Monday, May 23, 2016
15:30 - 17:00

<< Back to Schedule

15:30 - 15:45: / 313 VARIABLE SELECTION WITH RANDOM FOREST: BALANCING STABILITY, PERFORMANCE, AND INTERPRETATION IN ECOLOGICAL AND ENVIRONMENTAL MODELING

5/23/2016  |   15:30 - 15:45   |  313

VARIABLE SELECTION WITH RANDOM FOREST: BALANCING STABILITY, PERFORMANCE, AND INTERPRETATION IN ECOLOGICAL AND ENVIRONMENTAL MODELING Random forest (RF) is popular in ecological and environmental modeling, in part, because of its insensitivity to correlated predictors and resistance to overfitting. Although variable selection has been proposed to improve both performance and interpretation of RF models, it is uncertain how selection affects final predictions. We used forward and backward variable selection on 212 landscape predictors from the EPA’s StreamCat Dataset (anthropogenic and natural metrics) to produce four models of benthic condition (good vs. poor condition as a binary response). Variable selection produced models with 10-15 predictors, and evaluations suggested excellent performances (e.g., AUC = 0.82-0.86, 78% correct classification). Selection improved AUC values by up to 5 points compared to the 212-predictor model. Despite similar performances, correspondence of predicted probabilities among models varied greatly (r2 = 0.55-0.75) and produced markedly different maps. Moreover, removal or addition of predictors to these reduced-set models substantially altered predicted values. This instability reduced both confidence and interpretability in candidate models; therefore, we suggest the use of all predictors or a set of uncorrelated predictors in RF modeling with minimal selection.

Ryan Hill (Primary Presenter/Author), US Environmental Protection Agency, hill.ryan@epa.gov;
Ryan Hill is an aquatic ecologist with the U.S. EPA Office of Research and Development. He is interested in how watershed conditions drive differences in freshwater diversity and water quality across the United States. He has worked extensively with federal physical, chemical, and biological datasets to gain insights into the factors affecting water quality and biotic condition of freshwaters across the conterminous US. He has also worked to develop and distribute large datasets of geospatial watershed metrics for streams and lakes for the Agency (EPA’s StreamCat and LakeCat datasets). Ryan completed his PhD in Watershed Ecology at Utah State University in 2013 with Dr. Chuck Hawkins. He was an ORISE postdoctoral fellowship at the U.S. EPA from 2014-2019 before joining the EPA in 2019.

Eric Fox ( Co-Presenter/Co-Author), US EPA, Western Ecology Division, Corvallis, OR, fox.ericw@epa.gov;


Scott Leibowitz ( Co-Presenter/Co-Author), US EPA, Pacific Ecological Systems Division, Corvallis, OR, leibowitz.scott@epa.gov;


Anthony Olsen ( Co-Presenter/Co-Author), EPA, olsen.tony@epa.gov;


Darren Thornbrugh ( Co-Presenter/Co-Author), ORISE c/o US EPA, Western Ecology Division, Corvallis, OR , thornbrugh.darren@epa.gov;


Marc Weber ( Co-Presenter/Co-Author), US EPA, Pacific Ecological Systems Division, Corvallis, OR, weber.marc@epa.gov;


Presentation:
This presentation has not yet been uploaded.

15:45 - 16:00: / 313 INFORMATION THEORY AND THE ASSESSMENT OF HYDROLOGICAL CHANGE IN FRESHWATER ECOSYSTEMS

5/23/2016  |   15:45 - 16:00   |  313

INFORMATION THEORY AND THE ASSESSMENT OF HYDROLOGICAL CHANGE IN FRESHWATER ECOSYSTEMS Hydrologic variability is a fundamental component of the hydrological regime and can be altered by climate change and land use practices in watersheds. However, the available metrics of hydrologic variability are difficult to correlate with changes in forest cover that could occur over short time frames (~1 year). Here, we explore information theory as a framework to understand changes in hydrologic variability as a watershed’s response to disturbances (e.g. floods, droughts, debris flows, and clear-cuts). We analyzed discharge time series (1968-2013) from H. J. Andrews watersheds 9 (Reference) and 10 (100% logged in 1975). Contrary to our expectations, the reference watershed seems hydrologically more sensitive and less stable to a varied drought regime compared to the treatment. Future research will explore the implications of these changes in hydrologic variability on water quality (e.g. nutrient and sediment transport). Information theory allows the study of changes in variability as an ecosystem response. We consider that this approach would provide fundamental insight about the complex behavior of aquatic ecosystems.

Francisco Guerrero (Primary Presenter/Author), Oregon State University, francisco.guerrero@oregonstate.edu;


Jeff Hatten ( Co-Presenter/Co-Author), Oregon State University, jeff.hatten@oregonstate.edu;


Victor Penaranda ( Co-Presenter/Co-Author), Universidad Nacional-Medellin, hidroingvmpv@gmail.com;


Presentation:
This presentation has not yet been uploaded.

