Abstract:
Abstract: The selection of sensitive band center and optimal band width is of great significance to improve accuracy of crop biomass estimation based on hyperspectral data. As one of main food crops, winter wheat yield is critical for food safety and winter wheat biomass is the base of crop productivity, so accurate estimation of winter wheat biomass is particularly important. Objective of the study was to determine sensitive spectral band centers and their band widths which were best suited for characterizing agricultural crop biophysical variables. The experiment data included winter wheat canopy hyperspectral reflectance data between 350 nm and 1 000 nm in critical crop growth stages and field-measured fresh crop biomass. In order to achieve above purposes, firstly linear models were established between fresh winter wheat biomass and narrow band normalized difference vegetation indexes (N-NDVI) derived from crop canopy hyperspectral reflectance. Then two-dimensional distribution of R2 values was drawn through analyzing correlations between winter wheat fresh biomass and N-NDVI of any two bands. In order to select optimal band width, area weight of R2 maximum values was regarded as the hyperspectral sensitive band-pair center because of the non-uniform of R2 distribution. After that, band widths of sensitive band centers were extended with a step length of ±1 nm (±3 nm when band width exceeded 50 nm). Finally, the results of band extension were validated and the optimal band widths of sensitive band centers were ultimately determined at a higher accuracy level. On this basis, winter wheat fresh biomass were estimated based on the optimal band widths of sensitive band centers and the accuracy of the winter wheat biomass estimation results were validated. The results indicated that five band-pairs centered at 401 nm/692 nm, 579 nm/698 nm, 732 nm/773 nm, 725 nm/860 nm, and 727 nm/977 nm were the best combinations for fresh crop biomass estimation as the weight of an area with R2≥0.65 was selected as a sensitive band center for fresh crop biomass estimation from hyperspectral data. For each extension, crop biomass was estimated with N-NDVI constructed by average reflectance of band-pairs, the optimal band widths were ±21, ±5, ±51, ±40 and ±23 nm respectively for the above mentioned sensitive band-pair centers when NRMSE (normalized root mean square error) and RE (relative error) were both less than or equal to 10%. The estimated winter wheat biomass based on optimal band width showed significant correlation with field measured fresh biomass data at P<0.01 level, and RE were in the range of 8.15%-9.14%, NRMSE were in the range of 8.69%-9.65%. These indicated that the method of determining hyperspectral sensitive band centers and the optimal band widths based on areas weight of R2 maximum values between N-NDVI and winter wheat fresh biomass had certain feasibility and effectiveness in the study. The method could provide a new thought thread of crop hyperspectral band selection in crop monitoring, and it also could provide a certain basis for band settings of broadband multispectral imaging spectrometer and for evaluating potential applications of remote sensing data.