A Novel Remote Sensing Method for Detecting Cyanobacteria-dominated Harmful Algal Blooms in Constructed Shallow Urban Waterbodies

Abstract:

Cyanobacterial harmful algal blooms (cyanoHABs) are an increasing threat to the management of freshwaters globally, and a number of satellite remote sensing algorithms have been developed and used for monitoring algal blooms in large inland and near-coastal waters. However, the surveillance and quantification of cyanoHABs that occur in constructed shallow urban waterbodies, via the interrogation of satellite remote sensing data, is challenging due to a number of factors. This includes the small size of these assets, the limited number of freely-available satellite products with high spatial resolution at the scale of tens of meters, local variations in water colour and field sampling limitations. Here we propose a machine-learning method, the Self-Organizing Map, to classify remotely-sensed images into different ‘blooming’ categories. We compare this novel method to existing empirical algorithms for cyanoHAB detection.

The in situ cyanobacterial biovolumes were sampled in the Melbourne region (south-eastern part of Australia) from 2009 to 2020. Remotely-sensed images were used from both Sentinel 2 and Landsat 8 satellites that cover 200 shallow waterbodies over the same study period. This novel method can be applied to other shallow waterbodies and past events without field data to predict or hindcast cyanoHAB blooming categories. This approach can help us to understand changes in cyanoHABs over time with far higher spatial and temporal resolution than field data alone. We will demonstrate how the method can support future operational monitoring and decision-making for stormwater managers.