Ecological Forecasting Initiative 2019 Conference
Speaker & Poster Abstracts
May 13-15, 2019
AAAS Headquarters 1200 New York Ave, NW Washington, DC 20005
@eco4cast #efi2019
The Ecological Forecasting Initiative (EFI) is a broad, interdisciplinary effort aimed at promoting the use of forecasts to understand, manage, and conserve ecosystems and the services they provide. The EFI 2019 meeting is aimed at bringing together scientists, agencies, industry, and stakeholders to build a community of practice and advance research, applications, and collaboration around near-term (subdaily to decadal) ecological forecasts.
EFI2019 is sponsored by the Alfred P. Sloan Foundation, the Frederick S. Pardee Center for the Study of the Longer-Range Future at Boston University, and the National Science Foundation’s Office of International Science and Engineering.
@ecoforecast #efi2019
Speaker Abstracts
Session 1: Theory and Synthesis Monday, May 13 9:00 am
S1.1: Keynote
Ecological forecasting: the role of observations in models
David Schimel1
1NASA JPL
[abstract]
2 @ecoforecast #efi2019
Session 1: Theory and Synthesis Monday, May 13 9:00 am
S1.2: Lightning Talk
Understanding the uncertainties in estimating post-fire recovery of biomass using the Ecosystem Demography (EDv2.2) model
Karun Pandit1, Hamid Dashti1, Nancy F. Glenn1, Alejandro N. Flores1, Kaitlin C. Maguire2, Douglas J. Shinneman2, Gerald N. Flerchinger3, Aaron W. Fellows3 [email protected]
1Boise State University; 2U.S. Geological Survey; 3ARS
Frequent wildfires in the sagebrush-steppe ecosystems in the Western United States lead to heavy biodiversity loss, and changes in vegetation composition. It is debated in literature whether fires are beneficial or not for maintaining sagebrush ecosystems, it is essential that we understand the impact of these fires in terms of recovery time and potential changes in vegetation composition. The objective of this study is to estimate post-fire recovery of vegetation in the Ecosystem Demography (EDv2.2) model and quantify uncertainties by comparing with satellite-derived information. This study builds upon previous work in biomass loss and post-fire recovery from the 2015 Soda Fire in the Great Basin, Western United States. We initialized EDv2.2 with the existing vegetation ecosystem using mean metrics based on a field inventory dataset along with fire disturbance from Landsat 8 data and the 2015 fire. We ran the EDv2.2 model for thirty years to observe changes in above ground biomass and PFT distributions in the study area through the years. We also compared overall biomass growth and its spatial patterns projected by EDv2.2 with relevant indices from satellite data from the year 2019. These included NDVI and GPP derived from the Landsat 8 data and PFT classifications based on AVIRIS-NG data from the study area. Preliminary results from EDv2.2 modeling shows spatial patterns of vegetation regrowth being significantly influenced by precipitation and elevation. We observed that the initially grass-dominated areas post-fire are gradually replaced by the shrub PFT in later years. We also found that the volume of biomass regrowth during initial four years post-fire from EDv2.2 matches well with that derived from Landsat 8 images.
3 @ecoforecast #efi2019
Session 1: Theory and Synthesis Monday, May 13 9:00 am
S1.3: Lightning Talk
Integrating evolutionary history into forecasts of species assembly
Amanda Gallinat1, William D. Pearse1
1Utah State University
As climate change alters habitats and species distributions, there is a strong need for predictive tools that describe where species will occur and potentially interact under future conditions. Species’ present-day ecological responses are constrained by their evolutionary history, so our understanding and forecasts of those responses may be improved with the use of phylogeny. Species’ phylogeny reflects their shared biogeography, the processes by which functional traits and environmental sensitivities evolved, and how environmentally constrained different groups of species are. With the overarching goal of integrating evolutionary history into forecasts of species assembly, we have been using novel eco-phylogenetic tools to analyze regional species occurrence data for plants, birds, and mammals collected by the National Ecological Observatory Network (NEON), Forest Inventory & Analysis, Breeding Bird Survey, and Thibault et al. 2011. We contrasted the relative importance of species’ traits, phylogeny, and environmental responses for predicting species occurrences in plants, birds, and mammals. Our preliminary results show that environmental tolerances are strongly conserved in all three groups; in other words, related species share similar environmental sensitivities, which could be leveraged to impute sensitivities for species that lack data. And while functional traits also exhibit phylogenetic signal in all taxa, we find that the signal of environmental tolerances is stronger than that of traits alone. By combining information about species’ evolution, functional traits, and present-day environmental responses, we can produce models that make more powerful predictions about the evolution of species’ niches, and their responses to future environmental change.
4 @ecoforecast #efi2019
Session 1: Theory and Synthesis Monday, May 13 9:00 am
S1.4: Lightning Talk
Estimating Body Temperature and Thermal Performance at fine spatial and temporal scales
Francis Choi1, Tarik Gouiher1, Fernando Lima, Gil Rilov2, Rui Seabra, Brian Helmuth1 [email protected]
1Northeastern University; 2Israel Oceanographic & Limnological Institute
The rocky intertidal is one of the most dynamic and thermally variable ecosystems, where the joint influences of solar radiation, temperature and topography can lead to differences of up to 20°C over centimeter scales. However, the ecological importance of this variation in the face of climate change remains poorly understood. Here, we present a novel technique for modelling microhabitat heterogeneity and patterns of thermal physiology among interacting organisms. We used drone photogrammetry to re-create virtual topographic maps at a resolution of 400 cm2, which are then fed as inputs to a heat budget model estimating hourly surface temperature. These body temperature layers are then converted to thermal performance layers for organisms using thermal performance curves, creating physiological “landscapes” that display spatially-explicit patterns of “microrefugia”. Our analyses show how nonlinear interactions between these layers lead to distinct predictions about organismal performance and survivorship from those made using any individual layer alone. For instance, thermal performance layers reveal that microrefugia are variable through time and space, showing how mobile species must continue to migrate between microhabitats in order to maintain optimal performance. This approach provides a method for exploring the role of micro-topographic variability in driving organismal vulnerability to environmental change.
5 @ecoforecast #efi2019
Session 1: Theory and Synthesis Monday, May 13 9:00 am
S1.5: Lightning Talk
Forecasting Population Impacts of Chronic Wasting Disease
Alison Ketz1, Daniel J. Storm2, Michael Samuel1, Robin E. Russell3, Daniel P. Walsh3
1University of Wisconsin, Madison; 2Wisconsin Department of Natural Resources; 3U.S. Geological Survey, National Wildlife Health Center
Forecasts of wildlife disease typically focus on predictions of geographic spread, temporal dynamics of disease transmission, and the impact of management actions on epidemiological processes. Forecasting the population impacts of an epidemic are further complicated by species- specific demography, inter-specific interactions of host and pathogen, genetics, and spatial- temporal factors influencing patterns of infection, recruitment, and mortality. Chronic wasting disease (CWD) is a contagious prion disease affecting four species of free-ranging and captive cervids in North America. Geographic detection and distribution of CWD noticeably increased after 2002, although the disease has been present in North America since the 1960s. The prolonged course of infection has led to lengthy epizootics that last for decades, making population effects difficult to determine. We are in development of a model to forecast population effects of CWD on white-tailed deer in Wisconsin. As prevalence and spread continue to accelerate, management actions to mitigate CWD impacts will be challenging, costly, and will likely require changes in how we manage cervid populations.
6 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.1: Keynote
Combining Forecasts
Yael Grushka-Cockayne1, Jason Merrick2 [email protected]
1Harvard University, 2Virginia Commonwealth University
Early research in forecast aggregation focused on Bayesian approaches that account for forecast accuracy and correlation among forecasts. Other research has examined why in practice such theoretically sound methods are often outperformed by simpler techniques such as the average. In recent years, the wisdom of the crowd literature has encouraged the exploration of alternative heuristics that performed better than even the simple average in some cases. In this talk, we will review alternative methods and consider how they are evaluated. We will provide guidance on when various methods perform well, and how one might use forecast aggregation in practice.
7 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.2: Lightning Talk
Ecological forecasts for integrated socio-environmental systems
Jaime Ashander1, Kailin Kroetz1, Yusuke Kuwayama1
1Resources for the Future (RFF)
Human activities influence almost every aspect of the earth system, but these influences vary widely in their strength and immediacy. Furthermore, the rapidity with which humans assess and respond to the state of their environment varies with the spatial scale and environmental context of interaction. In integrated socio-environmental systems (SESs), feedbacks between human actions and the environmental state occur at multiple temporal and spatial scales, resulting in potentially complex dynamics. Systems like this include lake networks where recreational activities contribute to the spread of already-established aquatic invasive species but interventions to control invasive species differ among lakes, coastal communities where individuals participate in multiple fisheries subject to varying regulations, and water-limited ecosystems coexisting with irrigated agriculture subject to varying regulations on water use via ground or surface water. In each of these cases, human actors fall into broad categories of policymaker and resource-user and there are multiple pathways for human influence, direct and indirect, on the ecosystem state. However, the degree to which feedbacks between ecological systems and human actions should be modeled explicitly in near-term forecasts is an open question. We present a generalized theoretical model of an integrated SES, including the ecosystem state and dynamics, the policymaker decision criterion, and the dynamic response of resource users to both ecosystem and policy. Using this framework, we develop a typology that categorizes whether and how an ecological forecast includes dynamics due to human behavior. For several integrated SESs, we then discuss the relevant temporal and spatial scales for management and the corresponding type of forecast that is needed.
