Using convolutional neural network for predicting cyanobacteria concentrations in river water

Authors: PYO, JONGCHEOL - Ulsan National Institute Of Science And Technology (UNIST); PARK, LAN JOO - National Institute Of Environmental Research; Pachepsky, Yakov; BAEK, SANGSOO - Ulsan National Institute Of Science And Technology (UNIST); KIM, KYUNGHYUN - National Institute Of Environmental Research; CHO, KYUNGHWA - Ulsan National Institute Of Science And Technology (UNIST); KIM, SEONGYUN - University Of Maryland

Submitted to: Water Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/26/2020
Publication Date: 8/26/2020
Citation: Pyo, J., Park, L., Pachepsky, Y.A., Baek, S., Kim, K., Cho, K., Kim, S. 2020. Using convolutional neural network for predicting cyanobacteria concentrations in river water. Water Research. https://doi.org/10.1016/j.watres.2020.116349.
DOI: https://doi.org/10.1016/j.watres.2020.116349

Interpretive Summary: Harmful aquatic blooms are a matter of concern around the world. Predicting the timing and magnitude of those blooms has become a challenging problem of environmental modeling. Machine learning is becoming a tool of choice for creating successful predictive models. Recently one type of such models, convolutional neural networks (CNNs) have proven their suitability for predictions of the complex system behavior. CNNs require a large volume of training and validation data that cannot be collected with existing sampling technologies in applications to harmful blooms across large water bodies. New sensing methods hold promise to provide much more water quality data to the extent that the application of CNN can become feasible. The objective of this work was to generate high spatial and temporal density data and to use them for the evaluation of CNN as a model for the development of the cyanobacteria Microcystis. The high-density data were generated with the coupled hydrodynamic and water quality model for the 29-km reach of the Nakdong River in Korea. We found that CNN can be trained to provide a good nowcast prediction of the Microcystis biomass. Three and seven-day forecasts were obtained with reasonable accuracy. The results of this work can be useful for water quality monitoring professionals in that the opportunities are demonstrated of using the improved data collection for creating accurate machinel earning-based predictive models.

Technical Abstract: Machine learning modeling techniques have emerged as a potential means for predicting algal blooms. In this study, synthetic spatio-temporal water quality data for a river section were generated with a 3D water quality model and used to investigate the capability of a convolutional neural network (CNN) for predicting harmful cyanobacterial blooms. The CNN model displayed a reasonable capacity for short-term predictions of cyanobacteria (Microcystis) biomass. In the nowcasting of Microcystis, the CNN performance had a Nash-Sutcliffe Efficiency (NSE) of 0.87. An increase in the forecast lead time resulted in a decrease in the prediction accuracy, reducing the NSE from 0.87 to 0.58. As the spatial observation density increased from 20% to 100% of the input image grids, the CNN prediction NSE had improved from 0.70 to 0.84. Adding noise to the data resulted in accuracy deterioration, but even at the noise amplitude of 10%, the accuracy was acceptable for some applications, with NSE = 0.76. Visualization of the CNN results characterized its performance variations across the studied river reach. Overall, this study successfully demonstrated the capability of the CNN model for cyanobacterial bloom prediction using high temporal frequency images.