Computer-Aided Crop Yield Forecasting Techniques - Systematic Review Highlighting the Application of AI

Accurate yield forecasts can assist decision-makers in developing plans to bridge the food demand gap in the context of changing climatic conditions. Literature shows a variety of methodologies for forecasting crop yield; however, it is difficult to find a general methodology/model or a better one among the available literature. This review provides insight into the yield forecasting techniques available for agricultural crops, highlighting that most of the work has focused on wheat and rice crops. Most studies have mainly concentrated in Asia, Europe, the USA, and Africa. Of all the 54 selected publications, 70% of the papers have developed models by AI techniques. The statistical indices commonly used to compare the developed models are RMSE and correlation coefficient. From the standpoint of model performance and reliability of outcomes, the hybrid model (integrated approach of ML, namely, CNN/XGBoost and CSM and other crop models) has improved overall efficiency compared to standalone models. The AI tools can improve the accuracy of simulations by considering the effects of variables and processes that are not simulated in crop models. A range of input datasets, including meteorological parameters, crop characteristics, and hydro-geological properties, have been used for the model development. The results demonstrate that maximum temperature is the influencing parameter in model development. This study also demands the inclusion of local/ regional variables as inputs for such modeling studies.