Assessing the need for artificial data in offshore operability calculations

Offshore operations are frequently delayed by weather conditions that exceed operational limits. While such delays are expected, it is challenging to estimate them at an early point in a project. Accurate estimates are beneficial for bidding contracts, planning vessel schedules, optimizing operation sequences, and reducing commercial risk and costs. Models that generate artificial but representative time series data have previously been proposed to increase the statistical confidence in weather waiting estimates. In this study, two existing methods that rely on artificially generated data, a Markov Chain model and a VAR (Vector Auto-Regressive) model, were studied and compared against a method that directly relies on 48 years of NORA3 hindcast data. The modelling errors that are introduced with the artificial data are difficult to quantify and are found to vary with different schedules and the season. Such methods are not recommended for cost estimation on projects that may be included in a portfolio strategy. Instead, the weather waiting may be calculated directly from hindcast data, and the confidence intervals may be quantified using the Bootstrap Method, which was found to perform better than analytical alternatives. However, if sufficient effort is spent to validate the results, the use of artificial data may be informative for cases where the stakes and the uncertainties are particularly high, such as for operations performed during the late fall and winter season in the North Sea region.