Predicting water quality in unmonitored watersheds using artificial neural networks

Area of Science:

• Environmental Science
• Hydrology
• Data Science

Background:

• Land use and land cover (LULC) significantly influence watershed water quality (WQ).
• Predicting WQ parameters is crucial for effective water resource management.
• Existing methods often require extensive WQ data, limiting their application.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting WQ parameters in watersheds lacking historical WQ data.
• To establish relationships between LULC, climate data, and WQ parameters.
• To assess the model's performance across different LULC types and watershed sizes.

Main Methods:

• An artificial neural network (ANN) model was employed.
• Inputs included LULC percentages, temperature, and stream discharge.
• The model was trained, validated, and tested on 18 watersheds in west Georgia, USA.

Main Results:

• ANN models demonstrated good to very good performance in predicting WQ parameters like TDS, Cl, NO3, SO4, Na, K, and DOC.
• Average performance for TSS and TP was rated as 'good'.
• Models performed best in pastoral and forested watersheds ('very good') and well in urban watersheds ('good').

Conclusions:

• ANN models can successfully predict WQ parameters using LULC and basic environmental data, even without historical WQ measurements.
• This approach enables the prediction of LULC change impacts on WQ in nearby watersheds with similar physiographic properties.
• The methodology offers a valuable tool for water resource management and environmental monitoring.