Extracting Value from Ensembles for Cloud-Free Forecasting
Abstract
The Air Force Weather Agency (AFWA) is currently producing cloud-free forecasts for several agencies, but operational forecasts do not incorporate forecast uncertainty. Uncertainty can be forecasted via an ensemble created with perturbed initial conditions. We combine AFWA's global cloud analysis and cloud advection model with the National Centers for Environmental Prediction's global weather ensemble to study the potential for ensemble cloud-free forecasting in support of space-based image collection. A year of ensemble forecasts forms the evaluation dataset. The operationally relevant cloud-free forecast threshold (cloud cover less than 30%) is evaluated over sets of 24-km grid boxes in three climatologically different regions. The analyses and forecasts favor cloud-cover values near 0% and 100% cloud cover, making skill metrics that assume normal statistics mostly inappropriate. Thus we focus on contingency table metrics at the 30% threshold and argue that the odds ratio is most appropriate. Because costs of satellite image collection are largely unknown or classified, and typical cost/loss models may not apply, we also invoke utility theory to quantify operator benefits obtainable from the ensemble. Ensemble skill is apparent, and utility for risk-averse users in persistently clear, cloudy, and variable regions/seasons yields up to a 20% increase in operational efficiency.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2011
- Accession Number
- ADA551927
Entities
People
- Cedrick L. Stubblefield
Organizations
- Naval Postgraduate School