Scientist in the Environmental and Societal Impacts Group, National Center for Atmospheric Research. His research interests include the application of statistics to the environmental sciences. He has published articles in Journal of Applied Probability, Journal of Forecasting, Technometrics, The American Statistician and The Review of Economics and Statistics, as well as in various meteorological journals. He is co-editor of the book Probability, Statistics, and Decision Making in the Atmospheric Sciences.
Quality/value relationships for imperfect weather forecasts in a prototype multistage decision-making model
Article first published online: 23 SEP 2006
Copyright © 1990 John Wiley & Sons, Ltd.
Journal of Forecasting
Volume 9, Issue 1, pages 75–86, January/February 1990
How to Cite
Katz, R. W. and Murphy, A. H. (1990), Quality/value relationships for imperfect weather forecasts in a prototype multistage decision-making model. J. Forecast., 9: 75–86. doi: 10.1002/for.3980090107
- Issue published online: 23 SEP 2006
- Article first published online: 23 SEP 2006
- Manuscript Received: SEP 1988
- Dynamic programming;
- Markov decision process;
- Value of information;
- Weather forecasts
Some theoretical results concerning the nature of the relationship between the scientific quality and economic value of imperfect weather forecasts are obtained. A prototype multistage decision-making model is considered, involving only two possible actions and two possible states of weather. This particular form of model is motivated by a real-world application known as the fruit-frost problem. For an infinite-horizon, discounted version of this model it is shown that economic value remains zero below a forecast quality threshold and then rises monotonically but nonlinearly above this threshold. In particular, the relative sensitivity of economic value to changes in the quality of forecasts increases as perfect information is approached. This curve is compared with quality/value relationships that have been obtained for other versions of the model; namely, a single-stage model and a multistage, finite-horizon model.