Analog-Based Methods

FY2017 Accomplishments

Figure 1 RMSE (top), bias (middle), and correlation (bottom), for O3 (left) and PM2.5 (right), vs. lead time of forecasts from the community Multi-scale Air Quality (CMAQ) Chemical Transport Model (CTM) CMAQ (black) and AnEn mean (red).  Calculations are averages over all sites of AIRNow Environmental Protection Agency (EPA) network (1045 and 458 sites for O3 and PM2.5, respectively).
Figure 1 RMSE (top), bias (middle), and correlation (bottom), for O3 (left) and PM2.5 (right), vs. lead time of forecasts from the community Multi-scale Air Quality (CMAQ) Chemical Transport Model (CTM) CMAQ (black) and AnEn mean (red).  Calculations are averages over all sites of AIRNow Environmental Protection Agency (EPA) network (1045 and 458 sites for O3 and PM2.5, respectively).

The analog ensemble (Delle Monache et al. 2013) has been applied with success to the following applications:

  • The  (AnEn) technique has been applied to generate gridded probabilistic forecasts of 10-m wind speed up to six days ahead (Sperati et al. 2017). The novel aspect with respect to past applications of the AnEn consists in producing gridded fields instead of point forecasts, and the use of analysis model data instead of observations as the ground-truth.  The AnEn forecasts are generated using data from the European Centre for Medium-Range Weather Forecasts (ECMWF) deterministic and analysis model. Also, data from the ECMWF Ensemble Prediction System (EPS) are used for comparison. Given that the AnEn predictions are generated independently at any locations and lead time, the resulting spatial and temporal correlation may be degraded by noise. A reordering technique called the Schaake Shuffle (SS) is then applied to the ensemble members to recover the spatio-temporal correlation. The AnEn outperforms a calibrated version of EPS for the first two days ahead prediction. During the third forecast day the AnEn remains competitive with the calibrated EPS, while the latter is more skillful from 72 to 144 hours ahead. The AnEn uses about 1/6 of the computational resources necessary to generate the real-time EPS prediction.

  • The AnEn has been applied to generate real-time probabilistic predictions of rapid intensification of tropical storms in the 0-72 hour time interval. The AnEn had exhibited the best reliability on the East Pacific basin after a comparison to other ensemble systems for 2016 season.

  • The application of the AnEn to generate probabilistic predictions of O3 and PM2.5 surface concentrations over the US (Delle Monache et al. 2018) has been completed. The deterministic predictions of O3 and PM2.5 concentrations by Community Multi-scale Air Quality (CMAQ) Chemical Transport Model (CTM) over AIRNow Environmental Protection Agency (EPA) network across the US are used as input to the AnEn. This novel application of an analog-based method to AQ predictions demonstrates several advantages: A drastic reduction of both systematic and random errors of model-based (CMAQ) deterministic predictions (Figure 1), while considerably increasing the correlation between the predictions and the observations; Intrinsic ability to produce reliable probabilistic forecasts (i.e., no ensemble calibration is required).

 

FY2018 PLANS

In FY18 the potential of the analog ensemble technique will be further explored for several applications:

  • the prediction of tropical cyclones rapid intensification, structure, and track on both East Pacific and Atlantic dataset on the new dataset including storms from 2017.

  • AnEn predictions of rare events will be improved through a bias correction technique introduced in Sperati et al. (2017). This technique will be extended to solar power and wind power predictions.

 

REFERENCES

Alessandrini, S., Delle Monache L., Rozoff C., Lewis W., 2016: Probabilistic prediction of hurricane intensity with an analog ensemble.  Submitted Monthly Weather Review.

Davò, F., Alessandrini, S., Sperati, S., Delle Monache, L., Airoldi, D., & Vespucci, M. T., 2016: Post-processing techniques and principal component analysis for regional wind power and solar irradiance forecasting. Solar Energy, 134, 327-338.

Delle Monache L, F. A. Eckel, D. L. Rife, B. Nagarajan, K. Searight, 2013. Probabilistic weather prediction with an analog ensemble. Mon. Weather. Rev., 141, 3498–516.

Delle Monache, L., Alessandrini, S., Djalalova, I., Wilczak, J., and Knievel, J, 2018: Probabilistic air quality predictions with an Analog Ensemble. Submitted to Atmospheric Chemistry and Physics.

Sperati S., Alessandrini S., Delle Monache L., 2016: Gridded probabilistic forecasts of weather parameters with an analog ensemble. Quarterly Journal of Royal Meteorological Society (2017, in press).