Estimation of grain filling rate and thousand-grain weight of winter wheat (Triticum aestivum L.) using UAV-based multispectral images

Abstract

Estimating grain filling rate (GFR) and thousand-grain weight (TGW) plays an important role in evaluating yield and guiding the selection of varieties and cultivation strategies of winter wheat (Triticum aestivum L.). However, the current GFR and TGW monitoring methods mainly rely on destructive sampling, which can not achieve rapid estimation in a large area of farmland. This study aims to establish a method for estimating GFR and TGW of winter wheat using multispectral UAV images. Initially, grey correlation analysis method was used to evaluate the contributions of Leaf Area Index (LAI), Chlorophyll Content (SPAD), Aboveground Biomass (AGB) to GFR. A new comprehensive indicator, called LAI-SPAD-AGB index (LSA), was proposed to characterize GFR by establishing a linear regression model between LSA and GFR. Subsequently, UAV-based multispectral images were used to estimate LAI, SPAD, AGB, employing the methods such as Partial Least Squares Regression (PLSR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Using the linear regression equation between LSA and GFR along with estimated LSA values, GFR was estimated and mapped. TGW was estimated based on GFR and grain-filling duration (GFD). Results showed the high GFR estimation accuracy (R²: 0.89, RMSE: 0.29 g/d, NRMSE: 10.0 %) and remarkable TGW estimation precision (R²: 0.92, RMSE: 4.20 g, NRMSE: 8.1 %). The parcel-scale distribution maps of estimated GFR and TGW were generated. The novel and non-destructive method of estimating GFR and TGW of winter wheat using UAV-based images can offer strong support for water and fertilizer management in the field.