A comparative study of data-driven models for runoff, sediment, and nitrate forecasting

Cluster-based downscaling of precipitation using Kolmogorov-Arnold Neural Networks and CMIP6 models: Insights from Oman

Accurate precipitation predictions are crucial for addressing climate change impacts on water resources, especially in arid regions like Oman. Therefore, this study evaluates three machine learning models-Random Forest (RF), Multilayer Perceptron Neural Networks (MLP-ANN), and Kolmogorov-Arnold Neural Networks (KANNs)-to downscale and predict precipitation patterns under climate scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5. We assessed each model's ability to reproduce past trends and predict future precipitation using historical data from 1995 to 2014 and projections from 2020 to 2099. The KANN model demonstrated exceptional proficiency in forecasting extreme precipitation occurrences, especially in the most severe scenario (SSP5-8.5). The MLP-ANN model offered a balanced methodology, yielding dependable forecasts that are adaptive to fluctuating situations, even amongst small increases in precipitation and uncertainty. The RF model generated the most reliable forecasts, suggesting small increases in future precipitation while closely correlating with historical data. The study indicates distinct seasonal patterns, with peak precipitation occurring during the monsoon season from June to August. The RF model predicted more intense and uniformly distributed precipitation during this period, demonstrating its advanced data processing capabilities. The geographical patterns predicted by each model exhibited temporal stability, highlighting their consistent reliability, which is essential for precise climate predictions. This comparative research highlights the significance of choosing a suitable machine learning model according to distinct forecasting requirements. Effective downscaling methodologies are essential for informed water resources management, particularly in areas susceptible to cyclones and water shortages. These results provide essential direction for policymakers to improve climate resilience, optimize water infrastructure, and formulate adaptation measures in Oman and other dry locations.

Evaluation and source identification of water pollution

Maintaining good surface water quality is essential for protecting ecosystems and human health. Henan Province has long faced challenges related to water scarcity and severe water pollution. To support effective management of water pollution in Henan Province and provide insights for regional water pollution management, we collected extensive water quality monitoring data and applied spatial autocorrelation along with random forest to analyze the sources of heavily polluted areas. Results indicate that the spatial pollution pattern of surface water quality in Henan Province can be generally classified as insignificant pollution in the north, heavy pollution in the central regions, and light pollution in the south. Heavily polluted areas are mainly located in Zhengzhou, Luoyang, and Kaifeng. Key indicators affecting water quality in these regions are chemical oxygen demand (CODMn), dissolved oxygen (DO), ammonia nitrogen (NH3-N), and total phosphorus (TP), with urban sewage and industrial wastewater identified as the main causes of deterioration. These results not only provide a scientific basis for the systematic management of surface water quality pollution in Henan Province but also provide a reference for regional water pollution management.

Comparative efficiency of the SWAT model and a deep learning model in estimating nitrate loads at the Tuckahoe creek watershed, Maryland

Modeling nitrate fate and transport in water sources is an essential component of predictive water quality management. Both mechanistic and data-driven models are currently in use. Mechanistic models, such as SWAT, simulate daily nitrate loads based on the results of simulating water flow. Data-driven models allow one to simulate nitrate loads and water flow independently. Performance of SWAT and deep learning model was evaluated in cases when deep learning model is used in (a) independent simulations of flow series and nitrate concentration series, and (b) in both flow rate and concentration simulations to obtain nitrate load values. The data were collected at the Tuckahoe Creek watershed in Maryland, United States. The data-driven deep learning model was built using long-short-term-memory (LSTM) and three-dimensional convolutional networks (3D Convolutional Networks) to simulate flow rate and nitrate concentration using weather data and imagery to derive leaf area index according to land use. Models were calibrated with data over training period 2014-2017 and validated with data over testing period. SWAT Nash-Sutcliffe efficiency (NSE) was 0.31 and 0.40 for flow rate and -0.26 and -0.18 for the nitrate load rate over training and testing periods, respectively. Three data-driven modeling scenarios were implemented: (1) using the observed flow rate and simulated nitrate concentration, (2) using the simulated flow rate and observed nitrate concentration, and (3) using the simulated flow rate and nitrate concentration. The deep learning model performed better than SWAT in all three scenarios with NSE from 0.49 to 0.58 for training and from 0.28 to 0.80 for testing periods with scenario 1 showing the best results. The difference in performance was most pronounced in fall and winter seasons. The deep learning modeling can be an efficient alternative to mechanistic watershed-scale water quality models provided the regular high-frequency data collection is implemented.

