Hourly disaggregation of daily wind projections: an analogue-based, spatially coherent approach to support energy applications

As the global energy transition accelerates, planning for resilient and reliable power systems increasingly depend on the spatiotemporal dynamics of variable renewable energy (VRE) generation. However, climate projections often lack the necessary high temporal resolution required to balance supply and demand, limiting their utility in robust energy system planning and risk assessment.

In this work, we present a novel and computationally inexpensive temporal disaggregation approach to generate plausible hourly time series from coarse daily climate model projections over multiple sites or regions, with a focus on wind power generation. The approach picks an analogue day from a bank of historical observations for the candidate day to disaggregate. The choice of analogue is based on squared Euclidean distance between candidate day and historical observations, taking into account all sites and conditions before and after the candidate day. Hourly values from the analogue day are then employed across sites and rescaled to match the daily data to disaggregate. Wind speed values are then converted into hourly capacity factor time series.

We validate the framework using a 71-year open-source ERA5 reanalysis record for onshore near-surface wind speed and wind power generation across the twelve NUTS1 regions of the United Kingdom, which we split between training and validation data sets (15 years).

Our applications shows that the model is highly efficient, requiring less than one minute to downscale 15 years of daily mean data into hourly series. Our approach successfully captures the full probability distribution of the real hourly data and preserves high autocorrelation – up to 0.95 – at midnight when the analogue day changes, which has previously been a challenge for these downscaling methods. Resulting hourly wind power time series also successfully reproduce key energy-modelling-relevant characteristics. For wind drought analysis, the reconstructed time series closely follow the observed event-duration distribution, particularly for the longer, system-critical events. Similarly, the model accurately reproduces the observed rapid change distribution, confirming its ability to capture both the frequency and magnitude of wind power ramp events across different timescales. These results hold for both uniform and area-proportional spatial weights, and for different values of the algorithm’s hyperparameters. 

The proposed analogue-based approach provides an efficient, reliable, and stochastically consistent tool for generating the high-resolution VRE time series needed to assess energy-climate interactions and inform critical investment and policy decisions for future decarbonized energy systems.