A NEURAL-NETWORK SYSTEM FOR THE PROTECTION OF CITRUS CROPS FROM FROSTDAMAGE

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