Sunflower yield forecast using ARIMA time series models

Abstract

The forecast of sunflower yield was based on the analysis of the time series of yield data of this crop at its cultivation in the Northern Steppes of Ukraine against the background of natural fertility for 1971-2019. The true average yield value of sunflower ranged from 2.15 ± 0.17 t/ha, the average variation of yield data for the study period was: coefficient of variation – Cv = 24 %, standard deviation – s = 0.516 t/ha. Analysis of the scattering graph of the series showed a tendency to increase the sunflower yield over a given period of time. An adequate linear model with an increasing trend of yield data is obtained. According to the forecast results by this method for the period up to 2025, the sunflower yield is expected at the level of 2.59–2.67 t/ha. Forecasting with ARIMA (Autoregressive Integrated Moving Average) was carried out by reduction of the yield data series to a stationary form, which was achieved by first order differentiation D (-1). The selection of the most adaptive model was carried out by varying the values of p and q, according to the type of autocorrelation (ACF) and partial autocorrelation functions (PACF). It was found that the best model is D (-1) ARIMA model: (2,0,0), the stationarity of which was achieved by first order differentiation, the residuals are not autocorrelated and normally distributed, and the regression coefficients corresponded to the values of residual probabilities less (p <0, 05). According to the short-term forecast based on the chosen model, it was found that the maximum of sunflower yield against the background of natural fertility in 2023 should be expected up to 3.56 t/ha. Keywords: forecast, yield, sunflower, model, time series, ARIMA model.