Mapping reservoir water quality from Sentinel-2 satellite data based on a new approach of weighted averaging: Application of Bayesian maximum entropy

In regions like Oman, which are characterized by aridity, enhancing the water quality discharged from reservoirs poses considerable challenges. This predicament is notably pronounced at Wadi Dayqah Dam (WDD), where meeting the demand for ample, superior water downstream proves to be a formidable task. Thus, accurately estimating and mapping water quality indicators (WQIs) is paramount for sustainable planning of inland in the study area. Since traditional procedures to collect water quality data are time-consuming, labor-intensive, and costly, water resources management has shifted from gathering field measurement data to utilizing remote sensing (RS) data. WDD has been threatened by various driving forces in recent years, such as contamination from different sources, sedimentation, nutrient runoff, salinity intrusion, temperature fluctuations, and microbial contamination. Therefore, this study aimed to retrieve and map WQIs, namely dissolved oxygen (DO) and chlorophyll-a (Chl-a) of the Wadi Dayqah Dam (WDD) reservoir from Sentinel-2 (S2) satellite data using a new procedure of weighted averaging, namely Bayesian Maximum Entropy-based Fusion (BMEF). To do so, the outputs of four Machine Learning (ML) algorithms, namely Multilayer Regression (MLR), Random Forest Regression (RFR), Support Vector Regression (SVRs), and XGBoost, were combined using this approach together, considering uncertainty. Water samples from 254 systematic plots were obtained for temperature (T), electrical conductivity (EC), chlorophyll-a (Chl-a), pH, oxidation-reduction potential (ORP), and dissolved oxygen (DO) in WDD. The findings indicated that, throughout both the training and testing phases, the BMEF model outperformed individual machine learning models. Considering Chl-a, as WQI, and R-squared, as evaluation indices, BMEF outperformed MLR, SVR, RFR, and XGBoost by 6%, 9%, 2%, and 7%, respectively. Furthermore, the results were significantly enhanced when the best combination of various spectral bands was considered to estimate specific WQIs instead of using all S2 bands as input variables of the ML algorithms.

Research on machine learning hybrid framework by coupling grid-based runoff generation model and runoff process vectorization for flood forecasting

One of the important non-engineering measures for flood forecasting and disaster reduction in watersheds is the application of machine learning flood prediction models, with Long Short-Term Memory (LSTM) being one of the most representative time series prediction models. However, the LSTM model has issues of underestimating peak flows and poor robustness in flood forecasting applications. Therefore, based on a thorough analysis of complex underlying surface attributes, this study proposes a framework for distinguishing runoff models and integrates a Grid-based Runoff Generation Model (GRGM). Simultaneously considering the time series characteristics of runoff processes, including rising, peak, and recession, a runoff process vectorization (RPV) method is proposed. In this study, a hybrid deep learning flood forecasting framework, GRGM-RPV-LSTM, is constructed by coupling the GRGM, RPV, and LSTM neural network models. Taking the Jialu River in the Zhongmu station control basin as an example, the model is validated using 18 instances of measured floods and compared with the LSTM and GRGM-LSTM models. The study shows that the GRGM model has a relative error and average coefficient of determination for simulating runoff of 8.41% and 0.976, respectively, indicating that considering the spatial distribution of runoff patterns leads to more accurate runoff calculations. Under the same lead time conditions, the GRGM-RPV-LSTM hybrid forecasting model has a Nash efficiency coefficient greater than 0.9, demonstrating better simulation performance compared to the GRGM-LSTM and LSTM models. As the lead time increases, the GRGM-RPV-LSTM model provides more accurate peak flow predictions and exhibits better robustness. The research findings can provide scientific basis for coordinated management of flood control and disaster reduction in watersheds.

