Modeling the drivers of interannual variability in cyanobacterial bloom severity using self-organizing maps and high-frequency data

Abstract

It is well established that cyanobacteria populations in shallow lakes exhibit dramatic fluctuations on both interannual and intraannual timescales; however, despite extensive research, disentangling the drivers of interannual variability in bloom severity has proved challenging. Critical thresholds of abiotic drivers such as wind, irradiance, air temperature, and tributary inputs may control the development and collapse of blooms, but these thresholds are difficult to identify in large and complex datasets. In this study, we compared high-frequency estimates of oxygen metabolism in a shallow bay of Lake Champlain to concurrent measurements of physical and chemical parameters over 3 years with very different bloom dynamics. We clustered the data using supervised and unsupervised self-organizing maps to identify the environmental drivers associated with key stages of bloom development. We then used threshold analysis to identify subtle yet important thresholds of thermal stratification that drive transitions between bloom growth and decline. We found that extended periods with near-surface temperature differentials above 0.20 °C were associated with the initial development of bloom conditions, and subsequent frequency and timing of wind mixing events had a strong influence on interannual variability in bloom severity. The methods developed here can be widely applied to other high frequency lake monitoring datasets to identify critical thresholds controlling bloom development.

Keywords:

• Artificial neural network
• cyanobacterial bloom
• Lake Champlain
• self-organizing map

Acknowledgements

Thanks to Saul Blocher of Johnson State College for nutrient analyses, and to Trevor Gearhart, DongJoo Joung, and Courtney Giles for assistance with field data collection.