J1.5 Can Machine Learning Provide A Shortcut To Fog Prediction

Wednesday, 9 January 2019: 9:30 AM
North 124B (Phoenix Convention Center - West and North Buildings)
Yuchao Jiang, Weathernews America Inc., Norman, OK; and D. Makino and K. Sakamoto

Fog is a boundary layer phenomenon which involves microphysical and dynamic processes. The accurate prediction of fog formation and duration is still challenging for the scientific community (Gultepe, Tardif et al. 2007). Though the liquid water content (LWC)-based visibility parameterization, suggested by Kunkel (1984), has been widely used as the postprocessor for current operational systems such as GFS, Ryerson and Hacker (2014) identified the largest error source came from parameterization of subgrid-scale process. Very recently, Met Office scientists reported a research model Meso-NH with 100-m grid and 148 vertical layers below 1400 m (Price, Lane et al. 2018). Such a high-resolution model has significantly improved the forecasting performance. However, it is computationally expensive and not ready for operational use for hundreds of sites.

Here we report a machine learning approach that is computationally fast and reasonably accurate. Variables from our operational NWP models are carefully selected as input features. Ryerson and Hacker (2014) pointed out WRF models usually had a warm bias that lead to forecasts of zero cloud water. We found that this drawback can be compensated by machine learning models via hyperparameter tuning. As shown in Figure 1, a combination of machine learning and physical approach has achieved results better than human forecaster, especially for the recall score.

References:

Gultepe, I., R. Tardif, S. C. Michaelides, J. Cermak, A. Bott, J. Bendix, M. D. Müller, M. Pagowski, B. Hansen, G. Ellrod, W. Jacobs, G. Toth and S. G. Cober (2007). "Fog Research: A Review of Past Achievements and Future Perspectives." Pure and Applied Geophysics 164(6): 1121-1159.

Kunkel, B. A. (1984). "Parameterization of Droplet Terminal Velocity and Extinction Coefficient in Fog Models." Journal of Climate and Applied Meteorology 23(1): 34-41.

Price, J. D., S. Lane, I. A. Boutle, D. K. E. Smith, T. Bergot, C. Lac, L. Duconge, J. McGregor, A. Kerr-Munslow, M. Pickering and R. Clark (2018). "LANFEX: a field and modeling study to improve our understanding and forecasting of radiation fog." Bulletin of the American Meteorological Society 0(0): in press.

Ryerson, W. R. and J. P. Hacker (2014). "The Potential for Mesoscale Visibility Predictions with a Multimodel Ensemble." Weather and Forecasting 29(3): 543-562.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner