The Application and Evaluation of Analog Ensemble Method to Day-Ahead Temperature Forecasting in California

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 7 January 2015: 9:15 AM
123 (Phoenix Convention Center - West and North Buildings)
Steven H. Young, MESO, Inc., Troy, NY; and J. Zack

Temperature is the most significant factor in the weather-related component of electric load variability. This is particularly true in coastal California, where variations in the inland penetration of the marine layer can result in large day-to-day swings in temperature. In order to address the significant impact of temperature on load, temperature forecasts are an important input to day-ahead load forecasting tools employed by the California Integrated System Operator (CAISO). MESO has applied the analog ensemble (AE) method (Delle Monache, et. al., 2013) to the day-ahead temperature-forecasting problem in California.

The AE method selects similar cases by picking numerical weather prediction (NWP) model forecasts from an historical sample that most closely resemble the current NWP model forecast. There are several steps to this process. First, a set of observed or simulated “case-matching” variables is chosen. A case-matching variable may be a single variable at a model grid point or it may be a derived variable computed from one or more variables over a range of grid points. Second, the case-matching score components are computed. A score component is the difference between a case-matching variable for the forecast case and that same variable from an historical case. Finally, the case-matching score components are combined into a case matching score that measures the composite difference between all of the case-matching variables from the current case and those from the historical case. The observed temperatures from historical cases with the smallest case-matching scores are selected as ensemble members.

The AE method was initially applied to predict the hourly temperature of the following day at the 24 National Weather Service observing stations in California used as inputs to CAISO's load model. The forecasts were generated over an historical sample using data that would have been available at 0700 Pacific Prevailing Time, when CAISO's load model is run to generate the day-ahead load prediction. In this case, all input data were extracted from the 0600 UTC run of the National Center for Environmental Prediction's North American Mesoscale (NAM) model. Several case-matching variables were explored including 2-m temperature, temperature averaged over various near-surface layers, wind vector components at single levels or shallow-layer averages between 10 m above the ground to 700 hPa, the planetary boundary layer height, net long-wave radiative flux at the surface, and maximum relative humidity over layers typical of various cloud types.

Deterministic forecasts consisting of the AE mean and 50% probability of exceedance were compared to the raw NAM forecast, the NAM, the NAM Model Output Statistics (MOS) forecast, the NCEP Global Forecast System MOS forecast, and the National Digital Forecast Database forecast.

The conference presentation will provide a brief overview of the AE method and show the dependence of AE forecast performance on (1) case matching variable selection, (2) ensemble size, (3) time-of-day and seasonal windows for ensemble member selection, and (4) weather regime-specific case matching variables.

Delle Monache, L., F.A. Eckel, D.L. Rife, B. Nagarajan and K. Searight, 2013: Probabilistic Weather Prediction with an Analog Ensemble, Mon. Wea. Rev., 141, 3498–3516.