Sei sulla pagina 1di 7

J Indian Soc Remote Sens (March 2012) 40(1):1117 DOI 10.

1007/s12524-011-0144-9

RESEARCH ARTICLE

CO-Kriging Approach for Cartosat-1 Height Product with ICESat/GLAS Data for Digital Elevation Surface Generation
Anil Kumar & Parul Trivedi

Received: 10 March 2011 / Accepted: 15 June 2011 / Published online: 1 July 2011 # Indian Society of Remote Sensing 2011

Abstract Nowadays, Geostatistics and its various interpolation techniques have become a major threshold area in the field of research in GIS. In this research work poorly sampled (less accurate height data relative to ICESat/GLAS height data) Cartosat-1 height data has been used with well sampled (more accurate height data relative to Cartosat-1 height data) ICESat/GLAS LiDAR (Light Detection and Ranging) height point data using Cokriging Interpolation technique, to study the effect of ICESat/GLAS on Cartosat-1 height data. Space borne LiDAR data has led researchers to explore its utilities in many applications. Space borne LiDAR data can be acquired through space borne LiDAR sensors also, like; GLAS (Geoscience Laser Altimeter System) system onboard ICESat (Ice, Cloud and land Elevation Satellite) satellite. In this study, it has been tried to apply Cokriging interpolation on two different sources of data sets, with a common variable (elevation) to generate DES and assessment of this surface has been conducted by DGPS data. After optimizing Cokriging parameters, results of digital elevation surface (DES) generated using Cokriging showed that RMSE has been second least than global polynomial in comparison to Kriging interpolation RMSE after being evaluated by GPS values. So, global polynomial as well as cokriging interpolation technique out performs while comparing with kriging technique for DES generation.
A. Kumar (*) : P. Trivedi Indian Institute of Remote Sensing, Dehradun, India e-mail: anil@iirs.gov.in

Keywords ICESat/GLAS satellite data . Geostatistical analysis . DES . Cokriging

Introduction The basic premise of Geostatistics is that observations closer together are more similar on an average than those farther apart. Although geostatistics was developed independently of GIS, it has become an integral part of GIS (Barton et al. 1999). The geostatistical techniques quantify the spatial autocorrelation among measured points and account for the spatial configuration of the sample points around the prediction location (Goulard and Voltz 1992). It creates surfaces incorporating the statistical properties of the measured data. Geostatistical techniques are also applied for generation of Digital Elevation Surface (DES) using height point data (Krige 1994, 1999). One of the geostatistical procedures is Cokriging, which allows one or more secondary or ancillary variables to be included in the model and assuming that the primary and ancillary variables are moderately correlated, the estimation accuracy of the primary variable should increase. Cokriging requires a model of spatial continuity (Myers 1982, 1983). The most common models of spatial continuity are the variogram or semivariogram which is used to model the average degree of similarity or dissimilarity as a function of both distance and direction (Stein and Corsten 1991; Pelletier et al. 2004). In this study DES result are obtained by using poorly sampled (relatively less height accurate compared to ICESat/GLAS-LiDAR data) Cartosat-1 height data with well sampled (rela-

12

J Indian Soc Remote Sens (March 2012) 40(1):1117 Table 2 Accuracy of GCPs acquired from geodetic single frequency GPS in relative mode GCP no. 1 2 3 4 5 6 7 8 9 10 11 PDOP 2.46 2.06 1.44 1.84 1.72 1.94 1.97 2.11 1.84 2.47 2.02 Position quality (M) 0.009 0.020 0.007 0.015 0.007 0.008 0.008 0.031 0.005 0.008 0.014 Height quality (M) 0.017 0.040 0.010 0.024 0.012 0.010 0.018 0.052 0.010 0.014 0.018

Fig. 1 Study area: ICESat ground tracks displayed in the study area (yellow dots are footprint of ICESat/GLAS-LiDAR data)

tively more height accurate compared to Cartosat-1 data) ICESat/GLAS LiDAR data through cokriging interpolation technique. This study has been conducted to analyze the interpolation technique for deriving DES from poorly sampled Cartosat-1 stereo data and well sampled ICESat/GLAS data to find out level of confidence for Cokriging interpolation technique, whether it can generate better interpolated continuous surface compared to kriging method and improve the height accuracy of DES from Cartosat-1 points data added with ICESat/GLAS point set.

