The 2023 EFI Unconference, hosted by the Ecological Forecasting Initiative Research Coordination Network (EFI RCN) and supported by the National Science Foundation, brought together 45 passionate individuals at the National Ecological Observatory Network (NEON) headquarters in Boulder, Colorado on June 21-23, 2023 to work on a diverse range of projects that were nominated and selected by the participants. With a focus on collaborative problem-solving, the Unconference fostered a unique environment for participants to exchange knowledge, generate new approaches, and advance the field of ecological forecasting.
In addition to project development, activities included a warm welcome from Kate Thibault, NEON Science Lead, icebreaker activities, expertly facilitated by Cayelan Carey from Virginia Tech that helped participants connect and form meaningful relationships, a tour of NEON facilities, and a poster session and social hour, where participants showcased their research and projects. Through these activities, Unconference participants and NEON staff were able to engage with one another, exchange feedback, and forge new collaborations.
To ensure a productive and focused Unconference, participants engaged in a review of project ideas and subsequent project selection. This process allowed attendees to propose projects aligned with their interests and expertise and fostered a sense of ownership and investment in the outcomes. Ten project groups developed out of the 24 that were initially proposed as part of the pre-meeting preparation.
Summaries provided by each project working group are listed below. Some groups will provide additional details in forthcoming blog posts, so be sure to watch for those future posts.
This was the first in-person EFI event since 2019 and it was absolutely lovely to be in the same room to meet new people and to see in-person people we had only seen on Zoom before. We appreciate the Unconference participants’ willingness to share their time, talents, and perspectives. As you will read below, there were a number of accomplishments over the three days of the meeting and we look forward to seeing future outcomes from what was developed at the Unconference!
List of Projects
- Spatially Explicit Forecasting
- Forecast Uncertainty
- Forecasting Impacts: Measuring the Current and Future Impacts of EFI
- Reenvisioning the NEON Ecological Forecasting Challenge Dashboard Visualization
- Transporting Models Between NEON and Non-NEON Systems
- ML-based Uncertainty in the NEON Ecological Forecasting Challenge
- Forecasting Ground Beetles: Avoiding Pitfalls
- Towards Principles for Designing Inclusive Ecological Forecasts
- A proactive step toward decision-ready forecasts: Fusing iterative, near-term ecological forecasting and adaptive management
- Disease Forecasting
Spatially Explicit Forecasting
Participants: John Smith, David Durden, Emma Mendelsohn, Carl Boettiger
To date, the NEON Ecological Forecasting Challenge has been focused on generating near term forecasts for specific sites. However, many interesting ecological phenomena occur across both time and space. At the EFI 2023 Unconference, our group prototyped a forecasting challenge that is also spatially explicit. For our prototype forecasting challenge, we focused on Leaf Area Index (LAI) recovery in post-burn areas. Our focal sites so far include the California August complex fire and the Colorado East Troublesome fire. Our work at the Unconference focused on building cyber-infrastructure to ingest and aggregate data, build target files, assess models using proper scoring rules, and build baseline climatological forecasts. Current progress, including an example notebook and a detailed workflow diagram, are available on GitHub: https://github.com/eco4cast/modis-lai-forecast/. Current and future work includes building additional baseline models, setting up a submission portal using GitHub actions, and integrating additional sites to include a variety of ecoclimatic domains.
Participants: Noam Ross, Eli Horner, Ashley Bonner, Mike Dietze, Chris Jones
Interest and use of ecological forecasting have increased in recent years due in large part to the efforts of EFI, including the NEON Ecological Forecasting Challenge. However, only a small percentage of ecological forecasts published have fully quantified and partitioned their forecast uncertainties. Quantifying and validating model predictions and uncertainties allows for understanding the degree of uncertainty in forecasts and how much we understand the underlying ecological system (our ability to predict them). Partitioning forecast uncertainties allows for increased focus on data collection efforts that could lead to improved model performance and reduction in uncertainty. Our group worked toward creating a tutorial for how to quantify and partition forecast uncertainties and validate model predictions with uncertainty by using the NEON Phenology Forecasting Challenge. We are using an ARIMA model and a random forest model as examples. During the Unconference we were able to get both models working and partition uncertainties. We are finishing up the code base, tutorial, and discussing challenges with each type of model when it comes to performing uncertainty quantification and partition.
