Forecasting Meteorological Variables and Anticipating Climatic Aberrations of an Oceanic Buoy Using A Neighbour Buoy

Abstract

Weather buoys or oceanic buoys are floating sensors that measure meteorological data. They are crucial decision-making infrastructures but are often a single point of failure. For instance, the Smart Atlantic Herring Cove Buoy (SMA-H) is owned by the Centre for Ocean Ventures and Entrepreneurship (COVE) and used by the Port of Halifax and Atlantic Pilotage Authority to support operational efficiency, safety, and situational awareness for marine transportation. The variables from the SMA-H buoy are judged by the marine pilots to safely drive the vessels into the port, considering the wind and tidal effects. The efficient movement of vessels is critical to the sustainability and reliability of the port, as delayed vessel movement results in high wasted costs and possible negative reputational impacts. Such an important buoy will be taken down for maintenance for a period of six weeks or may temporarily fail. Therefore, a redundancy model is required for the buoy during downtime to avoid loss and negative reputation. The previous work developed proof-of-concept machine learning models (SVM, Random Forest, and Neural Network) to check the feasibility of predicting one buoy using another buoy. The authors gathered information from different data sources and concluded ECCC buoy as the reliable source of input for the machine learning models to predict significant wave height and wind speed. The results from the prototype models were encouraging and motivated the present work to develop optimally functioning, production-ready models for additional variables, namely maximum wave height and wave period.

The main aim of the current study is to develop enhanced, optimal, and production-ready models for predicting meteorological variables of the SMA-H buoy, in addition to anticipating anomalies like strong winds and snowstorms. Five machine learning models were developed, four for predicting four variables (normal conditions) of the buoy and one for anticipating anomalies. The normal conditions model produced good results as compared to the state-of-art results in identifying regular weather conditions. Furthermore, the anomaly detection model produced superior results, accurately identifying 95 out of 100 anomaly instances.