Home
Research
People
Papers and presentations
Links
Conference

• Penalized Spline Regression Estimation of Finite Population Totals under Two-Stage Sampling Mark Delorey (Colorado State University) A nonparametric regression estimator for the finite population total in two-stage sampling with complete stage-one auxiliary information is developed. The estimator is model-assisted, incorporating auxiliary information through penalized spline regression, thus extending the single-stage estimator in Breidt, Claeskens, and Opsomer (2003). The estimator is asymptotically design-unbiased and design consistent under mild assumptions and its variance can be consistently estimated. The two-stage model-assisted estimator is contrasted with the estimator in Zheng and Little (2003), which uses penalized spline regression in a model-based approach. A series of simulations demonstrate that the model-assisted estimators generally fare no worse than model-based estimators when the model is correctly specified and generally are superior to model-based estimators when the model is incorrectly specified.
  
  This is joint work with F. Jay Breidt, Colorado State University, and Jean D. Opsomer, Iowa State University.


• A generalized model for in-season forecast of salmon return Saang-Yoon Hyun (Columbia River Inter-Tribal Fish Commission), David Salinger (University of Washington), Stuart Ellis (Columbia River Inter-Tribal Fish Commission) We developed a generalized model for making an in-season forecast of salmon return size when data are available of fish run proportions from the past years, and daily fish run collected during in-season (fish run season). Regardless of setting, managers for anadromous fish populations need to know fish run size in advance to achieve management objectives of ensuring optimum escapement, maximizing harvest, and keeping fish run timing. Considerable year-to-year variability in fish run timing is the main obstacle in making an accurate forecast early during in-season. The objective of this research is to develop a generalized and programmatic forecast model whose output is predicted run size and forecast interval (say, 95% interval). We developed new parametric distributions of run size (say, inverted-beta and location-gamma densities), and optimization forecast model (say, pattern-matching algorithm). Also we explored possible other parametric distributions. Using data of Columbia River chinook salmon runs and Alaska Bristol Bay sockeye salmon runs, we demonstrate and present our results.


• Integration of Spatial Information and Spatial Statistics: Forecasting Wetlands Along the Texas Coastal Area Mohammed A. Kalkhan (Colorado State University), Ricardo D. Lopez Lopez, (USEPA) We present a primly results ongoing study that utilizes a landscape approach to assessing ecological/hydrologic functions and related human values of depressional wetlands along coastal Texas, considered to be vulnerable to human disturbance. Our research approach integrates geospatial information (remotely sensed, GIS, and GPS) data, field data, geospatial statistics, and a priori knowledge of depressional wetlands to estimate their extent; connectivity to other waterbodies and ecosystems; and their ecological/hydrologic functions. This approach focuses on locating and quantifying the cumulative area of depressional wetlands, invasive species, and geospatial model-map the ecological/hydrologic functions and services of wetlands. This research will increase the cooperative exchange of knowledge between governmental agencies [e.g., U.S. Environmental Protection Agency (USEPA), NASA- Invasive Species Program, USGS- National Institute of Invasive Species, and the National Park Service (NPS- Inventory and Monitoring Program (I & M)) - USGS- NPS Vegetation Mapping Program], and educational/ research institutes [e.g., Natural Resource Ecology Laboratory (NREL) at Colorado State University, and others. The study uses remotely sensed data types, such as Landsat ETM+ and other high spectral/spatial resolution sensors to assess wetlands at a landscape scale. Further, this study will be used as the foundation of long-term monitoring of wetlands, specifically focusing on depressional wetlands throughout the USA. These techniques are particularly applicable to the detection and monitoring of invasive species in wetlands, riparian habitats, carbon-nitrogen cycling in the biosphere, forest characteristics, forest health, and forest fuel variability parameters, and they provide an ideal ecosystem type for hypothesis-testing regarding the influence of human activities on these four main topics. This approach is a unique departure from other traditional statistical models of the environment, in that it provides a unique tool that links field data, remote sensing, and geospatial-statistical analyses in a way that could be used in a client-oriented model (e.g., a decision support system and a self-guided web-link system with a windows-based graphic user interface). The tool that will result would facilitate the production of predictive spatial maps, would be accessible by a variety of end-users, and could be used from any location connected to the Internet. Current empirical statistical models may be limited with respect to: (1) providing a clear spatial description of the environmental characteristics of interest; (2) flexibility of use with new and emerging remote sensing data types; and (3) accessibility for individuals who do not possess specialized geostatistical backgrounds/education. Thus, the research team will focus on providing an educational opportunity to new scientists and graduate students who can provide knowledge and perspective on the development of these tools.


