Nowcasting is a modern technique in weather prediction that seeks to produce highly accurate analysis and short-term forecasts of up to six hours. Challenges to nowcasting include numerical forecasting spatial and temporal resolution and data availability, especially in data-denied or limited regions. Nowcasting cloud ceiling height and horizontal visibility is a specific example of a challenging nowcasting problem.

A nowcast system is applied and tested on summertime conditions from June to August 2017 over the Monterey Regional Airport in California. The system post-processes 12 km North American Mesoscale Model (NAM) data from a local grid point to produce short-term multivariate probabilistic predictions of ceiling of height and visibility. Bayesian Estimation (BE) and Monte Carlo Markov Chain (MCMC) methods are used to train the system from a set of past predictor variables and observations.

The approach demonstrates error reduction and skill improvement over the raw NAM ceiling height and visibility forecasts. The computationally cheap system also explicitly communicates uncertainty and requires a relatively limited training data set compared to other statistical post-processing techniques. Using short training periods and/or analog techniques, this system can be used to nowcast in regions with limited or no observational data availability.

Subjects: operational nowcasting; cloud forecasting; Bayesian estimation; statistical post-processing; supervised machine learning; ceiling height; visibility; probabilistic weather forecasting