MODELING AND FORECASTING CYANOBACTERIAL HARMFUL ALGAL BLOOMS (CYANOHABS) IN LAKE PONTCHARTRAIN

Harmful algal blooms (HABs) frequently occur in coastal waters worldwide, adversely affecting economies, human health, aquatic ecosystems, and recreational activities. Lake Pontchartrain, an oligohaline estuary, experienced frequent HABs from 2018 to 2023. While NOAA’s NCCOS and USEPA’s CyAN currently monitor cyanobacteria harmful algal blooms (CyanoHABs) in coastal, freshwater, and riverine systems, there is a notable lack of systems monitoring phycocyanin concentrations in the lake. This thesis utilizes satellite remote sensing environmental stressors to model and predict CyanoHABs in Lake Pontchartrain using machine learning techniques. A key focus of this study is assessing the forecasting capabilities of the random forest method in terms of spatial and temporal analysis, with up to a 10-day lead period using various lagged environmental stressors. Eight random forest models are developed based on attribute importance to evaluate the efficacy of the random forest approach within the WEKA software. Level 3 gap-free environmental stressors, such as chlorophyll-a anomaly, chlorophyll-a concentration, sea surface temperature anomaly, cyanobacteria abundance index (CIcyano), light attenuation diffusion coefficient, secchi disk depth, sea level anomaly, and suspended particulate matter, derived from the visible infrared imaging radiometer suite (VIIRS), are employed to reconstruct past harmful algal bloom (Phycocyanin concentration) events from 2018 to 2023. This analysis shows that HABs in Lake Pontchartrain are strongly influenced by the chlorophyll-a anomaly, cyanobacteria abundance index (CIcyano), sea level anomaly, sea surface temperature anomaly, and light attenuation diffusion coefficient. The outcomes of the developed random forest models are compared with NOAA’s NCCOS and USEPA’s CyAN nowcasting cyanobacterial abundance index (CIcyano) to assess their forecasting capabilities against historical HABs events. The models achieved a correlation of 0.8379, 0.8256, and 0.7904 for the 1-day, 5-day, and 10-day lead period, respectively. The analysis of individual time-lagged environmental stressors on HABs demonstrates the models' ability to forecast past HABs events with sufficient lead time, up to 10 days in advance, providing efficient and effective management tools for proactive response and protection of public health in Lake Pontchartrain.