| Crop biomass is the basis of crop yield formation, and accurate biomass information has vital significance for guiding agricultural production, ensuring national food security, promoting agricultural sustainable development and global carbon cycle, etc. Winter wheat is one of the main food crops, and the yield is critical for food safetyand its biomass information is particularly important. Biomass estimation involves a variety of disciplines and technologies, and compared to the traditional estimation methods which are time-consuming, the method of remote sensing has a great advantage of simultaneous observation in large area. Further more, compared to the traditional optical remote sensing, hyperspectral remote sensing has abundant spectral information which provides more data to estimate biomass. However hyperspectral data has considerable redundancy, so it is important to utilize hypespectral information efficiently and futher to guide multispectral remote sensing use. So, this paper selected sensitive band centers and optimal band widths based on winter wheat biomass hyperspectral estimation, on this basis, to guide regional winter wheat multispectral remote sensing estimation.This paper selected Hengshui city of Hebei province as the study area, which is the major grain producing areas in China, and winter wheat biomass, canopy spectra and GPS informations were collected in experimental samples. Then, linear models were established between winter wheat biomass and narrow band vegetation indices(N-VIs) derived from crop canopy hyperspectral reflectance. After that, two-dimensional distribution of R2 values was drawn through analyzing correlations between winter wheat biomass and N-VIs of any two bands. In order to select optimal band width, the maximum area was determined by setting the threshold value and area weight of R2 maximum values was regarded as the hyperspectral sensitive band-pair center because of the non-uniform of R2 distribution and band widths of sensitive band centers were extended. 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, Hyperion, GF-1 and Landsat8 which have different spatial resolution or spectral resolution were used to estimate regional winter wheat biomass under the direction of sensitive band centers. The main conclusions were follows:(1) Sensitive band centers were screened through correlation analysis between winter wheat biomass and N-VIs derived from canopy hyperspectral reflectance, and optimized band widths through band extension under the condition of RE≤10%, NRMSE≤10%. The results showed that, 401 nm/692 nm, 579 nm/698 nm and 732 nm/773 nm etc 5 fresh biomass estimation sensitive band-pairs based on N-NDVI were selected, and the band widths between 10 nm and 102nm; 387nm/840 nm, 465nm/500 nm and 527nm/963 nm etc 8 dry biomass estimation sensitive band-pairs based on N-NDVI were selected, and the band widths between 24 nm and 62nm; 818nm/614 nm, 821nm/734 nm and 986nm/844 nm 3 fresh biomass estimation sensitive band-pairs based on N-DVI were selected, and the band widths between 28 nm and 194nm; 502 nm/454 nm, 623 nm/428 nm and 947 nm/593 nm etc 4 dry biomass estimation sensitive band-pairs based on N-DVI were selected, and the band widths between 24 nm and 34nm; 502 nm/454 nm, 623 nm/428 nm and 947 nm/593 nm etc 4 dry biomass estimation sensitive band-pairs based on N-DVI were selected, and the band widths between 24 nm and 34nm; 398nm/672 nm, 551 nm/865 nm and 577 nm/699 nm etc 9 fresh biomass estimation sensitive band-pairs based on N-RVI were selected, and the band widths between 18 nm and 146nm; 439 nm/623 nm, 506 nm/461 nm and 538 nm/965 nm etc 9 dry biomass estimation sensitive band-pairs based on N-RVI were selected, and the band widths between 22 nm and 86 nm.The estimated winter wheat biomass based on optimal band width showed significant correlation with field measured fresh biomass data at p<0.01 level(2) Regional winter wheat biomass inversion based on Hyperion band showed good result. The result showed that, RVI constructed by Band 79 and Band 37 showed the best result for fresh biomass estimation, R2 values is 0.6665, RE and NRMSEwere 10.71% and 11.99%, respectively; RVI constructed by Band 79 and Band 19 showed the best result for dry biomass estimation, R2 values is 0.6848, RE and NRMSEwere 9.24% and 10.45%, respectively.(3) Winter wheat biomass was estimated based on simulated remote sensing data and real remote sensing data. At the regional scale, winter wheat biomass was estimated based on NDVI, DVI and RVI derived from Hyperion hyperspectral remote sensing data and GF-1, Landsat8 wide band multispectral remote sensing datas and the accuracy of biomass remote sensing inversion were verified. Compared to NDVI and DVI, RVI achieved the best results of biomass remote sensing inversion in the major winter wheat growing period. For the result of regional winter wheat biomass remote sensing inversion which based on simulated GF-1 and Landsat8 broad band by canopy hyperspectral, the simulated remote sensing data had better performance than real remote sensing data in biomass estimation. Futher, GF-1 data showed better accuracy than Landsat8 data in winter wheat biomass estimation based on both simulated and real data. The results showed that, RVI performed best, and the rank of biomass estimation accrucy based on simulated remote sensing data was: GF-1 > Landsat8, and the rank of biomass estimation accrucy based on real remote sensing data was: Hyperion > GF-1 > Landsat8.The results showed that, fresh biomass estimation based on Hyperion hyperspectral data, R2 value was 0.6665, RE, NRMSE were 10.71% and 11.99%, respectively, and R2 was 0.6988, RE, NRMSE were 9.40% and 10.10 % respectively for dry biomass estimation. For fresh biomass estimation based on simulated GF-1 data, R2 value was 0.7124, RE, NRMSE were 8.33% and 10.38%, respectively, and R2 was 0.7292, RE, NRMSE were 11.59% and 11.58 % respectively for dry biomass estimation. For fresh biomass estimation based on simulated Landsat8 data, R2 value was 0.7126, RE, NRMSE were 10.15% and 11.22%, respectively, and R2 was 0.7028, RE, NRMSE were 13.44% and 13.47 % respectively for dry biomass estimation. For fresh biomass estimation based on GF-1 data, R2 value was 0.6972, RE, NRMSE were 13.48% and 14.06%, respectively, and R2 was 0.6821, RE, NRMSE were 10.31% and 15.07 % respectively for dry biomass estimation. For fresh biomass estimation based on Landsat8 data, R2 value was 0.7098, RE, NRMSE were 12.83% and 14.84%, respectively, and R2 was 0.6383, RE, NRMSE were 12.70% and 15.74 % respectively for dry biomass estimation.(4) According to the regional biomass inversion results of comparison analysis of Hyperion, GF-1 and Landsat8, Hyperion data achieved the highest accuracy, which showed hyperspectral data has greater potential for regional biomass inversion; compared higher spatial resolution GF-1 data and Landsat8 data, GF-1 data had a better ability of biomass eatimation.In conclusion, the method of determining hyperspectral sensitive band centers and the optimal band widths based on areas weight of R2 maximum values between N-VIs and winter wheat biomass had certain feasibility, which provided a new thought thread of crop hyperspectral band selection in crop monitoring. On this basis, to achieve biomass high-precision inversion baesd on hyperspectral and multispectral remote sensing, this provided basis to effective application of hyperspectral data, setting multispectral band and the evaluation of remote sensing data potential in application. |
PDF Full Text Request