Predicting dam inflow is necessary for effective water management. This study created machine learning algorithms to predict the amount of inflow into the Soyang River Dam in South Korea, using weather and dam inflow data for 40 years. A total of six algorithms were used, as follows: decision tree (DT), multilayer perceptron (MLP), random forest (RF), gradient boosting (GB), recurrent neural network–long short-term memory (RNN–LSTM), and convolutional neural network–LSTM (CNN–LSTM). Among these models, the multilayer perceptron model showed the best results in predicting dam inflow, with the Nash–Sutcliffe efficiency (NSE) value of 0.812, root mean squared errors (RMSE) of 77.218 m³/s, mean absolute error (MAE) of 29.034 m³/s, correlation coefficient (R) of 0.924, and determination coefficient (R²) of 0.817. However, when the amount of dam inflow is below 100 m³/s, the ensemble models (random forest and gradient boosting models) performed better than MLP for the prediction of dam inflow. Therefore, two combined machine learning (CombML) models (RF_MLP and GB_MLP) were developed for the prediction of the dam inflow using the ensemble methods (RF and GB) at precipitation below 16 mm, and the MLP at precipitation above 16 mm. The precipitation of 16 mm is the average daily precipitation at the inflow of 100 m³/s or more. The results show the accuracy verification results of NSE 0.857, RMSE 68.417 m³/s, MAE 18.063 m³/s, R 0.927, and R² 0.859 in RF_MLP, and NSE 0.829, RMSE 73.918 m³/s, MAE 18.093 m³/s, R 0.912, and R² 0.831 in GB_MLP, which infers that the combination of the models predicts the dam inflow the most accurately. CombML algorithms showed that it is possible to predict inflow through inflow learning, considering flow characteristics such as flow regimes, by combining several machine learning algorithms.

Rainfall erosivity factor (R-factor) is one of the Universal Soil Loss Equation (USLE) input parameters that account for impacts of rainfall intensity in estimating soil loss. Although many studies have calculated the R-factor using various empirical methods or the USLE method, these methods are time-consuming and require specialized knowledge for the user. The purpose of this study is to develop machine learning models to predict the R-factor faster and more accurately than the previous methods. For this, this study calculated R-factor using 1-min interval rainfall data for improved accuracy of the target value. First, the monthly R-factors were calculated using the USLE calculation method to identify the characteristics of monthly rainfall-runoff induced erosion. In turn, machine learning models were developed to predict the R-factor using the monthly R-factors calculated at 50 sites in Korea as target values. The machine learning algorithms used for this study were Decision Tree, K-Nearest Neighbors, Multilayer Perceptron, Random Forest, Gradient Boosting, eXtreme Gradient Boost, and Deep Neural Network. As a result of the validation with 20% randomly selected data, the Deep Neural Network (DNN), among seven models, showed the greatest prediction accuracy results. The DNN developed in this study was tested for six sites in Korea to demonstrate trained model performance with Nash–Sutcliffe Efficiency (NSE) and the coefficient of determination (R²) of 0.87. This means that our findings show that DNN can be efficiently used to estimate monthly R-factor at the desired site with much less effort and time with total monthly precipitation, maximum daily precipitation, and maximum hourly precipitation data. It will be used not only to calculate soil erosion risk but also to establish soil conservation plans and identify areas at risk of soil disasters by calculating rainfall erosivity factors.

For effective water management in the downstream area of a dam, it is necessary to estimate the amount of discharge from the dam to quantify the flow downstream of the dam. In this study, a machine learning model was constructed to predict the amount of discharge from Soyang River Dam using precipitation and dam inflow/discharge data from 1980 to 2020. Decision tree, multilayer perceptron, random forest, gradient boosting, RNN-LSTM, and CNN-LSTM were used as algorithms. The RNN-LSTM model achieved a Nash–Sutcliffe efficiency (NSE) of 0.796, root-mean-squared error (RMSE) of 48.996 m³/s, mean absolute error (MAE) of 10.024 m³/s, R of 0.898, and R² of 0.807, showing the best results in dam discharge prediction. The prediction of dam discharge using machine learning algorithms showed that it is possible to predict the amount of discharge, addressing limitations of physical models, such as the difficulty in applying human activity schedules and the need for various input data.

