Geometric Validation of Hyperion Data at Coleambally Irrigation Area

Transcription

1 Geometric Validation of Hyperion Data at Coleambally Irrigation Area Tim McVicar, Tom Van Niel, David Jupp CSIRO, Australia Jay Pearlman, and Pamela Barry TRW, USA

2 Background RICE SOYBEANS The Coleambally Irrigation Area (CIA) is approximately 95,000 hectares in size, and comprises over 500 farms. Fields are large (up to 70 hectares) and well maintained, making them ideal for instrument validation. The principal summer crops grown at the CIA are rice, corn, and soybeans, while winter crops include wheat, barely, oats, and canola. CORN

3 Background The CIA falls completely within two Landsat scenes (Path 92 Row 84, and Path 93 Row 84) allowing for an 8 day repeat cycle for the acquisition of both LANDSAT and EO-1 imagery. The nearby Uardry is a NASA Earth Observing System (EOS) land validation core site. Radiosonde and visibility data are acquired by the Australian Bureau of Meteorology at Wagga Wagga.

4 Datasets / Landsat ETM+ 19 Landsat ETM+ images were acquired over the summer growing season (October 2000 May 2001)

5 Datasets / Hyperion Seventeen EO-1 acquisitions over the summer growing season (December 2000 May 2001). Eight of these were clear (green). Two more were partly clear (blue). The other seven are cloudy (grey).

6 Satellite-based -EO-1 (ALI, Hyperion) -TERRA (Aster, MODIS) -Landsat ETM+ Coleambally Datasets Airborne -Hymap -Photography Ground-based -ASD Spectra -GER Spectra -LAI -Albedo -Photography -Producer (land use) Survey -Chemistry -Biomass -Chlorophyll Meter -Field Survey GIS-based -DGPS ground-based measurements -DGPS roads/ road intersections -Paddock (field) boundaries -Rice bay boundaries Green = Clear Blue = Partly Clear Grey = Cloudy

7 Datasets / Reference Aerial Photography High resolution (2m) digital aerial photographs acquired January 2001 used for creating positionally accurate field and rice bay GIS datasets. Over 466km of linear road network and 129 well-defined points digitised with a Differential Global Positioning System (DGPS). These datasets can be used for geo-referencing Hyperion time series

8 Datasets / Producer Survey Land use information within the focus Hyperion acquisition boundary of the CIA is being collected. Information about crop type, variety, and management practices are linked to individual paddocks within the GIS. Also, Coleambally Irrigation Limited (CIL), have provided water use information by farm. This database provides the validation of remotely sensed site characteristics for this study %DUH*URXQG &RUQ 5LFH &RUQ &URSW\SH5LFH 9DULHW\$PDURR 3UHYLRXV&URS6R\EHDQV +DUYHVWGDWH$SU )HUWLOLVHU$SSOLFDWLRQ8UHD6HS 5LFH 6R\EHDQV

9 Datasets / Plant Samples Rice Samples Rice samples were collected in association with both the 24 December 2000 and 02 January 2001 E0-1 overpasses. In both cases, 38 1m 2 samples were analysed for nitrogen (N) and biomass. Twenty-three of these same locations were measured with a chlorophyll meter (SPAD-502) in the field on 03 January Other Samples Some other crop plant samples were collected in association with the 19 February 2001 EO-1 overpass, including sorghum, soybean, corn, and sunflower. Three common rice weed samples were collected in association with 07 March 2001 EO-1 overpass and ASD spectra collection.

10 Datasets / Ground Spectral Measurements Spectral measurements were made in situ to the 03 February 2001, and the 07 March 2001 EO-1 overpasses with an ASD spectroradiometer. Rice, soybean, soil, and bright stubble targets were measured both days. Mean reflectance of the targets is shown below for the 03 February collection.