8 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.3: Lightning Talk
Embrace the uncertainty: What self-driving cars can teach us about ecological forecasting
Carl Boettiger1 [email protected]
1University of California, Berkeley
How do we make good decisions with only mediocre forecasts? The field of decision theory provides a framework for selecting the best course of action in face of uncertainty about the future -- but it's rigid optimal control policies can fail dramatically when those inevitable 'simplifying assumptions' are violated by real-world complexity. Drawing on examples from fisheries, I will illustrate how a pivot from the stylized approach favored by natural resource economists to an engineering approach that has underscored advances in autonomous vehicle navigation can change policies and outcomes for decision making in managing complex ecological systems.
9 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.4: Lightning Talk
A Structural Dynamic Econometric Approach to Incorporating Fisher Behavior into Policy Design
Kailin Kroetz1, James N. Sanchirico2, Daniel K. Lew3
1Resources for the Future; 2University of California, Davis; 3NOAA Fisheries, Alaska Fisheries Science Center
Ecological system components can be part of an integrated socio-environmental system (SES) where human interventions may drastically alter near term ecological outcomes. Furthermore, the human and ecological components may be endogenous and coevolve over time, and therefore a joint model may be needed for near-term forecasting and policy design.
We develop a dynamic discrete choice model of fisher behavior, including ecological and economic factors and policy constraints as model covariates. The model is forward-looking, forecasting fisher decisions over multiple time periods based on current and expected future ecological and economic conditions. Parameters of the model are identified using confidential data from the Alaska halibut and sablefish fisheries on individual fisher behavior. This data includes whether fishers participate in a fishery, and if so, how much they fish, as well as information on determinants of fisher profit including ex-vessel price, vessel characteristics including vessel length, areas fished, and attributes of quota purchased. We include simple differential equation models of fish stocks and total allowable catch (TAC) in the model, although more complex ecological model forecasts could easily be integrated.
Once the parameters are identified, the model can be used to forecast individual fisher entry and exit from the fisheries, quantities fished, profit, and aggregate outcomes including total fishery participation and quota (tradable permit) prices under different conditions (e.g. ecological, policy). We focus on estimating individual fisher and aggregate responses and outcomes under multiple counterfactual policy designs. Using the model to develop counterfactual scenarios allows us to quantify the economic efficiency differences between the alternative policy designs. We also examine community and cultural goals (e.g. following National Standard 8 of the Magnuson Stevens Act). Specifically, we focus on how program design influences the geographic distribution of program participation and how participation varies by community size and fishery reliance.
The sign and significance of the model parameters yield insight into factors determining how fishers respond to exogenous changes such as changes to stock size, total allowable catch (TAC), and fuel oil price. We find that two ecologically-linked variables, stock and TAC, are both significant factors in determining fisher behavior and therefore we expect changes in these factors would result in changes in fisher behavior. Similar models could be constructed to explore fisher responses to ecological shocks and/or impacts of other future proposed policy changes and to identify the socioeconomic consequences of these changes.
10 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.5: Lightning Talk
Population Dynamics and Ecological Forecasting in the Niger Delta
Daniel Odigie Abumere1, Omoyemen Lucia Odigie-Emmanuel2
1Centre for Human Rights and Climate Change Research; 2Global Alliance for Justice Education
National resources in the Niger Delta have been severely negatively impacted by over exploitation for the development of Nigeria. Understanding and predicting changes in human demand, natures capacity, consumption and production rates have a role to play in ecosystem balance. This is a critical issue and ought to be central to decision making including planning for implementation and monitoring of sustainable development goals.
No known study has currently measured the demand on natural resources and the quantity of nature it takes to support Sustainable Development in the Niger Delta.
This study shall focus on broadening understanding of Population Dynamics and Ecological Forecasting in the Niger Delta using a multidisciplinary approach.
11 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.6: Lightning Talk
A Value-of-Information Framework to Estimate the Societal Benefits of Improved Near-Term Ecological Forecasts
Yusuke Kuwayama1
1Resources for the Future
Ecological forecasts can yield societal benefits by helping decisionmakers take actions that lead to improved outcomes for people and the environment. However, further analysis is needed to connect the decision processes that are informed by ecological forecasts with direct societal benefits, and there is a need to substantiate the benefits of improved forecasts in socially and economically meaningful terms. Quantifying the socioeconomic benefits of ecological forecasts is important for several reasons. First, these estimates can demonstrate return on investment in projects that seek to improve ecological forecasts. Second, estimating the potential benefits of different types of forecasts can help program managers make informed choices about how to invest limited resources in improving forecasts. Third, benefit estimates can provide ecologists with an effective tool for communicating the value of their work in socioeconomically meaningful terms. Fourth, benefits quantification can increase the likelihood that the forecast produces socioeconomic benefits by requiring teams to think about how to evaluate the outputs of their project during the planning phase.
In this lightning talk, we will present a value-of-information (VOI) framework that can be applied to ecological forecasts. This framework was developed by economists and decision scientists from the Consortium for the Valuation of Applications Benefits Linked with Earth Science (VALUABLES), a cooperative agreement between Resources for the Future (RFF) and the National Aeronautics and Space Administration (NASA) that is working to quantify and communicate how the use of satellite information in decisions can improve outcomes for people and the environment. The consortium brings together economists and decision scientists, NASA scientists and remote sensing experts, and members of the wider Earth science community.
This lightning talk will contribute to the Ecological Forecasting Initiative’s objective of integrating the ecological sciences with decision, governance, and economic sciences, to better understand how ecological forecasts can influence decision making. The talk will also describe how ecological forecasters and decisionmakers in industry, state and federal agencies, and non- profit organizations can work together to co-produce ecological forecasts that are likely to maximize socioeconomic benefits.
12 @ecoforecast #efi2019
Session 2: Decision Support Monday, May 13 10:45 am
S2.7: Lightning Talk
Learning About and From Ecological Forecasting Models: A Decision Science Approach
Michael Gerst1, Melissa Kenney2, Eva Regneir3 [email protected]
1University of Maryland; 2University of Minnesota; 3Naval Postgraduate College
Models are ubiquitous in science and science-based aiding of decision-making. Despite the frequency of their use, little progress has been made in articulating how modelers and model users learn from model results; specifically, there is a paucity of frameworks that are theoretically-grounded and independent of scientific discipline. For ecological forecasting and other model-dependent disciplines, this can be a problem for two reasons.
First, models are often used as a means of integrating knowledge across disciplines or subject areas (e.g., ecology and economics). If modeling conventions and epistemic norms are not transparent, then differing practices can easily be misinterpreted, leading to inaccurate model insights. Second, model-building and interpretation is, conceptually, a learning process that consists of a chain of decisions, from choosing the system boundary to deciding how to validate the model to testing model robustness. This framing is important because experts and non-experts alike can exhibit biases and heuristics during the process of learning, especially from model- based evidence. In aggregate, these learning biases and heuristics can lead to systematic errors in how (i) prior knowledge is elicited and incorporated, (ii) information or evidence is sought out, and (iii) model-based evidence is used to learn and provide insight.
Using decision science as a foundation, we develop a framework that guides model builders and model users through the process of learning about and from a model in a way that minimizes error introduced by biases and heuristics and aids in choosing a model that maximizes learning. This is accomplished through an application of Bayesian inference that treats models as constructed evidence-generating processes that are used in conjunction with modeler and user judgment of how adequate the model is for learning. Following this framework generates a comprehensive set of questions to ask in the learning process that is theoretically-grounded and independent of scientific discipline. Importantly, questions are linked to mathematical terms with foundations in Bayesian inference, which allows for use of well-established elicitation techniques. Thus all the steps of generating model-based insight are clearly delineated and their links to final insight are transparent.
13 @ecoforecast #efi2019
Session 3: Education and Inclusion Monday, May 13 3:30 pm
S3.1: Keynote
Bridging engagement and practice: pathways toward an education in ecological forecasting
Jaclyn Matthes1 [email protected]
1Wellesley College
Ecological forecasting asks big questions that can easily engage broad audiences: what will happen in the near future, and how does that impact us? But in practice, ecological forecasting requires a large set of technical skills that can be daunting for students to learn and apply. Teaching conceptual systems thinking, rooted in systems that students care about, alongside technical forecasting skills could facilitate persistence in pathways to ecological forecasting and improve learning outcomes. Many forecasting skills can be introduced in introductory undergraduate courses in biology, environmental science, and associated fields, which can increase familiarity with core forecasting concepts. This talk will outline the broad set of skills connected to ecological forecasting, along with pathways for acquiring, practicing, and applying these skills. In particular, we will focus on characteristics of equitable educational pathways that can build an inclusive community of diverse ecological forecasters.