Forecasting Impacts: Measuring the Current and Future Impacts of EFI
Participants: Rebecca Finger-Higgens, Jessica Burnett, Alexis O’Callahan, Ayanna St. Rose
It turns out getting-to-know-you style ice breakers can provide more than just a few new friends, they can also demonstrate group priorities and motivations for coming together in the first place. On a sunny morning at the NEON Headquarters in Boulder, CO, Cayelan Carey (Virginia Tech) asked the group of EFI Unconference participants to organize themselves based on whether they individually felt that the goal of forecasts were for understanding or decision making. As the participants shuffled around and considered the question before them, the final results revealed a pattern among the group that resembled a skewed desire for forecasts to inform decision making versus broadening the understanding of ecological systems. However, the ability of ecological forecasts to effectively inform decision making has not clearly been measured. Besides directly impacting decision making processes, how do we, as a grassroots organization, recognize and measure the other societal impacts that EFI might be, or capable, of creating?
This led our group to think through ways that EFI could measure impacts, to ask: what are the impact goals and achievements of EFI, what does the community want out of EFI, and what is the best way to measure these often hard to measure metrics? Using five categories of societal impacts (instrumental applications, connectivity impacts, conceptual impacts, capacity building, and socio-ecological impacts), we developed a poll for Unconference participants to assess the priorities and current thoughts of this representative group. The poll results suggest that EFI community goals emphasize conceptual impacts (i.e. improve ecological understanding), connectivity impacts (i.e. maintaining and developing community and partnerships) and instrumental applications (i.e., applications for decision making). We also found that EFI has made the greatest advancements in capacity building (i.e., curriculum development, short courses), conceptual impacts (i.e., working groups), and connectivity impacts (i.e., newsletters and conference sessions). These discoveries have allowed us to identify a space for the creation of a concrete link between the connectivity of forecasting and the desired application outcomes of the group. It has allowed us to develop a number of recommendations for the steering committee and the EFI community. Some of these recommendations include focusing on if, how, and why a created forecast product achieves one of the five predefined societal benefits. Together, we hope to continue to build on the vision statement of EFI to build forecasts to understand, manage, and conserve ecosystems in a measurable and remarkable way.
Figure 1: Word cloud generated from Unconference participant responses to the question “describe the potential, importance, or value of the community of EFI”.
Reenvisioning the NEON Ecological Forecasting Challenge Dashboard Visualization
Participants: Melissa Kenney, Michael Gerst, Toni Viskari, Austin Delaney, Freya Olsson, Carl Boettiger, Quinn Thomas
With the growth of the NEON Ecological Forecasting Challenge, we have outgrown the current Challenge Dashboard, which was designed to accommodate a smaller set of forecasts and synthesis questions. Thus, we have reenvisioned the next stage NEON Forecast Challenge Dashboard in order to facilitate the ability to answer a wider range of questions that forecast challenge users would be interested in exploring. The main audience for this dashboard is NEON forecasters, the EFI community, Forecast Synthesizers, and students in classes or teams participating in the NEON Ecological Forecasting Challenge. Given this audience, we have identified 3 different dashboard elements that will be important to include:
- forecast synthesis overview,
- summary metrics about the Forecast Challenge, and
- self diagnostic platform.
Transporting Models Between NEON and non-NEON Systems
Participants: Brendan Allison, Olufemi Fatunsin, & Jeff Mintz
A community of practice is increasingly active in developing models and forecasts for NEON sites. We asked: how can we take models trained on NEON data and refine them for use in another context? Similarly, how can we take models trained on non-NEON data and refine them on NEON data? This goal of transplanting models can empower a range of applications, including local field studies, adaptive management, and data fusion from multiple monitoring networks, enabling greater statistical power for big ecological questions. Whether transporting a model to or from NEON, the challenges are effectively the same. These included unbalanced data, different monitoring protocols, different predictors, and different site selection criteria. To focus our efforts, we picked the particular case study of bringing together NEON vegetation survey data with similar datasets generated under the Forest Inventory Analysis (FIA) program. Our first product was the development of a Bayesian multilevel model with the capacity to scale to the integration of multiple sets of continental or global-scale monitoring networks, or shrink to the job of predicting outcomes at a single site, but informed by a shared global layer. With this case study in mind, we have been building a codebase for processing the relevant NEON and FIA forestry data and for joint modeling of residual error across monitoring systems in Stan, a popular probabilistic programming language. Find more details about this project in this blog post.