• Acid Neutralizing Capacity CDF Estimation In The Northeastern Lakes Survey: A Nonparametric Model Calibrated Pseudo Empirical Maximum Likelihhood Approach. M.Giovanna Ranalli (University of Perugia) The National Surface Water Survey sponsored by the United States Environmental Protection Agency (EPA) between the years of 1984 and 1986 estimated 4.2% of the lakes in the northeastern region of the US to be acidic (Stoddard et al., 2003). These acid-sensitive Northeastern lakes were among the concerns addressed by the Clean Air Act Amendment (CAAA) issued by EPA in 1990, which placed restrictions on industrial sulfur and nitrogen emissions in an effort to reduce the acidity of these waters. Between 1991 and 1996, the Environmental Monitoring and Assessment Program (EMAP) of EPA conducted a survey of lakes in the Northeastern states of the U.S. These data were collected in order to determine the effect that restrictions put in place by the CAAA had on the ecological condition of these waters. The survey is based on a population of 21,026 lakes from which 334 lakes were surveyed, some of which were visited several times during the study period. Estimation of quantities of interest at the population level can be carried on using a model-assisted approach. Models can be built relying on complete auxiliary information in the form of spatially-referenced data maintained in a geographic information system (GIS). Satellite images, in fact, can provide the values of variables thought to influence the process under study for each frame location: land cover, ecosystem typology, elevation can be obtained at little or no extra cost from GIS maps. In this work we explore the possibility of employing nonparametric techniques, and in particular Multi Adaptive Regression Splines (MARS, Friedman, 1991) to build a model-assisted estimator of the distribution function (cdf) of Acid Neutralizing Capacity (ANC) in the northeastern lakes survey. ANC is a common measure of acidity defined as a water’s ability to buffer acid. Here concern is mainly with the assessment of how many lakes are at (high) risk of acidification or are acidified already. The estimator of the cdf and the corresponding confidence intervals are based on an extension of Model Calibrated Pseudo Empirical Maximum Likelihood (MCPEML) estimation proposed in Chen and Wu (2002) and Wu and Rao (2005). Nonparametric model calibration has been introduced in Montanari and Ranalli (2005) and used to estimate totals and means also for environmental populations; although it could be applied as is to cdf estimation, it would have the drawback of possibly taking values outside the interval [0; 1] and of not always being a monotone function of the response variable. Therefore, if on one side nonparametric regression allows a more flexible modeling of ANC with respect to remote sensed auxiliary variables, on the other side MCPEML assures the achievement of a genuine distribution function. Finally, the estimation of confidence intervals through pseudo empirical likelihood has been shown to be superior over normal confidence bounds (Wu and Rao, 2005) especially for cdf estimation. Notes: This is joint work with G.E. Montanari (University of Perugia) and was developed while I was appointed as a post-doc at CSU References: Chen J. and Wu C. (2002) Estimation of distribution function and quantiles using the model-calibrated pseudo empirical likelihood method, Statistica Sinica, 12, 1223-1239. Friedman J.H. (1991) Multivariate adaptive regression splines, with discussion, The Annals of Statistics, 19, 1-67. Montanari G.E. and Ranalli M.G. (2005) Nonparametric model calibration estimation in survey sampling, Journal of the American Statistical Association, in print. Stoddard J.L., Kahl J.S., Deviney F.A., DeWalle D.R., Driscoll C.T., Herlihy A.T., Kellog J.H., Murdoch P.S., Webb J.R. and Webster K.E. (2003) Response of surface water chemistry to the clean air act amendments of 1990, Technical Report EPA/620/R- 93/001, U.S. Environmental Protection Agency. Wu C. and Rao J.N.K. (2005) Pseudo empirical likelihood ratio confidence intervals for complex surveys, Working paper, 2004-06, Department of Statistics and Actuarial Science, University of Waterloo.




• Nonparametric Small Area Estimation Using Penalized Spline Regression, J. D. Opsomer, (Iowa State University), G. Claeskens (Katholieke Universiteit Leuven), M. G. Ranalli (Colorado State University), G. Kauermann (Universit¨at Bielefeld), F. J. Breidt, (Colorado State University)
  
  Key Words: mixed model, best linear unbiased prediction; bootstrap inference, natural resource survey. We propose a new small area estimation approach that combines small area random effects with a smooth, nonparametrically specified trend. By using penalized splines as the representation for the nonparametric trend, it is possible to express the small area estimation problem as a mixed effect model regression. This model is readily fitted using existing model fitting approaches such as restricted maximum likelihood. We develop a corresponding bootstrap approach for model inference and estimation of the small area prediction mean squared error. The applicability of the method is demonstrated on a survey of lakes in the Northeastern US.