14 @ecoforecast #efi2019
Session 3: Education and Inclusion Monday, May 13 3:30 pm
S3.2: Keynote
Strategic Planning for Diversity--Effective Strategies for Broadening Participation in the Environmental Sciences
Diana Dalbotten1 [email protected]
1University of Minnesota
Let's talk about ways to break down the issues related to the lack of racial and ethnic diversity in the Environmental Sciences. This talk takes a look at barriers to participation and examines how strategic planning can help your organization, department, center or institution be a part of the solution. We'll examine some effective programs and practices that can support diversity at every stage along the pathway towards a career in environmental science.
15 @ecoforecast #efi2019
Session 4: Cyberinfrastructure Tuesday, May 14 9:00 am
S4.1: Keynote
An Ecosystem of Cyberinfrastructure
Kenton McHenry1
1University of Illinois at Urbana-Champaign, National Center for Supercomputing Applications
A need to better navigate and utilize data within the scientific community has been clear for some years now. Within fields such as ecology this spans the need to work with large datasets to effectively run analyses, navigate unstructured and/or very heterogeneous datasets, work with real-time data streams, QA/QC, navigate different naming/unit conventions, model execution over a range of computational resources, and so on, all while maintaining data provenance to ensure scientific reproducibility. To address this there has been an increased emphasis in establishing strong working collaborations with those supporting/developing cyberinfrastructure and the scientific community in order to build out and best utilize tools addressing these scientific needs as requirements continuously emerge and evolve over time (e.g. as they have recently with the wide adoption of machine learning). I will provide an overview of some of the large movements on the cyberinfrastructure side to support these needs, from the evolution of a national data infrastructure, to the Research Software Engineer movement, to specific tools being developed.
16 @ecoforecast #efi2019
Session 4: Cyberinfrastructure Tuesday, May 14 9:00 am
S4.2: Lightning Talk
Sandboxing the pipeline: developing automated forecasting systems that facilitate model development
Juniper Simonis1, Glenda Yenni2, Shawn Taylor2, Erica Christensen2, Ellen Bledsoe2, Hao Ye2, Ethan White2, SK Morgan Ernest2
1University of Florida, DAPPER Stats; 2University of Florida
Accurate and up-to-date population forecasts are needed in a variety of fields including endangered species management, disease epidemiology, and invasive species control. However, we still lack many tools necessary for operationalizing ecological forecasts. To help address that gap, we are developing a near-term forecasting system for a population ecology study of desert rodents outside Portal, Arizona. To date, we have built a fully functioning automated iterative forecasting system for our study using continuous integration (CI) to run the pipeline and produce new forecasts every week. Although our infrastructure was designed to allow the contribution of new models, we noticed a bottleneck limiting the addition of new models to the pipeline caused by the fact that while the codebase functioned well within the CI system it was not easy enough to explore the data and develop new models outside of the CI application of the pipeline.
The process of code development and exploration outside of a production pipeline is known as sandboxing and is key to scientific progress in a forecasting setting, as it substantially lowers the bar for new users to contribute models. We addressed our bottleneck by creating an R package (portalcasting) that contains the code underlying the CI pipeline while also being portable, thereby allowing users to set up a fully-functioning replica repository on a local or remote machine. portalcasting formalizes the processes of creating, filling, and cleaning the repository; running the models (as forecasts or hindcasts); and evaluating the model output into simple yet powerful API functions by using a flexible directory tree for organization and a set of hierarchical options lists that integrate user-controls across the pipeline. We designed the functions in portalcasting to work out-of-the-box with minimal function calls and no need to input or alter arguments while also providing full control of the setup and execution of the pipeline. portalcasting has been deployed and is underpinning the CI application.
Although portalcasting is designed for the Portal Project forecasting system specifically, it lays the groundwork for a generalized approach to forecasting pipelines that work seamlessly in production and sandbox environments, thus facilitating scientific development in a forecasting setting.
17 @ecoforecast #efi2019
Session 4: Cyberinfrastructure Tuesday, May 14 9:00 am
S4.3: Lightning Talk
Cross-forecast synthesis and cyberinfrastructure in Near-term ecological forecasting
John Foster1, Michael Dietze1, Colin Averill1, Jennifer Bhatnagar1, Shannon LaDeau2, Kathleen Weathers2, Zoey Werbin1, Kathryn Wheeler1, Katherine Zarada1
1Boston University; 2Cary Institute of Ecosystem Studies
The Near-term Ecological Forecasting Initiative has set out to forecast a variety of ecological systems to analyze the uncertainties within individual forecasts (initial condition, model parameters, ecological drivers, and process error), and perform cross-forecast synthesis across disciplines. Focus areas where temporal forecasts are currently operational (automated and iterative) include 16-day carbon and water fluxes and phenology at a broadleaf forest site in Willow Creek, Wisconsin. More sites are coming soon and forecasts for tick-borne diseases (ticks and their small mammal hosts) will come on-line this summer. In addition, spatial forecasts of soil fungal and bacterial communities have been calibrated using community data and validated using out-of-sample data from NEON. All models are in a Bayesian state-space framework, which allows for the explicit separation of observation error from the process of interest.
To automate the forecasting process we have prototyped a generalized forecast cyberinfrastructure where any forecaster can drag-and-drop their model code, automatically schedule repeated forecasts, and archive results. Upon completion, the forecast is returned to an R shiny website for the user to explore results. This general workflow allows for scalable, repeatable research where competing forecasts can directly be compared.
18 @ecoforecast #efi2019
Session 4: Cyberinfrastructure Tuesday, May 14 9:00 am
S4.4: Lightning Talk
Integrating high-frequency environmental sensors, overlay virtual private networks, and simulation models in an end-to-end workflow to generate real-time iterative water quality forecasts
Cayelan Carey1, R. Quinn Thomas1, Renato Figueiredo2, Vahid Daneshmand2, Bethany Bookout1, Francois Birgand3 [email protected]
1Virginia Tech; 2University of Florida; 3North Carolina State University
Our team has developed a robust, open-source cyberinfrastructure (FLARE: Forecasting Lake And Reservoir Ecosystems) that seamlessly connects remote water quality sensors, data QA/QC and processing, and simulation models to generate 16-day water quality forecasts for managers in Falling Creek Reservoir, a drinking water supply in Roanoke, Virginia, USA. Our cyberinfrastructure uses both hardware (high-frequency environmental sensors connected with low-cost sensor “gateway” devices) and software (overlay virtual private networking) as part of our end-to-end workflow. The sensor data update daily simulations of a numerical simulation water quality model running in the cloud to create 16-day forecasts that will be autonomously published with digital object identifiers and searchable through the Environmental Data Initiative repository. Preliminary forecasts developed from FLARE to date demonstrate that the model can successfully predict water temperature within <0.5oC and the date of fall turnover, which has provided important information to managers on which depths to withdraw water from the reservoir for drinking. By developing new network computing methods for connecting distributed sensors and cloud infrastructures through virtual private networking and generating new computational methods for automated model-data fusion, our overarching goal is to develop easy- to-implement cyberinfrastructure that will address the rapidly-growing need within the ecological research community for tools that create near-term iterative forecasts to support environmental decision-making in the face of global change.
19 @ecoforecast #efi2019
Session 4: Cyberinfrastructure Tuesday, May 14 9:00 am
S4.5: Lightning Talk
Australia's Environmental Prediction System
Mark Grant1 [email protected]
1TERN Australia
Australia is endowed with precious natural assets and sustaining them requires ongoing monitoring, assessment and foresighting of changes in environmental conditions. The necessary field of view is large, encompassing the atmospheric, marine, terrestrial and hydrologic domains. We have strong foundations, comprised of institutions, expert teams, data sets and supporting analytic tools. Yet, end-users and providers of environmental information both see the need to build on those foundations so that better environmental management decisions can be made.
Australia’s National Environmental Prediction System (NEPS) is conceived as a federated form of national research infrastructure, enabling integration of environmental observations with predictive modelling to produce data and information products and services to be used by decision-makers to sustain natural assets.
A scoping study is currently deciding why a NEPS is necessary, what it is to produce, where to focus effort, and who should be involved in building and maintenance. The NEPS Scoping Study involves undertaking targeted consultations with key experts and stakeholders, including relevant areas of the existing National Collaborative Research Infrastructure Strategy (NCRIS) network.
This presentation will introduce Australia’s NEPS and outline its evolution and proposed establishment. It will also give conference participants the opportunity to engage with the scoping study’s consultation process.
