Using the DJPheno software for model calibration and validation of apple phenological stages.

Bourgeois, G., Plouffe, D., Beaudry, N., Choquette, D., Chouinard, G., and Bellerose, S. (2015). "Using the DJPheno software for model calibration and validation of apple phenological stages.", Acta Horticulturae (ISHS), 1068, pp. 117-124.

Abstract

Data on apple phenological stages and insect captures have been collected for over 30 years in many orchards across the province of Quebec in Canada. These biological data were used to develop a number of bioclimatic models, based on the cumulative degree-day (DD) approach. These models are being used mainly as decision tools for plant protection purposes. Since the available database can provide between 200 to 300 sets of data and since each of them can be coupled to measured temperature data, the DJPheno software (Degree-days estimator to predict phenological stages) was developed to determine the most appropriate lower threshold temperature (Tbase) and DD requirements for given phenological events in the process of DD model development and updating. Statistics, such as the root mean square error (RMSE) and the forecasting efficiency (EF), were used to determine the most appropriate model for the selected data sets. The overall approach was applied to the prediction of eight apple phenological stages, from green tip to fruit set. Data were collected from 1977 to 2005 in 13 pilot orchards covering the major apple growing regions in Quebec, for a total of 286 sets of data. These data were compiled and analyzed using the DJPheno software to generate the following DD model: beginning computation on 1 March with a Tbase of 5°C using the DD single sine method. To complete the model evaluation, a cross-validation method was implemented by creating seven subsets corresponding to different geographical regions of the province. Comparative statistics RMSE and EF obtained from the validation for each region were similar to those from the calibration process. According to these results, the selected DD model provided excellent generalization ability when used with weather data from apple growing regions of Quebec.

Date modified: