Development of wavelet-ANN models to predict water quality parameters in Hilo Bay, Pacific Ocean

• We developed wavelet-ANN models for ocean water quality prediction.
• Accurate predictions of dissolved oxygen, temperature and salinity achieved.
• The hourly models provide more accurate predictions than the daily models.
• The best model included the previous time steps of the target as input variables.
• The predictive models have few input parameters which are time and cost effective.

The main objective of this study is to apply artificial neural network (ANN) and wavelet-neural network (WNN) models for predicting a variety of ocean water quality parameters. In this regard, several water quality parameters in Hilo Bay, Pacific Ocean, are taken under consideration. Different combinations of water quality parameters are applied as input variables to predict daily values of salinity, temperature and DO as well as hourly values of DO. The results demonstrate that the WNN models are superior to the ANN models. Also, the hourly models developed for DO prediction outperform the daily models of DO. For the daily models, the most accurate model has R equal to 0.96, while for the hourly model it reaches up to 0.98. Overall, the results show the ability of the model to monitor the ocean parameters, in condition with missing data, or when regular measurement and monitoring are impossible.