R.K. Gupta

Comparative analysis of red-edge hyperspectral indices

Using 3 nm bandwidth observations over the growth cycle of wheat in 680–820 nm range, the point of inflection was identified through differentiation and it was at 729 nm corresponding to peak observed at 64 days after sowing (DAS). To provide physical basis for various red edge hyperspectral indices reported in literature, the ratio indices (RI) corresponding to 1dB, 2dB, 3dB and half-power bandwidths around 729 nm were computed. These ratio indices had significant relationship with VOGa (after Vogelmann) = R740R720 and RESP (Red Edge Spectral Parameter) = R750R710. Here ‘Rλ’ refers to 3 nm bandwidth reflectance centered at wavelength λ. The λ used in GMI (Gitelson and Merzylak Index) = R750R700 and Inverse-Cartera Index (In-CI) = R760R695 wer found located at one-third and one fourth stage of the differential peak response at 64 DAS. The regression relationship between LAI and all the red-edge spectral indices (executed independently) was with r2 = 0.88 − 0.95 range for the decline phase of LAI while statistically significant relationship during growth phase of LAI was only for RESP, GMI, CI and RIhalf . The 3 nm bandwidth hyperspectral indices for ratio in 670–759 nm range to 673 nm had higher sensitivity as compared to corresponding normalized indices during peak vegetative and reproduction phenological stages.

Relationship between LAI and NDVI for IRS LISS and Landsat TM bands

Abstract Owing to a highly non-linear relationship between leaf area index and normalized difference vegetation index, polynomials of orders 1 to 4 were fitted to LAI and NDVI for wheat and onion crops using the data collected over their crop growth cycles. It was found that fourth or higher order curve fit better for wheat data whereas polynomials of the order 2 or higher fit the data well for onion. Using the coefficients of the polynomials, NDVI values were computed for LISS-I/II and correlated with actual NDVI values and results were significant at 99% confidence level.

New hyperspectral vegetation characterization parameters

Abstract The Near Infrared (NIR) bands for Indian Remote Sensing (IRS)/ Linear Imaging Self Scanning (LISS), Landsat TM and NOAA AVHRR are in 770–860, 760–900 and 725–1100 nm region. These NIR bands include water vapour absorption bands. To improve the biomass related intelligence of Normalized Difference Vegetation Index (NDVI), spectral region in NIR has to be chosen by excluding water vapour absorption bands in a manner that the existing dynamic range is maintained. Based on 3 nm bandwidth measurements over Wheat, it was found that reflected radiance in NIR had peaks at 747 nm and 777 nm, respectively. New Vegetation Index (NVI) was constructed by ratioing the difference of peak radiances at 777 and 747 nm with reflected radiance at 673 nm in Red region by using 3, 10 and 20 nm bandwidths. The 3 and 10 nm bandwidths gave nearly similar values of NVI. Thus estimation of this NVI is possible from 10 nm bandwidth data sets becoming available from space-based platforms in the near future. The rate of change of this NVI for wheat crop during 49 – 84 Days After Sowing (DAS) was comparatively very high than that for Ratio Vegetation Index (RVI) which conveys about the high sensitivity of NVI during peak vegetative through reproductive stages for wheat. The ratio of difference in radiance from chlorophyll-β (centered at 640 nm) and chlorophyll-α (centered at 673 nm) to that for chlorophyll-α depicted high sensitivity during maturity (and initial) stage.

Characterization of red-near infrared transition for wheat and chickpea using 3 nm bandwidth data

Abstract Enhancement of space based capabilities to discriminate different crops and different varieties of a particular crop needs measurement of (i) the shift in red edge and (ii) the slope of the sudden rise of reflectance in 680 – 760 nm spectral region as a function of Days After Sowing (DAS). To develop the knowledge base for catering to the analysis of future space-based hyperspectral measurements, ground based measurements in 3 nm bandwidth in visible - near Infrared region together with corresponding Leaf Area Index (LAI) observations were taken over the Crop Growth Cycle (CGC) of Wheat and Chickpea. The red edge for wheat crop was at 679 nm for 25 DAS and reached the upper limit i.e., 693.7 nm at 84 DAS and thereafter shifted backward to 679 nm at 108 DAS. There was no change in red edge value of 684.9 nm during 40 to 49 DAS and of 687.8 nm during 55 to 71 DAS. The slope of Red to NIR transition for wheat varied from 0.457 to peak value of 0.784 during 25 to 71 DAS and thereafter decreased to 0.073 at 108 DAS. The peak of Red to Near Infrared (NIR) transition slope and Ratio Vegetation Index (RVI) occurred at the same time i.e., 71 DAS. However, the upper most value of red edge shift occurred at 84 DAS. Paper discusses the above aspects including role of mid point of Red to NIR transition, interrelationships among the Red-NIR transition Slope, Red Edge, LAI and RVI for wheat and chickpea.