• Spatial Statistical Models that Use Flow and Stream Distance, Jay M. Ver Hoef (National Marine Mammal Laboratory), Erin Peterson and David Theobald (Colorado State University) We develop spatial statistical models for stream networks that can estimate relationships between a response variable and other covariates, make predictions at unsampled locations, and predict an average or total for a stream or a stream segment. There have been very few attempts to develop valid spatial covariance models that incorporate flow, stream distance, or both. The application of typical spatial autocovariance functions based on Euclidean distance, such as the spherical covariance model, are not valid when using stream distance. In this paper we develop a large class of valid models that incorporate flow and stream distance by using spatial moving averages. These methods integrate a moving average function, or kernel, against a white noise process. By construction they are valid models. We show that with proper weighting, many of the usual spatial models based on Euclidean distance have a counterpart for stream networks. Using chemical concentrations and fish counts from an example data set, the Maryland Biological Stream Survey (MBSS), we develop models using flow and stream distance. For the MBSS data set, we use restricted maximum likelihood to fit a valid covariance matrix, and then we use this covariance matrix to estimate fixed effects and make kriging and block kriging predictions.


• Functional Linkage of Watersheds and Streams Using Landscape Networks of Reach Contributing Areas David M. Theobald, John Norman, and Erin Peterson (Colorado State University) Landscape metrics are often used to support statistical analyses of aquatic resources, particularly to examine the possible consequences of land use change on water quality. Better understanding of these linkages requires explicit representation of hydrologic and ecological processes within a spatially-explicit framework. Our approach is to represent watersheds as a tessellation of networked reach contributing areas. We describe this framework, the basic landscape network data structure that is the key to this framework, and indicators that are generated. We illustrate this framework and ArcGIS tools with examples that have supported our project (STARMAP) efforts.


• Spatial prediction of Coho salmon counts on stream networks Dan Dalthorp and Lisa Madsen (Oregon State University) Two challenges of modeling count-valued processes on stream networks are that the response variable is non-normal (often incorribly so) and that sites that are quite close together may be on entirely different stream segments and be substantially less similar than more widely separated sites on flow-connected stream segments. We use a latent process regression model to account for the discrete response and use a variety of "distance measures" in modeling the spatial correlations. Model performances are assessed and compared using cross-validation.


• Cholesky Factorization Models for Within-Stream Network Dependence William Coar (Colorado State University) Recent literature on methods of estimation with autocorrelated data focuses on the use of the modified Cholesky decomposition of the covariance matrix to obtain covariance estimates without restriction. This Cholesky factorization models the dependence within reaches by reparameterizing the covariance matrix such that the new parameters have a meaningful interpretation. In aquatic data from stream networks, autocorrelation can arise from the flow of water from reach to reach. The Cholesky factorization suggests a covariance model for such data that is analogous to an autoregressive process in time series. In the simplest version of this model, any given reach is correlated to each of the contributing up-stream reaches, yet conditionally dependent only on the two immediate up-stream reaches. Extensions of this model can allow for different types of spatial inhomogeneities, including flow dependence. Maximum likelihood estimation of model parameters is considered for various stream network models. Joint work with F. Jay Breidt, Colorado State University.


• Quantifying Fragmentation of Freshwater Systems Using a Measure of Discharge Modification and a Hydrological Spatial Scan Statistic David M. Theobald, John Norman, and David M. Merritt (Colorado State University and USDA Rocky Mountain Research Station) Dams and diversion structures affect the condition of freshwater resources in watersheds through modification of natural flow regime. Attempts to measure the potential effects of this type of aquatic fragmentation have been based typically on a summary of the number of structures in a watershed. We have developed an alternative GIS-based measure that quantifies, for each stream reach, the proportion of discharge (flow volume) that is likely modified by upstream dams. We also identify “hot spots” or significant clusters of low modification area, using an implementation of a modified spatial scan statistic that uses hydrologic (not straight-line) distance, within the FLoWS tools for ArcGIS v9. We will illustrate our method and results using data for Upper Colorado watershed (1:100,000) and nationwide with coarse-scale hydrology (ERF1.2). This analysis is helpful in targeting high priority locations for conservation action.

