skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


This content will become publicly available on June 1, 2026

Title: Dual‐Transformer Deep Learning Framework for Seasonal Forecasting of Great Lakes Water Levels
Abstract The Great Lakes of North America form one of the largest freshwater systems on Earth, and their lake‐wide average water levels (lake levels) can fluctuate by more than 0.5 m on a seasonal scale. These fluctuations pose substantial challenges for coastal resilience, flood risk management, and navigation planning. Accurate seasonal forecasting of lake levels using traditional mechanistic models is challenging due to the complex physical mechanisms and coupled hydroclimatic processes involved. Recently, deep learning has gained prominence in geoscience applications for its ability to recognize intricate patterns within multiphysical data sets. Here, we introduce a novel Dual‐Transformer deep learning framework, tested on the Great Lakes. This architecture integrates two modified Transformer models: the Prophet, which predicts underlying trends, and the Critic, which refines the Prophet's predictions. The final lake level prediction is derived by weighting the outputs of both models through a multi‐layer perceptron, jointly trained with the Prophet and Critic to enhance overall accuracy. Our results demonstrate that the innovative learning framework achieves the highest prediction accuracy compared to established deep learning models when using identical input features. It attains a root mean square error of 4–7 cm in predicting lake levels up to 6 months in advance across the lakes. Additionally, the Dual‐Transformer model runs six orders of magnitude faster than conventional mechanistic models, producing results in less than one second on a typical personal computer. These findings suggest that our deep learning framework has strong potential to advance lake level prediction and carries important implications for water management and disaster mitigation, thereby enhancing the quality of life in coastal regions.  more » « less
Award ID(s):
2438826
PAR ID:
10659495
Author(s) / Creator(s):
 ;  
Publisher / Repository:
AGU
Date Published:
Journal Name:
Journal of Geophysical Research: Machine Learning and Computation
Volume:
2
Issue:
2
ISSN:
2993-5210
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; (, Shore & Beach)
    null (Ed.)
    Coastal storms are an important driver of geomorphic change along Great Lakes shorelines. While there is abundant anecdotal evidence for storm impacts in the region, only a handful of studies over the last few decades have quantified them and addressed system morphodynamics. Annual to seasonal lake-level fluctuations and declining winter-ice covers also influence coastal response to storms, yet relationships between hydrodynamics and geomorphology are poorly constrained. Given this, the Great Lakes region lags behind marine coasts in terms of predictive modeling of future coastal change, which is a necessary tool for proactive coastal management. To help close this gap, we conducted a year-long study at a sandy beach-dune system along the western shore of Lake Michigan, evaluating storm impacts under conditions of extremely high water level and absent shorefast ice. Drone-derived beach and dune topography data were used to link geomorphic changes to specific environmental conditions. High water levels throughout the year of study facilitated erosion during relatively minor wave events, enhancing the vulnerability of the system to a large storm in January 2020. This event occurred with no shorefast ice present and anomalously high winter water levels, resulting in widespread erosion and overwash. This resulted in 20% of the total accretion and 66% of the erosion documented at the site over the entire year. Our study highlights the importance of both antecedent and present conditions in determining Great Lakes shoreline vulnerability to storm impacts. 
    more » « less
  2. ; (, University of Michigan)
    The Laurentian Great Lakes (hereafter the Great Lakes) comprise the world’s largest surface freshwater system. Over the past two decades, water levels in the Great Lakes have fluctuated drastically, reaching both record highs and lows. Accurate water level forecasting is critical due to the extensive ecosystem and millions of US and Canadian citizens that rely on this valuable resource. One of the most dominant variables for water supply in any freshwater system is surface runoff, which is directly impacted by precipitation amount, type, magnitude, and timing across the system’s land surfaces. Lake Superior, the most upstream of the Great Lakes, receives the greatest amount of seasonal snowfall annually out of all the great Lakes. This snowfall affects both the timing and quantity of runoff into the Great Lakes system and impacts the water supply of the Great Lakes. In this study, I analyzed the patterns of snow water equivalent and its effect on surface runoff in the Lake Superior basin. My results indicate important changes in snow water equivalent and runoff patterns over time. Specifically, I found that, as of 1971, maximum seasonal snow water equivalent is occurring on average 12 days earlier in the spring season. I also found that maximum seasonal runoff is occurring earlier; however, the change in the timing of peak runoff occurred in 1983 and is found to now be on average 11 days earlier than it was before 1983. By advancing an understanding of these relationships and ensuring they are reflected in state-of-the-art modeling systems, I provided critical information for improving the skill of water level forecasts and preparing water managers and communities for future hydrologic changes, including those associated with climate change. 
