Probabilistic forecasting of ceiling and visibility: Skill improvement through blending of rapid refresh, LAMP and regression

Current and emerging ceiling and visibility (C&V) forecast resources comprise an increasingly diverse and skilled population of probabilistic forecast tools. Examples include LAMP, SREF, VSREF, MOS, the North American Rapid Refresh Ensemble, time-lagged ensembles built out of today's RR, HRRR and NAM, and observations-based statistical forecast models. While this diversity is desirable, it raises the question how can the complementary strengths within a population of skillful resources be effectively harnessed and used?

This paper describes the development of performance-guided blending techniques which seek to address the challenge outlined above. We report on the status of current work toward a probabilistic 1-10 hour C&V forecast guidance product built through blending of Rapid Refresh time-lagged ensembles, LAMP probabilistic forecast data, and probabilistic forecasts derived from an observations-based statistical forecast model. Most often, the blended product delivers probabilistic forecast skill greater than that of its input components. We outline the key elements of our development approach and lessons learned to date, focusing on

• Automated methods to guide bias removal and advantageous blending among input forecasts

• Formulation of a customized forecast metric to guide the forecast generation process

• Development of a logistic regression-based forecast method supplementing forecasts from 1-8 hr.

Finally, we present verification data that illustrate skill improvements achieved through use of our resource-blending approach.

This research is in response to requirements and funding by the Federal Aviation Administration (FAA). The views expressed are those of the authors and do not necessarily represent the official policy or position of the FAA.