Worldwide Cloud Prediction Algorithms
Abstract
This report describes the major algorithms included in the Worldwide Cloud Prediction Model (WCPM). A lack of data supplied by DSWA precluded code development beyond a feasibility level. Algorithm performance is presented in the project final report (Poehls, Crandall, O'Rourke and Heikes; 1997). The forecast is designed around a unified neural network with weather inputs representing advection, persistence, and evolution of clouds. Cloud observation data is taken from SERCAA level 3 nephanalysis. NOGAPS forecasts are used for the meteorological parameter inputs. The adopted pixel-by-pixel approach assumes that a forecast is possible based solely upon the past, current and approaching clouds. Each pixel is only loosely connected to surrounding pixels through geographic inputs. No formal synoptic weather inputs are employed. The major pieces are the neural network itself and the advection algorithm utilized to locate data in space and time. All other algorithms provide either neural network input or training data. The general form, the training process, and the final input vectors to the neural network are detailed. The persistence and evolution algorithms actually represent the final input choices for specific space time data. Although separately described, there was never any intention that the algorithms would perform as stand alone modules.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1998
- Accession Number
- ADA359020
Entities
People
- David M. Crandall
- Kenneth A. Poehls
- Kenneth E. Heikes
- Kevin O'rourke