NDSI map and IPLS using hyperspectral data for assessment of plant and ecosystem variables: With a case study on remote sensing of grain protein content, chlorophyll content and biomass in rice [Oryza sativa]
2008
Inoue, Y.(National Inst. for Agro-Environmental Sciences, Tsukuba, Ibaraki (Japan)) | Miah, G. | Sakaiya, E. | Nakano, K. | Kawamura, K.
Hyperspectral reflectance is an important data source for remote sensing of vegetation and ecosystems, especially in assessment of ecophysiological and physiochemical properties. Nevertheless, major advantages of hyperspectra such as waveband richness, sharpness of wavebands, and spectral continuity have not been well utilized to date. Here, we investigated the ability of two methods 1) normalized difference spectral index (NDSI) using thorough combinations of two wavebands (NDSI [i, j]=[Rj-Ri]/[Rj+Ri] using reflectance values Ri and Rj at i and j nm wavelengths), and 2) partial least squares regression (PLS) with waveband selection based on a case study in rice canopies. Three important rice variables (grain protein content: GPC, chlorophyll content: CHL, and aboveground biomass: AGB) were analyzed with airborne and ground-based hyperspcetral data. Results showed that the predictive-ability map of NDSI was useful for extracting effective wavebands and bandwidth for specific variables. NDSI [970, 570], NDSI [710, 550], and NDSI [710, 630] were selected as the best NDSIs for predicting GPC, CHL and AGB, respectively. The predictive ability of the best NDSIs was much higher than that of NDVI, especially in GPC and AGB, and comparable with those of the multiple linear regression (MLR) using four selected bands or PLS using all bands. The PLS with waveband selection used only 20-50% of all wavebands, but showed much higher predictive ability than the other methods. Number of wavebands selected commonly for the three variables was only 2% and 10-20% for any pairs among the three, whereas 18% was never used. Results suggest that NDSI map and IPLS are useful for advanced use of hyperspectral data, especially in parallel assessment of multiple vegetation/ecosystem variables.
显示更多 [+] 显示较少 [-]