Characterizing Industrial Pond Ecology Timeline in DISCOVR Cultivation Trials for Early Detection of Pond Crashes

For sustainable algal biomass cultivation, we need substantial improvement in annualized productivity by reducing the frequency of crop failure and improved growth in open raceway pond systems. In this study, high-performing strains were identified and optimized for biomass productivity. We utilized next-generation sequencing methods to quantify the ecological features of open raceway systems cultivated at in Arizona. We utilized data from several months of cultivation runs to construct a rich time-series of the ecology dynamics using amplicon sequencing and used custom anomaly detection, “PondSentry”, for the early prediction of pond crashes. PondSentry uses tensor decomposition of higher-order joint moments to detect incipient anomalies in multivariate data and displays significant improvements from standard knowledge-based anomaly detection methods. The PondSentry strategy identifies signs of deteriorating pond health at an average of three days before an actual crash event, with rank order of the ecological features plausible for crop failures driven by organisms such as Amoeboaphelidium occidentale FD01. These findings are independently confirmed with PCR and microscopy studies at an Arizona cultivation site. PondSentry’s time-series-based anomaly detection of crashes provides a suitable monitoring strategy for eukaryotic crash agents in unialgal culture. The early warnings can be used to time interventions or harvests to prevent biomass loss. The PondSentry strategy strengthens the role of data science and data-driven methods in algal cultivation and can increase the feasibility of algal-biomass based products.