• Spatial Lasso with Application to GIS Model Selection F. Jay Breidt (Colorado State University) Geographic information systems (GIS) organize spatial data in multiple two-dimensional arrays called layers. In many applications, a response of interest is observed on a set of sites in the landscape, and it is of interest to build a regression model from the GIS layers to predict the response at unsampled sites. Model selection in this context then consists not only of selecting appropriate layers, but also of choosing appropriate neighborhoods within those layers. We formalize this problem and propose the use of Lasso to simultaneously select variables, choose neighborhoods, and estimate parameters. Spatial smoothness in selected coefficients is incorporated through use of a priori spatial covariance structure, and this leads to a modification of the Lasso procedure. The LARS algorithm, which can be used in a fast implementation of Lasso, is also modified to yield a fast implementation of spatial Lasso. The spatial Lasso performs well in numerical examples, including an application to prediction of soil moisture.
  
  This is joint work with Hsin-Cheng Huang ,Academica Sinica, Taiwan, Nan-Jung Hsu, National Tsing-Hua University, Taiwan, and Dave Theobald Colorado State University


• Geostatistical Modeling:  Model Selection and Parameter Estimation Jennifer Hoeting (Colorado State University) We will provide an overview of a number of issues at the interface between theory and practice for geostatistical modeling.  We will compare three likelihood-based methods for parameter estimation: Bayesian, maximum likelihood, and REML estimation.  We will also describe a number of issues related to model selection for geostatistical models.  This talk is based on work with several research groups which include Richard Davis, Alix Gitelman, Megan Dailey, Kathi Georgitis, and Andrew Merton.


• Bayesian Selection of Geostatistical Regression Models Devin Johnson (University of Alaska, Fairbanks) The problem of covariate selection for spatial regression models is considered. Previous research has shown that failure to take spatial correlation into account can influence the outcome of standard model selection methods. Often, these standard criteria suggest models that are too complex in an effort to compensate for spatial correlation ignored in the selection process. Here calculation of posterior model probabilities for regression models through a Markov Chain Monte Carlo (MCMC) method is investigated. In addition, the proposed MCMC algorithm is modified for covariate selection in spatial generalized linear mixed models (GLMM). The GLMM analysis makes use of Langevin-Hastings updates for random effects. These methods are demonstrated with two data sets, one normally distributed and the other a Poisson spatial GLMM.


• Connecting Correlated GIS Predictors Using Graphical Models Alix I. Gitelman and Kathryn Georgitis (Oregon State University) We consider the problem of habitat association modeling when habitat data are collected across multiple scales. Specifically, with the ready abundance of remotely sensed data we can obtain land-cover information that is correlated both within and across scales, and address questions about how that land-cover information is associated with the presence or absence of a particular animal species. Principal component regression and partial least squares regression have been used to reduce the dimensionality of correlated predictors in similar settings. We use graphical models as an alternative in which connections between correlated predictors are modeled in an effort to (a) understand those connections and (b) reduce the number of those predictors needed to model their association with a response. The results from the graphical models afford an interpretation of the interrelationships between land-cover variables, and, having accounted for those interrelationships, provide estimates for the associations between the land-cover and the presence/absence of an animal species. We illustrate the methods on a dataset of forest bird species in the western Great Lakes region.


• Structural Break Detection in Time Series, Richard Davis, Thomas Lee and Gabriel Rodriguez-Yam (Colorado State University)
  
  To be supplied.

• Spatial Prediction on a River Network Noel Cressie (The Ohio State University) The Ecosystem Health Monitoring Program (EHMP) provides a means of monitoring the health of the streams and waterways in South East Queensland (SEQ), Australia, on an annual basis. In this talk, I shall develop methods for spatially predicting the value of a variable (dissolved oxygen change) at both sampled locations (134 freshwater sites in 2002 and 2003) and other locations of interest throughout the river network in SEQ. In order to deal with the relative sparseness of the monitoring locations in comparison to the number of locations where one might want to make predictions, we use the notions of stratification from survey sampling and (ordinary and constrained) kriging from geostatistics. River networks offer special challenges because of their tree structure. A complete approach to spatial prediction on a river network is given, with special attention paid to environmental exceedances. This research is joint with Jesse Frey, Bronwyn Harch, and Mick Smith.


