Winter wheat yield estimation model based on information diffusion and remote sensing data at major growth stages

Developing high accuracy models for crop yield estimation using remote sensing data is of great significance in decision making for national food policy and food security. Information diffusion methodology was introduced to construct yield estimation model with remote sensing data in the paper. Firstly, Remote sensing data at key stages and ground survey data were diffused into multi-dimensional control space and a fuzzy synthetic method was proposed to construct the relationship between remote sensing data and ground survey data. Secondly, cross validation was used to estimate the model’s stability and forecasting ability. Finally, the performance of information diffusion yield estimation model was compared with multiple linear regression model and BP neural network model. The results showed that information diffusion yield estimation model could obviously increase the precision and stability of yield prediction. The determination coefficients were increased by 0.180 and 0.491, respectively, while the root mean squared errors were decreased by 173.10 kg/hm2 and 487.79 kg/hm2 compared with the multiple linear regression model and BP neural network model. The proposed yield estimation model can simulate the non-linear relationship between NDVI and winter wheat yield with excellent generalization ability, which is an effective model to estimate crop yield with multi-temporal remote sensing data.