Remote sensing provides full-coverage and dynamic water quality monitoring with high efficiency and low consumption. Deep learning (DL) has been progressively used in water quality retrieval because it efficiently captures the potential relationship between target variables and imagery. In this study, the multimodal deep learning (MDL) models were developed and rigorously validated using atmospherically corrected Landsat remote sensing reflectance data and synchronous water quality measurements for estimating long-term Chlorophyll- a (Chl- a ), total phosphorus (TP), and total nitrogen (TN) in Lake Simcoe, Canada. Since TP and TN are non-optically active, their retrievals were based on the fact that they are closely related to the optically active constituents (OACs) such as Chl- a . We trained the MDL models with one in-situ measured data set (for Chl- a , N = 315, for TP and TN, N = 303), validated the models with two independent data sets ( N = 147), and compared the model performances with several DL, machine learning, and empirical algorithms. The results indicated that the MDL models adequately estimated Chl- a (mean absolute error (MAE) = 32.57%, Bias = 10.61%), TP (MAE = 42.58%, Bias = −2.82%), and TN (MAE = 35.05%, Bias = 13.66%), and outperformed several other candidate algorithms, namely the progressively decreasing deep neural network (DNN), a DNN with trainable parameters similar to MDL but without splitting input features, the eXtreme Gradient Boosting, the support vector regression, the NASA Ocean Color two-band and three-band ratio algorithms, and another empirical algorithm of Landsat data in clear lakes. Using the MDL models, we reconstructed the historical spatiotemporal patterns of Chl- a , TP, and TN in Lake Simcoe since 1984, and investigated the effects of two water quality improvement programs. In addition, the physical mechanism and interpretability of the MDL models were explored by quantifying the contribution of each feature to the model outputs. The framework proposed in this study provides a practical method for long-term Chl- a , TP, and TN estimation at the regional scale. 

Performance of deep learning in mapping water quality of Lake Simcoe with long-term Landsat archive

Water is one of the most essential elements in nature that forms the basis of human life and contributes to the economic growth and development of societies. Safe water is closely related to environmental health and activities. The lives of all the animals on our planet depend on water and oxygen. Moreover, sufficient dissolved oxygen (DO) is crucial for the survival of aquatic animals. In the present research, temperature (T) and flow (Q) variables were used to predict DO. The time series were monthly and data were related to the Cumberland River in the southern United States from 2008 to 2018. Support vector regression (SVR) was employed for prediction of the model in both standalone and hybrid forms. The employed hybrid models consisted in SVR combined with metaheuristic algorithms of chicken swarm optimization (CSO), social ski-driver (SSD) optimization, Black widow optimization (BWO), and the Algorithm of the innovative gunner (AIG). Pearson correlation coefficient was utilized to select the best input combination. Box plots and Taylor diagrams were employed in the interpretation of the results. It was observed that all the four hybrid models achieved better results. Also according to the evaluation criteria among the models used the following were found: SVR–AIG with the coefficient of determination (R2 = 0.963), the root mean square error (RMSE = 0.644 mg/l), the mean absolute value of error (MAE = 0.568 mg/l), the Nash–Sutcliffe coefficient (NS = 0.864), and bias percentage (BIAS = 0.001). Overall the research showed that hybrid models increased the accuracy of the single SVR model by 6.52–1.75%.

Dissolved oxygen concentration predictions for running waters with using hybrid machine learning techniques

The optimization of Low Impact Development placement considering life cycle cost using Genetic Algorithm.

Long-term continuous monitoring (LTCM) of water quality can bring far-reaching influences on water ecosystems by providing spatiotemporal data sets of diverse parameters and enabling operation of water and wastewater treatment processes in an energy-saving and cost-effective manner. However, current water monitoring technologies are deficient for long-term accuracy in data collection and processing capability. Inadequate LTCM data impedes water quality assessment and hinders the stakeholders and decision makers from foreseeing emerging problems and executing efficient control methodologies. To tackle this challenge, this review provides a forward-looking roadmap highlighting vital innovations toward LTCM, and elaborates on the impacts of LTCM through a three-hierarchy perspective: data, parameters, and systems. First, we demonstrate the critical needs and challenges of LTCM in natural resource water, drinking water, and wastewater systems, and differentiate LTCM from existing short-term and discrete monitoring techniques. We then elucidate three steps to achieve LTCM in water systems, consisting of data acquisition (water sensors), data processing (machine learning algorithms), and data application (with modeling and process control as two examples). Finally, we explore future opportunities of LTCM in four key domains, water, energy, sensing, and data, and underscore strategies to transfer scientific discoveries to general end-users.

