Analog Ensembles


Figure 1. Schematic representation of the process for finding four members of the analog ensemble (AnEn) at one forecast lead time.
Figure 1. Schematic representation of the process for finding four members of the analog ensemble (AnEn) at one forecast lead time.

The analog of a forecast for a given location and time is defined as the observation (or analysis grid point) that verified when a past prediction matching selected features of the current forecast was valid. A novel ensemble design, called analog ensemble (AnEn), was proposed in 2011. The best analogs of a deterministic Numerical Weather Prediction (NWP) are combined to form an ensemble and to generate skillful and reliable probabilistic predictions (Delle Monache et al. 2011, 2013). Work to improve understanding and applications has continued. 

As shown in Fig. 1, the AnEn method generates ensemble members for a prediction at a given location and forecast lead time via three main steps using a history of cases, called the analog training period, in which both the NWP deterministic prediction and a verifying observation are available. Analogs are sought independently at each location and for each lead time (black square in step 1). The best-matching historical forecasts for the current prediction are selected as the analogs (blue boxes in step 1). An analog may come from any past date within the training period, i.e., a day, week, or several months ago. Next, each analog’s verifying observation is selected as a member of AnEn (green boxes in step 2). Taken all together, these observations constitute the ensemble prediction for the current forecast (orange circles in step 3).

FY2013 Accomplishments

An in-depth analysis of the AnEn performance for the 0-48 h prediction of 10-m wind speed and 2-m temperature was presented by in Delle Monache et al. (2013). The salient aspects of this technique are:

  • The use of a higher resolution model (since only one real-time forecast is needed);
  • No need for initial condition and model perturbation strategies to generate an ensemble;
  • Intrinsically reliable forecasts (i.e., no postprocessing required);
  • Ability to capture the flow-dependent error characteristics;
  • Superior skill in predicting rare events when compared to state-of-the-science postprocessing methods.

The analog ensemble has also been applied with success to the prediction of hub-height winds and power, for wind energy applications (Alessandrini et al. 2013). When compared to advanced power prediction systems such as the one based on the European Centre for Medium-Range Weather Forecasts (ECMWF), the AnEn exhibited superior skill at a lower computational cost. For power generation, it exhibits a superior skill for the probabilistic prediction of rare events when compared to other state-of-the-science calibration methods.

FY2014 Plans

In FY14 the potential of the analog ensemble technique will be further explored for several applications, including wind and solar power forecasting, the generation of probabilistic weather predictions over a 2/3D grid, and the prediction of tropical cyclones intensity.


Alessandrini, S., L. Delle Monache, S. Sperati, J. Nissen, 2013: A novel application of an analog ensemble for short-term wind power forecasting. Submitted to Renewable Energy.

Delle Monache, L., T. Nipen, Y. Liu, G. Roux, and R. Stull, 2011: Kalman filter and analog schemes to post-process numerical weather predictions. Mon. Wea. Rev., 139, 35543570.

Delle Monache, L., F. A. Eckel, D. Rife, B. Nagarajan, and K. Searight, 2013: Probabilistic weather predictions with an analog ensemble. Mon. Wea. Rev., 141, 3498–3516.