Estimation of maize yield by assimilating biomass and canopy cover derived from hyperspectral data into the AquaCrop model

Abstract The accurate and timely estimation of temporal and spatial changes in crop growth and yield before harvesting is essential for ensuring global food security. The integration of remote sensing data and crop models is a potential approach for the estimation of key crop growth parameters and crop yields. Therefore, the aim of this study was to assimilate biomass and canopy cover (CC) derived from vegetation indices into the AquaCrop model using the particle swarm optimization (PSO) algorithm in order to obtain a more accurate estimation of CC, biomass, and yield for maize. The results show that, compared to other vegetation indices, the enhanced vegetation index (EVI) and the three-band water index (TBWI) can be used to obtain a better estimation of CC (R2 = 0.78 and root-mean-square error (RMSE) =9.84%) and biomass (R2 = 0.76 and RMSE = 2.84 ton/ha), respectively. Additionally, it was found that the data assimilation approaches in which only CC was used as a state variable (scheme SVcc) and only biomass was used as a state variable (scheme SVbio) can be used to obtain more accurate estimations of CC (R2 = 0.83 and RMSE = 8.12%) and biomass (R2 = 0.81 and RMSE = 2.51 ton/ha), respectively; however, larger differences were found between the measured and estimated values of one variable (i.e., CC or biomass) when the other variable (i.e., biomass or CC) was used as the only state variable during the data assimilation. The data assimilation approach in which both CC and biomass were used as state variables (scheme SVcc+bio) produced a robust result, with the estimation accuracy being fairly close to that obtained using the single-variable (SVcc or SVbio) data assimilation approaches. The estimation accuracy for maize yield was slightly better when using a double-variable data assimilation approach (R2 = 0.78 and RMSE = 1.44 ton/ha) than when using a single-variable data assimilation approach. In summary, this study presents a robust approach for increasing the estimation accuracy for maize CC, biomass, and yield, and for optimizing field management strategies, by assimilating remote sensing data into the AquaCrop model at a regional scale.