Search for: All records

Physics-guided machine learning (PGML) has become a prevalent approach in studying scientific systems due to its ability to integrate scientific theories for enhancing machine learning (ML) models. However, most PGML approaches are tailored to isolated and relatively simple tasks, which limits their applicability to complex systems involving multiple interacting processes and numerous influencing features. In this paper, we propose a Physics-Guided Foundation Model (PGFM) that combines pre-trained ML models and physics-based models and leverages their complementary strengths to improve the modeling of multiple coupled processes. To effectively conduct pre-training, we construct a simulated environmental system that encompasses a wide range of influencing features and various simulated variables generated by physics-based models. The model is pre-trained in this system to adaptively select important feature interactions guided by multi-task objectives. We then fine-tune the model for each specific task using true observations, while maintaining consistency with established physical theories, such as the principles of mass and energy conservation. We demonstrate the effectiveness of this methodology in modeling water temperature and dissolved oxygen dynamics in real-world lakes. The proposed PGFM is also broadly applicable to a range of scientific fields where physics-based models are being used. 

Accurate prediction of dissolved oxygen (DO) concentrations in lakes requires a comprehensive study of phenological patterns across ecosystems, highlighting the need for precise selection of interactions amongst external factors and internal physical-chemical-biological variables. This paper presents the Multi-population Cognitive Evolutionary Search (MCES), a novel evolutionary algorithm for complex feature interaction selection problems. MCES allows models within every population to evolve adaptively, selecting relevant feature interactions for different lake types and tasks. Evaluated on diverse lakes in the Midwestern USA, MCES not only consistently produces accurate predictions with few observed labels but also, through gene maps of models, reveals sophisticated phenological patterns of different lake types, embodying the innovative concept of “AI from nature, for nature”. 

Abstract As a key water quality parameter, dissolved oxygen (DO) concentration, and particularly changes in bottom water DO is fundamental for understanding the biogeochemical processes in lake ecosystems. Based on two machine learning (ML) models, Gradient Boost Regressor (GBR) and long‐short‐term‐memory (LSTM) network, this study developed three ML model approaches: direct GBR; direct LSTM; and a 2‐step mixed ML model workflow combining both GBR and LSTM. They were used to simulate multi‐year surface and bottom DO concentrations in five lakes. All approaches were trained with readily available environmental data as predictors. Indices of lake thermal structure and mixing provided by a one‐dimensional (1‐D) hydrodynamic model were also included as predictors in the ML models. The advantages of each ML approach were not consistent for all the tested lakes, but the best one of them was defined that can estimate DO concentration with coefficient of determination (R²) up to 0.6–0.7 in each lake. All three approaches have normalized mean absolute error (NMAE) under 0.15. In a polymictic lake, the 2‐step mixed model workflow showed better representation of bottom DO concentrations, with a highest true positive rate (TPR) of hypolimnetic hypoxia detection of over 90%, while the other workflows resulted in, TPRs are around 50%. In most of the tested lakes, the predicted surface DO concentrations and variables indicating stratified conditions (i.e., Wedderburn number and the temperature difference between surface and bottom water) are essential for simulating bottom DO. The ML approaches showed promising results and could be used to support short‐ and long‐term water management plans. 

Near‐term, iterative ecological forecasts can be used to help understand and proactively manage ecosystems. To date, more forecasts have been developed for aquatic ecosystems than other ecosystems worldwide, likely motivated by the pressing need to conserve these essential and threatened ecosystems and increasing the availability of high‐frequency data. Forecasters have implemented many different modeling approaches to forecast freshwater variables, which have demonstrated promise at individual sites. However, a comprehensive analysis of the performance of varying forecast models across multiple sites is needed to understand broader controls on forecast performance. Forecasting challenges (i.e., community‐scale efforts to generate forecasts while also developing shared software, training materials, and best practices) present a useful platform for bridging this gap to evaluate how a range of modeling methods perform across axes of space, time, and ecological systems. Here, we analyzed forecasts from the aquatics theme of the National Ecological Observatory Network (NEON) Forecasting Challenge hosted by the Ecological Forecasting Initiative. Over 100,000 probabilistic forecasts of water temperature and dissolved oxygen concentration for 1–30 days ahead across seven NEON‐monitored lakes were submitted in 2023. We assessed how forecast performance varied among models with different structures, covariates, and sources of uncertainty relative to baseline null models. A similar proportion of forecast models were skillful across both variables (34%–40%), although more individual models outperformed the baseline models in forecasting water temperature (10 models out of 29) than dissolved oxygen (6 models out of 15). These top performing models came from a range of classes and structures. For water temperature, we found that forecast skill degraded with increases in forecast horizons, process‐based models, and models that included air temperature as a covariate generally exhibited the highest forecast performance, and that the most skillful forecasts often accounted for more sources of uncertainty than the lower performing models. The most skillful forecasts were for sites where observations were most divergent from historical conditions (resulting in poor baseline model performance). Overall, the NEON Forecasting Challenge provides an exciting opportunity for a model intercomparison to learn about the relative strengths of a diverse suite of models and advance our understanding of freshwater ecosystem predictability. 