20 @ecoforecast #efi2019
Session 5: Methods and Tools Tuesday, May 14 10:45 am
S5.1: Keynote
A hierarchical spatiotemporal analog forecasting model for non-Gaussian data
Chris Wikle1, Patrick McDermott2, Joshua Millspaugh3 [email protected]
1University of Missouri; 2Jupiter Intelligence; 3University of Montana
Analog forecasting is a mechanism-free nonlinear method that forecasts a system forward in time by examining how past states deemed similar to the current state moved forward. Previous applications of analog forecasting have been successful at producing robust forecasts for a variety of ecological and physical processes, but it has typically been presented in an empirical or heuristic procedure, rather than as a formal statistical model. The methodology presented here places analog forecasting within a fully hierarchical statistical framework that can accommodate non-Gaussian observations. Using a Bayesian approach, the hierarchical analog model is able to quantify rigorously the uncertainty associated with forecasts. Forecasting waterfowl settling patterns in the northwestern United States and Canada is conducted by applying the hierarchical analog model to a breeding population survey dataset. Sea surface temperature (SST) in the Pacific Ocean is used to help identify potential analogs for the waterfowl settling patterns.
21 @ecoforecast #efi2019
Session 5: Methods and Tools Tuesday, May 14 10:45 am
S5.2: Lightning Talk
A novel method of forecasting lake water quality using a combined approach of catchment modelling and modelling in-lake DOC dynamics
Tadhg Moore1 [email protected]
1Dundalk Institute of Technology, Ireland
The PROGNOS project is developing a modelling framework which will combine 1-D hydrodynamic models with biogeochemical model to provide 7-day water quality forecasts for lakes and reservoirs. The focus is on predicting algal blooms and dissolved organic carbon (DOC) fluxes within the lake. Here, we demonstrate the approach that is currently being developed to utilise weather forecasts and a combination of a catchment model with a lake biogeochemical model to make short-term predictions. This novel method demonstrates the value of collecting high frequency data as this data is collected in near-real time and can be assimilated into the model to improve model performance. This tool would inform water managers on the possible changes which could occur and allow them to make proactive decisions rather than reactive decisions in responding to varying levels of water quality within the lake.
22 @ecoforecast #efi2019
Session 5: Methods and Tools Tuesday, May 14 10:45 am
S5.3: Lightning Talk
Forecasting forest productivity by assimilating regional observations from plot networks, ecosystem experiments, and satellites into a process-based model
R. Quinn Thomas1, Annika Jersild, Evan Brooks, Valerie Thomas, Randolph Wynne [email protected]
1Virginia Tech
Forecasting how forest ecosystems will be altered by a changing environment can help society anticipate changes to the ecosystem services upon which it depends. However, producing a forecast requires quantifying and propagating different components of uncertainty in ecosystem model predictions. Here we present an ecological forecast of forest productivity for loblolly pine plantations across the Southeastern U.S. through the mid-21st century. To quantify uncertainty in ecosystem model parameters and processes, we used a hierarchical Bayesian approach (DAPPER; Data Assimilation to Predict Productivity for Ecosystems and Regions) to assimilate 35 years of global change research across the region into a process-based ecosystem model. The data assimilated included biometric, ecophysiological, and flux observations from regional plot networks by industrial forest research cooperatives, CO2 fertilization experiments, throughfall exclusion experiments, nutrient addition experiments, and water addition experiments. By combining the parameter and process uncertainty with uncertainty in climate projections, we found that 1) productivity is forecasted to increase across the region between 2010 and 2055; 2) there is considerable uncertainty in the forecast that overlaps zero (i.e., no change or negative change) in the southern and western extents of the region; and 3) the ecosystem process model was the single largest source of uncertainty. Overall, beyond providing information for resource managers in the region, our study demonstrates how diverse time-series observations from regional plot networks and ecosystem experiments can be combined to advance ecosystem forecasting under global change.
23 @ecoforecast #efi2019
Session 5: Methods and Tools Tuesday, May 14 10:45 am
S5.4: Lightning Talk
Seasonal forecast verification: from raw atmospheric data to hydrologic and lake applications
Daniel Mercado-Bettín1, Rafael Marcé, Sixto Herrera, Maialen Iturbide, María Frías [email protected]
1Catalan Institute for Water Research (ICRA)
Linking seasonal forecasting and impact models related to water quality and ecology in lakes remains a challenging task for the stakeholders and the scientific community. The current global databases of seasonal forecasting allow the identification of seasonal climate anomalies for the next months (usually, between 3 and 9 months), which influence the surface water balance in any watershed. However, a feasible and robust tool is needed to connect this available information with the hydrology of the watershed and, the water quality and ecology in downstream freshwater systems. Here we present such a tool following a work-flow from seasonal forecasting information through hydrologic modeling to lake modeling. We applied this methodology in Sau Reservoir (Spain) associated with the Ter River (~1700 km2), as a practical example, yet the tool is designed to be applied in any system, using as forcing two seasonal forecasting databases: CFSv2 and System4. The seasonal forecasting data were corrected (climate4R packages) and then used as meteorological input in the hydrologic model. We used the hydrologic model (mHMv5.9) to obtain streamflow, which was used as input in the lake model (GOTM). Finally, the lake model is forced by the streamflow and seasonal forecasting data. Due to the lack of predictability in Europe, skill among the forecasting systems used was not found in the meteorological data and hydrologic approach. In spite of that, due to the inertia associated with approximately close systems, significant skill was found in the reservoir. This skill improves as the depth increases. The outputs and analysis from this methodology are useful to identify the most pressing needs to develop usable seasonal water quality and ecological predictions.
24 @ecoforecast #efi2019
Session 5: Methods and Tools Tuesday, May 14 10:45 am
S5.5: Lightning Talk
Global Scale Crop Yield and Condition Forecasting System Using Multiple Earth Observation Datasets
Ritvik Sahajpal1, Inbal Reshef, Brian Barker, Joanne Hall, Jie Zhang, Estefania Puricelli, Mike Humber [email protected]
1University of Maryland
Earth observations (EO) can play a key role in monitoring progress made towards the achievement of the sustainable development goal of ending hunger through achieving food security and promotion of sustainable agriculture. By enabling the timely dissemination and usage of EO data, the Group on Earth Observations Global Agricultural Monitoring (GEOGLAM) crop monitor has contributed to the efficient functioning of global markets by enabling the international community to collaboratively produce operational crop condition assessments. Based on user feedback, we have developed the Global Earth Observation based Crop Yield and Condition Forecasting (GEOCIIF) system to use EO data to produce in-season alerts to assess crop yields and conditions globally for maize, soybean, rice, spring wheat and winter wheat. We have designed GEOCIIF to flexibly accommodate any number of EO data sources, with the current EO inputs including NDVI, LAI, evaporative stress index, soil moisture, precipitation and temperature. This is also the first study to use the best available crop-specific maps and crop calendars produced in collaboration with GEOGLAM partner organizations globally. By applying a suite of machine learning algorithms on EO data, GEOCIIF produces in- season crop yield forecasts and uses them to derive crop conditions by considering the varying response of each crop to abiotic factors, geography and phenological growth stage. We apply the algorithm to > 80% of global maize and soybean producing areas, > 60% of rice producing regions, and > 65% of wheat producing regions and assess performance by comparing our yield forecasts to regional scale observed yields from 2000 – 2016. In back testing, we find that our global yield forecasts have errors ranging between 3 – 5%, with the quality of crop maps playing a key role in determining forecast quality. We compare the performance of various machine learning algorithms to forecast crop yields and provide practical tips on the merits of each. Our results indicate the utility of EO data in monitoring progress towards achieving global food security by producing timely and useful crop yield and condition forecasts.
25 @ecoforecast #efi2019
Session 6: Partnerships & Public Engagement Wednesday, May 15 8:30 am
S6.1: Keynote
NOAA's Ecological Forecasting Roadmap: Lessons Learned and Future Opportunities
Lonnie Gonsalves1
1NOAA, National Ocean Service, National Centers for Coastal Ocean Science
The NOAA Ecological Forecasting Roadmap was created in 2015 in order to coordinate cross- agency efforts to develop regional-scale forecasting capabilities across the U.S. Currently, NOAA’s ecological forecasting efforts primarily focus on 4 key areas; Harmful Algal Blooms (HABs), hypoxia, habitat (predominantly marine fisheries and coral reefs), and aquatic pathogens. These efforts are highly interdisciplinary and require integration of NOAA’s expansive observation networks, data assimilation platforms, and forecast product dissemination capabilities. NOAA’s forecast products provide tailored decision support tools with specific applications to protection of public health, supporting aquaculture production, and management of natural resources. Robust engagement with regional stakeholders and partners to create tailored forecasts to meet specific stakeholder needs is a keystone of NOAA’s approach to ecological forecasting. This presentation will focus on describing NOAA’s approach to developing user- driven forecast products, address lessons learned for transitioning applied research into operational products, and highlight potential opportunities for partnership and collaboration to advance this rapidly evolving area of science. You will learn about the various specific uses for the portfolio of decision support tools that NOAA's forecasts provide and have the opportunity to discuss the future direction of these efforts.