• Variability among Continuous-Domain Samples Characterized by the Point Pattern of a Non-Probability Sample Cynthia Cooper (Oregon State University) A fundamental difference between model- and design-based sampling and estimation is the variance addressed by each approach. Design-based variance addresses variance induced by the sampling process. Model-based variance quantifies variance explicitly due to the response structure. On a population or response surface with covariance, as sampling resolution increases, the magnitude of covariance within the sample increases. The resulting effective covariance within the sample impacts the estimator variance over the sample process. For a purposive sample, technically, purposive elements do not contribute anything to the sample-process variance (as they would be in every sample). Nevertheless, a stakeholder might reasonably ask, using information inferred from observations about an assumed-stationary response covariance structure, how much would an estimate derived from other "similarly arranged" patterns of elements vary? Sample inclusion densities impose a joint density on the spatial arrangement of elements included in the sample. A class of "similarly arranged" elements or a sample process can be characterized by joint inclusion densities. This research examines the derivation of the joint inclusion densities from estimated joint densities of collections of elements observed on a continuous domain. The joint density of the elements is conveniently characterized by densities on the point-pair distances, or functions related to these point-pair distance densities. The characterization of the arrangement of points and estimated response covariance structure are applied subsequently to characterize the variation in an estimate derived from a class of similarly arranged patterns of elements. The process by which the practitioner quantifies the variance of the estimate does not in any way eliminate the possibility of bias that exists whenever sampling is non-random and/or there is frame error.


• Utility of Site-Based Sampling for Characterizing Watersheds and Fish Distribution Patterns R. E. Gresswell (USGS-NRMSC), A. R. Olsen (USEPA NHEERL, Western Ecology Division), D. S. Bateman (Oregon State University), D. P. Larsen, (USEPA NHEERL, Western Ecology Division), Christian Torgersen (USGS- FRESC), David Hockman-Wert (USGS- FRESC) In order to characterize the ecological condition of Pacific Northwest watersheds and their aquatic ecosystems, interagency teams have developed the Aquatic and Riparian Effectiveness Monitoring Plan. Monitoring is targeted at the subwatershed scale (6th-field Hydrologic Unit Code), and a methodology for selecting a statistical sample of subwatersheds has been implemented. In addition, this monitoring effort requires a rigorous sampling protocol that can be used for variables that will be measured at a subset of sites within each watershed. In order to evaluate a site-selection procedure based on a probability sampling, we used Monte Carlo simulations of random samples of stream sections that were approximately 40 channel-widths in length. Physical and biological data collected from continuous censuses in 22 third-order watersheds provided a known population from which samples were drawn. Estimates from 5, 10, 15, and 20 sample sections per watershed were compared to census information. Simulated samples and census data were further evaluated by examining spatial relationships using semivariograms. This simulation procedure provided measures of uncertainty associated with estimates at site, watershed, and population scales, and results suggest that a probability-based sampling protocol provides precise estimates that can be used to characterize watersheds for monitoring at the regional scale. Spatial analysis, however, underscored the limitations site-based data for describing fish and habitat distributions in individual stream networks.


• Estimates of Year-to-Year Variability for Physical Habitat, Riparian Vegetation, and Macroinvertebrate Attributes, Rick Henderson (US Forest Service - PIBO Effectiveness Monitoring Program) The PIBO Effectiveness Monitoring Program samples wadeable streams on Forest Service and Bureau of Land Management lands within the Interior Columbia River basin. The program goal is to assess whether land management activities are maintaining, degrading, or restoring the condition of aquatic and riparian resources. In addressing this goal, we wanted to determine how year-to-year variation affects our ability to detect trends. Our objectives were to describe the 1) synchronous year-to-year variation expressed by all sites together (concordant variation), 2) the average independent year-to-year variation at individual sites, and 3) whether the results differ between reference and managed sites. We used data from 48 sites that were sampled either 3 (13 sites), 4 (20 sites), or 5 (15 sites) times between 2000 and 2004. We first used a random effects Analysis of Variance (ANOVA) model to estimate the site, year, site*year, and residual variance components for 10 physical habitat, three riparian vegetation, and two macroinvertebrate attributes. The results will be used to describe the relative importance of concordant variation, how conditions vary between years at individual sites, the effects of land management activities on annual variation, and how these results differ between attributes


