Search for: All records

ABSTRACT The water quality of drinking water reservoirs is critical for human and ecosystem health. In this study, we examined the drivers of three metals, aluminum (Al), barium (Ba), and copper (Cu), across two drinking water reservoirs in southwestern Virginia, USA, over 4 years. One reservoir has a hypolimnetic oxygenation system; the other does not. We used time series modeling and multivariate analysis of water column chemistry, suspended sediment, inflow, and precipitation data to assess the relative roles of hydrologic and geochemical drivers of metal behaviors in the two reservoirs. Results suggest that Al concentrations were primarily influenced by high-flow events, consistent with the mobilization of clays from physical weathering. In contrast, Ba showed stronger sensitivity to geochemical drivers, specifically redox conditions. Drivers of Cu behavior were obscured by low Cu concentrations. For all metals, patterns varied among years. Our findings highlight the importance of long-term monitoring and integrated approaches to evaluate the drivers of metal dynamics in reservoir ecosystems and inform strategies for maintaining safe drinking water supplies. 

Modeling Irregularly-sampled and Multivariate Time Series (IMTS) is crucial across a variety of applications where different sets of variates may be missing at different time-steps due to sensor malfunctions or high data acquisition costs. Existing approaches for IMTS either consider a two-stage impute-then-model framework or involve specialized architectures specific to a particular model and task. We perform a series of experiments to derive insights about the performance of IMTS methods on a variety of semi-synthetic and real-world datasets for both classification and forecasting. We also introduce Missing Feature-aware Time Series Modeling (MissTSM) or MissTSM, a simple model-agnostic and imputation-free approach for IMTS modeling. We show that MissTSM shows competitive performance compared to other IMTS approaches, especially when the amount of missing values is large and the data lacks simplistic periodic structures–conditions common to real-world IMTS applications. 

Abstract Lakes and reservoirs play a key role in the global carbon cycle, representing important carbon sinks and sources within the terrestrial landscape under different environmental conditions. Changes in climate and land use have led to increased air and surface water temperatures; increased occurrence and duration of hypolimnetic anoxia; and altered hydrology and nutrient loading, which have the potential to affect how these freshwater ecosystems receive and process carbon. To assess how interacting environmental drivers influence carbon cycling in lakes and reservoirs, we used a 5-year whole-ecosystem experiment to investigate the effects of variable catchment, meteorology, and in-lake drivers on epilimnetic and hypolimnetic dissolved organic carbon in a small reservoir. Using a combination of whole-ecosystem models and time-series analyses, we found that primary production and other internal sources contributed a mean of 29% (range: 7–49%) of the dissolved organic carbon in the reservoir’s epilimnion over the 5-year period. We also found that sinking epilimnetic primary production, dissolved organic carbon from the sediments, and other factors were likely important sources of hypolimnetic dissolved organic carbon, especially during periods of anoxia. Both the epilimnion and hypolimnion were found to be intermittent sinks, yet net sources, of dissolved organic carbon. Overall, water temperature was identified as the most important environmental predictor for water-column dissolved organic carbon, with higher concentrations observed under seasonally elevated temperatures during the late summer and early fall. Our results suggest that lakes and reservoirs may become larger sources of dissolved organic carbon to downstream ecosystems in a warmer, more anoxic future. 

Depth profiles of fluorescence-based phytoplankton biomass were sampled using a bbe Moldaenke FluoroProbe (Schwentinental, Germany) during 2014 to 2024 in five drinking water reservoirs in southwestern Virginia, USA. These reservoirs are: Beaverdam Reservoir (Vinton, Virginia), Carvins Cove Reservoir (Roanoke, Virginia), Falling Creek Reservoir (Vinton, Virginia), Gatewood Reservoir (Pulaski, Virginia), and Spring Hollow Reservoir (Salem, Virginia). Beaverdam, Carvins Cove, Falling Creek, and Spring Hollow Reservoirs are owned and operated by the Western Virginia Water Authority as primary or secondary drinking water sources for Roanoke, Virginia, and Gatewood Reservoir is a drinking water source for the town of Pulaski, Virginia. The dataset consists of depth profiles of fluorescence-based phytoplankton biomass measured at the deepest site of each reservoir adjacent to the dam, except in Falling Creek Reservoir, where depth profiles were also taken at four upstream sites ranging from the riverine to the lacustrine zone during 2016-2019 and 2024. Casts were taken approximately weekly from May-October and monthly from November-April. Casts were collected at Beaverdam and Falling Creek Reservoirs during all years (2014-2024); casts were collected at Carvins Cove Reservoir during 2014-2016 and 2018-2023; casts were collected at Spring Hollow Reservoir during 2014-2016 and 2019; and casts were collected at Gatewood Reservoir in 2015-2016. 

