Leveraging satellite products for enhanced season-ahead water quality prediction

Prediction of water quality characteristics has traditionally received significantly less attention than water quantity prediction - particularly at the seasonal scale - yet with growing evidence of degradation and concerns regarding negative effects on human health, this field is growing. Given the high complexity of lake water dynamics, feedbacks, and interactions at multiple time-scales, physically-based models are challenging to construct, parameterize, and run in a predictive mode. In lieu of this, we propose statistical models for seasonal prediction of cyanobacteria abundance and associated beach closings across the June-August summer season on Lake Mendota in Wisconsin to aid in local lake and health management. Statistical approaches, however, are highly dependent on sufficiently long and diverse historical records for model fitting. This can be problematic for applications void of rich temporal and spatial observations, including water quality variables in lake systems. To augment observational records in Lake Mendota, high resolution remote sensing products are leveraged, including a cyanobacteria index (comparable to abundance), phycocyanin, and chlorophyll-a derived from instruments on past and current satellites, 1984-present, at 30-300-meter resolution and 4-16-day frequency. While some products are currently available through the Wisconsin DNR and their participation in the Cyanobacteria Assessment Network Project, other satellite images are processed with existing and proposed algorithms to derive cyanobacteria products and compare with observational records using GIS spatial analysis tools and spatial performance metrics. In general, the corrected satellite-derived products reasonably agree with observational records for Lake Mendota, and serve to increase seasonal prediction performance of expected summertime cyanobacteria abundance and beach closings.