资源环境科技发展态势分析平台

Coral bleaching in the North Atlantic Ocean was modeled based on the historical relationship between reef observations (1987-2011), El Nino indicators and thermal seawater anomalies. The model components were hierarchically organized into a Bayesian network structure according to their level of influence on coral bleaching to generate seasonal predictions to be confirmed (or not) by the near-real-time forecast at similar to 5 km scale. Validations and score steps, used as a criterion for comparison between competing models, proved the viability of the Bayesian approach to perform seasonal forecasts of the bleaching occurrence, achieving an overall hit rate of 84%. Custom models with databases restricted to specific situations are presented as an alternative to improve the accuracy levels but at the cost of the loss of predictive ability. The models were developed to be conceptually simple and useful tools to assist environmental management through an early warning system for coral bleaching.