26 @ecoforecast #efi2019
Session 6: Partnerships & Public Engagement Wednesday, May 15 8:30 am
S6.2: Lightning Talk
Opportunities for forecasting drought and restoration in dryland ecosystems
Robert Shriver1, John Bradford1 [email protected]
1U.S. Geological Survey, Southwest Biological Science Center
Drought, driven by warming temperatures and increasing precipitation extremes, is expected to become more prevalent in the 21st century. Planning for when and where drought may occur in making natural resource decisions is one of the biggest challenges for land managers in arid and semi-arid regions of the western US. Near-term ecological forecasting offers significant opportunities to anticipate and plan for weather variability in natural resource decision making, including planning restoration treatments or adjusting grazing pressure. I will outline urgent forecasting needs for natural resource management in the western US, and efforts at the USGS to apply ecological forecasting. Specifically, I will highlight results suggesting how forecasting approaches can be used to improve the timing and efficacy of restoration treatments following wildfire across shrublands in the Great Basin.
27 @ecoforecast #efi2019
Session 6: Partnerships & Public Engagement Wednesday, May 15 8:30 am
S6.3: Lightning Talk
Building capacity for applied ecological forecasting: Natural resource planning and management
Jake Weltzin1, Staff of the USA National Phenology Network National Coordinating Office [email protected]
1U.S. Geological Survey
Natural resource managers need timely and accurate information on plants, animals and habitats across a variety of spatial and temporal scales to evaluate outcomes of potential management actions, to better understand the effectiveness of management practices, and to efficiently target limited time and resources to accomplish desired management goals. For example, timely information regarding the activity, or phenological status, of species of management concern (e.g., disease vectors, invasive species, pest species) is critical for efficient planning and management (e.g., detection and control efforts of invasive species) at spatiotemporal resolutions that match decision-making (daily- and kilometer-scale). The USA National Phenology Network (USA-NPN; www.usanpn.org) is developing workflows designed to efficiently produce, distribute and validate real-time and short-term forecasts of the activity of native and invasive species of management concern, including emerald ash borer, winter moth, hemlock woolly adelgid and buffelgrass. Components of the workflow include stakeholder engagement, data collection, model development, product development, communication and validation. These particular steps in this conceptual workflow are flexible enough to guide these and other applications of ecological forecasting, with consideration of goals along a research to operations continuum. This presentation will use several new and emerging case studies under development by the USA-NPN. Based on the success and utility of these case studies to real-world resource management issues, the workflow established for generating short-term forecasts will be extended to additional applications such as human (environmental) health, wildlife disease, biological invasions, adaptive management, and restoration planning.
28 @ecoforecast #efi2019
Session 6: Partnerships & Public Engagement Wednesday, May 15 8:30 am
S6.4: Lightning Talk
Collaborating to develop actionable phenology forecast products
Katharine Gerst1, Alyssa Rosemartin1, Erin Posthumus1, Theresa Crimmins1, Ellen Denny1 [email protected]
1USA National Phenology Network
Through the collection and curation of professional and citizen scientist-derived phenology datasets, a primary aim of the USA National Phenology Network is to inform decisions across a diversity of realms, from natural resource management and conservation to public health and human well-being. The Network does this by developing and delivering quality-controlled data and products designed to improve decision-making at local to regional to continental scales. These include short-term forecast maps of heat accumulation, spring onset, and invasive and pest seasonal activity. We facilitate informed decision-making by engaging stakeholders early and often in the product development process, following best practices of trust, communication and equity. We garner feedback on products in a consultative framework, derive ideas for new products in a collaborative framework, and enable and empower partners to build their own data collection and application effort in a collegial framework. In this presentation, we share challenges and lessons learned from our experiences engaging and collaborating with key stakeholders across a broad array of applications that depend on phenological forecasting. Additionally, we demonstrate how engaging citizen scientists in data collection and validation through our “Springcasting” and “Pest Patrol” campaigns improve both community support and model evaluation and improvement. We found that one size doesn’t fit all, but instead, successful engagement is context-specific and scale-dependent. As the Network looks to the future, we identify next steps for achieving “actionable” science through implementing robust evaluation methods and fostering enduring partner collaborations.
29 @ecoforecast #efi2019
Session 6: Partnerships & Public Engagement Wednesday, May 15 8:30 am
S6.5: Lightning Talk
Data-Model Integration for Forecasting Carbon Sequestration in Coastal Wetland Soils
James Holmquist1, The Coastal Carbon Research Coordination Network’s Soils Working Group
1Smithsonian Environmental Research Center
Coastal wetlands, including marshes, mangroves and tidal forests, are among the world's most productive ecosystems. They can continuously bury carbon as sea-levels rise and they form new soil as a dynamic response to increased inundation. Forecasting carbon cycle properties of coastal wetlands can provide general assessments of wetland resilience to climate change, how plant productivity contributes to food webs, and the carbon market potential of wetland preservation and restoration. Carbon burial rates have wide uncertainties in coastal carbon accounting at the scale of the contiguous U.S. because, first, researchers tend to report carbon burial rates from dated sediment cores, which include live root and labile carbon, in addition to the slower decaying pool of organic matter that contributes to long-term carbon sequestration. Second, relevant datasets are heterogeneous, with varying time frames and inconsistent data quality surrounding 137Cs, 210Pb, and 14C dating. Third, responses of soil formation to sea-level rise can be non-linear and dynamic, which limits the utility of extrapolating recent trends to future conditions. To advance understanding of uncertainty relative to these myriad issues, the Coastal Carbon Research Coordination Network (CCRCN) convened a Soils Working Group with 17 internationally-based experts who met in December 2018 to organize data-model synthesis work focusing on these issues. On-going activities of the working group include: coding an open source R version of a commonly used wetland response to sea-level rise model, the marsh equilibrium model (MEM), incorporating MEM into a probabilistic framework and partitioning variance using a bayesian hierarchical model, generalizing model inputs based on a review of plant traits, and calibrating the model using a library of over a dozen sites and 200 dated cores synthesized from previous studies. The CCRN work will provide an example of how to quantify carbon sequestration given uncertain data that encompases, but does not fully represent, long term burial of a stable carbon pool. We also hope to develop recommendations for future research priorities for the blue carbon research community including improved consistency in collecting and analysing sediment core data, more consistent and effective monitoring of associated environmental data, and modeling improvements. This talk will discuss progress made and lessons learned as the Soils Working Group both quantifies past trends in carbon sequestration, and forecasts future changes given localized sea-level rise scenarios.
30 @ecoforecast #efi2019
Poster Abstracts
Listed alphabetically by presenting author
Tuesday, May 14 5:00 pm
P1
Exploring Satellite Observations for Ecosystem Studies
Sheekela Baker-Yeboah1, Paul DiGiacamo [email protected]
1University of Maryland
Tracking sea level rise and warming surface temperatures of the World’s largest upwelling systems in conjunction with phytoplankton chlorophyll/ PAR and surface wind conditions using multiple satellite observations can provide ongoing monitoring tools for ecosystem forecasting. In light of the upcoming release of a West Coast Operational Forecast System (WCOFS), we explore a set of regionally focused cross-sensor analysis maps for the five major upwelling regions. These products are applicable for exploring coastal transition zone linkages to our changing Large Marine Ecosystems (LMEs) in an attempt to advance decision making science applications.
31 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P2
Short-term ecological forecasting in Chesapeake Bay using a mechanistic-empirical modeling approach
Christopher Brown1, Marjorie Friedrichs2, Raleigh Hood3 [email protected]
1NOAA; 2Virginia Institute of Marine Science; 3University of Maryland, Horn Point Laboratory
The Chesapeake Bay is the largest and most productive estuary in the continental US, providing crucial habitat and natural resources for a suite of native and migratory species, and a multitude of services and opportunities for enjoyment and commerce. Yet several types of natural and human-induced changes in water quality conditions in the Bay, such as the occurrence of hypoxia and harmful algal blooms, have jeopardized its ecosystem services and economic productivity. In order to enhance the ability of management agencies to respond to and mitigate the deleterious impacts of these events, we are developing the capability to predict the timing, location, and intensity of harmful biotic events by implementing a mechanistic – empirical modeling approach. We use this hybrid approach to routinely generate daily nowcasts and 3-day forecasts of several environmental variables, such as sea-surface temperature and salinity, the concentrations of chlorophyll, nitrate, and dissolved oxygen, and the likelihood of encountering several noxious species, including harmful algal blooms and water-borne pathogens, in Chesapeake Bay for the purpose of monitoring its ecosystem. The physical and biogeochemical variables are forecast using an implementation of the Regional Ocean Modeling System in the Chesapeake Bay with fully mechanistic physical and biogeochemical components. The species predictions are generated by using the real-time output from the coupled physical – biogeochemical model to drive multivariate empirical habitat models of the target species. These predictions guide recreational, management, and research activities in the Bay and its tributaries.