ML-based Uncertainty in the NEON Ecological Forecasting Challenge
Participants: Marcus Lapeyrolerie, Caleb Robbins
How can machine learning (ML) provide a solution to estimating forecast uncertainty across NEON Ecological Forecasting Challenge? We generated a proof-of-concept workflow combining two machine learning approaches to make probabilistic forecasts. Random forests were used to learn relationships between forecast challenge variables and past NOAA weather data and to make predictions. While these models were able to make forecasts that perform well in approximating the future target time series, they were not implemented to provide estimates of uncertainty. We explored how we could use past data along with these deterministic forecasts to generate probabilistic forecasts. Our approach was to train another machine learning model to make probabilistic forecasts on the residual errors from the previous Random Forest models. We then used these predicted residual error forecasts to modify the Random Forest-based forecasts. This combined approach holds potential as it could be used in a plug-n-play manner, where this method could correct the deterministic (or even probabilistic) forecasts from any model to account for temporal trends in the residual error and provide uncertainty estimates. In our next steps, we will work on creating an automated workflow to generate residual error forecasts for the Eco4Cast challenge.
Forecasting Ground Beetles: Avoiding Pitfalls
Participants: Eric Sokol, Glenda Wardle, Vihanga Gunadasa, Juniper Simonis, Alison Gerken, Meghan Beatty, Khum Thapa-Magar
Ground beetles are a versatile species with which to measure biodiversity, yet they lack behind other EFI NEON Ecological Forecasting Challenge themes in terms of forecasts and models. Our group at the Unconference wanted to figure out why forecasters were not submitting to the NEON Ecological Forecasting Challenge Beetle Communities Theme and how we could remove those barriers to increase forecast submission. We created a tutorial (in progress) that describes general goals for forecasting ecological communities, a how-to on submitting a forecast, some of the challenges in forecasting ecological community data, and examples of forecasts people might submit to begin to address those challenges. We first reviewed the underlying data structure of the pre-made targets file that had been developed for the forecasting challenge. We then combined currently available code for a null model, an ARIMA model, and an available tutorial for working with data from the Aquatics Challenge into a workable tutorial to prepare and submit forecasts to the Beetle Challenge. Our goal is to finalize the tutorial by adding a random walk model and more detail on how to add additional covariates to the model, including climate variables. We are also designing a new targets file that has different variables of interest at finer spatial scales at a given NEON site (e.g. plot or habitat information, survey effort). The beetles community data provides an example of when patterns in non-continuous or seasonal data may be poorly capture by a simple model (e.g., ARIMA). When there is latency or gaps in the data more data processing is often required than when using continuous sensor-captured data. Knowing the experimental design is also critical to be able to design a model to build understanding. We hope that this tutorial increases overall interest in submitting forecasts to the beetle forecasting challenge and removes barriers that may prevent forecasters at all levels from submitting. Further information and development on community ecology and biodiversity data is critical for understanding many different biological systems, can help researchers broaden their understanding of how and why communities change over time, and can better provide decision-making tools for ecosystem monitoring.
Towards Principles for Designing Inclusive Ecological Forecasts
Participants: Anna Sjodin, Mary Lofton, Sean Dorr, Jody Peters, Jason McLachlan, Cazimir Kowalski, Melissa Kenney, Katie Jones
Our group is interested in exploring opportunities for improving inclusivity in ecological forecasting. Through discussion, we identified the ten principles of Design Justice (Box. 1) as a potential mechanism for evaluating the inclusivity of forecast products, services, and systems.
Box 1: Design Justice Network Principles, reproduced from https://designjustice.org/read-the-principles, license CC BY-ND 4.0.
To learn more, please visit the Design Justice Network website (https://designjustice.org) or see Design Justice: Community led practices to build the world we need by Sasha Costanza-Chock (https://designjustice.mitpress.mit.edu/;
open access pdf version is available here: https://library.oapen.org/bitstream/handle/20.500.12657/43542/1/external_content.pdf).
|1||We use design to sustain, heal, and empower our communities, as well as to seek liberation from exploitative and oppressive systems.|
|2||We center the voices of those who are directly impacted by the outcomes of the design process.|
|3||We prioritize design’s impact on the community over the intentions of the designer.|
|4||We view change as emergent from an accountable, accessible, and collaborative process, rather than as a point at the end of a process.|
|5||We see the role of the designer as a facilitator rather than an expert.|
|6||We believe that everyone is an expert based on their own lived experience, and that we all have unique and brilliant contributions to bring to a design process.|
|7||We share design knowledge and tools with our communities.|
|8||We work towards sustainable, community-led and -controlled outcomes.|
|9||We work towards non-exploitative solutions that reconnect us to the earth and to each other.|
|10||Before seeking new design solutions, we look for what is already working at the community level. We honor and uplift traditional, indigenous, and local knowledge and practices.|
As a first step towards applying the Design Justice principles to EFI-created products, services, and systems, our team evaluated to what degree the ten principles were evident in the design of the NEON Ecological Forecasting Challenge. We identified several ways in which the design of the Challenge was well-aligned with Design Justice Principles (e.g., Principle 4: We view change as emergent from an accountable, accessible, and collaborative process, rather than as a point at the end of a process.), as well as areas in which we thought we could improve (e.g., Principle 3: We prioritize design’s impact on the community over the intentions of the designer).