Study Area and Data Used The study area, which is moderately hilly, lies in the north western part of Dehradun, Uttarakhand State, India, with boundary coordinates as 77509.30E 775216.32E and 302323.15N302535.00N as shown in Fig. 1. The data used in this study was
Table 1 Test data from ground based GPS point observation

Cartosat-1 height point data, which was derived from Cartosat-1 Stereo pairs. The Cartosat-1 stereo data used was of 2nd October 2005. The ICESat height point data set considered was acquired from February 2003 to February 2008 and were of release 28, was ordered from the NSIDC (National Snow and Ice Data Center) site for this study. The Geoscience Laser Altimeter System (GLAS) instrument on the Ice, Cloud, and land Elevation Satellite (ICESat) provides global measurements of polar ice sheet elevation to discern changes in ice volume (mass balance) over time. Secondary objectives of GLAS are to measure sea ice roughness and thickness, cloud and atmospheric properties, land topography, vegetation canopy heights, ocean surface topography, and surface reflectivity. GLAS has a 1,064 nm laser channel for surface altimetry and dense cloud heights, and a 532 nm LIDAR channel for the vertical distribution of clouds and aerosols. This covers seven data sets: a) GLA01: Level-1A altimetry data include the transmitted and received waveforms from the instrument.

Point Latitude no 1 2 3 4

Longitude

Height

ICESat/GLASS ICESat/GLASS height height (m) difference (m) 494.7730 509.2990 520.3230 525.9470 1.3246 1.4246 2.3645 1.4477

775042.36744 302348.81819 493.4484 m 775111.37477 302352.01163 507.8744 m 775132.03173 302353.67582 522.6875 m 775144.23666 302346.61847 524.4993 m

J Indian Soc Remote Sens (March 2012) 40(1):1117 Fig. 2 Methodology adopted

13

ICESat/GLAS Data

Point map Extraction

Cartosat-1 Points

GLAS Point map and Cartosat - 1 Points

GLAS Point maps

Point map sets: 2 variables: GLAS Point map and Cartosat - 1 Point map

Experiment (2 variables) Cokriging

Digital Elevation Surface

DGPS Points

Accuracy Assessment

b) GLA05: Level-1B waveform parameterization data include output parameters from the waveform characterization procedure and other parameters required to calculate surface slope and relief characteristics. c) GLA06: Level-1B elevation data include surface elevation, surface roughness assuming no slope, surface slope assuming no roughness, and geodetic and atmospheric corrections for range measurements. Both Level-1B products are geolocated to the center of the laser footprint.

d) GLA12GLA15: GLA06 is used in conjunction with GLA05 to create the Level-2 altimetry products. Level-2 altimetry data provide surface elevations for ice sheets (GLA12), sea ice (GLA13), land (GLA14), and oceans (GLA15). Data also include the laser footprint geo-location and reflectance, as well as geodetic instrument and atmospheric corrections for range measurements. GLA06 Global Elevation Data Product and GLA14 global land surface altimeter data product

14

J Indian Soc Remote Sens (March 2012) 40(1):1117

Fig. 3 Crossvariogram model of predictand and covariable

Fig. 5 Semivariogram model for covariable

was used for ground point height information from ICESat/GLAS-LiDAR data.