32 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P3
Integrating environmental sensor networks and real-time iterative ecological forecasting to adaptively manage water quality
Cayelan Carey1, R. Quinn Thomas1, Renato Figueiredo2, Vahid Daneshmand2, Bethany Bookout1, Francois Birgand3, Heather Wander1
1Virginia Tech; 2University of Florida; 3North Carolina State University
Freshwater ecosystems around the globe are facing unprecedented levels of anthropogenic stress, resulting in increased toxic phytoplankton blooms, metal contaminants, and low oxygen concentrations that threaten water quality. To ensure safe drinking water in the face of global change, managers need real-time environmental data and ecological forecasts to detect and predict when water quality thresholds are crossed so they can act rapidly to mitigate threats. In response, we have developed a “smart water system” by embedding a secure, wireless sensor network in Falling Creek Reservoir, a drinking water reservoir in Roanoke, Virginia, USA, to improve water quality and freshwater management. High-frequency data on reservoir hydrodynamics, chemistry, and algal conditions collected by novel sensor technology are being used to create real-time water quality forecasts for drinking water managers. The sensor data update daily simulations of an open-source water quality model running in the cloud to create 16- day forecasts that will be autonomously published with digital object identifiers and searchable through the Environmental Data Initiative repository. These forecasts are being co-designed in partnership with Roanoke’s water utility to ensure that they successfully translate water quality model output into decision support tools useful to managers. Preliminary forecasts developed from the smart water system to date demonstrate that the model can successfully predict water temperature within <0.5 degrees C and the date of fall turnover, which has provided important information to managers on which depths to withdraw water from the reservoir for drinking. By developing new network computing methods for connecting distributed sensors and cloud infrastructures through virtual private networking; generating new computational methods for automated model-data fusion; and providing greater understanding of how global change and management interact to control water quality, we envision that our smart water system will serve as a prototype for ecological forecasting systems in other drinking water supply lakes and reservoirs globally.
33 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P4
Incorporating parameter estimability into model selection to select useful, predictive models
Jake Ferguson1, Mark Taper [email protected]
1University of Hawai'i
We investigate a class of information criteria based on the informational complexity criterion (ICC), which penalizes model fit based on the degree of dependency among parameters. In addition to existing forms of ICC, we develop a new complexity measure that uses the coefficient of variation matrix, a measure of parameter estimability, and a novel compound criterion that accounts for both the number of parameters and their informational complexity. We compared the performance of ICC and these variants to more traditionally used information criteria (i.e., AIC, AICc, BIC) in selecting models with the best out-of-sample predictive ability.
34 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P5
Applying Ecological Function in Environmental Decision Making
Richard Fulford1, Jim Hagy1, Marc Russell1
1US EPA
The Final ecosystem goods and services (FEGS) concept has become increasingly valuable for identifying and evaluating important trade-offs in estuarine management, yet the translation of FEGS science into policy is limited by a need for meaningful reference points that facilitate forecasting policy outcomes. A research priority for the United States Environmental Protection Agency (EPA) is to develop methods for incorporation of FEGS into decision making to protect human health and the environment. Here we present a case study application of methods to translate available science into decision thresholds based on important FEGS in estuarine systems. Thresholds in function delivery are a necessary part of making scientific information useful to decision makers as a tool for forecasting impacts. Such thresholds can be defined based on functional equivalency (FE) of ecosystem components. For any given FEGS, we can identify ecosystem components that contribute to its production and then define that contribution as a target function. When management decisions are predicted to result in change in FEGS production, the function has changed. A threshold can be defined beyond which a change in function results in a loss of FEGS production enough to say the ecosystem is no longer functionally equivalent to the desired state. Decision makers need both a measure of the reduction in function, as well as the meaningful reference point to make use of scientific information. Forecasting changes in FE is well established in fishery management and has been linked implicitly to resource sustainability. In the broader ecological case, the FE paradigm is used but not well defined as an operational concept. Our goal is to provide a conceptual map for the FE paradigm to questions of habitat management and nutrient load management and explore meaningful and measurable thresholds for forecasting change in coastal estuaries.
35 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P6
TERN Australia
Mark Grant1
1TERN Australia
TERN is Australia’s land ecosystem observatory.
TERN observes, records and measures critical terrestrial ecosystem parameters and conditions for Australia over time from continental scale to field sites at hundreds of representative locations.
This information is standardised, integrated and transformed into model-ready data, enabling researchers to discern, interpret and predict changes in land ecosystems.
Understanding ecosystem change, the rate of change, and underlying causes is essential for effectively protecting and managing Australia’s environment and the many services it provides.
36 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P7
Identifying spatial and temporal drivers of waterborne pathogen dynamics
Tao Huang1 [email protected]
1Cary Institute of Ecosystem Studies
Pathogens are a significant source of water quality impairment. Developing a model to predict the spatial and temporal of pathogen concentration is critical to estimate waterborne disease risks. Most studies focuses on Escherichia coli, the indicator for fecal contamination. Watershed characteristics (e.g., drainage area) and hydrologic factors (e.g., runoff) are important predictors for pollutant concentrations, but their effects on pathogens have not been well studied. The aim of this study is to apply machine learning for identifying important predictors of pathogen concentration across flow conditions. This study analyzed the spatial and temporal dynamics of E. coli and other pathogens, including Cryptospirosis, across the United States. Machine learning techniques, including random forests, boosted regression tree, and logistic regression were applied. This study found that the storm runoff ratio index, derived from hydrographs, is a better predictor than precipitation and runoff to characterize hydrologic impact on pathogen concentrations. Watershed characteristics, including urbanized area and drainage area, are important predictors to explain the spatial variation of pathogen concentrations. These results suggest that watershed hydrologic factors are critical to pathogen concentration forecasting.
37 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P8
Forecasting water quality in a drinking water reservoir: an ensemble model approach
Chris Jones1, Pest or Pathogen Forecasting Project1 [email protected]
1NC State University, Center for Geospatial Analytics
Invasive insects and pathogens are a significant threat to forested and agricultural ecosystems worldwide. However, dynamic species distribution models (SDMs) are frequently black boxes and not readily available for quick use by decision makers. We have built a new simulation framework and software called PoPS (Pest or Pathogen Spread) Model - an open source dynamic SDM that includes the necessary functions for data preparation, calibration, and validation. The software is available in multiple programming languages (R and Grass GIS) or via a web-based interface. We are co-developing the model with the USDA Animal and Plant Health Inspection Service (APHIS) and iteratively testing it with real world management plans, making changes to the model, and adding new decision analytics based on stakeholder feedback. We are proposing to build off of the success of this collaboration by combining this model and APHIS’s current detection network with (i) real-time in field sensors measuring pest abundance in fields, (ii) minituratized loop mediated isothermal amplification (LAMP) assays based on a smartphone sensor and application, (iii) swine movement and infection data from the Swine Health Information Center, and (iv) citizen science based detections from iNaturalist. For initial prototyping of the complete system, we are focused on 4 major pest and pathogen systems with stakeholders/decision makers already on board to use and implement management based on the results of these forecasts: 1. Porcine Epidemic Diarrhea Virus (PEDv) in swine across North Carolina, 2. Spotted Lanternfly across the Northeast US in partnership with USDA APHIS, 3. Phytophthora infestans and Phytophthora nicotianae in tomatoes and potatoes in North Carolina in partnership with extension agents and growers, and 4. Phytophthora ramorum in Oregon in partnership with the Oregon Department of Forestry. The modeling platform is currently being used in two (P. ramorum and spotted lanternfly) of the four systems, the other two are new partnerships based on new and emerging technology. Here we layout a diagram of our plan and are looking for feedback and advice from others that have previously implemented iterative automated forecasts of a similar nature.
38 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P9
The Art and Science of Databasing: Synthesis and Archival for Forecasting by the Coastal Carbon RCN
David Klinges1, James Holmquist1, Patrick Megonigal1, Michael Lonneman1 [email protected]
1Smithsonian Environmental Research Center (SERC)
The Coastal Carbon Research Coordination Network (CCRCN) is a consortium of biogeochemists, ecologists, pedologists, and coastal land managers with the goal of accelerating the pace of discovery in coastal wetland carbon science by providing our community with access to data, analysis tools, and synthesis opportunities. We are accomplishing this goal by facilitating the sharing of open data and analysis products, offering training in data management and analytics, and leading topical working groups aimed at quantitatively reducing uncertainty in coastal greenhouse gas emissions and storage. Here, we will provide insight on the databasing methodologies of the CCRCN: our systems for approaching the sometimes awkward situations of data outreach/collection (i.e. better approaches than “send me your data please”), data curation, and data storage. This will inform others performing data synthesis-- an integral component of ecological forecasting-- at various scales, and will encourage best practices of data networking and enhanced collaboration. As some of the primary scientific users of our database, we will also outline the suite of questions that the CCRCN is actively engaged in addressing (e.g., forecasting rates of carbon storage and sequestration of coastal wetlands across the contiguous United States), as well as other research aims that can be taken using this rich dataset. Our goal is to encourage EFI 2019 participants to engage in data exchange in a collaborative and progressive manner in order to improve our scientific practices.
