A neural network system for the protection of citrus crops from frost damage

The island of Sicily depends heavily on its agricultural produce for economic prosperity, in particular its harvest of citrus crops. During most of the year the climate in this region is ideal for the cultivation of this type of fruit, but during some winter months, night frosts can occasionally occur. This is very damaging to the crop, resulting in loss of the harvest, some trees, or in extreme cases a whole orchard. There are methods available for both protecting against and preventing frost, but they need to be implemented before the frost actually occurs. In this paper the use of neural networks to predict the occurrence of frost from meteorological data is investigated. A range of different network architectures are trained and tested using data collected in Sicily between 1980 and 1983. The results indicate that the ability of a network to accurately predict minimum temperature varies greatly with the structure of the network. The best performance from a network using an unseen data set of 50 patterns resulted in correct predictions of overnight frost on all but one occasion.