• The Concept of a Master Sample: An Oregon Case Study, David P. Larsen and Anthony R. Olsen (USEPA), and Don L. Stevens, Jr. (Oregon State University) Over the past several years there has been a significant increased interest among Federal and State agencies and Tribal nations about the condition of aquatic resources at regional scales. This interest has spawned a need for survey designs that allow valid statistical inferences to be drawn from representative samples of the resources when a census cannot be conducted. The tendency has been to apply a recently developed spatially balanced design to each agency’s needs independent of other agency needs or designs even if the aquatic resource domains overlap and common indicators of condition are used. This can create some statistical challenges when combining data and is inefficient if agency’s collect redundant data. As a potential solution, we propose a statistical survey design consisting of a dense, spatially balanced master sample of sites on stream networks that can be classified to meet individual agency needs and allow combining data across agency domains. This master sample consists of an ordered list of sites from which an agency selects the subset that meets its design needs (i.e., sample size, geographic domain, strata or subpopulation). We illustrate a prototype master sample using the stream networks in Oregon as an example and show how this master sample can be subset and integrated to meet a variety of monitoring needs.

• Methods for Trend Analysis of Water Quality in Coastal Environments Anders Grimvall and Claudia Libiseller (Linköping University, Sweden) Trend assessment of water quality in coastal areas and archipelagos is often a difficult task. The complex hydrography of coastal systems can make process-based modelling impracticable. Moreover, the randomness in the mixing of saline and fresh water can long conceal important anthropogenic trends or create spurious trends. Here, we present statistical approaches in which data from several sites and depths are pooled into a single analysis, and salinity is treated as a covariate. In particular, we discuss monotonic and semi-parametric regression models and algorithms to estimate such models. The theory is illustrated with a case study of nitrogen trends in the Stockholm archipelago. In the case of monotonic regression, the concentration of total nitrogen was assumed to decrease with time and salinity. A dummy variable was introduced to distinguish the inner from the outer archipelago, and a recently developed algorithm for monotonic regression in two or more explanatory variables was employed to estimate the model. In addition, the statistical significance of detected downward tendencies in total nitrogen were assessed using partial Mann-Kendall tests in which salinity was treated as a covariate.


• Detection of Short Term Trends In Water Quality Data Lieven Clement and Olivier Thas (Ghent University) In order to evaluate the health of river ecosystems, river monitoring networks have been established to assess the water quality. In Europe, the Water Framework Directive triggers local authorities to improve the water quality by environmental regulations which are commonly expressed in terms of threshold levels. In Flanders, the Flemish Environmental Agency (VMM) is dedicated to develop a long term vision in order to comply with these standards. To evaluate and refine their strategy it is important to detect short term trends in the frequency of violating water quality standards. Due to the inherent variability of water quality data, difficulties arise when differentiating between natural variability and trends. Therefore, research has mainly focussed on the detection of long term trends. The complex nature of water quality data typically leads to the violation of the distributional assumptions made in classical parametrical statistical tests. Hence, distribution-free tests should be adopted for the detection of trends. This can be accomplished by transforming the data towards binary data using the water quality standard as a threshold. In this way the data are Bernoulli distributed and trends in the compliance frequency can still be modelled. Although the data complexity is reduced, the spatio-temporal dependence still remains.
  
  In this paper, a logistic state space model for the probability of violating a standard is presented. In contrast to common ad hoc methods, this model explicitly incorporates the dependence structure of the data and uses a mean model for the trend and seasonal variation. The trend detection on the level of individual sampling locations only implies the temporal dependence to be modelled. This is done by using a first order autoregressive process. When assessing the trend on a more regional scale, the spatial dependence has to be incorporated in the model as well. However, an important distinction with classical spatial dependence structures has to be made: the water flows only in one direction within the river reaches, therefore a causal interpretation can be given to the correlations. Furthermore, as opposed to time, rivers can join or split. This implies a more general branched unidirectional structure which can be represented by using a Directed Acyclic Graph (DAG). Generalized estimation equations are used for the parameter estimation. They rely on the correlation structure of the data, which is deduced from the state space model. The model is applied on real data to assess short term trends on the level of individual sampling locations and on a more regional scale.


• Developments in Trend Detection in Aquatic Surveys Scott Urquhart (Colorado State University)
  
  To be supplied


• An Application of Estimating Equations and Quadratic Inference Functions in Complex Surveys Leigh Ann Harrod and Virginia Lesser (Oregon State University), Estimating equations are applied in longitudinal studies to account for the correlation between repeated measures taken on the same subject in clinical trials. This methodology requires the specification of a working correlation matrix and yields estimates that are consistent even when the correlation matrix is misspecified. Quadratic inference functions are used to combine estimating equations to obtain regression estimators with favorable asymptotic qualities when the dimension of the score vector is greater than the dimension of the parameter vector to be estimated. An application of quadratic inference function inference to ecological data will be discussed.


