Statistical Ensemble Weather Forecasts Using The Geostatistical Output Perturbation (GOP) Method
by Joseph Ryan Glover © 2009

Traditional weather forecasting is conducted using Numerical Weather Prediction (NWP) models that simulate the physics of the atmosphere. These dynamic models are deterministic and their forecasts are point estimates of the future state of the atmosphere. Ensemble forecasting collects a number of forecasts for the same time frame and develops a probability distribution function to describe future weather. The Geostatistical Output Perturbation (GOP) method is a statistical tool for generating ensemble forecasts based on historical weather model performance and a current point forecast. The method is computationally inexpensive when compared with other ensemble methods. Research at the University of Washington has shown that the simpli ed GOP model is e ective for forecasting surface temperature in the American Paci c Northwest and this study sought to improve upon this research. This project's modeling exercises explored the e ectiveness of the GOP method under sparse data conditions, the use of Generalized Least Squares (GLS) instead of Ordinary Least Squares as the underlying regression method and the e ect of a spatio-temporal variogram in modelling the previously ignored temporal correlation in the data set. Results from experiments for data sparseness and the spatio-temporal variogram were encouraging while the inclusion of the GLS regression method did not produce satisfactory results.