Data Analysis and Applications 1

13Frost Prediction in Apple Orchards Based upon Time Series Models

The scope of this work was to evaluate the autoregressive integrated moving average (ARIMA) model as a frost forecast model for South Tyrol in Italy using weather data of the past 20 years that were recorded by 150 weather stations located in this region. Accurate frost forecasting should provide growers with the opportunity to prepare for frost events in order to avoid frost damage. The radiation frost in South Tyrol occurs during the so-called frost period, i.e. in the months of March, April and May during calm nights between sunset and sunrise. In case of a frost event, the farmers should immediately switch on water sprinklers. The ice cover that builds on the trees protects the buds and blossoms from damage. Based on the analysis of time series data, the linear regression (LR) and ARIMA models were compared and evaluated. The best result was achieved by the ARIMA model, with the optimal value of 1.0 for recall in case of forecast of 95% confidence intervals. This means that all frost cases could be correctly predicted. Despite the encouraging results for recall, the rate of false positives with a sensitivity of 21% is too high, such that further investigations are desirable (e.g. testing VARIMA models, which are a multivariate extension of ARIMA models). The graphical illustration of the 95% confidence intervals of the ARIMA model forecast and the linear models forecast should be helpful in frost prediction ...