39 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P10
Using Metapopulation Modeling to Understand and Predict the Spread of an Emerging Infectious Disease
Kristjan Mets1
1Stony Brook University
The emergence of a new pathogen can cause the decline or even extinction of a species. Preemptive surveillance of potential disease reservoirs provides information about the possible trajectory of a disease, but in many cases, it is expensive and impractical for managers to adequately survey all possible spread routes. Relating physical landscape structures to pathogen incidence has the potential to estimate the risk of disease spread on a broad geographic scale and identify specific paths in a landscape that a pathogen can take. Despite the promise of these spatial models, the need for information on the vital rates of host species and lack of a means to test and evaluate model results challenge their application. White-nose syndrome, an emerging fungal disease in multiple species of North American bats, presents a clear test case for the use of predictive disease spread models. This project combines metapopulation modeling with SIS disease spread modeling to provide insight into ways that landscape structure and the spatial position of populations impedes or promotes the spread of an emerging infectious disease. The model is applied to cases within the Ozark plateau region as well as broadly throughout North America. This project seeks to identify features of populations at greater risk of spreading disease and to inform decision-making by managers.
40 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P11
Development of state-space Bayesian models for near-term forecasts of phytoplankton blooms in a large, north temperate lake
Mary Lofton1, Whitney Beck2, Ruchi Bhattacharya3, Jennifer Brentrup4, Ludmila Brighenti5, Sarah Burnet6, Ian McCullough7, Simon Stewart8, Jacob Zwart9, Cayelan Carey1, Kathryn Cottingham4, Shannon LaDeau10, Kathleen Weathers10 [email protected]
1Virginia Tech; 2Colorado State University; 3University of Missouri – Columbia; 4Dartmouth College; 5UEMG; 6University of Idaho; 7Michigan State University; 8University of Waikato; 9U.S. Geological Survey; 10Cary Institute
Phytoplankton blooms alter lake ecosystems through the production of surface scums, toxins, and taste and odor issues, and their decomposition may deplete oxygen in the water. Near-term forecasts of phytoplankton blooms would help lake managers preemptively manage water quality, allow waterfront property owners to plan for bloom events, and provide advance warning to the public of potential recreational water closures. However, blooms are notoriously challenging to predict, often arising or disappearing within days or even hours and varying spatially within a lake. Our project, which includes a collaboration among members of the Ecological Forecasting Initiative (EFI), the Global Lake Ecological Observatory Network (GLEON), and the Lake Sunapee Protective Association (LSPA), aims to produce near-term forecasts of phytoplankton densities in Lake Sunapee, NH, USA, a clear-water lake that experiences blooms of Gloeotrichia echinulata, which causes unsightly surface scums. We created and assessed the efficacy of Bayesian state-space models to predict density of G. echinulata by calibrating and validating candidate models using subsets of a 14-year, ongoing time series of weekly G. echinulata densities collected during the summer months at each of four sampling sites in Lake Sunapee. We compared candidate models to a Poisson random walk using one-week ahead predictions and quantile plots and evaluated covariates inclusion using DIC. Candidate model covariates included wind direction, site differences, seasonality and temperature. Our priors were informed by studies from other lakes in the northeast U.S. that experience G. echinulata blooms. Preliminary uncertainty partitioning of the predictive intervals for an initial model revealed that both observation error and process error constituted substantial proportions of total prediction uncertainty, suggesting that we need more frequent G. echinulata observations and should explore further candidate models in order to improve models for near-term bloom prediction. Our collaborative partnership between EFI, GLEON, and LSPA advances methods and tools for producing near-term predictions and measures of uncertainty for an important ecological phenomenon in aquatic systems.
41 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P12
National-extent water temperature prediction at the U.S. Geological Survey
Samantha Oliver1, Jordan Read1, Alison Appling1 [email protected]
1U.S. Geological Survey
Temperature is an important factor for many processes in aquatic systems, from metabolism to evapotranspiration to greenhouse gas emissions. In many instances, the need for temperature data extends to times or places where observations do not exist. Water temperature predictions (hindcasts, forecasts, and projections) are therefore important for a wide range of applications and stakeholders. The U.S. Geological Survey is working to provide national-extent water temperature predictions, leveraging its vast network of water temperature observations and its expertise and infrastructure for modeling of hydrologic processes at the national extent. The near- term goals (1-3 years) of the water temperature prediction project at the USGS are to 1) inventory and assemble the nation’s water temperature observations, 2) engage stakeholders to determine what water temperature products are of highest value, 3) create a simple but national-extent baseline model to measure against more complex models and to prototype workflows for operationalizing predictions, 4) assess an existing national-extent mechanistic water temperature model, 5) use process-guided machine learning algorithms to improve process models and reduce the need for site-specific parameterization, and 6) identify additional processes and drivers whose inclusion will improve our ability to predict temperature in headwater systems. These objectives will serve the long-term goal of providing national-extent temperature forecasts for the continental United States.
42 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P13
Echo state network forecasting of animal movement trajectories
Toryn Schafer1, Christopher Wikle1
1University of Missouri
The ability of linear models to forecast multivariate time series is limited for inherently non-linear systems. Trajectory data, such as describe animal movement, represent a non-linear system of interest to ecologists. Improving forecasts of individual movement and providing measures of forecast uncertainty increase our ability to manage populations affected by changes in the landscape such as due to climate change, habitat loss and fragmentation. Movement data typically are prone to missingness and therefore, good prediction models should be able to accurately fill in the data gaps. In addition, as datasets get larger, computationally efficient methods are necessary. The echo state network is a type of machine learning recurrent neural network with parsimonious parameterization of dynamics, but it provides accurate forecasts in spatio-temporal settings. The echo state network learns non-linear dynamics through a reservoir computing, which is very computationally efficient. Realistic uncertainty bounds can be accounted for through an ensemble parametric bootstrap implementation.
43 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P14
Contributions of structural and parametric uncertainty to centennial-scale projections of community succession in Upper Midwest temperate forests
Alexey N. Shiklomanov1, Jeff Atkins2, Christopher M. Gough2, Ben Bond-Lamberty1 [email protected]
1Joint Global Change Research Institute, Pacific Northwest National Laboratory; 2Virginia Commonwealth University
The value of representing vegetation demography in terrestrial ecosystem models is increasingly apparent. Demographic processes are closely linked to biogeochemistry, and models that ignore demography often fail to accurately predict the responses of biogeochemical fluxes to disturbances. Moreover, demographic models allow researchers to explore questions about ecosystem composition and structure that are often just as, if not more, relevant to policymakers and stakeholders than questions about the carbon cycle. However, explicitly representing plant demography in models comes at a cost: Compared to their simpler "big-leaf" counterparts, demographic models have higher data requirements and are much more computationally intensive, and it is unclear whether this cost is worth the benefit of demographic realism. In this work, we explore this topic in the context of the ecological succession of plant communities. Our guiding research questions are: (1) What processes are most important to accurately model community succession of Upper Midwest temperate forests? (2) What are the largest drivers of uncertainty in model predictions, and what measurements should we prioritize to reduce those uncertainties? To answer these questions, we explore community succession in Upper Midwest temperate forests using the Ecosystem Demography v2 (ED2) model run at sites near the University of Michigan Biological Station (UMBS). To assess structural uncertainty, we run ED2 with different formulations of key processes important to succession. Specifically, we run ED2 with the factorial combination of two different submodels of canopy structure (complete shading of shorter trees by taller ones vs. partial shading through a finite canopy radius model) and four forms of N limitation (just photosynthesis, just respiration, both, and neither). To assess parametric uncertainty, we run ED2 ensembles that sample over parameter uncertainties (based on existing trait data and best estimates in the literature). We then compare the magnitude of structural (across submodels) and parametric (within each submodel) uncertainty, and assess which (if any) of the submodel-parameter combinations could reproduce observed C fluxes and 100-year succession at UMBS sites.
44 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P15
Bayesian Parameter Estimation for Ecosystem State Space Models with Linear Autoregressive Process Models
John Smith1 [email protected]
1Virginia Tech
Our society depends greatly on services provided by ecosystems. Forest ecosystems, for example, help to store carbon dioxide, provide timber, and regulate water cycling. Our understanding of ecosystems is instrumental in planning for the future. With the high cost of planning and executing experiments, the primary method of understanding ecosystems has shifted towards a model driven approach that is calibrated by experimental data rather than purely driven by experimental data. The question that we ask about ecosystems are intrinsically about the future and therefore it is important that we are able to predict how these ecosystems will evolve over time, with uncertainty, contingent on scenarios that we expect to see. This is the premise of the field of Ecological Forecasting (EF).
