Predicting risk in missions under sea ice with Autonomous Underwater Vehicles

Autonomous Underwater Vehicles (AUVs) have afuture as effective platforms for multi-disciplinary scienceresearch and monitoring in the polar oceans. However, operationunder ice may involve significant risk to the vehicle. A riskassessment and management process that balances the riskappetite of the responsible owner with the reliability of thevehicle and the probability of loss has been proposed. A criticalstep in the process of assessing risk is based on expert judgmentof the fault history of the vehicle, and what affect faults orincidents have on the probability of loss. However, this subjectiveexpert judgment is sensitive to the nature of sea ice cover. Incontrast to the simple, yet high risk, case of operation under anice shelf, sea ice offers a complex risk environment. Furthermore,the risk is modified by the characteristics of the support vessel,especially its ice-breaking capability. We explore how theASPeCt sea ice characterization protocol and probabilitydistributions of ice thickness and concentration can be usedwithin a rigorous process to quantify risk given a range of sea iceconditions and with ships of differing ice capabilities. A solutionfounded on a Bayesian Belief Network approach is proposed,where the results of the expert judgment elicitation is taken as areference. The design of the network topology captures the causaleffects of the environment separately on the vehicle and on theship, and combines these to produce the output. Complementaryexpert knowledge is included within the conditional probabilitytables of the Bayesian Belief Network. Using expert judgment onthe fault history of the Autosub3 vehicle and sea ice datagathered in the Arctic and Antarctic by its predecessor,Autosub2, examples are provided of how risk is modified by thesea ice environment.