With the ever-increasing share of Variable Renewable Energy (VRE) sources in the European energy mix, mitigating larger production forecast errors becomes increasingly challenging, with impacts on intra-day and balancing markets. From a market perspective, Day-Ahead (DA) and imbalance prices have become more volatile over recent years, with the intensification of extreme price events, including a higher number of positive spikes, as well as negative prices. In the long term, adequate reserves need to be planned and allocated accordingly to ensure grid stability. Therefore, accurately modeling weather-driven forecast uncertainties from day-ahead to hour-ahead for wind and solar generation over several decades provides valuable insights for energy system planning, market price projection (especially intra-day and balancing), and adequacy assessment in Europe.

With the recent rise of AI-driven weather models, simulating weather forecasts over long periods has become more accessible compared to traditional physical modeling approaches. Among all AI models, the ECWMF launched AIFS in 2022. With its version “single 1.0”, both wind and solar weather variables can be forecasted based on ERA5 with a 6-hourly resolution

This paper develops a methodology based on AIFS to simulate VRE generation forecasts from day-ahead to one-hour ahead for available power (real-time generation), covering multiple decades. The paper investigates in detail the spatial and temporal forecast error patterns in Denmark and Germany. Coupled with a methodology to convert weather variables into power generation (CorRES), VRE forecast errors are simulated over multiple years from AIFS and aggregated at the bidding zone level.

Results are compared to a previously established stochastic simulation approach for Denmark and Germany. The new methodology is expected to outperform during weather-driven extreme events by capturing better heterogeneous spatiotemporal correlation structure.

Potential use case for enhancing pan-European adequacy assessment by considering VRE generation forecast uncertainties and projecting future intra-day and balancing market price trends is discussed.