Worldwide Cloud Forecasts with Neural Networks

Abstract

Most approaches to weather and cloud forecasting entail the use of a large numerical weather prediction code. This project investigated an alternative approach to cloud forecasting based upon using a neural network (NN) to analyze and combine the basic meteorological elements of persistence, advection and evolution. The Worldwide Cloud Prediction Model (WCPM) is based upon a pixel by pixel implementation of a NN. The temporal evolution and advection are estimated from past satellite and numerical weather prediction data. Persistence of cloud properties at a pixel is estimated from past data. The forecast is based upon this pixel level analysis and is almost independent of changes in neighboring pixels. Over the limited data available, the NN performed somewhat better in tropical regions than the current HRCP model. The approach demonstrated the ability to predict both the advection and evolution of clouds. Performance was best in regions of significant cloud cover, regions of scattered clouds were smeared. RMS prediction errors of about 20% were typical for the WCPM as compared to rms errors of about 30% for tropical HRCP predictions. An alternative, fully object oriented approach to the NN is outlined to improve the performance and forecast sharpness in regions of scattered clouds.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 01, 1998
Accession Number
ADA343697

Entities

People

  • David M. Crandall
  • Kenneth A. Poehls
  • Kenneth E. Heikes
  • Kevin O'rourke

Tags

Communities of Interest

  • Energy and Power Technologies
  • Materials and Manufacturing Processes
  • Space

DTIC Thesaurus Topics

  • Advection
  • Air Force
  • Algorithms
  • Artificial Satellites
  • Cloud Cover
  • Data Sets
  • Geographic Regions
  • Geography
  • Grids
  • Military Research
  • Neural Networks
  • Pattern Recognition
  • Regions
  • Remote Sensing
  • Tropical Regions
  • Two Dimensional
  • Weather Forecasting

Fields of Study

  • Environmental science

Readers

  • Atmospheric Science/Meteorology
  • Image Processing and Computer Vision.
  • Ocean-Atmosphere Mesoscale Modeling, Data Assimilation, and Flux Boundary Layers

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks
  • Space