Proximal Hyperspectral Imaging Detects Diurnal and Drought-Induced Changes in Maize Physiology

Hyperspectral imaging is a promising tool for non-destructive phenotyping of plant physiological traits, which has been transferred from remote to proximal sensing applications, and from manual laboratory setups to automated plant phenotyping platforms. Due to the higher resolution in proximal sensing, illumination variation and plant geometry result in increased non-biological variation in plant spectra that may mask subtle biological differences. Here, a better understanding of spectral measurements for proximal sensing and their application to study drought, developmental and diurnal responses was acquired in a drought case study of maize grown in a greenhouse phenotyping platform with a hyperspectral imaging setup. The use of brightness classification to reduce the illumination-induced non-biological variation is demonstrated, and allowed the detection of diurnal, developmental and early drought-induced changes in maize reflectance and physiology. Diurnal changes in transpiration rate and vapor pressure deficit were significantly correlated with red and red-edge reflectance. Drought-induced changes in effective quantum yield and water potential were accurately predicted using partial least squares regression and the newly developed Water Potential Index 2, respectively. The prediction accuracy of hyperspectral indices and partial least squares regression were similar, as long as a strong relationship between the physiological trait and reflectance was present. This demonstrates that current hyperspectral processing approaches can be used in automated plant phenotyping platforms to monitor physiological traits with a high temporal resolution.

Automated summary

Hyperspectral imaging is a useful tool for non-destructive phenotyping of plant physiological traits, especially for studying responses to drought, development, and day-night cycles. With better understanding of spectral measurements, it has been demonstrated to accurately predict physiological changes in plants under different conditions using advanced processing techniques such as partial least squares regression.