Forward-Looking Roadmaps for Long-Term Continuous Water Quality Monitoring: Bottlenecks, Innovations, and Prospects in a Critical Review

Remote sensing retrieval of inland water quality parameters using Sentinel-2 and multiple machine learning algorithms

A generalized machine learning approach for dissolved oxygen estimation at multiple spatiotemporal scales using remote sensing.

An ensemble machine learning model for water quality estimation in coastal area based on remote sensing imagery.

Abstract Atmospheric correction (AC) is a crucial step in the quantitative analysis of inland and coastal waters. In this study, the performances of six water-based AC methods (SeaDAS, ACOLITE-DSF/EXP, C2RCC, iCOR) were evaluated by using multiple global datasets (N = 139). Four evaluation strategies were applied including spectral similarity and Chlorophyll-a retrieval. The results showed that SeaDAS and ACOLITE-DSF performed the best in terms of analytical match, band ratios, Chlorophyll-a retrieval, and spectral similarity. SeaDAS had the lowest RMSE in the blue-green bands and showed good consistency across the spectral with the lowest median spectral angle of 7°. It should be noted that ACOLITE-DSF outperforms SeaDAS in high turbidity waters. SeaDAS coupled with Chlorophyll-a retrieval algorithms of OC3 and Clark had the lowest RMSE, which showed the advantage of SeaDAS in Chlorophyll-a retrieval. This study provides scientific basis for choosing AC methods of Landsat-8 data for aquatic environment monitoring.

Comprehensive comparison performances of Landsat-8 atmospheric correction methods for inland and coastal waters

Information regarding Chlorophyll-a (ChlA) and total nitrogen (TN) is critical for early warning of algal blooms. However, reliable models for accurate forecasting of the ChlA and TN are still lacking due to the optical complexity of coastal waters. To address this issue, we proposed a novel hybrid model named the CEEMDAN-VMD-XGBOOST-SARIMA (CVXS) model to forecast ChlA and TN concentrations. The model performance was validated at three hydrological monitoring stations in Hong Kong, China. Four independent models including extreme gradient boosting (XGBoost), support vector regression (SVR), deep learning (DL), and Seasonal autoregressive integrated moving average (SARIMA), and three hybrid models including complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)-XGBoost, CEEMDAN-SVR, and CEEMDAN-DL were developed to compare their performance with the CVXS model. In addition, the physical mechanisms of the CVXS model were further explored through correlation analysis between the decomposed time series of water quality parameters. The result indicated that (1) the CVXS model had the best accuracy among all models for forecasting ChlA and TN, and all the NSEs remained above 0.97 at three hydrological monitoring stations. For forecasting ChlA, the performance of the eight models is ranked as CVXS > CEEMDAN-XGBoost > CEEMDAN-DL > CEEMDAN-SVR > XGBoost > DL > SARIMA > SVR. For forecasting TN, the performance of the eight models is ranked as CVXS > CEEMDAN-XGBoost > CEEMDAN-DL > CEEMDAN-SVR > XGBoost > SVR > SARIMA > DL; (2) the optimal forecasting time horizons of the CVXS model were one to two months; and (3) the variability of ChlA and TN concentrations induced by hydrologic factors has been inherently embedded in the decomposed time series data, thus providing the theoretical basis for the CVXS model forecasting water quality parameters. The results of this study are promising with respect to forecasting algal blooms and coastal water resource management. 

Shang Tian

Papers

A generalized machine learning approach for dissolved oxygen estimation at multiple spatiotemporal scales using remote sensing.

Remote sensing retrieval of inland water quality parameters using Sentinel-2 and multiple machine learning algorithms

An ensemble machine learning model for water quality estimation in coastal area based on remote sensing imagery.

Comprehensive comparison performances of Landsat-8 atmospheric correction methods for inland and coastal waters

A Hybrid Decomposition and Machine Learning Model for Forecasting Chlorophyll-a and Total Nitrogen Concentration in Coastal Waters