Discussion on the requirements of VNIR hyperspectral data for agricultural applications according to PHI data and CSAM model

According to the application of PHI data to agricultural information extraction both in Changzhou, China and Nagano. Japan. we found that the results closely depend on the hyperspectral remote sensing data quality. As the spectral features of vegetation are particular with high shape similarity but with high intensity variation, the quality requirements are special. The general factors of pixel size, band combination and noise limit for different information objects are discussed in this paper. According to the two models of CSAM and IARS, with rice as example, it shows that a pixel size of 1.2 meter is suitable in the areas such as Changzhou and Nagano. The top-priority range of VNIR should be the red edge of 670-780 nm and fine-division is necessary with a suggestion of about 40 bands with 3 nm resolution. While in belts of 400-670 and 780-900 nm. wider and less bands are acceptable. According to the application of IARS and the linear property of rice spectra. it was suggested that S/N in order of /spl times/10 is enough for plant recognition. but for species classification for a certain plant, it should be square with all order up to /spl times/10/sup 2/-/spl times/ 10/sup 3/.