An Experiment in Probabilistic Weather Forecasting

View More View Less
  • 1 University of Stockholm and The Economic Research Institute at the Stockholm School of Economics, Sweden
© Get Permissions
Full access

Abstract

Probabilistic forecasts were given for temperature and precipitation for 2 and 5-day periods. They were evaluated by a scoring rule and the scores were given as feedback to the participants together with the actual outcomes. On the average the forecasts did not improve on forecasts based on climatology, and the study indicates some explanations for this. It also shows how the meteorologists' knowledge could be used to produce probabilistic forecasts that are better than climatology. The study also covers empirical evaluation of different scoring rules and of techniques for aggregating sets of forecasts.

Abstract

Probabilistic forecasts were given for temperature and precipitation for 2 and 5-day periods. They were evaluated by a scoring rule and the scores were given as feedback to the participants together with the actual outcomes. On the average the forecasts did not improve on forecasts based on climatology, and the study indicates some explanations for this. It also shows how the meteorologists' knowledge could be used to produce probabilistic forecasts that are better than climatology. The study also covers empirical evaluation of different scoring rules and of techniques for aggregating sets of forecasts.

Save