Methodology Initially height point information from ICESat/GLAS data and Cartosat-1 stereo data sets were generated. At preprocessing stage all data sets were brought to same reference frame. Out of 15 products of ICESat data, GLA06 and GLA14 were used in this study. These were transformed from TOPEX/Poseidon to WGS84 Ellipsoid/Datum using a transformation routine written in JAVA, as well as height point data from Cartosat-1 stereo data set was also transformed into WGS84 reference frame. Rational function coefficients along with Ground Control Points (GCPs) were used for stereo model generation to

extract height point information from Cartosat-1 stereo data set. The orthoimage from Cartosat-1 stereo was also generated to depict ICESat/GLAS data while overlaying on it, to be used during ground truthing with the help of geodetic single frequency global positioning system in differential mode. The GPS data was collected to be used as GCPs for generation of a stereo model from Cartosat-1 stereo data sets as well as evaluating the DES generated using ICESat/GLAS data as well as Cartosat-1 stereo data sets in this research work. It has been tried to observe the GPS data in DGPS mode with phase measurement techniques to have accuracy of GPS data within 1 cm to 30 cm2 ppm. The Accuracy of GCPs acquired from Geodetic Single Frequency GPS in Relative mode is shown in Table 2. While generating the stereo model, seven ground points were used as GCPs and other four known points on the ground were used as check point. The observation time for rover was kept 1 h each and observation time for base was 72 h. Tables 1 and 2 shows the list of GPS points used for evaluation of DES generated from ICESat/GLAS data as well as Cartosat-1 stereo data sets.
Table 3 Cokriging parameters Semivariogram Nugget Sill model Predictand (cartosat-1) Wave Model Covariable (LiDAR) Crossvariogram Wave Model Wave Model 50 10 0 Range

8500 3000 7500 3000 3700 2000

Fig. 4 Semivariogram model for predictand

J Indian Soc Remote Sens (March 2012) 40(1):1117

15

Fig. 6 DES using Cokriging interpolation

After preprocessing data, 3 data sets were generated: Cartosat-1 height data, ICESat/GLAS height data, and GPS points that have been transformed into WGS84 reference frame. In this
Table 4 DES height from different data/interpolation techniques relative to GPS data GPS points 1 2 The range height differences of Cokriging digital elevation surface are between 0.752 m and 3.409 m 3 4 RMSE() GPScartosat-1 (m) 0.452 11.876 4.138 0.119 6.29

research, Cokriging method was applied on ICESat/GLAS point map and Cartosat-1 points. Methodology adopted in this research work has been shown in Fig. 2.
GPSICESat (m) 1.325 1.423 2.365 1.448 1.69 GPSglobal polynomial (m) 2.347 1.274 1.475 0.620 1.55 GPSKriging (m) 3.345 3.639 2.458 2.462 3.02 GPSCokriging (m) 1.007 1.952 3.409 0.752 2.07

16

J Indian Soc Remote Sens (March 2012) 40(1):1117


DES Height of Cokriging Interpolation Technique Relative to GPS Data

Cokriging interpolation technique is applied using parameters like semivariogram models, limiting distance, etc. Different types of semivariogram models were analysed to obtain the best fit or goodness model. Further, various combinations of limiting distance and maximum and minimum number of points were used so as to obtain best DES output. Figure 3 shows crossvariogram model of predicted and covariable, while Figs. 4 and 5 here demonstrates the semivariogram model for predicted (Cartosat-1 point data) and covariable (ICESat/GLAS data) respectively. Cokriging parameters that were used to obtained these best fit variogram models have seen shown in Table 3.

Height Variations (M)

