTIVITY We used macroinvertebrate data collected from 3,265 streams in 68 ecoregions across the USA to assess how both assemblage predictability (precision of multi-taxon niche models) and beta diversity (compositional dissimilarity among sites) varied with regional environmental heterogeneity, isolation (drainage density), and productive capacity (total nitrogen [TN] and total phosphorus [TP]). For reference quality streams, assemblage predictability declined with increasing isolation (drainage density range = 0.1-0.7 km/km²) and TN (asymptote ~300 μg/L, range = 20-500), but was unrelated to environmental heterogeneity. A model that included degraded sites was ~10% less precise than a reference site model. For reference-quality streams, beta diversity (adjusted for alpha diversity, gamma diversity, and region size) increased with TN (asymptote ~300 μg/L), TP (range = 5-50 μg/L), isolation, and environmental heterogeneity. Across all streams, beta diversity increased with both TN (asymptote ~300 μg/L, range = 20-5000) and TP (asymptote ~200 μg/L, range = 5-800) but was only weakly related to isolation and environmental heterogeneity. These results imply that the importance of different processes in community assembly/disassembly vary with both natural environmental setting and human-caused increases in productivity.

Charles Hawkins (Primary Presenter/Author), Utah State University, chuck.hawkins@usu.edu;


Jacob Vander Laan (Co-Presenter/Co-Author), Department of Watershed Sciences, Western Center for Monitoring and Assessment of Freshwater Ecosystems, and the Ecology Center, Utah State University, Logan, UT, USA, jacob.vl@usu.edu;

5/18/2015  |   11:30 - 11:45   |  101B

ASSEMBLAGE PREDICTABILITY AND BETA DIVERSITY IN STREAM ECOSYSTEMS: MACRO-SCALE EFFECTS OF ENVIRONMENTAL HETEROGENEITY, ISOLATION, AND PRODUCTIVITY We used macroinvertebrate data collected from 3,265 streams in 68 ecoregions across the USA to assess how both assemblage predictability (precision of multi-taxon niche models) and beta diversity (compositional dissimilarity among sites) varied with regional environmental heterogeneity, isolation (drainage density), and productive capacity (total nitrogen [TN] and total phosphorus [TP]). For reference quality streams, assemblage predictability declined with increasing isolation (drainage density range = 0.1-0.7 km/km²) and TN (asymptote ~300 μg/L, range = 20-500), but was unrelated to environmental heterogeneity. A model that included degraded sites was ~10% less precise than a reference site model. For reference-quality streams, beta diversity (adjusted for alpha diversity, gamma diversity, and region size) increased with TN (asymptote ~300 μg/L), TP (range = 5-50 μg/L), isolation, and environmental heterogeneity. Across all streams, beta diversity increased with both TN (asymptote ~300 μg/L, range = 20-5000) and TP (asymptote ~200 μg/L, range = 5-800) but was only weakly related to isolation and environmental heterogeneity. These results imply that the importance of different processes in community assembly/disassembly vary with both natural environmental setting and human-caused increases in productivity.

Charles Hawkins (Primary Presenter/Author), Utah State University, chuck.hawkins@usu.edu;


Jacob Vander Laan (Co-Presenter/Co-Author), Department of Watershed Sciences, Western Center for Monitoring and Assessment of Freshwater Ecosystems, and the Ecology Center, Utah State University, Logan, UT, USA, jacob.vl@usu.edu;