Crop Yield Forecasting Using Bidirectional Gated Recurrent Unit (Bi-GRU) Networks

Crop yield forecasting is an important aspect of the sustainable agriculture to help farmers and policy makers anticipate both production quantities and qualities, thus providing a technical tool for decision making. In this paper, a new technique of using Bidirectional Gated Recurrent Unit (Bi-GRU) to forecast the growth rate of one or several crops with improved precision and credibility is introduced. The model used here utilizes sequence data of meteorological parameters, soil, satellite index, and yield data also to incorporate temporal dependency and non-linear trends. Imputation of missing values; normalization; feature scaling all helps to remove inconsistencies in data and improve model performance. Folically, the employment of the Bi-GRU model provided a significan improvement over traditional machine learning models and unidirectional models with RMSE of 460.3 kg/ha and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> of 0.89. Moreover, hyperparameters tuning enhanced the model revealing the importance of bidirectional architecture in managing and capturing the past and future temporal contexts. Similar trends of increased TVB-accuracy arias were observed in sub-Regional analysis considering different crop types and environmental conditions and the accuracy prediction of the yield increased up to 14.1%. This work suggests that Bi-GRU networks can be beneficial to predict crop yields with predictive power, reliability, and fine-grained architectures to handle problems such as climate fluctuations and dataset heterogeneity. The study provides a solution of how to incorporate deep learning algorithms into smart agriculture systems for development, posterity.