Use of high-resolution image data outperforms vegetation indices in prediction of maize yield

AGUATE, F.; TRACHSEL, S.; BURGUEÑO, J.; CROSSA, J.; BALZARINI, M.; DE LOS CAMPOS, G.

Victoria

Conferencia; XXVIIIth International Biometric Conference; 2016

Modern high-resolution cameras can provide reflectance data at potentially hundreds of wavelengths. This information can be used to predict physiological agronomic and disease traits. Traditionally, image data were used to derive vegetation indices (VI), which are predictive of traits. However, the data generated by high-resolution cameras contains more information than what can be summarized using VI. The objective of this study was to compare the predictive performance of regression methods using information from 62 bands to that of VI derived from the same reflectance data. We considered both ordinary-least square regressions and a Bayesian shrinkage/variable selection procedure. The data were generated by CIMMYT in 12 maize yield trials conducted in 2014 under irrigation and combined heat and drought stress. The trait analyzed was grain yield (ton/ha) and inputs were either VI or normalized reflectance at 62 bands, all collected at five different time points from flowering to pre-harvest. We show that using data from all bands leads to higher cross-validation prediction accuracy than using VI. Among the models that used data from a single time point, the ones using data collected at pre-harvest gave the highest prediction accuracy. Combining image data collected at different time points led to a small increase in prediction accuracy relative to models using data from a single time point.