Posters

Abstract

Seasonal wind forecasts allow informed and strategic decision-making in O&M wind farm planning by incorporating wind speed predictions for the months ahead. Yet, not only the forecast skill, but also inherent uncertainties may challenge the reliability of these predictions. To account for the atmospheric chaotic behaviour, and thus to capture the diverse potential outcomes of the future, the most robust and accepted approach to seasonal forecasting is in terms of probabilistic predictions. In particular, through different "realisations" of this possible future (themembersof the so-called forecast "ensemble"); seasonal forecasting allows the quantification of certain events' probabilities, as well as the spread of this range of possibilities. However, a critical challenge with seasonal models is their tendency towards over or under-dispersion, coupled with the issue that the probabilities derived from ensemble members often lack proper calibration. Hence, to derive proper confidence intervals or associated uncertainties directly from the ensemble is simply not possible. This divergence justifies the need for a methodology, not just for uncertainty quantification, but also for the establishment of confidence intervals that genuinely reflect statistical coverage and validity. To bridge this gap, we propose an innovative approach rooted in the principles of 'conformal prediction'. This technique allows for the adjustment of forecast uncertainties to align closely with actual observed data, thereby ensuring that the confidence intervals produced are not only statistically valid but also practically relevant. Through our methodologies, we aim to 'conformalize' the uncertainty inherent in the seasonal models to the observational reference. With this process we strive to ensure the confidence intervals are not just theoretical constructs, but robust, statistically-guaranteed and empirically-backed ranges so that stakeholders can rely on to make more informed decisions in wind farm operations and maintenance planning. We are confident this marriage of theoretical statistical rigour with practical applicability presents a significant step forward in the field of seasonal wind forecasting.