Abstract Harmful phytoplankton blooms driven by climate warming and nutrient pollution are a growing threat to freshwater ecosystems worldwide. Predicting these blooms is critical for managing water resources. However, process‐based models often struggle to capture the complex nonlinear dynamics of phytoplankton. Although machine learning (ML) offers powerful predictive capabilities, its “black‐box” nature has limited its adoption for management. In this study, we applied four ensemble ML algorithms, Extreme Gradient Boosting (XGBoost), Random Forest (RF), Gradient Boosting Machine (GBM), and Categorical Boosting (CatBoost), to model phytoplankton dynamics in two adjacent drinking‐water reservoirs. All models performed similarly; however, the XGBoost algorithm achieved the best performance in both reservoirs, with a root mean square error (RMSE) of 2.4–6.6 μg/L chlorophyll a and a Pearson correlation coefficient (r) of 0.83–0.86. To enhance the interpretability of these ML models, we applied explainable artificial intelligence (XAI) techniques, including Shapley Additive Explanations (SHAP) and partial dependence analyses (PDPs). Our XAI analysis revealed reservoir‐specific dynamics, with deep dissolved oxygen and water temperature more influential in one reservoir, whereas seasonality and water column stability were more critical in the other. We also observed opposing effects of thermal stratification, with high water temperatures and strong stratification stimulating phytoplankton in one reservoir and suppressing it in the other. Stability analyses confirmed that the SHAP explanations were robust to perturbations, boosting confidence in their interpretability. This study demonstrates the application of a stability‐constrained XAI framework to provide transparent and trustworthy ecological insights, offering a robust approach for data‐driven water quality management and forecasting. 

Abstract Data science skills (e.g., analyzing, modeling, and visualizing large data sets) are increasingly needed by undergraduates in the life sciences. However, a lack of both student and instructor confidence in data science skills presents a barrier to their inclusion in undergraduate curricula. To reduce this barrier, we developed four teaching modules in the Macrosystems EDDIE (for environmental data-driven inquiry and exploration) program to introduce undergraduate students and instructors to ecological forecasting, an emerging subdiscipline that integrates multiple data science skills. Ecological forecasting aims to improve natural resource management by providing future predictions of ecosystems with uncertainty. We assessed module efficacy with 596 students and 26 instructors over 3 years and found that module completion increased students’ confidence in their understanding of ecological forecasting and instructors’ likelihood to work with long-term, high-frequency sensor network data. Our modules constitute one of the first formalized data science curricula on ecological forecasting for undergraduates. 

Abstract. Water quality in lakes is an emergent property of complex biotic and abiotic processes that differ across spatial and temporal scales. Water quality is also a determinant of ecosystem services that lakes provide and is thus of great interest to ecologists. Machine learning and other computer science techniques are increasingly being used to predict water quality dynamics as well as to gain a greater understanding of water quality patterns and controls. To benefit the sciences of both ecology and computer science, we have created a benchmark dataset of lake water quality time series and vertical profiles. LakeBeD-US contains over 500 million unique observations of lake water quality collected by multiple long-term monitoring programs across 17 water quality variables from 21 lakes in the United States. There are two published versions of LakeBeD-US: the “Ecology Edition” published in the Environmental Data Initiative repository (https://doi.org/10.6073/pasta/c56a204a65483790f6277de4896d7140, McAfee et al., 2024) and the “Computer Science Edition” published in the Hugging Face repository (https://doi.org/10.57967/hf/3771, Pradhan et al., 2024). Each edition is formatted in a manner conducive to inquiries and analyses specific to each domain. For ecologists, LakeBeD-US: Ecology Edition provides an opportunity to study the spatial and temporal dynamics of several lakes with varying water quality, ecosystem, and landscape characteristics. For computer scientists, LakeBeD-US: Computer Science Edition acts as a benchmark dataset that enables the advancement of machine learning for water quality prediction. 

Abstract Dissolved organic matter (DOM) plays an important role in aquatic carbon cycling and is a valuable metric of ecosystem functioning and water quality in freshwater ecosystems. Despite its importance for biogeochemical cycling and water quality, no near‐term iterative forecasts have previously been developed for freshwater DOM concentrations. To advance both our understanding of freshwater DOM dynamics and management, we developed 1–34 days‐ahead forecasts of fluorescent DOM (fDOM) in three drinking water reservoirs. These temperate reservoirs are co‐located in Virginia, USA and experience variable DOM dynamics (range: 5–27 QSU (quinine sulfate units)). We developed six different forecasting models to predict fDOM in each reservoir. Three models were time series models based on forecasted drivers (water temperature and meteorology) that were updated daily from high‐frequency fDOM sensors. The other forecast models included a neural network machine learning model and two baseline reference models (day‐of‐year mean and persistence). Altogether, our forecasts were able to capture observed dynamics over a year in all three reservoirs, with one time series model outperforming the baseline models across the full 34‐day forecast horizon. Aggregated across reservoirs and models over a year, forecast RMSE increased from 0.7 to 4.1 QSU over the 1–34 days‐ahead forecast horizon. Forecast skill varied substantially across seasons, with greatest accuracy in the spring and winter compared to the summer and fall across reservoirs. These forecasts can help improve our understanding of the predictability of DOM and inform management in freshwater ecosystems as carbon dynamics become more variable due to global change.