Assessing and forecasting water quality in the Danube River by using neural network approaches

Area of Science:

• Environmental Science
• Water Resource Management
• Artificial Intelligence in Environmental Monitoring

Background:

• Danube River ecosystems are impacted by nutrient loads, hazardous substances, and altered flow patterns.
• Water Quality Index (WQI) is crucial for assessing ecosystem health but current scores may not reflect actual conditions.
• Early warning systems for water pollution are vital for public health protection.

Purpose of the Study:

• To develop and evaluate an Artificial Intelligence (AI) based forecasting scheme for Danube River's Water Quality Index (WQI).
• To predict WQI time series data using physical, chemical, and flow parameters.
• To compare the performance of Cascade-forward network (CFN) models against Radial Basis Function Network (RBF) models.

Main Methods:

• Utilized historical water quality data (2011-2017) from the Danube River.
• Developed Cascade-forward network (CFN) and Radial Basis Function Network (RBF) models.
• Employed Random Forest (RF) algorithm to select the eight most relevant input features for prediction.
• Forecasted WQI for 2018-2019, evaluating model accuracy using Mean Squared Error (MSE) and R-value.

Main Results:

• CFN models demonstrated superior performance over RBF models, with higher accuracy in forecasting WQI.
• Models utilizing the eight most relevant features showed effectiveness in predicting water quality time series.
• CFN models provided the most accurate short-term WQI forecasts, particularly for the colder seasons (Q1 and Q4).

Conclusions:

• Cascade-forward network (CFN) models are effective for short-term water quality forecasting in the Danube River.
• AI models can successfully learn historical patterns and non-linear relationships for accurate WQI prediction.
• The proposed AI forecasting scheme offers a valuable tool for early warning of water pollution events.