Assessment of digital camera-derived vegetation indices in quantitative monitoring of seasonal rice growth

A commercially available digital camera can be used in a low-cost automatic observation system for monitoring crop growth change in open-air fields. We developed a prototype Crop Phenology Recording System (CPRS) for monitoring rice growth, but the ready-made waterproof cases that we used produced shadows on the images. After modifying the waterproof cases, we repeated the fixed-point camera observations to clarify questions regarding digital camera-derived vegetation indices (VIs), namely, the visible atmospherically resistant index (VARI) based on daytime normal color images (RGB image) and the nighttime relative brightness index (NRBI_NIR) based on nighttime near infrared (NIR) images. We also took frequent measurements of agronomic data such as plant length, leaf area index (LAI), and aboveground dry matter weight to gain a detailed understanding of the temporal relationship between the VIs and the biophysical parameters of rice. In addition, we conducted another nighttime outdoor experiment to establish the link between NRBI_NIR and camera-to-object distance. The study produced the following findings. (1) The customized waterproof cases succeeded in preventing large shadows from being cast, especially on nighttime images, and it was confirmed that the brightness of the nighttime NIR images had spatial heterogeneity when a point light source (flashlight) was used, in contrast to the daytime RGB images. (2) The additional experiment using a forklift showed that both the ISO sensitivity and the calibrated digital number of the NIR (cDN_NIR) had significant effects on the sensitivity of NRBI_NIR to the camera-to-object distance. (3) Detailed measurements of a reproductive stem were collected to investigate the connection between the morphological feature change caused by the panicle sagging process and the downtrend in NRBI_NIR during the reproductive stages. However, these agronomic data were not completely in accord with NRBI_NIR in terms of the temporal pattern. (4) The time-series data for the LAI, plant length, and aboveground dry matter weight could be well approximated by a sigmoid curve based on NRBI_NIR and VARI. The results confirmed that NRBI_NIR was more sensitive to all of the agronomic data for overall season, including the early reproductive stages. VARI had an especially high correlation with LAI, unless yellow panicles appeared in the field of view.