Moving forward, we are soliciting broader participation from all EFI community members in small focus groups to continue our internal evaluation of the inclusivity of current EFI products, with the ultimate goal of furthering the inclusivity of ecological forecasting by developing recommendations towards a more complete alignment of EFI-designed products with design justice principles. If you are interested in participating in such a focus group, please provide your contact information in the Google Form linked here.
A proactive step toward decision-ready forecasts: Fusing iterative, near-term ecological forecasting and adaptive management
Participants: Jaime Ashander, LM Bradley, Mark Buckner, Nathan Byer, Cayelan Carey, Michael Gerst
This group aimed to improve the conceptual tools for co-production of ecological forecasts that aid in decision making. We identified that there is a need for tighter conceptual integration of the iterative, near-term ecological forecasting cycle (as practiced by the EFI community) with the adaptive management cycle (as practiced by communities of natural resource managers) and the broader context for management decisions. While prior frameworks have treated the iterative, near-term forecasting and adaptive management cycles as independent, with limited points of contact, a careful fusion of these processes may increase conceptual utility for co-production. As a first step towards a more useful framework, we then located iterative, near-term forecasting activities within the management decision making process, using the PrOACT (Problem, Objectives, Alternative Actions, Consequences, and Tradeoffs) tool from structured decision making. After creating this draft version of a framework, we explored several targeted case studies in ecological forecasting and adaptive management to evaluate its efficacy as a tool for fusing forecasting and adaptive management efforts. We will continue meeting to develop these ideas and work towards a manuscript.
Participants: Janet Agbaje, Kayode Oshinubi, Ethan Deyle (and thanks Ayanna St. Rose!)
Developing models to understand the transmission of pathogens in disease ecology is critical to understanding the spread of diseases and how to prevent them. A model study is relied on to simulate the spread of disease and predict the effectiveness of different control strategies. Model forecasting is also critical, both for planning and enacting public health interventions but also for building our understanding of the sometimes complex drivers of disease dynamics across space and time. Vector-transmitted diseases (e.g.,mosquito- or tick-borne) represent an exceptionally difficult case since key processes affecting spread and transmission are not directly reflected in typical public health monitoring. For example, the presence and behavior of the vector species themselves, but often there are infection reservoirs in wildlife populations as well. In this way, connecting ecological forecasts to human epidemiological forecasts is an important challenge to tackle.
The NEON Ecological Forecasting Challenge has already included a tick forecast challenge, although it has not yet been tied directly to tick pathogen status monitoring or human health. In this project, we worked on the West Nile Virus (WNV), which is a mosquito-borne disease in the family of flaviviruses. The primary host is birds (across a wide range of species), while humans are the dead-end host. WNV occurs and is commonly spread, especially in the summer, through mosquito bites. Our goal at the EFI Unconference was to examine the opportunities that NEON data could provide to create impactful forecasts for the public’s health from vector-borne diseases, focusing on WNV. Especially since humans are a dead-end host, understanding and forecasting the disease dynamics demands ecological, human, and human data. We intend to forecast the mosquito abundance as well as the infection rate in humans over time, incorporating the mosquito abundance, seasonality, drivers, and co-occurring bird abundances.
We built a preliminary bridge that connects National Ecological Observatory Network (NEON) and Center for Disease Control (CDC) data and, through preliminary visualization, demonstrated the potential to match between the NEON mosquito data (abundance and pathogen status) and CDC-reported human cases on a year-by-year and county-by-county level for 14 NEON sites located in counties with reported cases of WNV. A first look at the collected data set showed a relationship between the NEON bird and mosquito abundance that suggests large bird presence is one driving condition of large mosquito abundance in a summer sampling season. Given the relative rarity of WNV compared to some other vector-borne illnesses like Lyme disease, there are definitely some challenges to setting up a forecasting challenge for the full disease dynamics, although we may be able to cast a wider net for human cases in counties adjacent to NEON site counties. We’re excited to build this preliminary effort into a new neon4cast theme, and we’re also eager to dive into the lessons learned from one of the other Unconference projects that examined pitfalls in recruiting broad engagement in the beetle forecasting challenge.