Alternative states in microbial communities during artificial aeration: Proof of incubation experiment and development of recurrent neural network models

Abstract

Artificial aeration, a widely used method of restoring the aquatic ecological environment by enhancing the re-oxygenation capacity, typically relies upon empirical models to predict ecological dynamics and determine the operating scheme of the aeration equipment. Restoration through artificial aeration is involved in oxic-anoxic transitions, whether these transitions occurred in the form of a regime shift, making the development of predictive models challenging. Here, we confirmed the existence of alternative states in microbial communities during artificial aeration through aeration incubation experiment for the first time and considered its existence in neural network modeling in order to improve model performance. By aeration incubation experiment, it was confirmed that the alternative states exist in microbial communities during artificial aeration by two independent approaches, potential analysis and "enterotyping" approach. Comparing neural network models with and without considering the existence of alternative states, it was found that considering the existence of alternative states in modeling could improve the performance of neural network model. Our study provides a reference for the prediction of systems containing time series data where the current state will have an impact on later states. The developed model could be used for optimizing the operating scheme of the artificial aeration.

Keywords: Alternative state; Artificial aeration; Ecological state; Machine learning model; Recurrent neural networks.