Estimating the dynamic harvest index of winter wheat using the fraction of accumulated aboveground biomass after flowering

The harvest index (HI) has been served as a predictor of the crop yield in fields. An accurate and rapid HI estimation is highly required using remote sensing. However, the current estimation cannot fully consider the change of crop biomass and HI in the process of grain filling so far. It is very necessary to improve the accuracy of HI estimation in recent years. Taking the winter wheat as the research object, supported by the canopy hyperspectral data and aboveground biomass data in different growth periods after flowering, this study aims to estimate the Dynamic Harvest Index (DHI) based on the fraction of accumulated aboveground biomass after flowering. In this paper, the concept of DHI from grain filling to maturity was proposed, and the dynamic fG parameter (D-fG) was developed, which was defined as the ratio of the aboveground biomass accumulation after flowering to that in the given period. The fitting accuracy R2 two-dimensional diagrams were drawn between the Normalized Difference Spectral Index (NDSI) and D-fG. The sensitive band centers to the winter wheat D-fG were obtained by determining the centers of gravity of the local maximum region of R2 values. Then, on the basis of the statistical relationship model between the measured D-fG and DHI, the estimation and accuracy verification of the DHI of winter wheat based on the D-fG remote sensing parameter information was carried out. At the same time, the sensitive band width was determined when the normalized root mean square error (NRMSE) and the mean relative error (MRE) reached the maximum permissible error (15%). Furthermore, the maximum width of sensitive band was determined for the winter wheat D-fG estimation, further to realize the D-fG remote sensing estimation and DHI remote sensing acquisition under the maximum width of the sensitive band. The results showed that the high level of accuracy was achieved in the D-fG remote sensing estimations using the NDSI of selected five sensitive band centers (such as λ(366 nm, 489 nm), λ(443 nm, 495 nm), λ(449 nm, 643 nm), λ(579 nm, 856 nm) and λ(715 nm, 849 nm)). Among them, the root mean square errors (RMSE) were between 0.036 and 0.050, the NRMSEs were between 10.46% and 14.59%, and the MREs were between 9.49% and 12.78%, respectively. In the DHI estimation using D-fG based on the canopy hyperspectral sensitive band centers, the RMSEs were between 0.039 and 0.053, the NRMSEs were between 10.50% and 14.28%, and the MREs were between 9.27% and 13.25%, respectively. The maximum band widths of five sensitive band centers for D-fG estimation were 30, 68, 58, 20 and 86 nm, respectively. In the D-fG remote sensing estimation using NDSI constructed by the band centers with the maximum band widths, the RMSEs were between 0.051 and 0.052, the NRMSEs were between 14.85% and 14.98%, and the MREs were between 13.43% and 14.82%. In the DHI estimation using D-fG obtained from the NSDI constructed by the maximum width bands, the RMSEs were between 0.054 and 0.055, the NRMSEs were between 14.38% and 14.65%, and the MREs were between 12.95% and 13.70%. Consequently, the remote sensing estimation method of the winter wheat DHI was feasible. The finding can also provide a strong technical reference for the narrow band hyperspectral and wide band multispectral satellite remote sensing data to obtain the spatial information of crop harvest index at the regional scale.