Estimation of roughness length and sensible heat flux from WiFS and NOAA AVHRR data

Abstract Development of representative NOAA AVHRR data based Land Surface Processes (LSP) parameters, towards incorporation into GCMs in quasi-dynamic mode, was undertaken by using (i) Normalized Difference Vegetation Index (NDVI) for the computation of roughness length (Z o ), and (ii) Land Surface Temperature (T s ) along with air density, specific heat of air at constant pressure, air temperature and aerodynamic resistance data to compute Sensible Heat Flux (SHF). NDVI was corrected for sensor degradation and atmospheric errors. Surface temperature (T s ) was computed for the pixel by weighted mixing of the temperatures computed for the assumed bare soil (T g ) and full canopy (T v ) conditions, using respective emissivity values in AVHRR channels 4 and 5 and split window algorithm, based on relative normalized distance of NDVI (for the pixel) from NDVI (soil) and NDVI (full canopy). The results reported in this study compared well with the Z o and SHF computed from tower-based meteorological measurements. This paper also discusses Z o computed with 188 m resolution IRS-1C/WiFS data and validation of results obtained from the AVHRR data with those derived from WiFS data

Problems in upscaling of high resolution remote sensing data to coarse spatial resolution over land surface

Abstract Characterization of land cover type for a grid cell (50 to 100 km) of General Circulation Models has to be achieved through spatial integration of satellite data available at higher spatial resolution. Preservation of information integrity is required during such upscaling and a suitable algorithm is to be developed and validated. The 36.25 and 72.5 m resolution IRS Linear Imaging Self Scanner (LISS) data were used for studying (i) change in proportion of the area under given land cover classes with the degradation in resolution and (ii) correlation between upscaled 72.5 m resolution (from 36.25 m) and observations in 72.5 m resolutions, and further upscaled resolutions, for different land cover types using different upscaling algorithms. In general, the percentage proportion of land cover types decreased with degradation of spatial resolution and the fractal method gave best correlations for water and agricutlure classes.

Spectral signature variations as a function of bandwidth using 3 nm bandwidth observations

Abstract Study with 3 nm bandwidth observation in visible Near-IR spectral region revealed that 760–830 nm and 650–673 nm are optimum bands in NIR and red region for computation of NDVI over the crop growth cycle of wheat. The 760–850 nm could be used in place of 760–830 nm if high spatial resolution is needed.

Upscaling aspects of multi-resolution satellite data in spatial and frequency domains

Abstract To incorporate intelligence from NOAA/AVHRR data for defining land surface boundary in GCMs, 1.1 km resolution data have to be upscaled to 50 or 100 km. The 36.25 m resolution LISS-II data was upscaled to 72.5 m and then correlation of this upscaled 72.5 m data was studied with the observed 72.5 m resolution LISS-I data. Research results reported in this paper address to the preservation of maximum information content in the upscaled data by reducing the impact of random noise component during upscaling process. For frequency domain upscaling, 72.5 m upscaled image was obtained from 36.25 m resolution IRS LISS-II data by (i) removing high frequency component in Fourier Transform of the 36.25 m resolution image and thereafter performing inverse Fourier Transform and (ii) after applying mean, median, k-average, mode and fractal operators on the image. This method gave very high correlations with LISS-I image even in heterogeneous regions and thus holds high promise in handling the heterogeneity, which occurs in 50×50 or 100×100 square km grids over land surface. This paper also addresses to upscaling of 188 m resolution NDVI derived from WiFS data to 1100 m and its correlation with 1100 m resolution NDVI derived from AVHRR.