About

William Currie is a faculty member affiliated with the School for Environment and Sustainability and the College of Literature, Science, and the Arts at the University of Michigan. He teaches courses such as Environ 305.004: Sustainability Issues in the Great Lakes Region. His work is associated with the Program in the Environment, and he is involved in the Currie Lab Group. His research and academic focus are centered on sustainability issues, particularly related to the Great Lakes region, contributing to environmental education and sustainability practices through his teaching and research activities.

Research topics

• Environmental science
• Ecology
• Biology
• Computer Science
• Fishery
• Geography
• Economics
• Environmental resource management
• Environmental protection
• Natural resource economics

Selected publications

• A Vision for Machine Learning and Artificial Intelligence in Great Lakes Research and Management

  2026-02-06

  articleOpen access

  The Laurentian Great Lakes are a vital freshwater resource and a regionally significant natural system facing complex, persistent, and compounding challenges from climate change, nutrient loading, and invasive species. The increasing availability of observational data, coupled with advances in computational power and machine learning (ML) and artificial intelligence (AI) methods, presents an opportunity to address these challenges by improving data integration and enabling powerful data-driven models. This perspective article outlines a broad vision for applying AI in Great Lakes research and management. We review the current state of AI efforts across several key topic areas and propose a cross-disciplinary roadmap focused on advanced modeling, multi-modal data fusion, and operational forecasting. Realizing this vision will require sustained investment in open data infrastructure, shared computational resources, and inter-institutional collaboration. If successful, this roadmap will accelerate research progress, improve decision-support tools, and enhance the resilience and sustainability of the Great Lakes region’s interconnected ecological and economic foundations.

  PublisherOA PDFDOI

• Identifying Trustworthiness Challenges in Deep Learning Models for Continental-Scale Water Quality Prediction

  ArXiv.org · 2025-03-13 · 1 citations

  preprintOpen access

  Water quality is foundational to environmental sustainability, ecosystem resilience, and public health. Deep learning offers transformative potential for large-scale water quality prediction and scientific insights generation. However, their widespread adoption in high-stakes operational decision-making, such as pollution mitigation and equitable resource allocation, is prevented by unresolved trustworthiness challenges, including performance disparity, robustness, uncertainty, interpretability, generalizability, and reproducibility. In this work, we present a multi-dimensional, quantitative evaluation of trustworthiness benchmarking three state-of-the-art deep learning architectures: recurrent (LSTM), operator-learning (DeepONet), and transformer-based (Informer), trained on 37 years of data from 482 U.S. basins to predict 20 water quality variables. Our investigation reveals systematic performance disparities tied to process complexity, data availability, and basin heterogeneity. Management-critical variables remain the least predictable and most uncertain. Robustness tests reveal pronounced sensitivity to outliers and corrupted targets; notably, the architecture with the strongest baseline performance (LSTM) proves most vulnerable under data corruption. Attribution analyses align for simple variables but diverge for nutrients, underscoring the need for multi-method interpretability. Spatial generalization to ungauged basins remains poor across all models. This work serves as a timely call to action for advancing trustworthy data-driven methods for water resources management and provides a pathway to offering critical insights for researchers, decision-makers, and practitioners seeking to leverage artificial intelligence (AI) responsibly in environmental management.

  PublisherOA PDFDOI

• Identifying Trustworthiness Challenges in Deep Learning Models for Continental-Scale Water Quality Prediction

  Research Square · 2025-03-25

  preprintOpen accessSenior author
  PublisherOA PDFDOI

• Identifying trustworthiness challenges in deep learning models for continental-scale water quality prediction