    more » « less
  3. ; ; (, Acarologia)
    Water mites are aquatic arachnids that have been used in Europe and Central America as bioindicators of ecological health in various freshwater ecosystems (including bogs). Water mites can be found in high densities in the Laurentian Great Lakes and adjacent habitats. Although they are abundant, water mites are generally not used in the assessment of aquatic habitats in the Great Lakes and are usually assigned to the “other” category in macroinvertebrate assessments. This is despite evidence of their utility as aquatic bioindicators. In the present study we consider water mites as bioindicators of the environmental health of Point Rosa marsh, a threatened marsh found on the US side of transboundary Lake St. Clair. The abundance of water mites in Point Rosa Marsh increased from 2017 to 2019 as lake water levels increased. Although increasing water levels in Lake St. Clair can be considered a negative event due to loss of irreplaceable coastal habitat by erosion with potential economic impacts, this present study indicates that water mite populations in Point Rosa Marsh increased during the same period (2017 to 2019). As a result of our study we: update the biodiversity of water mites from Lake St. Clair with new records compared to the last report from the lake over 45 years ago, first report on water mite assemblages at Point Rosa marsh at the Lake St. Clair Metropark on Lake St. Clair and the first demonstration of water mites used as bioindicators in the habitats of the Laurentian Great Lakes. 
    more » « less
  4. ; (, Bulletin of the American Meteorological Society)
    A remarkable snow climate exists on the Japanese islands of Honshu and Hokkaido near the Sea of Japan. Mean annual snowfall in this “gosetsu chitai” (heavy snow area) exceeds 600 cm (235 in.) in some near-sea-level cities and 1,300 cm (512 in.) in some mountain areas. Much of this snow falls from December to February during the East Asian winter monsoon when frequent cold-air outbreaks occur over the Sea of Japan. The resulting sea-effect precipitation systems share similarities with lake-effect precipitation systems of the Laurentian Great Lakes of North America, but are deeper, are modulated by the regional coastal geometry and topography, and can sometimes feature transversal mode snowbands. Snowfall can maximize in the lowlands or the adjoining mountains depending on the direction and strength of the boundary layer flow. Remarkable infrastructure exists in Japan for public safety, road and sidewalk maintenance, and avalanche mitigation, yet snow-related hazards claim more than 100 lives annually. For winter recreationists, there is no surer bet for deep powder than the mountains of Honshu and Hokkaido near the Sea of Japan in January, but the regional snow climate is vulnerable to global warming, especially in coastal areas. Historically, collaborative studies of sea- and lake-effect precipitation systems involving North American and Japanese scientists have been limited. Significant potential exists to advance our understanding and prediction of sea- and lake-effect precipitation based on studies from the Sea of Japan region and efforts involving meteorologists in North America, Japan, and other sea- and lake-effect regions. 
    more » « less
  5. ; ; ; ; ; (, Scientific Reports)
    Abstract Himalayan lakes represent critical water resources, culturally important waterbodies, and potential hazards. Some of these lakes experience dramatic water-level changes, responding to seasonal monsoon rains and post-monsoonal draining. To address the paucity of direct observations of hydrology in retreating mountain glacial systems, we describe a field program in a series of high altitude lakes in Sagarmatha National Park, adjacent to Ngozumba, the largest glacier in Nepal. In situ observations find extreme (>12 m) seasonal water-level changes in a 60-m deep lateral-moraine-dammed lake (lacking surface outflow), during a 16-month period, equivalent to a 5$$\times 10^6$$ × 10 6 m$$^3$$ 3 volume change annually. The water column thermal structure was also monitored over the same period. A hydraulic model is constructed, validated against observed water levels, and used to estimate hydraulic conductivities of the moraine soils damming the lake and improves our understanding of this complex hydrological system. Our findings indicate that lake level compared to the damming glacier surface height is the key criterion for large lake fluctuations, while lakes lying below the glacier surface, regulated by surface outflow, possess only minor seasonal water-level fluctuations. Thus, lakes adjacent to glaciers may exhibit very different filling/draining dynamics based on presence/absence of surface outflows and elevation relative to retreating glaciers, and consequently may have very different fates in the next few decades as the climate warms. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email