Constructing a Consistent and Continuous Cyanobacteria Bloom Monitoring Product from Multi-Mission Ocean Color Instruments

Abstract

Satellite-based monitoring of cyanobacterial harmful algal blooms (CyanoHABs) heavily utilizes historical Envisat-MERIS and current Sentinel-OLCI observations due to the availability of the 620 nm and 709 nm bands. The permanent loss of communication with Envisat in April 2012 created an observational gap from 2012 until the operationalization of OLCI in 2016. Although MODIS-Terra has been used to bridge the gap from 2012 to 2015, differences in band architecture and the absence of the 709 nm band have complicated generating a consistent and continuous CyanoHAB monitoring product. Moreover, several Terra bands often saturate during extreme high-concentration CyanoHAB events. This study trained a fully connected deep network (CyanNet) to model MERIS-Cyanobacteria Index (CI)—a key satellite algorithm for detecting and quantifying cyanobacteria. The network was trained with Rayleigh-corrected surface reflectance at 12 Terra bands from 2002–2008, 2010–2012, and 2017–2021 and validated with data from 2009 and 2016 in Lake Okeechobee. Model performance was satisfactory, with a ~17% median difference in Lake Okeechobee annual bloom magnitude. The median difference was ~36% with 10-day Chlorophyll-a time series data, with differences often due to variations in data availability, clouds or glint. Without further regional training, the same network performed well in Lake Apopka, Lake George, and western Lake Erie. Validation success, especially in Lake Erie, shows the generalizability of CyanNet and transferability to other geographic regions.