The jelly report: Forecasting jellyfish using email and social media

Abstract

Ecosystem forecasting has potential societal value, for industry, recreation, and human health applications to name a few. The complexities of ecological systems, the expenses associated with monitoring them, and the suddenness at which forecasts become needed often make forecasts impractical. We tested a novel rapid spin-up daily forecasting system for jellyfish — Cyanea capillata (lion’s mane jellyfish), Aurelia aurita (moon jellyfish), and Staurostoma mertensii (whitecross jellyfish) — in the Gulf of Maine. The system blended satellite data with citizen reports collected via email and social media. The forecasting system took 1–3 weeks of tuning before performance plateaued, after which forecast performance was consistently high. Good model performance did not always correspond with good forecast performance, and predictor variables whose contribution improved model performance in some cases had the opposite effect on forecast performance. An adaptive learning mode provided a very modest improvement in performance. In a test of forecast range, forecast performance decreased significantly at a forecast range of around 1 week. Overall, the approach appeared to be a promising avenue toward rapid spin-up of forecasts for undermonitored systems.