Coupled data-driven and process-based model for fluorescent dissolved organic matter prediction in a shallow subtropical reservoir

Abstract

Monitoring and understanding the dissolved organic matter (DOM) cycle in a drinking water reservoir is crucial to water authorities, since most water treatment practices aim to remove DOM to prevent the formation of potentially harmful disinfection by-products. A vertical profiling system (VPS) installed in reservoirs can continuously detect the fluorescent DOM (fDOM) and determine the fDOM transport process. Although the VPS can interprete fDOM concentrations, water treatment operators still collect and rely upon DOM datasets that are manually sampled throughout the year. A long-term historical database provides an opportunity to develop a three-dimensional fDOM prediction model. In the present study, we collected and analysed VPS and sampling data and developed and assessed an innovative coupled data-driven and process-based model. These models were able to forecast future fDOM in both temperate and extreme weather conditions. Modelling scenario analysis concluded that deeper layers of the reservoir as well as areas close to the riverine zone had higher fDOM concentrations than any other zones during storm events. Simulated fDOM can be a proxy for dissolved organic carbon concentration. The model also determined that inflow creeks were predominant fDOM sources during storm events and continuing winds transported the fDOM from bottom to surface water layers. This study has implications for reservoir and water treatment plant operators seeking to gain a better understanding of the DOM cycle in a reservoir and to more efficiently manage DOM removal.