University of Leicester
63_2020_CEA_Tian_doi_10.1016j.compag.2019.105180.pdf (2.08 MB)

An IPSO-BP neural network for estimating wheat yield using two remotely sensed variables in the Guanzhong Plain, PR China

Download (2.08 MB)
journal contribution
posted on 2020-05-19, 11:04 authored by Huiren Tian, Pengxin Wang, Kevin Tansey, Shuyu Zhang, Jingqi Zhang, Hongmei Li
Early and accurate information of crop growth condition is vital for agricultural industry and food security, which gives rise to a strong demand for timely monitoring crop growth condition and estimating crop yields. This study selected the remotely sensed leaf area index (LAI) and vegetation temperature condition index (VTCI) which closely relate to crop growth and crop water stress as two key variables for indicating crop growth condition and estimating crop yields in the Guanzhong Plain, PR China. The single VTCI, the single LAI and the combination of VTCI and LAI at four growth stages of winter wheat (the turning green, jointing, heading-filling, and dough stages) were used as three input variable schemes of the back propagation (BP) neural network and the improved particle swarm optimization algorithm (IPSO)-BP neural network using a nonlinear decreasing inertia weight, respectively. The relative importance of the input variables to the output variable, yield of winter wheat, was used to determine the weight values of input variables at each growth stage. Based on the weights, the integrated index (I) was established, and then three linear regression models (weighted VTCIs, weighted LAIs, and I values) were established with yield data to estimate winter wheat yields. By calculating several statistical functions, i.e., coefficient of determination (R2) and probability value (P), the model between the I values and wheat yield performed better than those between the weighted VTCIs or weighted LAIs and wheat yields. The yield estimation model of I values by using the IPSO-BP neural network (R2 = 0.342) was found to be better than that using the BP neural network (R2 = 0.310). Therefore, we applied the model with better performance (R2 = 0.342) to map the regional winter wheat yields pixel by pixel in the Guanzhong Plain during 2011–2018, and analyzed the spatial and temporal characteristics of the estimated yields. Regarding the spatial distribution, the yields in the west part of the plain are the highest, followed by the central part, and the yields in the east part are lowest, consistent with previous studies. The estimated yields showed inter-annual fluctuations along with an increasing trend on the whole. Winter wheat yields were most depleted in 2013 and most abundant in 2015. These results were consistent with the actual situation of winter wheat production in the plain, which indicated that I can be used to provide a better quantification for monitoring regional winter wheat growth conditions and estimating crop yield. Thus, the approach of this study can provide significant benefit for regional crop production monitoring.


This work was supported by the National Natural Science Foundation of China under Grants 41871336 and 41811530303. This work was supported by a UK Science & Technology Facilities Council(STFC) Agri-Tech in China Newton Network+ (ATCNN) grant administered through Rothamsted Research. The work was further supported by a Royal Society-Newton Mobility Grant (UK).



Computers and Electronics in Agriculture Volume 169, February 2020, 105180


  • VoR (Version of Record)

Published in





105180 (10)







Acceptance date


Copyright date


Available date


Publisher version

Spatial coverage

Guanzhong Plain, PR China