16:00 - 16:15: / 313 PROGNOSIS, DIAGNOSIS AND TESTING OF HYDROECOLOGICAL INDICATORS IN A LONG-TERM MONITORING OF ALPINE RIVERS

5/23/2016  |   16:00 - 16:15   |  313

PROGNOSIS, DIAGNOSIS AND TESTING OF HYDROECOLOGICAL INDICATORS IN A LONG-TERM MONITORING OF ALPINE RIVERS Alpine freshwaters are particularly sensitive to climate change because hydroecological processes respond to even small changes in climate and alter ecosystem properties and function. For a long-term monitoring program in the Hohe Tauern National Park in the Austrian Alps we have been studying environmental conditions, nutrients and benthic invertebrates in glacier-fed and spring-/groundwater-fed stream segments in four glaciated catchment. Besides the recording and measuring of hydroecological conditions an evaluation of quality and adequacy of specific indicators is a necessity in monitoring programs. The main aim of this study therefore was, i) from available abiotic and biotic data to model a prognosis of benthic macroinvertebrate assemblages, their abundances and diversities, ii) to elaborate new data in the monitoring sites, and iii) to compare results from i) and ii) in order to find out relevant and adequate tools and indicators for climate change effects in alpine river ecosystems. These comparisons showed that most measures fell into the amplitudes of proposed values and therefore proofed that defined and implemented measures in our alpine river monitoring program were relevant.

Leopold Füreder (Primary Presenter/Author), University of Innsbruck, Austria, leopold.fuereder@uibk.ac.at;


Georg Niedrist ( Co-Presenter/Co-Author), University of Innsbruck, Austria, g.niedrist@student.uibk.ac.at;


Stefan Schütz ( Co-Presenter/Co-Author), University of Innsbruck, Austria, stefan.schuetz@student.uibk.ac.at;


Presentation:
This presentation has not yet been uploaded.

16:15 - 16:30: / 313 COMBINING POINT OCCURRENCES AND EXPERT RANGE INFORMATION TO MODEL FINE-GRAIN DIVERSITY PATTERNS OF NORTH AMERICAN FRESHWATER FISHES

5/23/2016  |   16:15 - 16:30   |  313

COMBINING POINT OCCURRENCES AND EXPERT RANGE INFORMATION TO MODEL FINE-GRAIN DIVERSITY PATTERNS OF NORTH AMERICAN FRESHWATER FISHES The North American freshwater fish fauna represents a unique and species-rich assemblage, whose coarse grain distribution and diversity patterns are well known. However, detailed knowledge about the fine-grain distributions and species~environment relationships are less well understood. Fish occurrence information consists in general of heterogeneous data, ranging from point records that provide accurate information in geographic and environmental space, to coarse expert range maps accounting for dispersal barriers or historical biogeographic limits. We combined both data types in a species distribution model (SDM) framework using newly-developed freshwater-specific environmental variables to make fine-grain (1 km) estimates of fish distributions across North America. The predictions highlight diversity patterns and hotspots along the stream network, further contributing to the understanding of the current-day environmental factors that shape the distribution of North American freshwater fish ranges, with the potential in ultimately aiding conservation and management efforts.

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


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


Presentation:
This presentation has not yet been uploaded.

16:30 - 16:45: / 313 EVIDENCE FOR NEGATIVE EFFECTS OF DROUGHT ON BAETIS SP. (SMALL MINNOW MAYFLY) ABUNDANCE IN A SOUTHERN CALIFORNIA STREAM

5/23/2016  |   16:30 - 16:45   |  313

EVIDENCE FOR NEGATIVE EFFECTS OF DROUGHT ON BAETIS SP. (SMALL MINNOW MAYFLY) ABUNDANCE IN A SOUTHERN CALIFORNIA STREAM Benthic macroinvertebrate (BMI) sampling was conducted at two sites in Topanga Creek from 2003-2014. During this period, Southern California experienced extreme drought conditions (US Drought Monitor 2014). Examining trends in species composition over this period allows for a relatively long-term analysis of potential effects of drought on BMI communities. The following trends were observed during the course of this study: 1) Wet year rainfall positively correlated to springtime abundance of Baetis sp., relative abundance of Simulium sp. up to 31” rain, and negatively correlated to relative abundance of Chironomidae n.d., 2) percent algae cover in April and May positively correlated to abundance per sq. ft. Baetis sp. and Simulium sp., and 3) multiple regression analysis revealed a negative relationship between Chironomid n.d. and Baetis sp. abundance. BMI are an important food source for endangered steelhead trout and other native aquatic and terrestrial insectivorous species of special concern; significant changes to the BMI community could have trophic repercussions for these and other wildlife. Long-term monitoring is important for tracking the influence of drought or climatic changes on BMI community.

Elizabeth Montgomery (Primary Presenter/Author), Resource Conservation District Santa Monica Mountains / Michigan Technological University, montgomerylizzy@gmail.com;


Presentation:
This presentation has not yet been uploaded.