4 3 2 1 0 -1 -2 -3
1 2 3 4

GPS Points

Fig. 7 DES height of Cokriging interpolation technique relative to GPS data

Result and Discussion Digital elevation surface model generated with Cokriging interpolation techniques using Cartosat-1 and ICESat/GLAS can be seen in Fig. 6 wherein limiting distance was 1,500 and minimum and maximum numbers of points used were 1 and 30 respectively. While digital elevation surface model generated with Cokriging interpolation technique using ICESat/ GLAS height data with Cartosat-1 Point Map was compared with sample ground coordinates from DGPS data along with results of Global Polynomial and Kriging (Kumar and Vyas 2009) and differences are shown in Table 4. The GPS point data used here to evaluate relative height differences between the Global Polynomial interpolation, Kriging interpolation and Cokriging interpolation techniques output and GPS data. Height of check point 1 and 4 observed by GPS were lying in fallow agriculture fields, check point 3 was lying in area and check point 2 was raod crossing surrounded by tall sal tree forest. As the check point 2 was surrounded by tall sal tree forest that is why while comparing its height with GPS was found large deviation. While evaluating height data generated from Cartosat-1 stereo data, the height variation has been observed from 2 to 12 m while comparing it with GPS height data (Kumar 2006). This variation in height has been found due to variation in contract in ground features as well geometrical variation due to +26 tilt in fore scene. Geometrical variation due to +26 tilt in fore scene, comparion to aft scene has been found more critical while observing the cartosat-

1 data sets of Badrinath rigion, where height variation are of the range of 500800 m within very small span. From Figs. 7 and 8 it can be seen while applying Cokriging with detail parameter inputs, the RMSE obtained was 2.07 m in comparison to Kriging interpolation RMSE height difference which has been found to be 3.02 m. Also Cokriging interpolated heights of DES were close to interpolation techniques like; Global Polynomial applied on point height data generated from Cartosat-1 data.

Conclusion Based on the analysis about the elevation difference result of interpolation type, the range of height difference of Cokriging DES is between 0.752 m and 3.409 m. In line with this result, Cokriging interpolation method that gives the least height differences of 0.75 m compared with GPS that uses the data of Cartosat-1 point map and ICESat/GLAS point map for Cokriging interpolation, the highest elevation difference comes from Kriging method that uses only Cartosat-1 point map. The average height difference DES generated using ICESat/GLAS added with Cartosat-1 data with different kriging interpolation techniques were between

Fig. 8 DES height from different interpolation techniques relative to GPS data

J Indian Soc Remote Sens (March 2012) 40(1):1117

17 Goulard, M., & Voltz, M. (1992). Linear coregionalization model: Tools for estimation and choice of cross-variogram matrix. Mathematical Geology, 24, 269286. Krige, D. G. (1994). An analysis of some essential basic tenets of geostatistics not always practiced in ore valuations. 1st Regional APCOM Symposium, Bled, Slovenia Krige, D. G. (1999). Conditional bias and uncertainty, APCOM99 symposium, Colorado School of Mines, Golden Kumar, A. (2006). Cartosat 1 (IRS P5) stereo data processing a case study of Dehradun area, GIS development magazine Kumar, A., & Vyas, S. (2009). Cartosat-1 height product and ICESat/GLAS data for digital elevation surface generation. Journal of the Indian Society of Remote Sensing, 37, 553 559. Myers, D. E. (1982). Matrix formulation of cokriging. Mathematical Geology, 14(3), 249257. Myers, D. E. (1983). Estimation of linear combinations and cokriging. Mathematical Geology, 1(5), 633637. Pelletier, B., Larocque, G., & Fyles, J. W. (2004). Fitting the linear model of Coregionalization by generalized least squares. Mathematical Geology, 36(3), 323343. Stein, A., & Corsten, L. C. A. (1991). Universal kriging and cokriging as a regression procedure. Biometrics, 47(2), 575587.

1.57 m and 4.88 m (Kumar and Vyas 2009). From this work it can be concluded that Cokriging interpolation method outperform, kriging method. This work concludes that ICESat/GLAS space based data, freely available, which is also more accurate than height data generated from Cartosat-1 stereo data, if fused with height data generated from Cartosat-1 stereo data, can produce more accurate DES. There is also need for understanding each aspect and concept of all the parameters that are used for any interpolation technique, as there might occur error which further can lead to improper conclusion.

References
Barton, J. M. H., Buchberger, S. G., Lange M. J. (1999). Estimation of error and compliance in survey by kriging. Journal of Survey Engineering, 87107

Potrebbero piacerti anche