Development of an early warning system for cyanobacterial harmful algal blooms

This paper presents the development of a CyanoHAB early warning system (CyanoHAB-EWS) for proactive response to potential CyanoHABs in Lake Pontchartrain Estuary in Louisiana and beyond. The CyanoHAB-EWS consists of three groups of Random Forest (RF)-based forecasting models, each with a different lead time of 1 day, 5 days, and 10 days. The models were developed within the R statistical computing environment by using 1 year of field sampling-based phycocyanin (PC) concentration data and corresponding satellite remote sensing data, and validated with additional 6 years of data. Each model group consists of 8 models involving 1-8 input variables, including cyanobacteria index (CI-cyano), chlorophyll-a anomaly (Chl-A), sea surface temperature anomaly (SSTA), sea surface height anomaly (SLA), light attenuation diffusion coefficient (Kd490), secchi disk depth (SDD), chlorophyll-a (Chl), and suspended particulate matter (SPM), providing optional models, based on the data availability, for diverse users. The most advanced model of each group involves the time lags (10-39) of the eight environmental predictors as input variables and is capable of explaining about 98 % variations in observed PC concentrations. The time lags provide new insights into the importance of antecedent environmental conditions, rather than current conditions, to the abundance of CyanoHABs, enabling the early warning of CyanoHAB events with a sufficient lead time and thereby allowing government agencies to take proactive measures for protecting public health. Therefore, the key novelty of this work is the development of CyanoHAB-EWS that is capable of making daily forecasts of CyanoHABs.