Hierarchical deep learning model to simulate phytoplankton at phylum/class and genus levels and zooplankton at the genus level

Authors: BAEK, SANG-SOO - ULSAN NATIONAL INSTITUTE OF SCIENCE AND TECHNOLOGY (UNIST); EUN-JUNG, YONG - ULSAN NATIONAL INSTITUTE OF SCIENCE AND TECHNOLOGY (UNIST); PYO, JONGCHEOL - COLLABORATOR; Pachepsky, Yakov; SON, HEE-JONG - COLLABORATOR; CHO, KYUNGHWA - ULSAN NATIONAL INSTITUTE OF SCIENCE AND TECHNOLOGY (UNIST)

Submitted to: Water Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/20/2022
Publication Date: 4/23/2022
Citation: Baek, S., Eun-Jung, Y., Pyo, J., Pachepsky, Y.A., Son, H., Cho, K. 2022. Hierarchical deep learning model to simulate phytoplankton at phylum/class and genus levels and zooplankton at the genus level. Water Research. https://doi.org/10.1016/j.watres.2022.118494.
DOI: https://doi.org/10.1016/j.watres.2022.118494

Interpretive Summary: Harmful algal blooms (HABs) threaten aquatic ecosystems and make the water dangerous to the health of humans and animals. Prediction of HABs has been difficult because of the multiplicity of interactions among algae and between algae, other aquatic organisms, and the environment. Artificial intelligence techniques can discover and mimic multiple complex interactions, and therefore look as good candidates for creating HAB predictions. We applied the artificial intelligence attention algorithm from the deep learning algorithm category to predict HABs of individual groups of plankton that were monitored in the large river. Using the attention algorithm substantially improved predictions of HAB caused by several types of phytoplankton but was less successful in estimation populations of zooplankton the predates on phytoplankton. This work will be beneficial for water managers who are involved in detecting, monitoring, and predicting HABs.

Technical Abstract: As harmful algal blooms (HABs) have been escalated from a regional issue to the global scale, public health and water industries in numerous countries have suffered from HABs. With the limitation of HAB monitoring, the model development could be an alternative way to understand and mitigate HABs. However, the modeling remains a major challenge in improving simulation performances and a limitation in simulating phytoplankton in genus level (e.g., Microcystis, Aulacoseira, and Eudorina) and zooplankton. The traditional modeling approach was also restricted from applying the interaction effect between phytoplankton and zooplankton. Recently, deep learning models have been proposed for solving these modeling problems owing to big-data handling capabilities and model structure flexibilities. Here, we evaluated the applicability of deep learning (DL) for phytoplankton in phylum/class level and genus level and zooplankton. Our developed model consisted of hierarchical deep learning (DL) with the classification and regression transformer (TF) models. These DL models were hierarchically connected; the output of the phylum/class level model transferred to the genus level simulation model, and the output of the genus level model was fed into the zooplankton simulation model. The classification TF model decided the initiation of plankton while the regression TF model generated the cell concentration of plankton. The hierarchical DL showed a potential capacity to simulate phytoplankton in phylum/class and genus level, by showing average R2 and RMSE value of 0.42 and 0.83 [log(cells mL-1)], respectively. There was still limited to follow the dramatic change of zooplankton cell concentration using our model. Using an attention map from the TF model, water temperature and nutrients were significant variables influencing to phytoplankton and zooplankton bloom. Overall, our study successfully demonstrated the capability of DL for HABs on inland water.