Changes in hydrological characteristics and increases in various pollutant loadings due to rapid climate change and urbanization have a significant impact on the deterioration of aquatic ecosystem health (AEH). Therefore, it is important to effectively evaluate the AEH in advance and establish appropriate strategic plans. Recently, machine learning (ML) models have been widely used to solve hydrological and environmental problems in various fields. However, in general, collecting sufficient data for ML training is time-consuming and labor-intensive. Especially in classification problems, data imbalance can lead to erroneous prediction results of ML models. In this study, we proposed a method to solve the data imbalance problem through data augmentation based on Wasserstein Generative Adversarial Network (WGAN) and to efficiently predict the grades (from A to E grades) of AEH indices (i.e., Benthic Macroinvertebrate Index (BMI), Trophic Diatom Index (TDI), Fish Assessment Index (FAI)) through the ML models. Raw datasets for the AEH indices composed of various physicochemical factors (i.e., WT, DO, BOD₅, SS, TN, TP, and Flow) and AEH grades were built and augmented through the WGAN. The performance of each ML model was evaluated through a 10-fold cross-validation (CV), and the performances of the ML models trained on the raw and WGAN-based training sets were compared and analyzed through AEH grade prediction on the test sets. The results showed that the ML models trained on the WGAN-based training set had an average F1-score for grades of each AEH index of 0.9 or greater for the test set, which was superior to the models trained on the raw training set (fewer data compared to other datasets) only. Through the above results, it was confirmed that by using the dataset augmented through WGAN, the ML model can yield better AEH grade predictive performance compared to the model trained on limited datasets; this approach reduces the effort needed for actual data collection from rivers which requires enormous time and cost. In the future, the results of this study can be used as basic data to construct big data of aquatic ecosystems, needed to efficiently evaluate and predict AEH in rivers based on the ML models.

Accurate prediction of dam inflows is essential for effective water resource management and dam operation. In this study, we developed a multi-inflow prediction ensemble (MPE) model for dam inflow prediction using auto-sklearn (AS). The MPE model is designed to combine ensemble models for high and low inflow prediction and improve dam inflow prediction accuracy. We investigated the impact of datasets assigned to flow regimes on the ensemble composition and compared the performance of the MPE model to an AS-based ensemble model developed using a conventional approach. Our findings showed that the MPE model outperformed the conventional model in predicting dam inflows during flood and nonflood periods, reducing the root mean square error (RMSE) and mean absolute error (MAE) by 22.1% and 24.9% for low inflows, and increasing the coefficient of determination (R²) and Nash–Sutcliffe efficiency (NSE) by 21.9% and 35.8%, respectively. These results suggest that the MPE model has the potential to improve water resource management and dam operation, benefiting both the environment and society. Overall, the methodology of this study is expected to contribute to the development of a robust ensemble model for dam inflow prediction in regions with high climate variability.

This study focuses on developing an hourly water level prediction model for small- and medium-sized agricultural reservoirs using the Tree-based Pipeline Optimization Tool (TPOT), an automated machine learning (AutoML) technique. The study area is the Baekhak Reservoir in South Korea, and various precipitation-related and reservoir water storage data were collected. Using these collected data, we compared widely used individual machine learning and deep learning models with the pipeline models generated by TPOT. The comparison showed that pipeline models, which included various preprocessing and ensemble techniques, exhibited higher predictive accuracy than individual machine learning and even deep learning models. The optimal pipeline model was evaluated for its performance in predicting water levels during an extreme rainfall event, demonstrating its effectiveness for hourly water level prediction. However, issues such as the overprediction of peak water levels and delays in predicting sudden water level changes were observed, likely due to inaccuracies in the ultra-short-term forecast precipitation data and the lack of information on reservoir operations (e.g., gate openings and drainage plans for agriculture). This study highlights the potential of AutoML techniques for use in hydrological modeling, and demonstrates their contribution to more efficient water management and flood prevention strategies in agricultural reservoirs.