Research advances and perspectives on crop yield estimation models

Although crop yield estimation is a necessary requirement of modern agriculture, it is one of the most difficult things to monitor in agriculture. Timely and accurate simulation of crop yield is important for national agricultural decision-making, agricultural production management, grain storage safety, etc. Model simulations of crop growth and yield formation are currently the most commonly method of crop yield estimation. Crop growth and yield formation models were divided into four categories after comparison on theoretical basis, which were empirical linear models, crop growth models, light use efficiency (LUE) models and coupled models. As so many different crop growth models existed, further classification of the models was necessary. The empirical linear models was further divided into four sub-groups according to their estimation methods, while the crop growth models were further divided into four sub-groups on the basis of the main or special driving factors. Then the paper analyzed the merits and demerits of each group of models. Although empirical linear models were simple and needed less data, they had poor generalization in space and time. Crop growth models were more comprehensive and reasonable as they were capable of simulating almost all plant physiological processes and even human disturbances. The shortages of these models were also obvious. The models required more parameters, most of which were not easily accessible. The models also had high software, hardware and professional (knowledge) requirements to accomplish operations. LUE models were capable of comprehensive simulation of light use and easily fitted for remote sensing data to improve simulation precision. The most obvious demerit of LUE models was their inability to simulate human disturbances, a non-ignorable factor, as farm environment in modern agriculture was highly subjected to human activity. Although the coupled models combined the merits of both crop and LUE models, they also shared the demerits of these models and with the theoretical basis widely questioned. This study also discussed and drew conclusions on the use of remote sensing data into the models. After concluded on the limiting factors of development of the models, hot spots of research on the models were discussed. The study finally summarized some possible development trends and prospects of the crop yield estimation models. It was concluded that the models had the potential to be more stable, efficient, accurate, practical and cost efficient as they were drivable on common software and hardware conditions and that even farmers could use them. The possible ways of resolving crop yield-estimation difficulties were optimizing crop models and innovatively using new remote sensing data such as radar data, hyperspectral data and high spatial resolution data.