Endurance Analytics

Our blog - covering sports analytics and EA products

Weather data for performance modelling

In cycling performance the most fast moving variables are always those attributed to weather. Courses don’t change, rider’s weight or power profiles are slow moving, but weather is relatively more dynamic. Understanding or predicting weather therefore has great importance in terms of modelling and optimising performance.

Advance but reliable knowledge about wind conditions can be pivotal in deciding race tactics, effective equipment or pacing strategies. Riders whose grand tour hopes have been dashed by unexpected side winds are a high profile example of getting the weather component of tactics wrong. Knowledge of air pressure fluctuations can be useful too where, for example, a rider planning an hour record attempt may use this data to determine the viability of a ride.

At CPL we use three different regimes of weather information, depending on how the date of the event relates to the date of modelling.

Pre-Event, 8+ days before

While “longer range” weather forecasts are often available more than a week before an event they tend to be unreliable. In this scenario we believe that the best prediction of event weather, the one we will use for modelling, should be based on historical data. What does the weather “normally” do in the time window of the event, at the event location, and what is the frequency distribution of weather at this time and place? “Wind Roses” give a great visual depiction of “normal” winds and we default our modelling assumptions to be based on the most frequent or probable wind vector. Air pressure is simply taken as “normal” for the location and elevation adjusted.

Pre-Event, 0-7 days before

In the week before an event weather forecasts become more accurate. Actual time-series weather (wind vectors and air density) for the day of the event can reasonably be expected to play out and represent the best basis for predictive modelling. We refresh event day weather forecasts every morning such that assumptions are defaulted to the most up-to-date view of the expected weather.


For post event modelling we have the benefit of hindsight and our modelling can “replay” the actual weather (wind vectors and air density) experienced during the event. This can be matched up with time stamps from a ride file, facilitating the most accurate modelling. One feature of our post-event models is estimation of CdA – rider’s aerodynamic drag metric – and the incorporation of local weather makes this sort of analysis applicable to days which cannot be considered “windless”.

As you use CPL to do Performance Modelling you may notice the weather chart that we display switching between a Wind Rose (frequency based or “normal weather”), a race day forecast, or actual race day history. This happens automatically so you can be sure that your modelling is based on the best assumptions.

Comments are closed