How to best combine statistical-empirical relationships to downscale seasonal forecasts

Abstract

Fine-scale seasonal average weather forecasts are produced by Weather Logistics from global weather predictors and measurements using a novel empirical model. Forecasts of this kind allow better management of agricultural risks and food supply chain operations. Uncertainties arise at each step in the forecast process and propagate through to affect the final fine-scale forecasts; quanitifying this uncertainty is crucial for allowing better decision making and cost-benefit analyses.

The ares of investigation posed to the study group were to identify the best global predictors to improve the forecasts and to quantify the uncertainty that propagtes through the down-scaling process of the fore- casting.

Kalmann filters were investigated to attempt to improve the predictive capabilities of the time series measurements of, for example, the El Nino index. This method reduces the influence of noise in the measurements and consequently the uncertainty in the data input into the empirical models.

The second area looked at was correlating various global measurements to the temperature in the North-west of England to attempt to identify the most important at different times of year. It was found that the large scale predictors that are most important in the summer may not be as efficient in the winter and a different set of predictors may need to be identified to predict the weather during the colder months. A weighted linear combination of a summer and winter predictor was derived which could offer a method of combining two different seasonal relations into one empirical relation valid throughout the year.