Rice yield forecasting using principal component regression and composite weather variables

Abstract

Reliable and timely forecasting helps policy makers and farmers to carry out proper crop planning. Weather based crop forecasting is having a paramount importance, since more than 50% of crop loss is due to uncertainties in weather. In this study two weather based statistical models, Principal component regression and composite weather variables were used to forecast yield of two different rice varieties i.e Jaya and Kanchana. Weather data were collected from principal Agromet observatory, Vellaikkara. Rice yield data during kharif season were collected from the experiment plot maintained in Agricultural Research Station, Mannuthy for a period of 2013 to 2019. Both PCR and CWV models explains the interactive effect of weather variables. The goodness of fit of these models were tested using t test. Calculated value of t was found to be less than that of t- critical value in both the models. Hence it was found out that the predicted yield was similar to that of actual yield. A comparison of model performance was done by estimating mean absolute percentage error (MAPE). The MAPE value calculated for PCR model in Jyothi was 0.76% and 6.19% and for CWV model the MAPE values were 3.69 for Jyothi and 2.95 for Kanchana. Eventhough by comparing MAPE value, PCR model was found to be performing better for the variety Jyothi and models based on composite weather variables performed better in the variety Kanchana. Both the models showed a good performance since the error percentage was in the acceptable limit of ±10%.