A Ramp Tool and Metric to Measure the Skill of Numerical Weather Prediction Models at Forecasting Wind Ramp Events

We present a Ramp Tool and Metric that was developed during the WFIP (Wind Forecast Improvement Project) experiment. This tool was developed out of recognition that relatively rare ramp events (large changes of power – Δp – in a short period of time – Δt) have a greater impact on grid integration costs for wind energy than do the quiescent periods between ramp events. A standard metric (such as MAE or RMSE) that does not give special consideration to ramp events may not give an adequate representation of model skill or model skill improvement. To test the Ramp Tool and Metric we used the NOAA/ESRL Rapid Refresh Numerical Model (RAP) which runs at 13km horizontal resolution. This ramp tool has three components: 1) The first is a process to identify ramp events in the time series of power. Four different ramp identification methods were developed to see if a consistent best method can be identified. 2) The second component is a method for matching in time each forecast ramp event with the most appropriate observed ramp event. 3) The third and last component of the ramp tool is a process through which a skill score of the forecast model is determined. The skill score is calculated from a utility operator's perspective, incorporates both phase (timing) and amplitude errors, and recognizes that up and down ramps can have significantly different impacts on grid operation. Since no single pair of power and time thresholds defines a ramp, and in fact some utilities may be interested in several different Δp, Δt definitions of a ramp at the same time, the ramp skill metric integrates skill over a range of Δp, Δt values.