The present study aims to identify the narrow spectral bands that are most suitable for characterizing rice biophysical parameters. The data used for this study come from ground-level hyperspectral reflectance measurements for five rice species at three levels of nitrogen fertilization during the growing period. Reflectance was measured in discrete narrow bands between 350 and 2500 nm. Observed rice biophysical parameters included leaf area index (LAI), wet biomass and dry biomass. The stepwise regression method was applied to identify the optimal bands for rice biophysical parameter estimation. This research indicated that combinations of four narrow bands in stepwise regression models explained 69% to 83% variability for LAI, 56% to 73% for aboveground wet biomass and 70% to 83% for leaf wet biomass. An overwhelming proportion of rice information was in a particular portion of near infrared (NIR) (1 100-1 150 nm), red-edge (700-750 nm), and a longer portion of green (550-600 nm). These were followed by the moisture-sensitive NIR (950-1 000 nm), the intermediate portion of shortwave infrared (SWIR) (1 650-1 700 nm), and another portion of NIR (1 000-1 050 nm).
Large-scale farming of agriculture crops requires real-time detection of disease for field pest management. Hyperspectral remote sensing data generally have high spectral resolution, which could be very useful for detecting disease stress in green vegetation at the leaf and canopy levels. In this study, hyperspectral reflectances of rice in the laboratory and field were measured to characterize the spectral regions and wavebands, which were the most sensitive to rice brown spot infected by Bipolaris oryzae (Helminthosporium oryzae Breda. de Hann). Leaf reflectance increased at the ranges of 450 to 500 nm and 630 to 680 nm with the increasing percentage of infected leaf surface, and decreased at the ranges of 520 to 580 nm, 760 to 790 nm, 1550 to 1750 nm, and 2080 to 2350 nm with the increasing percentage of infected leaf surface respectively. The sensitivity analysis and derivative technique were used to select the sensitive wavebands for the detection of rice brown spot infected by B. oryzae. Ratios of rice leaf reflectance were evaluated as indicators of brown spot. R669/R746 (the reflectance at 669 nm divided by the reflectance at 746 nm, the following ratios may be deduced by analogy), R702/R718, R692/R530, R692/R732, R535/R746, R521/R718, and R569/R718 increased significantly as the incidence of rice brown spot increased regardless of whether it’s at the leaf or canopy level. R702/R718, R692/R530, R692/R732 were the best three ratios for estimating the disease severity of rice brown spot at the leaf and canopy levels. This result not only confirms the capability of hyperspectral remote sensing data in characterizing crop disease for precision pest management in the real world, but also testifies that the ratios of crop reflectance is a useful method to estimate crop disease severity.