11 Ground-based Spectral Data Spectral measurements were made February 3, 2001, and March 7, 2001 during EO-1 overpasses with an ASD spectrometer. Data taken 2/3/01: Liang, Kaul and Van Niel reflectivity Rice Av 1-3 Rice Av 4-6 Soy av 4-6 Soy av 8-10 soil av soil av wavelength (nm)

12 Geo-Correction

13 Geocorrection Issues A one step process must address both: S V Geocorrection of VNIR and SWIR images Overlay of VNIR and SWIR images

14 Methods / Geo-correction Data acquired by Hyperion over the CIA from both the VNIR and SWIR arrays were geometrically assessed for three dates: 02 Jan 2001; 03 Feb 2001; and 07 Mar Hence, six Hyperion images are used in this study (VNIR and SWIR for each of the three). GCPs collected in each of the six Hyperion images and a highresolution digital aerial photography mosaic allowed the input Hyperion images to be geocorrected to a map base. Prior to fitting a transitive chain polynomial, outlier GCPs were identified and redefined. The order of the best polynomial was decided upon by using General Cross Validation and Predictive Error statistics. An affine (or linear) polynomial was selected.

15 Methods / Geo-correction For each EO-1 data acquisition event, the two images (VNIR and SWIR) were geocorrected using two methods: Firstly, we used the transitive ties, the basis of MOSMOD, to modify the input GCPs to provide geometric control through the stack of images (termed MOSMOD registration), and secondly we geometrically corrected two arrays independently to the map (termed MAP registration) Additionally, four generic shifts were applied to the SWIR image to geometrically match the SWIR image to the VNIR image. The SWIR image was shifted to the VNIR image by linearly interpolating across track. A constant 1 pixel shift was applied to the SWIR array in the along track direction. For example, to 1 and 1 shift means the SWIR image was shifted to the VNIR image by 0.25 lines at pixel 1 (the western edge of the array) and by 1 line at pixel 256 (the eastern edge of the array). The line shift between the end pixels (across track) was linearly interpolated. A constant 1 pixel shift was applied to the SWIR array in the line (along track) direction.

16 Results / Geo-correction Overall Statistics Total RMS residuals internal to the stack of input images(mosmod): X = m Y = m Polynomial from each of the 6 Hyperion images to base Aerial Photos: Average RMSx error = m, Standard deviation = m Average RMSy error = m, Standard deviation = m (Both slightly larger than 1/3 rd of the original 30 m pixel size) Supports the affine transitive polynomial coefficients, developed using MOSMOD

17 Results / Geo-correction 15 Hyperion Image Combination Statistics: Average RMSx error = m, Standard deviation = m Average RMSy error = m, Standard deviation = m Between any pair of Hyperion images, the error in the x (or pixel) direction is less than the error in the y (or line) direction. It is likely that optical alignment differences account for x-shift, and the VNIR-SWIR readout difference causes the y-shift.

18 Results / Geo-correction X s = X-stretch; Y s = Y-stretch γ = Earth rotation skew θ = Rotation of satellite track

19 X s and Y s Results MOSMOD Results: Results / Geo-correction Avg X s = m (SD 0.005m) and Avg Y s = m (SD 0.007m) TRW On-Orbit Results (Barry pers, comm.) : Avg X s = m (SD 0.25m) and Avg Y s = m (SD 0.73m)

20 VNIR/SWIR Spatial Structure defined from difference image calculated as band 57 minus band 78 Raw Data Merged then Registered to 1 and 1 Merged then Registered MOSMOD Registration Severe Edge Effects Best of the Shift then Register transformations still some edge effects mid scene Best results attained by registering VNIR and SWIR to map independently - still some localized edge effects and requires twice as many GCP s Results suggest that VNIR/SWIR shift is non-linear

21 Conclusions Geo-correction of VNIR and SWIR images independently provides best results. Results indicate that VNIR/SWIR shift is not linear However, a 0.25 to 1 cross track linear interpolation, and 1 offset along track provides a close approximation of the VNIR/SWIR offset. Independent geo-correction requires a full set of GCP s for both VNIR and SWIR, whereas shift then register transformation only requires one set of GCPs. The difference in methodologies for registration matching is ultimately a cost-benefit analysis of the user and will depend on the spatiotemporal data construct of the application.

