AI Models Transform Water Quality Assessment: A Study on Surface Water Monitoring

By Samudrapom DamReviewed by Susha Cheriyedath, M.Sc.Nov 12 2023

In an article recently published in the journal Water, researchers investigated the feasibility of using artificial intelligence (AI) for monitoring, evaluation, and assessment of surface water quality.

Background

Surface water sources, such as ponds, wetlands, lakes, and rivers, are crucial for maintaining human existence, economic activity, and environmental health. However, different natural processes and human activities can impact surface water quality. Specifically, industrial activities, the use of fungicides and manure, and substantial population growth have adversely impacted the surface water quality.

Human consumption is one of the major uses of surface water, specifically in regions where groundwater supplies are insufficient. Over two-thirds of the water consumed by humans worldwide primarily originates from surface water sources. Agriculture and industry are the other key users of surface water.

Thus, surface water quality evaluation and monitoring is essential for economic activities and human health sustainability. However, the water quality evaluation and monitoring have conventionally relied on manual sampling and time-consuming, expensive, and labor-intensive laboratory studies.

AI can identify patterns and trends in water quality data more quickly and accurately than conventional methods as they can efficiently and effectively process substantial amounts of data within a short duration. Thus, AI models can be used to improve surface water quality assessment, monitoring, and evaluation.

The Proposed Approach

In this study, researchers investigated the effectiveness of different AI models in monitoring surface water quality based on the simulation’s training and data quality. Specifically, long short-term memory (LSTM) and artificial neural networks (ANN) were utilized to categorize and properly predict the water quality index.

Researchers investigated the water samples from the River Ashwini in Sadhupul, Himachal Pradesh, a tourist hub in India, to assess the drinking water quality in the region. The Ashwini River is the primary water source for several towns and villages in the Solan district.

Twelve water quality measures were monitored for six months to determine the suitability of the water for various purposes. Among the twelve measures, six variables were selected, including temperature, pH, dissolved oxygen (DO), hardness, total dissolved solids (TDS), turbidity, and chlorides, as crucial factors in the study’s dataset based on their substantial influence on the water quality traits.

A small dataset containing 200 water samples for monitoring these selected parameters was utilized in this study. Additionally, several metrics, including mean squared error (MSE), coefficient of determination (R²), and mean absolute error (MAE), were used to create and assess the ANN and LSTM models.

Moreover, researchers also utilized correlation graphs and heat maps to investigate the connections between different water quality measures. The color-coded values of the selected parameters representing the water quality level of the sample were displayed on the heat map. Two classifiers, including the k-nearest neighbors (KNN) classifier and the decision tree (DT) classifier, were used to evaluate the ANN model accuracy.

Significance of the Study

The results demonstrated the effectiveness of the ANN and LSTM models in accurately assessing water quality and forecasting water quality index with high success rates. The correlation graph demonstrated the correlation between turbidity and TDS.

The MSE and root mean square error (RMSE) of the ANN model were 0.52 and 0.60, respectively, while the MSE and RMSE of the LSTM were 0.04 and 0.21, respectively, which indicated that the LSTM model was more accurate than the ANN model in predicting the output values.

However, the predictions of ANN and LSTM models were relatively close to the actual values on average, as the MSE of both models was less than one. Additionally, the accuracy of the ANN model using KNN and DT classifiers was 87.5% and 92.5%, respectively, significantly lower than the LSTM model’s accuracy of 95%.

Conclusion and Future Outlook

Overall, the study's findings demonstrated the feasibility of using AI models in monitoring water quality, specifically surface water quality, to ensure the management and protection of water resources.

LSTM models were the most accurate AI models for predicting surface water quality using small datasets collected in remote or rural areas where water quality monitoring resources are limited. ANN is a better alternative than LSTM for studies identifying detailed relationships between multiple water quality indicators.

However, more research is required to investigate other neural network topologies, such as deep belief networks (DBNs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs), for surface water quality prediction.