Latent variables are ubiquitous in EF problems. This makes the state space model one of the most powerful techniques in EF, for its flexibility in handling many forms of latent variables - such as random and systematic observation errors, missing data, unobserved variables, and proxy data. However, the ecological community has not fully adopted the state space framework, due to its computationally intensive nature and the inflexibility of pre-packaged MCMC software like JAGS and STAN to handle large ecosystem models. To address this challenge, we outline a general method for fitting these state space models in the Bayesian framework using MCMC, with tractable latent state full conditional distributions derive for a large class of process models, namely autoregressive process models that are linear at each time step. Focusing on carbon exchange models for forests, we assess the performance of these methods vs. standard approaches for several different models, for both computational time, accuracy of estimated parameters, and ability to operate in a range of observation gap sizes.
45 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P16
Developing short-term forecasts of marine mammal distributions in the Northeast United States
Julia E.F. Stepanuk1, Carolina Chong-Montenegry1, Janet A. Nye, Hyemi Kim, Jason J. Roberts2, Pat N. Halpin2, Debra L. Palka3, Ann Pabst4, William A. McLellan4, Susan G. Barco5, Lesley H. Thorne1
1Stony Brook University; 2Duke University; 3National Marine Fisheries Service; 4University of North Carolina – Wilmington; 5Virginia Aquarium and Marine Science Center Foundation
Large whale populations on the Northeast coast of the United States are impacted by anthropogenic activities such as ship strikes, entanglement in fishing gear, and offshore energy development. Understanding and predicting the distribution of these species is necessary to develop effective mitigation and management strategies. Here we build upon recently developed habitat models for large whales by: 1. incorporating prey distributions as covariates into predictive models for humpback whales (Megaptera novaeangliae) and fin whales (Balaenoptera physalus) in the Northeast U.S.; and 2. using forecasts from climate models to generate predictive spatial maps for these species. Recently developed extended-range hindcasts/forecasts (e.g., SubX) provide the opportunity to predict distributions of cetaceans and their prey on sub-seasonal time scales (lead times of days to weeks). Direct measurements of prey availability are not typically available when developing predictions of cetacean distribution, and as a result cetacean habitat models typically rely on environmental covariates such as temperature, salinity, bathymetry and oceanographic features, as proxies of prey distribution. While a small number of previous studies have modeled marine mammal abundance as a function of prey availability on small spatial scales, large-scale studies examining the distribution of marine mammals as a function of prey availability are lacking. We use data from standardized marine mammal line transects surveys (1992-present) to generate presences and pseudo-absences for humpback and fin whales and use data from the Northeast Fisheries Science Center’s bottom trawl survey (1992- present) to predict prey distributions. We then use these data to develop generalized additive models (GAMs) for humpback whales and fin whales, respectively, based on both environmental covariates and predicted prey distributions. Humpback whale GAMs that incorporate both environment and prey covariates perform similarly compared to GAMS developed from environmental covariates alone, but fin whale GAMs developed from environmental covariates alone perform better than models that incorporate both prey and environmental covariates. Lastly, we use output from the SubX sea surface temperature forecasts to generate probabilistic predictions of forage fish, humpback whale and fin whale distributions. We present forecasts with lead times of 5 to 30 days along with estimates of forecast skill for the northeast US. This work represents a first step towards the development of a sub-seasonal forecast for large whale distributions. These forecasts would provide fishermen and managers with information about times and areas where whales are likely to occur (i.e., high risk areas for ship strikes and entanglements).
46 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P17
Assessing stakeholder views of fisheries vulnerability: Applications for a decision support tool
Gretchen Stokes1, Jesse Wong2 [email protected]
1University of Florida; 2George Mason University
Inland fisheries are important contributors to global food security and poverty alleviation. However, unlike marine fisheries, there is no standardized method to monitor and assess the status of inland fisheries and thus far, global assessments of inland fisheries have not been feasible. Without assessments, it remains difficult track the effects of climate change, land use and human development on fisheries and prioritize conservation areas. The creation of an assessment method and effective decision support tools relies on stakeholder input. This study utilized participatory mapping and a questionnaire survey to assess stakeholder perceptions of threats to fisheries and potential application tools, using Lake Malawi as a case study due to its high importance of inland fisheries resources. The main objectives were to: 1) Determine the primary human, environmental and governmental threats to fishers, 2) evaluate the relationship between threats and barriers to effective management, and 3) determine the best fit interface for a risk assessment tool for Lake Malawi. A survey of fishers, fisheries managers, scientists and law enforcement officers (n = 26) representing non-profit, governmental and academic organizations was conducted and results suggest preferences toward an interface with the most up-to-date scientific data and focal selection of geographic fisheries areas but as an in-person format rather than digital technologies. Results will inform the creation of an assessment tool, contribute to a greater understanding of trends in global inland fisheries and aid in identifying priority hotspots for improved fisheries and environmental management.
47 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P18
Providing sustained delivery of scenario-based ecological forecasts
Beth Turner1, Mark Monaco1, Lonnie Gonsalves1, David Scheurer1 [email protected]
1NOAA National Centers for Coastal Ocean Science
Ecological modeling may be used to make predictions and analyses across a range of time and space scales from forecasts and hindcasts to scenario-based projections. Examples from shorter- term forecasts include predicting the time and location of landfall of a harmful algal bloom based on movement of an observed bloom, or forecasting the seasonal extent of hypoxia based on observed nutrient loads. At the other end of the spectrum, scenario-based projections provide estimates of how a system might respond to changing environmental (e.g. climate) or anthropogenic factors. Examples include forecasting changes in phytoplankton biomass or bottom water dissolved oxygen based on a range of potential nutrient management options. While these various approaches to ecological modeling efforts are related, their scale, scope, and applications – and therefore their requirements for delivery to managers, policy makers and scientists – can differ considerably.
Scenario-based ecological forecasts are often implemented in an adaptive management context or to address specific management questions. Therefore, forecasts may be delivered as a one-time product, annually, or on longer time scales as part of periodic assessment products. These differing delivery methods require an operational framework that will allow for sustained delivery with longer periodicity compared to forecasts provided on a regularly recurring basis (daily, sub- weekly, weekly, monthly).
This poster will review some of the challenges that NOAA has encountered in providing for long- term sustained delivery of scenario-based forecasts. Results will be presented from a survey of coastal manager end users. Case studies from NOAA forecasts will be reviewed, and recommendations for how to address some of the challenges will be introduced. We welcome a wider discussion around these ideas.
48 @ecoforecast #efi2019
Poster Session Tuesday, May 14 5:00 pm
P19
Forecasting water quality in a drinking water reservoir: an ensemble model approach
Whitney Woelmer1, Bethany Bookout1, Mary Lofton1, Ryan McClure1, R. Quinn Thomas1, Cayelan Carey1
1Virginia Tech
Lakes and reservoirs are increasingly threatened by eutrophication, a result of rapidly changing land use and climate. Consequently, there is a growing need to assess the current and future state of freshwater ecosystems by adopting iterative, near-term forecasting. Because the field of ecological forecasting is relatively new, there is not yet a consensus as to the best approach for predicting future water quality. For example, is it better to develop forecasts using empirical models based on historical time series data, or complex process-based models that integrate many variables but require extensive calibration? To assess these two distinct approaches, we compared an autoregressive integrated moving average (ARIMA) model developed using a suite of physical, chemical, and biological monitoring data and the General Lake Model (GLM), a highly parameterized one-dimensional hydrodynamic model for Falling Creek Reservoir (FCR), a drinking water reservoir in Vinton, VA, USA to hindcast chlorophyll-a during the past four years. Both models were then used to produce near-term (16-day) forecasts of chlorophyll-a using the Forecasting Lake and Reservoir Ecosystems (FLARE) framework. Both models yielded chlorophyll-a hindcasts and forecasts that generally captured observed chlorophyll-a dynamics. Our ARIMA model included discharge rates to the reservoir and shortwave radiation and hindcasted chlorophyll-a over 4 summers with an R2 of 0.44 and RMSE of 1.71 µg/L. In comparison, GLM, which included over 20 driver datasets and 500 parameters, hindcasted chlorophyll-a over 5 years with an R2 of 0.15 and RMSE of 3.42 µg/L. While GLM exhibited poorer performance than the ARIMA model, it modeled chlorophyll-a dynamics during time periods when driver data for the ARIMA model were unavailable. When applied to the FLARE framework, both models produced near-term iterative forecasts where the dominant form of uncertainty varied both by model type and through time. Ensemble forecasting allows us to compare the success of the two models at forecasting water quality, while also providing insight into the advantages and disadvantages of using empirical vs. numerical simulation techniques. For example, empirical models require only a few commonly available parameters; in comparison, numerical simulation models are data-hungry but provide information about the mechanisms underlying chlorophyll-a variability. Our research provides valuable information on how best to scale forecasting approaches from one waterbody to lakes and reservoirs globally.
49