Yield Adjustment Using GPR-Derived Spatial Covariance Structure in Cassava Field: A Preliminary Investigation

Abstract

Many processes concerning below-ground plant performance are not fully understood, such as spatial and temporal dynamics and their relation to environmental factors. Accounting for these spatial patterns is very important as they may be used to adjust for the estimation of cassava fresh root yield masked by field heterogeneity. The yield of cassava is an important characteristic that every breeder seeks to maintain in their germplasm. Ground-Penetrating Radar (GPR) has proven to be an effective tool for studying the below-ground characteristics of developing plants, but it has not yet been explored with respect to its utility in normalizing spatial heterogeneity in agricultural field experiments. In this study, the use of GPR for this purpose was evaluated in a cassava field trial conducted in Momil, Colombia. Using the signal amplitude of the GPR radargram from each field plot, we constructed a spatial plot error structure using the variance of the signal amplitude and developed GPR-based autoregressive (AR) models for fresh root yield adjustment. The comparison of the models was based on the average standard error (SE) of the Best Linear Unbiased Estimator (BLUE) and through majority voting (MV) with respect to the SE of the genotype across the models. Our results show that the GPR-based AR model outperformed the other models, yielding an SE of 9.57 and an MV score of 88.33%, while the AR1 × AR1 and IID models had SEs of 10.15 and 10.56% and MV scores of 17.37 and 0.00%, respectively. Our results suggest that GPR can serve a dual purpose in non-destructive yield estimation and field spatial heterogeneity normalization in global root and tuber crop programs, presenting a great potential for adoption in many applications.