  Nexus · 2025-10-30 · 3 citations

  articleOpen access

  Water quality is foundational to environmental sustainability, ecosystem resilience, and public health. Deep learning offers transformative potential for large-scale water quality prediction and scientific insights generation. However, their widespread adoption in high-stakes operational decision-making, such as pollution mitigation and equitable resource allocation, is prevented by unresolved trustworthiness challenges, including performance disparity, robustness, uncertainty, interpretability, generalizability, and reproducibility. In this work, we present a multi-dimensional, quantitative evaluation of trustworthiness benchmarking three state-of-the-art deep learning architectures: recurrent (LSTM), operator-learning (DeepONet), and transformer-based (Informer), trained on 37 years of data from 482 US basins to predict 20 water quality variables. Our investigation reveals systematic performance disparities tied to process complexity, data availability, and basin heterogeneity. Management-critical variables remain the least predictable and most uncertain. Robustness tests reveal pronounced sensitivity to outliers and corrupted targets; notably, the architecture with the strongest baseline performance (LSTM) proves most vulnerable under data corruption. Attribution analyses align for simple variables but diverge for nutrients, underscoring the need for multi-method interpretability. Spatial generalization to ungauged basins remains poor across all models. This work serves as a timely call to action for advancing trustworthy data-driven methods for water resources management and provides a pathway to offering critical insights for researchers, decision-makers, and practitioners seeking to leverage artificial intelligence (AI) responsibly in environmental management. Broader context: Water quality is a critical environmental and public health concern, with increasing pollution threatening ecosystems, drinking water supplies, and biodiversity. This study addresses a pressing challenge in environmental science: building trustworthy AI models that can reliably predict water quality at large scales. Current deep learning approaches focus on prediction accuracy, often lacking analysis of performance disparity, robustness, and interpretability, which are key elements needed for real-world decision-making in areas such as pollution control and resource management. By conducting a multi-dimensional quantitative evaluation of trustworthiness in continental-scale water quality prediction models, this work highlights key limitations and proposes actionable recommendations to address them.The research sets the stage for future advancements in AI-driven environmental monitoring, emphasizing the need for interdisciplinary collaboration among data scientists, hydrologists, ecologists, and policymakers. The findings offer insights into developing more reliable and responsible AI tools, enabling proactive water management under growing climate and socioeconomic pressures. In the long term, this work could support smarter data-driven policies and technologies that enhance water security and sustainability, critical steps toward safeguarding one of Earth’s most vital resources.

  PublisherDOI

• Temperate Forest

  Elsevier eBooks · 2025-10-15

  book-chapter1st authorCorresponding
  PublisherDOI

• Biotic resistance across a nutrient gradient in experimental wetland mesocosms

  Ecological Applications · 2025-06-01 · 2 citations

  article

  Abstract Anthropogenic degradation of wetlands often leads to regional biotic homogenization and reduced plant diversity. This reduction is often attributed to the proliferation and dominance of a few generalist, often non‐native, species. Biotic resistance from natives can sometimes impede the growth and spread of colonizers, but its dependence on environmental conditions is poorly understood. Based on field and modeling studies, we tested the predictions that (1) biotic resistance declines at higher nitrogen loading and (2) size influences colonization success. In a five‐growing season mesocosm experiment, we grew three cattail taxa: Typha latifolia (native, large), Typha angustifolia (non‐native, invasive, smallest), and Typha × glauca (hybrid, most invasive, large) as potential colonizers in the presence or absence of pre‐established resident vegetation. At two sites differing in climate and growing season length, biotic resistance treatments were crossed with 12 nitrogen levels (inflows 0–45 g N m −2 year −1 ). Each treatment combination was replicated twice, totaling 48 mesocosms per site. Without residents, colonizers (as total biomass of all three cattail taxa) persisted and expanded clonally across all nitrogen levels. However, their expansion was generally lower when colonizing a pre‐established resident community compare to bare ground. The magnitude of biotic resistance, measured as the effect of residents on colonizers' biomass, and its interaction with nitrogen differed between sites. As predicted, biotic resistance decreased with high nitrogen at the northern site, but at the southern site, residents nearly eliminated colonizers. As anticipated, smaller T. angustifolia was a poorer colonizer than the other taxa, while T. × glauca was the strongest colonizer, especially under high nitrogen conditions where biotic resistance was minimal. Our findings partially support the hypothesis that biotic resistance declines with nitrogen loading, indicating that additional research on the factors influencing the magnitude of biotic resistance is needed. Importantly, when combined with our finding that Typha can persist at all nutrient levels when natives are absent, this information could help identify wetlands particularly vulnerable to invasion, especially in environments experiencing concurrent nutrient enrichment and disturbances that expose bare ground.

  PublisherDOI

• Mapping Out How Machine Learning and Artificial Intelligence Will Change Great Lakes Observations, Modeling, and Forecasting in the Coming Decade

  Bulletin of the American Meteorological Society · 2024-12-26 · 1 citations

  article

  "Mapping out how machine learning and artificial intelligence will change Great Lakes observations, modeling, and forecasting in the coming decade" published on 26 Dec 2024 by American Meteorological Society.

