One of the foremost challenges in science is successfully predicting whether something will happen at a new place or time. Until recently, ecologists have assumed that reliable predictions would be possible only after we attained complete understanding, and so we did not try to predict the phenomena of interest. However, inspired by the push for iterative near-term ecological forecasting led by the Ecological Forecasting Initiative, ecologists are beginning to make predictions even with incomplete understanding – spurred by the idea that we will learn more by forecasting and failing, than by sticking with business as usual.
Near-term, iterative forecasts are urgently needed in fields where the phenomena of interest impact human health and well-being. For example, nuisance cyanobacterial blooms, which threaten the irreplaceable ecosystem services provided by freshwater lakes, are growing in frequency, magnitude, and duration worldwide. While the ultimate causes of blooms are characterized, identifying the proximate triggers of these events and forecasting incipient blooms are still major challenges due, in part, to the limited spatial and temporal resolution of the available data and the lack of real-time or near-real-time data integration.
This poster will describe several ongoing interdisciplinary, interconnected projects to predict cyanobacterial blooms using new tools and technologies.
Results/Conclusions
Work to date suggests that our partnerships, tools, and technologies will lead to scalable models for predicting cyanobacterial blooms. For example, we have identified the dominant sources of uncertainty in near-term predictions for a focal cyanobacterial species using relatively simple Bayesian state-space models fit to an ongoing weekly time series from oligotrophic Lake Sunapee, NH. This work provides the first estimate of uncertainty contributions for cyanobacterial predictions in an oligotrophic lake and suggests that more complex models, or model ensembles, that incorporate both abiotic and biotic factors are needed. This finding informs a new project to generate bloom forecasts from ‘big data’ collected across space and through time using sensors on fixed buoys, autonomous surface vehicles (robots), and unmanned aerial vehicles (drones). We will integrate these data streams in near real-time, then use machine learning and Bayesian modeling to improve bloom prediction capabilities, testing our models with data from traditional field sampling programs and state-of-the-art satellite remote sensing platforms. Our overarching goal is to generate bloom forecasts that will help lake managers preemptively manage water quality, provide advance warning of potential recreational water closures, and inform policy making.
Feedback and conversation at the poster are welcomed!