Weather Forecast Requirements and Evaluation Methods

Overview & Synopsis

PreFlexMS develops a dispatch optimizer and provides an optimized production schedule. The outcomes will be used to assess the absolute profitability of a new PreFlexMS CSP as well as the incremental value of the various innovative components of the PreFlexMS CSP. This deliverable has been prepared by DLR in collaboration with CENER and AEMET within WP4 ‘Weather forecasting and measurement for renewable energy predictability’. It aims at collecting all input needed for the WP4 from the market scenario analysis and the dispatch optimizer developers team in PreFlexMS, presents what is available in weather forecasts and how this needs to be converted. It also serves as a handbook for metrics in forecast evaluation. We reviewed and extracted the relevant information from previous PreFlexMS results on market scenarios about cost functions for a day-ahead spot market with hourly prices for different countries (chapter 1). Detailed discussions with the power plant simulation team helped to clarify use cases being planned for the optimizer. This interaction was needed to clarify about available information on uncertainty from numerical weather prediction (NWP) providers and how this could be used in the optimizer. Chapter 2 summarizes this information and defines the parameters direct normal irradiance, wind speed and ambient temperature as provided by NWP. Chapter 3 shortly introduces principles of deterministic numerical weather prediction. Afterwards a suite of evaluation metrics is defined, discriminating between indicators of dispersion, distribution similarity, variability, and temporal distortion. Chapter 4 shortly introduces principles of probabilistic numerical weather prediction. Afterwards it defines evaluation metrics which can be of use for the power plant simulation team in evaluating the ensemble of schedules being created in the optimizer based on ensemble numerical weather prediction. And finally, statistical post-processing for calibrating ensemble forecasts is discussed.