22 Preliminary Analysis

23

24

25 Temporal Sequence of Hyperion Images Coleambally Irrigation Area A B C D E SOIL CORN RICE SOY SOIL Day 001 Day 033 Day 065 Day 072 Julian calendar day of 2001

26 Courtesy: Jay Pearlman, TRW

27 Coleambally Chlorophyll Indices from Hyperion 24 Dec Jan Feb Feb Mar Mar 2001 Chl-a index value 2000/2001 growing season; Chl-a index = R 800 /R 635

28

29

30

31 Coleambally Irrigation Area (CIA) Summary of Activities CIA is located in the east-west overlap of two EO-1 paths, allowing for an 8-day repeat cycle For time series analysis accurate geolocation of both VNIR and SWIR arrays is required Within a growing season (4 months) crop variables change dramatically (e.g. rice LAI_max = 10) Time series of EO-1 data (Hyperion L1A, ALI) and L7 ETM are currently being established (GEO-CORR and AT-COR) for CIA Extensive ground based, including GIS, D-GPS harvested and meteorological data, are available for the 2000/2001 and 2001/2002 summer (Southern Hemisphere) growing seasons

32

33 X s and Y s Results MOSMOD Results: Results / Geo-correction Avg X s = m (SD 0.005m) and Avg Y s = m (SD 0.007m) TRW On-Orbit Results (Barry pers, comm.) : Avg X s = m (SD 0.25m) and Avg Y s = m (SD 0.73m) VNIR and SWIR γ and θ Results VNIR Avg θ = (SD ) and Avg γ = (SD ) SWIR Avg θ = (SD ) and Avg γ = (SD )

34 VNIR/SWIR Shifts - Cross Track Differences

35 VNIR/SWIR Shifts - Along Track Differences

36

37

38 CIA Study Area - Paddocks 32 and 33

39 Coleambally Irrigation Area Overview Large farms Irrigated and level modern management practices Variability among Farms in their Management Philosophy

40 Hyperion Data - Comments Level 1 data: nm and nm Bands 9-57 and ; SWIR is West of VNIR and rotated CCW by one pixel Band Center(nm) FWHM(nm) SWIR VNIR S V

41 Methods / Geo-correction Data acquired by Hyperion over the CIA from both the VNIR and SWIR arrays were geometrically assessed for three dates: 02 Jan 2001; 03 Feb 2001; and 07 Mar Hence, six Hyperion images are used in this study (VNIR and SWIR for each of the three). GCPs collected in each of the six Hyperion images and a highresolution digital aerial photography mosaic (which meets both the US National Map Accuracy Standards and the National Mapping Council of Australia mapping standards) allowed the input Hyperion images to be geocorrected to a map base. Prior to fitting a transitive chain polynomial, outlier GCPs were identified and redefined. The order of the best polynomial was decided upon by using General Cross Validation and Predictive Error statistics. An affine (or linear) polynomial was selected.

42 VNIR/SWIR Spatial Structure defined from difference image calculated as band 57 minus band 78 Raw Data Merged then Registered 0 to 1 and 1 Merged then Registered to 1.25 and 1 Merged then Registered to 1 and 1 Merged then Registered Regression Merged then Registered ENVI Registration MOSMOD Registration False Colour Composite Radiance (W m -2 sr -1 µm -1 ) 07 March 2001 cwl: band 57 = nm; band78 = nm

43 VNIR/SWIR Spatial Structure defined from difference image of Coleambally band 57 minus band 78 Raw Data Merged then Registered 0 to 1 and 1 Merged then Registered to 1.25 and 1 Merged to 1 and 1 Merged then Registered then Registered Regression Merged then Registered ENVI Registration MOSMOD Registration False Colour Composite Radiance (W m -2 sr -1 µm -1 ) 07 March 2001 cwl: band 57 = nm; band78 = nm