• Estimating Trend in Oregon Coastal Coho Salmon Populations Using a Multi-Panel Sampling Design William Gaeuman and Don L. Stevens, Jr. (Oregon State University) In 1998, the Oregon Department of Fish and Wildlife (ODFW) initiated a survey of coho salmon in Oregon coastal streams using a multi-panel survey, with panels having differing re-visit periods. One panel is visited every year, some after 3 years, and some after 9 years. In this paper, we explore design-based, model-assisted ways of estimating change and trend using data available after six years of sampling. At this point, the annual panel has been visited 6 times, 3 panels have been visited twice (in years (1, 4), (2, 5), and (3, 6)), and 6 panels have been visited only once (a different panel in each of years 1 through 6). The approach used here assumes that the response of interest is the site-specific least-squares estimate of slope based on 6 years of data. The response is observable only for the annual panel. We treat the slopes based on 2-year pairs as predictor variables, and use regression estimators to get a composite estimator. Results are illustrated using ODFW data on coho salmon density.

Posters

• A Hierarchical Bayesian Approach to Develop Predictive Maps of Coho Salmon (Oncorhynchus kisutch) Spawner Abundance in Oregon Coastal Streams Ruben Smith (Oregon State University), Don Stevens (Oregon State University) and Jeff Rogers (Oregon Department of Fish and Wildlife )
  
  Maps showing the freshwater Abundance and distribution of anadromous salmonoids are an important tool for ecosystem recovery planning. They are a useful visual tool to compare coho abundance and distribution across regions, to assess the impact of habitat management practices on coho populations, and to identify potential habitat restoration areas. The model used to generate prediction maps incorporates two random components. One captures the spatial variability of the abundance of Coho spawning, and the other accounts for any measurement error and small scale spatial variation not captured by the location of the monitoring sites. The model parameters are estimated using a Bayesian approach, which incorporates the uncertainty generated when estimating the parameters in the construction of the prediction intervals (Diggle et. al., 1998). Three advantages of this model over traditional methods used to produce prediction maps are: 1) it accounts for the discrete nature and correlation structure of the data; 2) all the model parameters are estimated simultaneously; and 3) it allows a map based depiction of estimates of the uncertainty surrounding the prediction maps.


• Review of Geostatistics in Aquatic Systems Joshua French and Scott Urquhart (Colorado State University)
  
  An extensive literature search was conducted of geostatistical work in aquatic systems.  The poster gives equal emphasis to three topics:  (1) Exploring Spatial Correlation in Rivers by French, (2) geostatistical developments related to aquatic systems, and (3) various applications of geostatistics in aquatic systems.  The literature summary areas include the choice of a distance metric, invalid covariance structures, prediction/estimation methods,  and sampling design and optimization.  Aquatic applications include oceans, seas and bays, estuaries, rivers and streams, and coastal systems. An annotated bibliography of relevant literature will be available as a handout.


• Predicting Water Quality Impaired Stream Segments UsingLandscape-scale Data and a Regional Geostatistical Model -- A Case Study in Maryland Erin E. Peterson and N. Scott Urquhart (Colorado State University)
  
  The Clean Water Act of 1972 requires states and tribes to identify impaired stream segments. This poster summarizes an effort to utilize and compare several geostatistical models for predicting where impaired stream segments are located. We use landscape data derived from a GIS as major covariates and apply the methodology to an extensive set of water quality data available from the Maryland Biological Stream Survey. The poster focuses on dissolved organic carbon, but the methodology could be applied to other regulated constituents. Predictions are depicted on a map, along with an indication of their precision.


