Future Projections of Iraq's Cereal Yields Using Wavelet-Arima Model

Abstract

This study presents an enhanced forecasting methodology for Iraq`s cereal production by integrating wavelet denoising with ARIMA modeling. Using annual production data (58 observations from 1961-2018), we demonstrate that preprocessing with the Daubechies wavelet of order 2 and soft thresholding significantly improves forecast accuracy by isolating noise while preserving trends. Our proposed wavelet-ARIMA hybrid model outperforms classical ARIMA across some statistical criteria (RMSE and AIC), achieving an increase in predictive precision through optimized signal decomposition. Empirical results reveal that wavelet analysis constructs latent patterns in fluctuating agricultural data that traditional methods overlook, enabling more reliable long-term projections. Based on this framework, Iraq`s cereal yields are forecasted to stabilize approximately at 4.03 million metric tons annually for the years 2019-2025, with a narrower confidence interval than ARIMA alone. These results provide mainly two important contributions. Firstly, a procedure that was validated for denoising of non-stationary agricultural time series, which is often affected by random noise. Secondly, it brings useful recommendations for decision-makers to mitigate the risks that are related to food security, especially in dry climate areas. The adaptability of the proposed method indicates that it can also be applicable to other product forecasting, where the volatility in data makes traditional techniques less effective.