GARLIC YIELD FORECASTING BY TIME SERIES MODELS

Abstract

The aim of the present study was to use time series models in order to realize forecasting garlic yield using data of bulb dry matter from six garlic accessions. The experiment was carried under completely randomized blocks design with 4 replicates. To adjust the time series models, data were collected in the periods: 60, 90, 120 and 150 days after planting (dap), being the latter the expected date for harvest. For this study, the plants were kept for 20 days in the field, 170 dap. The autocorrelation and partial autocorrelation functions were used in order to indicate the possible models order, which were chosen by AIC criterion. After obtaining the best model, one made the forecasting of the matter at 170 dap, and the predict values compared with the real ones. The linear regression models with residuals structure MA(1) was most adequate to forecasting garlic accessions yield 20 days after the expected time for harvest.