Multi-objective optimal reservoir operation considering algal bloom control in reservoirs

Comparing the performance of 10 machine learning models in predicting Chlorophyll a in western Lake Erie

Algal blooms, which have substantial adverse effects, are increasingly occurring worldwide in the context of global warming and eutrophication. Machine learning models (MLMs) are emerging as efficient and promising tools for predicting algal blooms. However, the performance of MLMs in directly simulating algal blooms has seldom been reported, particularly in the world's largest freshwater system, the Great Lakes. To address this gap, we compared the prediction performance of Chlorophyll a (Chl a, a proxy for algal biomass) concentration in western Lake Erie among 10 popular MLMs using 15 measured water quality data collected from 2012 to 2022. Results have shown that outlier removal is essential, as it can noticeably improve prediction accuracy such as increasing the coefficient of determination (R2) from 0.35 to 0.84 (140 %) for the optimal Gradient Boosting Decision Trees (GBDT) model. All 32,767 feature combinations of measured water quality parameters were exhaustively tested for each MLM and the best feature combinations are identified. MLMs benefit from this feature selection, with the Polynomial Regression model showing notable improvements: the R2 increased from 0.71 to 0.82 (15 %) compared to no feature selection. The tree-based ensemble models, including the GBDT (R2 = 0.84) and Random Forest (R2 = 0.82) models, show the top two performances in predicting Chl a. Based on feature importance analysis, particulate organic nitrogen (PON) is determined to be the most critical water quality parameter for predicting Chl a. These results establish a benchmark for the performance of common MLMs in predicting Chl a in western Lake Erie. The determined best feature combinations potentially make water quality observations more effective and targeted, thereby benefiting sustainable water quality management.

Integrated simulation-surrogate-optimization modeling framework for multiple tradeoffs among socioeconomic and ecological targets in reservoir operations

Integrated reservoir water quantity and quality management is significant for water supply security and river ecosystem health. However, the spatiotemporal heterogeneity of water quality and the nonuniform response of multiple indicators to operation changes make it difficult to determine optimal operation schedules. This study proposes a coupled simulation-surrogate-optimization modeling approach for compromising multiple water quantity and quality targets in reservoir operations. The Environmental Fluid Dynamics Code (EFDC) was used to simulate spatiotemporal reservoir water quality dynamics. Subsequently, an ecological damage assessment method was established, accounting for the spatiotemporal heterogeneity of multiple water quality indicators and the nonlinear relationship between the water quality deterioration and ecological damage. To quickly simulate the ecological damage, a surrogate model was developed using the nonlinear autoregressive network with exogenous inputs (NARX). Finally, the surrogate model was integrated into a reservoir operation optimization model for compromising socioeconomic and ecological targets. By applying the methods to China's Danjiangkou Reservoir as a case, it was shown that more even nutrient distribution in the reservoir increased water eutrophication area while reducing concentration peak values, which helped decrease the ecological damage. Operation changes could lead to opposite effects on in-reservoir and downstream ecological targets, increasing operation optimization complexity. Both ecological and socioeconomic benefits significantly increased (by 9.4%-16.4%) during dry years under the optimized operation scheme, implying that synergies were obtained. This study offers implications and a management tool for reservoir operations to address the multiple tradeoffs among socioeconomic and ecological benefits.

Modeling study on oil spill transport in the Great Lakes: The unignorable impact of ice cover

The rise in oil trade and transportation has led to a continuous increase in the risk of oil spills, posing a serious worldwide concern. However, there is a lack of numerical models for predicting oil spill transport in freshwater, especially under icy conditions. To tackle this challenge, we developed a prediction system for oil with ice modeling by coupling the General NOAA Operational Modeling Environment (GNOME) model with the Great Lakes Operational Forecast System (GLOFS) model. Taking Lake Erie as a pilot study, we used observed drifter data to evaluate the performance of the coupled model. Additionally, we developed six hypothetical oil spill cases in Lake Erie, considering both with and without ice conditions during the freezing, stable, and melting seasons spanning from 2018 to 2022, to investigate the impacts of ice cover on oil spill processes. The results showed the effective performance of the coupled model system in capturing the movements of a deployed drifter. Through ensemble simulations, it was observed that the stable season with high-concentration ice had the most significant impact on limiting oil transport compared to the freezing and melting seasons, resulting in an oil-affected open water area of 49 km2 on day 5 with ice cover, while without ice cover it reached 183 km2. The stable season with high-concentration ice showed a notable reduction in the probability of oil presence in the risk map, whereas this reduction effect was less prominent during the freezing and melting seasons. Moreover, negative correlations between initial ice concentration and oil-affected open water area were consistent, especially on day 1 with a linear regression R-squared value of 0.94, potentially enabling rapid prediction. Overall, the coupled model system serves as a useful tool for simulating oil spills in the world's largest freshwater system, particularly under icy conditions, thus enhancing the formulation of effective emergency response strategies.

Forecasting water quality variable using deep learning and weighted averaging ensemble models

Water quality variables, including chlorophyll-a (Chl-a), play a pivotal role in comprehending and evaluating the condition of aquatic ecosystems. Chl-a, a pigment present in diverse aquatic organisms, notably algae and cyanobacteria, serves as a valuable indicator of water quality. Thus, the objectives of this study encompass: (1) the assessment of the predictive capabilities of four deep learning (DL) models - namely, recurrent neural network (RNN), long short-term memory (LSTM), gated recurrence unit (GRU), and temporal convolutional network (TCN) - in forecasting Chl-a concentrations; (2) the incorporation of these DL models into ensemble models (EMs) employing genetic algorithm (GA) and non-dominated sorting genetic algorithm (NSGA-II) to harness the strengths of each standalone model; and (3) the evaluation of the efficacy of the developed EMs. Utilizing data collected at 15-min intervals from Small Prespa Lake (SPL) in Greece, the models employed hourly Chl-a concentration lag times, extending up to 6 h, as models' inputs to forecast Chla (t+1). The proposed models underwent training on 70% of the dataset and were subsequently validated on the remaining 30%. Among the standalone DL models, the GRU model exhibited superior performance in Chl-a forecasting, surpassing the RNN, LSTM, and TCN models by 8%, 2%, and 2%, respectively. Furthermore, the integration of DL models through single-objective GA and multi-objective NSGA-II optimization algorithms yielded hybrid models adept at effectively forecasting both low and high Chl-a concentrations. The ensemble model based on NSGA-II outperformed standalone DL models as well as the GA-based model across a range of evaluation indices. For instance, considering the R-squared metric, the study's findings demonstrated that the EM-NSGA-II stands out with exceptional effectiveness compared to DL and EM-GA models, showcasing improvements of 14% (RNN), 8% (LSTM), 6% (GRU), 8% (TCN), and 3% (EM-GA) during the testing phase.