Hybrid WT-CNN-GRU-based model for the estimation of reservoir water quality variables considering spatio-temporal features

Water quality indicators (WQIs), such as chlorophyll-a (Chl-a) and dissolved oxygen (DO), are crucial for understanding and assessing the health of aquatic ecosystems. Precise prediction of these indicators is fundamental for the efficient administration of rivers, lakes, and reservoirs. This research utilized two unique DL algorithms-namely, convolutional neural network (CNNs) and gated recurrent units (GRUs)-alongside their amalgamation, CNN-GRU, to precisely gauge the concentration of these indicators within a reservoir. Moreover, to optimize the outcomes of the developed hybrid model, we considered the impact of a decomposition technique, specifically the wavelet transform (WT). In addition to these efforts, we created two distinct machine learning (ML) algorithms-namely, random forest (RF) and support vector regression (SVR)-to demonstrate the superior performance of deep learning algorithms over individual ML ones. We initially gathered WQIs from diverse locations and varying depths within the reservoir using an AAQ-RINKO device in the study area to achieve this. It is important to highlight that, despite utilizing diverse data-driven models in water quality estimation, a significant gap persists in the existing literature regarding implementing a comprehensive hybrid algorithm. This algorithm integrates the wavelet transform, convolutional neural network (CNN), and gated recurrent unit (GRU) methodologies to estimate WQIs accurately within a spatiotemporal framework. Subsequently, the effectiveness of the models that were developed was assessed utilizing various statistical metrics, encompassing the correlation coefficient (r), root mean square error (RMSE), mean absolute error (MAE), and Nash-Sutcliffe efficiency (NSE) throughout both the training and testing phases. The findings demonstrated that the WT-CNN-GRU model exhibited better performance in comparison with the other algorithms by 13% (SVR), 13% (RF), 9% (CNN), and 8% (GRU) when R-squared and DO were considered as evaluation indices and WQIs, respectively.

The investigation of the binding ability between sodium dodecyl sulfate and Cu (II) in urban stormwater runoff

The simultaneous presence of heavy metals and surfactants in runoff induces complexation and ecological harm during migration. However, interactions between these pollutants are often overlooked in past studies. Thus, investigating heavy metal-surfactant complexes in runoff is imperative. In this work, Cu (II) and sodium dodecyl sulfate (SDS) were selected to investigate the interaction between heavy metals and surfactants due to the higher detected frequency in runoff. Through 1H NMR and FTIR observation of hydrogen atom nuclear displacement and functional group displacement of SDS, the change of SDS and Cu (II) complexation was obtained, and then the complexation form of Cu (II) and SDS was verified. The results showed that solution pH values and ionic strength had significant effects on the complexation of Cu (II). When the pH values increase from 3.0 to 6.0, the complexation efficiency of SDS with Cu (II) increased by 12.12% at low concentration of SDS, which may be attributed to the excessive protonation in the aqueous solution at acidic condition. The increase of ionic strength would inhibit the complexation reaction efficiency by 19.57% and finally reached the platform with concentration of NaNO3 was 0.10 mmol/L, which was mainly due to the competitive relationship between Na (I) and Cu (II). As a general filtering material in stormwater treatment measures, natural zeolite could affect the interaction between SDS and Cu (II) greatly. After the addition of SDS, the content of free Cu (II) in the zeolite-SDS-Cu (II) three-phase mixed system was significantly reduced, indicating that SDS had a positive effect on the removal of Cu (II) from runoff. This study is of great significance for investigating the migration and transformation mechanism of SDS and Cu (II) in the future and studying the control technology of storm runoff pollution.

Machine learning approaches for estimating interfacial tension between oil/gas and oil/water systems: a performance analysis

Interfacial tension (IFT) is a key physical property that affects various processes in the oil and gas industry, such as enhanced oil recovery, multiphase flow, and emulsion stability. Accurate prediction of IFT is essential for optimizing these processes and increasing their efficiency. This article compares the performance of six machine learning models, namely Support Vector Regression (SVR), Random Forests (RF), Decision Tree (DT), Gradient Boosting (GB), Catboosting (CB), and XGBoosting (XGB), in predicting IFT between oil/gas and oil/water systems. The models are trained and tested on a dataset that contains various input parameters that influence IFT, such as gas-oil ratio, gas formation volume factor, oil density, etc. The results show that SVR and Catboost models achieve the highest accuracy for oil/gas IFT prediction, with an R-squared value of 0.99, while SVR outperforms Catboost for Oil/Water IFT prediction, with an R-squared value of 0.99. The study demonstrates the potential of machine learning models as a reliable and resilient tool for predicting IFT in the oil and gas industry. The findings of this study can help improve the understanding and optimization of IFT forecasting and facilitate the development of more efficient reservoir management strategies.