  PublisherDOI

• Dark roads aid movement but increase mortality of a generalist herbivore in the American Southwest

  Ecosphere · 2023-05-01 · 7 citations

  articleOpen access

  Abstract Road networks pose many well‐documented threats to wildlife, from fragmenting habitats and restricting movement to causing mortality through vehicle collisions. For large, wide‐ranging mammals, home range requirements and seasonal migrations often necessitate road crossings, posing threats to human safety, property, and animal survival. Artificial nightlight, emanating from light posts and urban sky glow, is ubiquitous on and around road networks worldwide; however, its effects on road crossing behavior and the associated mortality risk for wildlife are not well understood. By integrating the latest NASA nightlight products with GPS collar data collected from 67 mule deer ( Odocoileus hemionus ) over a 7‐year period (2012–2018), we used a resource‐selection framework to assess factors influencing seasonal crossing behavior and road mortality in Salt Lake City, Utah, an expanding metropolitan area in the United States. We found deer preferred to cross the road where surrounding artificial nightlight was lower in both summer and winter seasons, especially during crepuscular and nighttime periods. However, lower nightlight levels also increased the risk of road mortality. Areas with more shrub cover and lower speed limits increased the likelihood of crossing as well as lowered the risk of road mortality. There were five times as many mortality events in winter as in summer, likely because of the combination of deer preference for dark roads mixed with proximity to both higher speed roads and increased human activity. Better understanding how a pervasive and expanding environmental pollutant like artificial nightlight may attract or repel human‐tolerant wildlife species from roadways presents an opportunity to mitigate collision risk while improving population management strategies for this abundant, generalist herbivore and many other economically and ecologically important species.

  PublisherOA PDFDOI

• Dark roads aid movement but increase mortality of a generalist herbivore in the American Southwest

  Figshare · 2023-01-01

  datasetOpen access

  The data provided here are the final datasets used as input in the primary models created for the manuscript “Dark roads aid movement but increase mortality of a generalist herbivore in the American Southwest”. These data were pre-processed as described in the manuscript. They include values derived from GPS-collared mule deer. GPS location data are owned by the Utah Division of Wildlife Resources. In Utah, mule deer are a legally protected species. As such, raw location data are considered proprietary and protected under state law. Data are not publically available except through a Government Records Access and Management Act request (GRAMA), under provisions stipulated by the Utah Division of Wildlife Resources (https://wildlife.utah.gov/grama.html). <br> Data files are provided as CSVs, which can be opened in a variety of programs, including Microsoft Excel, RStudio, or any basic text editor. <br> README files sharing the name of each CSV are provided to describe their columns. Data Files Included: Input to the winter seasonal road crossing models: road_xing_winter573m.csv (README_road_xing_winter573m.txt) Input to the summer seasonal road crossing models: road_xing_summer573m.csv (README_road_xing_summer573m.txt) Input to the winter seasonal road mortality models: road_mortality_winter573m.csv (README_road_mortality_winter573m.txt) Input to the summer seasonal road mortality models: road_mortality_summer573m.csv (README_road_mortality_summer573m.txt)

  PublisherDOI

• The natural capital of seagrass beds in the Caribbean: evaluating their ecosystem services and blue carbon trade potential

  Biology Letters · 2023 · 27 citations

  • Computer Science
  • Environmental science
  • Natural resource economics

  and $88.3 billion, respectively, highlighting their potential monetary importance for the region. Our results show that Caribbean seagrass beds are globally substantial pools of carbon, and our findings underscore the importance of such evaluation schemes to promote urgently needed conservation of these highly threatened and globally important ecosystems.

  PublisherOA PDFDOI

Recent grants

• Collaborative Research: Effects of Elevated CO2 on Forest N Cycling: Assessment with Large-Scale 15N Tracers and Modeling

  NSF · $149k · 2003–2007

Frequent coauthors

• Knute J. Nadelhoffer

  University of Michigan–Ann Arbor

  94 shared

• John D. Aber

  University of New Hampshire

  94 shared

• William H. McDowell

  Florida International University

  89 shared

• Alison H. Magill

  84 shared

• Steven G. McNulty

  Southern Research Station

  81 shared

• G. M. Berntson

  University of New Hampshire

  81 shared

• Mark Kamakea

  Chabot College

  81 shared

• Lindsey E. Rustad

  Michigan State University

  81 shared

Education

• PhD , Institute for the Study of Earth, Oceans, and Space

  University of New Hampshire

  1995