University of Twente Student Theses

Login

Downscaling leaf area index using downscaling cokriging on optical remotely sensed data

Singh, Ankur (2013) Downscaling leaf area index using downscaling cokriging on optical remotely sensed data.

[img] PDF
2MB
Abstract:In the field of remote sensing, scaling of data has become more practicing in different disciplines. Downscaling of data bring revolution for the usage of coarse spatial resolution data products. The data products were downscaled to desired fine resolution according to the usage. In this study LAI (leaf area index) is downscaled by using cokriging technique. The main aim of this research is to explore downscaling cokriging technique by studying the effect of resampling and point spread function (PSF). MODIS LAI product at 1000 m spatial resolution is used as primary variable and MODIS NDVI at 250 m is used as covariable to downscale LAI at 250 m. Cokriging is used as the technique for downscaling. The first step for downscaling cokriging involves the calculating of sample variogram and cross-variogram. To calculate cross-variogram both the variable LAI and NDVI should be on same scale. To bring MODIS LAI at 250 m it is resample at 250 m and cross-variogram is calculated between resample LAI and original NDVI at constant cut-off of 6000 m and bin of 15 and variogram and cross-variogram is modelled at common range of 3817.4 m using exponential model. To select optimal resampling from different resampling techniques (nearest neighbour, bilinear interpolation, cubic convolution and trivial method) variogram analysis has been executed. It was found that variance in trivial resampling was highest (sill = 3.78) which shows high spatial dependence. Trivial resampling is also selected for resampling because it resample’s pixel size with original pixel values. Gaussian and uniform PSF are used to study the effect of PSF on variogram. Standard deviation ߪ௫ and ߪ௬ were parameterized by experimenting with different value of standard deviation. ߪ௫ and ߪ௬ were parameterized at 250 m for MODIS LAI at resolution 1000 m and 122.5 m for MODIS NDVI at resolution 250m. Further, point support variogram and cross-variogram were estimated to see the effect of both (uniform and Gaussian) PSF by keeping rest of the parameter same and using nested exponential model. It was found that for LAI 1000 m using uniform PSF sill was 2.85 for the range 3040.5m which is more than that by using Gaussian PSF (2.06 for range 3857.0 m). It was found that variance is high using uniform PSF for LAI at 1000 m because distance from the pixel or the mean is more. It was observed that for NDVI at 250 m there is very less difference is observed by estimated point support variogram which was not observed by modelling of point support variogram in and variance for uniform PSF was found higher than by using Gaussian PSF. There was no change observed for the cross-variogram because PSF does not have much effect for different bands on same support. Due to above change the effect was also observed on the centre of downscaling cokriging weights. By using uniform PSF centre of the weight was high than that by using Gaussian PSF. There is change observed in the downscaled cokriging image. The downscaled image for Gaussian PSF was found smoother than that of uniform PSF. The standard deviation in downscaled image for uniform PSF (1.73) was found more than that of Gaussian PSF (1.69) which shows that variance is high for the uniform PSF than that of Gaussian PSF. Keywords: Downscaling, cokriging, MODIS LAI, MODIS NDVI, resampling, Point spread function (PSF), uniform PSF, Gaussian PSF, variogram
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/93814
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page