Probability forecasts for weather-dependent agricultural operations using generalized estimating equations

Abstract

In agricultural production many operations depend on the weather. In this paper, a model is investigated that calculates the probability for the execution of a given operation which depends on several meteorological parameters. The model is based on a 48-hr numerical weather forecast with hourly resolution. The probability forecasts are compared to the numeric forecasts for the operation based on the numeric weather forecast.

The model is a logistic regression model with generalized estimating equations. The Brier skill score, sharpness and reliability diagrams and relative operating characteristic curves are used to evaluate the model.

The model setup described is dynamic in the sense that on a given date, parameters are estimated based on history and these parameter estimates are used for calculating the probability forecasts. This means that parameter estimates adapt automatically to seasonal changes in weather and to changes in numerical weather forecasts following developments in the forecast models.

In this paper, we perform model output statistics, which tune the numeric weather forecast to an operation that depends on several meteorological parameters rather than only tuning a single weather parameter. Although some problems occurred, the model developed showed that the numerical forecast for such an operation could be improved.