1461 Probabilistic Cloud Cover Forecasting from an Ensemble

Wednesday, 15 January 2020
Hall B (Boston Convention and Exhibition Center)
Travis M. Harty, The Univ. of Arizona, Tucson, AZ; and S. McKinley, W. F. Holmgren, and A. T. Lorenzo

Solar power generation is inherently uncertain due to its reliance on the weather. To make efficient decisions, this uncertainty should be quantified with a probabilistic forecast. Here we present a probabilistic cloud cover forecasting method for intra-hour forecast horizons. The forecasting technique uses satellite images, numerical weather prediction (NWP), optical flow, and ensemble data assimilation.

Our method produces an ensemble of cloud index fields from which probabilities are calculated. The cloud index fields are derived from geostationary satellite images that are then advected using an ensemble of cloud motion fields that are the result of assimilating optical flow vectors and NWP winds. Probabilities of cloud cover are then estimated based on the proportion of ensemble members that contain clouds over a specified region. We explore different methods to calculate probabilities for a given point or region from the ensemble and the performance of forecasts over large and small areas. We analyze the performance of this forecasting method over Tucson, AZ using case studies, Brier scores, and reliability diagrams.

- Indicates paper has been withdrawn from meeting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