Abstract Species invasions can disrupt aquatic ecosystems by re‐wiring food webs. A trophic cascade triggered by the invasion of the predatory zooplankter spiny water flea (Bythotrephes cederströmii) resulted in increased phytoplankton due to decreased zooplankton grazing. Here, we show that increased phytoplankton biomass led to an increase in lake anoxia. The temporal and spatial extent of anoxia experienced a step change increase coincident with the invasion, and anoxic factor increased by 11 d. Post‐invasion, anoxia established more quickly following spring stratification, driven by an increase in phytoplankton biomass. A shift in spring phytoplankton phenology encompassed both abundance and community composition. Diatoms (Bacillaryophyta) drove the increase in spring phytoplankton biomass, but not all phytoplankton community members increased, shifting the community composition. We infer that increased phytoplankton biomass increased labile organic matter and drove hypolimnetic oxygen consumption. These results demonstrate how a species invasion can shift lake phenology and biogeochemistry. 

Abstract The rate of technological innovation within aquatic sciences outpaces the collective ability of individual scientists within the field to make appropriate use of those technologies. The process of in situ lake sampling remains the primary choice to comprehensively understand an aquatic ecosystem at local scales; however, the impact of climate change on lakes necessitates the rapid advancement of understanding and the incorporation of lakes on both landscape and global scales. Three fields driving innovation within winter limnology that we address here are autonomous real‐time in situ monitoring, remote sensing, and modeling. The recent progress in low‐power in situ sensing and data telemetry allows continuous tracing of under‐ice processes in selected lakes as well as the development of global lake observational networks. Remote sensing offers consistent monitoring of numerous systems, allowing limnologists to ask certain questions across large scales. Models are advancing and historically come in different types (process‐based or statistical data‐driven), with the recent technological advancements and integration of machine learning and hybrid process‐based/statistical models. Lake ice modeling enhances our understanding of lake dynamics and allows for projections under future climate warming scenarios. To encourage the merging of technological innovation within limnological research of the less‐studied winter period, we have accumulated both essential details on the history and uses of contemporary sampling, remote sensing, and modeling techniques. We crafted 100 questions in the field of winter limnology that aim to facilitate the cross‐pollination of intensive and extensive modes of study to broaden knowledge of the winter period. 

Abstract. Hypolimnetic oxygen depletion during summer stratification in lakes can lead to hypoxic and anoxic conditions. Hypolimnetic anoxia is a water quality issue with many consequences, including reduced habitat for cold-water fish species, reduced quality of drinking water, and increased nutrient and organic carbon (OC) release from sediments. Both allochthonous and autochthonous OC loads contribute to oxygen depletion by providing substrate for microbial respiration; however, their relative contributions to oxygen depletion across diverse lake systems remain uncertain. Lake characteristics, such as trophic state, hydrology, and morphometry, are also influential in carbon-cycling processes and may impact oxygen depletion dynamics. To investigate the effects of carbon cycling on hypolimnetic oxygen depletion, we used a two-layer process-based lake model to simulate daily metabolism dynamics for six Wisconsin lakes over 20 years (1995–2014). Physical processes and internal metabolic processes were included in the model and were used to predict dissolved oxygen (DO), particulate OC (POC), and dissolved OC (DOC). In our study of oligotrophic, mesotrophic, and eutrophic lakes, we found autochthony to be far more important than allochthony to hypolimnetic oxygen depletion. Autochthonous POC respiration in the water column contributed the most towards hypolimnetic oxygen depletion in the eutrophic study lakes. POC water column respiration and sediment respiration had similar contributions in the mesotrophic and oligotrophic study lakes. Differences in terms of source of respiration are discussed with consideration of lake productivity and the processing and fates of organic carbon loads.