Predicting wind speed well is a key step towards providing good forecasts for the power output of wind turbines. However, predicting wind speed is very challenging.

The plot below shows wind speeds recorded at a wind turbine hourly over a period of 45 days. Our objective is to predict the wind speed for the next 15 days

When we look at the plot, we cannot see much of a structure. There are no clear trends or cyclical patterns. If we fit a standard time series model to the data, we obtain the predictions shown below (we will look at the details of fitting time series models next week). The yellow area is a 95% prediction interval, i.e. the yellow area is indicative of the uncertainty of our predictions

However, after a few hours, the forecast of the wind speed just falls back to the long term average of the wind speed, with a rather wide prediction interval. A standard time series model does not seem to be of much use in this context.

We can improve our predictions a lot if we integrate weather forecasts (or more formally numerical weather predictions) into our time series model. The figure below shows that this allows for much better forecasts.

However the quality of our forecasts is now very much tied to accuracy of the numerical weather prediction. .

.