A coupled model to improve river water quality prediction towards addressing non-stationarity and data limitation

Accurate predictions of river water quality are vital for sustainable water management. However, even the powerful deep learning model, i.e., long short-term memory (LSTM), has difficulty in accurately predicting water quality dynamics owing to the high non-stationarity and data limitation in a changing environment. To wiggle out of quagmires, wavelet analysis (WA) and transfer learning (TL) techniques were introduced in this study to assist LSTM modeling, termed WA-LSTM-TL. Total phosphorus, total nitrogen, ammonia nitrogen, and permanganate index were predicted in a 4 h step within 49 water quality monitoring sites in a coastal province of China. We selected suitable source domains for each target domain using an innovatively proposed regionalization approach that included 20 attributes to improve the prediction efficiency of WA-LSTM-TL. The coupled WA-LSTM facilitated capturing non-stationary patterns of water quality dynamics and improved the performance by 53 % during testing phase compared to conventional LSTM. The WA-LSTM-TL, aided by the knowledge of source domain, obtained a 17 % higher performance compared to locally trained WA-LSTM, and such improvement was more impressive when local data was limited (+66 %). The benefit of TL-based modeling diminished as data quantity increased; however, it outperformed locally direct modeling regardless of whether target domain data was limited or sufficient. This study demonstrates the reasoning for coupling WA and TL techniques with LSTM models and provides a newly coupled modeling approach for improving short-term prediction of river water quality from the perspectives of non-stationarity and data limitation.

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.

Optimal waste load allocation in river systems based on a new multi-objective cuckoo optimization algorithm

Water pollution escalates with rising waste discharge in river systems, as the rivers' limited pollution tolerance and constrained self-cleaning capacity compel the release of treated pollutants. Although several studies have shown that the non-dominated sorting genetic algorithm-II (NSGA-II) is an effective algorithm regarding the management of river water quality to reach water quality standards, to our knowledge, the literature lacks using a new optimization model, namely, the multi-objective cuckoo optimization algorithm (MOCOA). Therefore, this research introduces a new optimization framework, including non-dominated sorting and ranking selection using the comparison operator densely populated towards the best Pareto front and a trade-off estimation between the goals of discharges and environmental protection authorities. The suggested algorithm is implemented for a waste load allocation issue in Jajrood River, located in the North of Iran. The limitation of this research is that discharges are point sources. To analyze the performance of the new optimization algorithm, the simulation model is linked with a hybrid optimization model using a cuckoo optimization algorithm and non-dominated sorting genetic algorithms to convert a single-objective algorithm to a multi-objective algorithm. The findings indicate that, in terms of violation index and inequity values, MOCOA's Pareto front is superior to NSGA-II, which highlights the MOCOA's effectiveness in waste load allocation. For instance, with identical population sizes and violation indexes for both algorithms, the optimal Pareto front ranges from 1.31 to 2.36 for NSGA-II and 0.379 to 2.28 for MOCOA. This suggests that MOCOA achieves a superior Pareto front in a more efficient timeframe. Additionally, MOCOA can attain optimal equity in the smaller population size.

Simulating daily PM2.5 concentrations using wavelet analysis and artificial neural network with remote sensing and surface observation data

Accurate PM2.5 concentrations predicting is critical for public health and wellness as well as pollution control. However, traditional methods are difficult to accurately predict PM2.5. An adaptive model coupled with artificial neural network (ANN) and wavelet analysis (WANN) is utilized to predict daily PM2.5 concentrations with remote sensing and surface observation data. The four evaluation metrics, namely Pearson correlation coefficient (R), mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE), are utilized to evaluate the performances of the artificial neural network (ANN) and WANN methods. From the predicting results, The WANN model has a higher R (R = 0.9990) during the testing period compared with R (R = 0.6844) based on the ANN model. Similarly, the WANN model has a lower MAPE (3.6988%), RMSE (1.0145 μg/m3), MAE (1.3864 μg/m3), compared with MAPE (80.0086%), RMSE (16.5838 μg/m3), MAE (12.2420 μg/m3) of the ANN. In addition, comparing the outcomes of the proposed WANN method with the ANN method, it was observed that the error during the training and verification period has decreased significantly. Furthermore, the statistical methods are used to analyze WANN and ANN, showing that WANN has higher training accuracy and better stability. Therefore, it is feasible to establish WANN to predict PM2.5 concentrations (1 day in advance) by using remote sensing and surface observation data.