• The Index Period Problem: Defining Seasons as They Apply to Florida's Water Quality Indicators Neal A. Doran, Rick Copeland, Paul Hansard, Jay Silvanima, Gail M. Sloane, and Margaret Murray
  
  The Florida Department of Environmental Protection has established status and trends monitoring networks in collaboration with the United States Environmental Protection Agency's (EPA's) Environmental Mapping and Assessment Program (EMAP) (Copeland, R. et al., 1999). The networks are designed to characterize Florida's ground and surface waters and determine changes in environmental quality over time (details may be found at http://www.dep.state.fl.us/water/monitoring/index.htm). Florida's 5-year rotating basin design originally incorporated index periods based on the greatest response to anthropogenic and climatic stressors. The index periods are divided along geographic lines, for peninsula and non-peninsula Florida, based on climate. One ongoing concern of a probabilistic monitoring design is the selection of index periods. A recent study suggests that Florida groundwater exhibits minimal seasonality, creating maximum flexibility for sampling options. Our current evaluation of the index periods is now focused on surface water; two considerations are primary. First, whether index period definitions in the state of Florida affect the most critical thresholds in our results; if so, index period definitions are of pivotal importance. Second, for consistency in comparison of the data along climatic gradients, we wish to find the seasonal definitions that minimize variability in chemical/biological indicators. The purpose is to allow comparability of probabilistic data across spatial and seasonal boundaries.


• Using a Spatially Balanced, Random Sampling Design to Assist Informed Management Decisions Sarah Lowe (San Francisco Estuary Institute (SFEI)) , J.R.M Ross (SFEI), C. Grosso (SFEI), A. Franz (SFEI), D. L. Stevens Jr. (Oregon State University)
  
  
  The San Francisco Estuary Regional Monitoring Program for Trace Substances (RMP) is the primary source for long-term contaminant monitoring information for the Estuary. The RMP was initiated in 1993, and is an innovative and collaborative effort between the scientific community, the San Francisco Bay Regional Water Quality Control Board (Water Board), and the regulated discharger community. The RMP includes a Status and Trends component for evaluating spatial and long-term temporal contaminant trends throughout the Estuary. Additional special studies address specific scientific questions relating to water quality and beneficial uses.
  
  The RMP Status and Trends sampling plan was redesigned in 2002 employing the Generalized Random Tessellation Stratified design (GRTS) utilized by the U.S. Environmental Protection Agency’s Environmental Monitoring and Assessment Program (EMAP) for monitoring the condition of the nation’s coasts and large estuaries. Each year, the Status and Trends program sequentially samples randomly selected water and sediment stations for trace contaminants and other water quality indicators providing increased spatial coverage of the Estuary over time. This design will provide environmental managers and regulators with statistically defensible information about the spatial and temporal distribution of regulated and emerging contaminants in the Estuary.
  
  The new Status and Trends monitoring design is well suited to address some of the new, more focused RMP management questions developed jointly, in 2005, by Bay Area scientists, environmental regulators, and the regulated community. Some of the management questions the new random sampling design will address include:
  • What are the spatial and temporal patterns of contamination in the Estuary and it’s sub-regions?
  • How does contamination in the various Estuary regions compare to each other, and to specific water and sediment effects thresholds?
  • What is the spatial and temporal extent of sediment toxicity in the Estuary?
  • What are possible pathways of contamination to the Estuary?
  • Is contamination in the shallow reaches of the Estuary different from the deeper channels?
  
  
• Aquatic Threat Pathways: A Spatially Explicit Framework For Assessing Neighboring Residential Development Effects On Aquatic Resources Alisa A. Wade (Colorado State University)
  
  Understanding the dynamic relationship between spatial arrangement of landscape characteristics and ecological flows remains a continuing challenge for landscape ecologists. This relationship is critical as it relates to anthropogenic landscape modification in light of the expanding and pervasive nature of residential development. Of particular importance are the effects of development on aquatic systems, which provide an integrative measure of broader ecosystem conditions. This research strives to synthesize mechanistic models and empirical data through quantitative modeling to create an analysis framework that can be used to predict how context and explicit arrangement of urban development relates to aquatic system condition. My research develops control models of the aquatic threat pathways – the mechanistic processes through which threats from residential land conversion cascade, spatially and temporally, to influence aquatic system condition. Bounded by these control models, I assess empirical relationships between threats from urbanization - human activities related to residential development that may cause impairment of natural processes - and indicators of aquatic resource condition. The model framework is based on a system of nested watersheds to enable analysis at multiple spatial scales to account for both general coarse-scale ecosystem constraints and fine-scale mechanistic processes, while including an analytical scale that is sufficiently fine to represent explicit location of development within the watershed. The nested network of watersheds is represented as a graph, dictating distances between watershed nodes based on ecological and hydrological connectivity. Using a regression process model, I use Bayesian techniques to gather strength across the different nested watershed scales. I conduct the research for development surrounding U.S. National parks. As relatively unmodified systems surrounded by landscapes that have been altered to varying degrees, parks provide natural laboratories for assessing influences on ecological condition.