Evaluating the WorldView-2, GeoEye-1, DMCII, THEOS and KOMPSAT-2 imagery for use in the CAP CwRS programme

Transcription

1 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Evaluating the WorldView-2, GeoEye-1, DMCII, THEOS and KOMPSAT-2 imagery for use in the CAP CwRS programme Joanna K. Nowak Da Costa and Agnieszka Walczyńska EC DG Joint Research Centre, IPSC, Monitoring Agricultural Resources Unit speaker: Pär J. Åstrand 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, We would like to thank: Acknowledgements - the Eurimage SpA personnel, in particular Bruno Biagini, Axel Oddone and Ilaria Tramacere, and Andrew Hanna and Gene Dial from GeoEye Inc., who kindly provided us with a sample of the GeoEye-1 test data and their technical assistance. - the European Space Imaging and DigitalGlobe personnel, and in particular Michaela Weber, Susanne Oberdorf and Edith Simon, who kindly provided us with a big sample of the WorldView-2 test data and technical assistance. - the SpotImage & Customer Relations Managers ~ KOMPSAT-2 personnel, and in particular Pierre Boubée, Brice Lamarre, Christelle Puech, Philomène Jacquier and Bruno Montfort, who kindly provided us with the Kompsat-2 test data over the JRC Maussane test site. - the Thai Geo-Informatics and Space Technology Development Agency GISTDA, and in particular Darasri Dowreang, Praneet Ditsariyakul and Damrongrit Niammuad, who kindly provided us with the THEOS data for testing. - Owen Hawkins from DMC International Imaging, and Julio C. López Bravo from Deimos Imaging who kindly provided us with the DMCII MS data for testing. - the Spacemetric company personnel, and in particular Torbjörn Westin, who kindly provided us with their THEOS quality control results and the set of orthorectified THEOS images.

2 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Outline Methodology 0.5 m sensors: GeoEye-1 and WorldView m sensors: Kompsat-2 and THEOS m sensors: DMCII and THEOS MS Key issues 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Image Geometric Quality Analysis - Methodology IMAGE GCPs MODELLING DEM INTERNAL QUALITY CONTROL GCPs RMSE ORTHORECTIFICATION EXTERNAL QUALITY CONTROL ICPs RMSE

3 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, m sensors: GeoEye-1 and WorldView-2 GE1 WV2 Orbit type Sun Synchronous Sun Synchronous Altitude 681 km 770 km Inclination (Sun synchronous) 98 [deg] 97.2 [deg] Launch Data September 6, 2008 October 8, 2009 Spectral bands PAN: nm Blue: nm Green: nm Red: nm Near-IR: nm PAN: nm Blue: nm Green: nm Red: nm Near-IR1: nm Coastal: nm Yellow: nm Red Edge: nm Near-IR2: nm Spatial resolution at nadir 0.41m (PAN); 1.65m (MS) 0.46m (PAN); 1.84m (MS) Swath width 15.2km (PAN, MS) 16.4km (PAN, MS) Product level Geo (2A), GeoProfessional (3A) 1A, 2A 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, WorldView-2 Geometric Quality Analysis (1) Fig. 1-D RMSE_ICP [m] measured on the final WV2 orthoimages as a function of the GCPs number used during the single scene correction applying the first order ERDAS LPS Rational Polynomial sensor model based on 1, 2, 3, 4, 6 and 9 well-distributed GCPs and a DTM with 0.6m accuracy. 3,5 2,5 1,5 RMS_E (off nadir 36 ) RMS_N (off nadir 36 ) RMS_E (off nadir 31.6) RMS_N (off nadir 31.6) RMS_E (off nadir 21.7) RMS_N (off nadir 21.7) 0, The 1-D RMS errors based on the measurements of the 12 well-distributed ICPs on the final WV2 orthoimage after the single scene correction applying RPC-based model: - are not sensitive to the number of GCPs used if they are well-defined, well-distributed and range between 4 and 9 (provided a DTM with 0.6m vertical accuracy); - are sensitive to the overall off-nadir angle and increase with increasing off-nadir angle. - The rigorous model requires a minimum 9 GCPs.

4 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, WorldView-2 Geometric Quality Analysis (2) Fig. 1-D RMSE_ICP [m] measured on the final WV2 orthoimages as a function of the DEM accuracy (RMSE_z) after the single scene correction applying the first (left) and zero (right) order ERDAS LPS Rational Polynomial sensor model based on 4, 6 and 9 well-distributed GCPs. 8,0 8,0 GCPs, RPC order 1, 36.0 GCPs, RPC order 1, 36.0 GCPs, RPC order 0, 36.0 GCPs, RPC order 0, m 1.9m 5.2m 8.1m 9.8m 15.3m 0.6m 1.9m 5.2m 8.1m 9.8m 15.3m The 1-D RMS errors based on the measurements of the 20 well-distributed ICPs on the final WV2 orthoimage after the single scene correction applying RPC-based model: - are not sensitive to the Rational Polynomial order (first or zero); - are sensitive to the DEM accuracy (RMSE_z) and increase with decreasing DEM accuracy; - are invariable with the GCPs number if points are well-distributed and range between th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, WorldView-2 Geometric Quality Analysis (3) Fig. 1-D RMSE_ICP [m] measured on the final WV2 orthoimages as a function of the DEM accuracy (RMSE_z) after the single scene correction applying the zero order ERDAS LPS Rational Polynomial sensor model based on 4, 6 and 9 well-distributed GCPs. On the left the WV2 overall off-nadir angle of 36.0, on the right off-nadir ,0 8,0 GCPs, RPC order 0, 36.0 GCPs, RPC order 0, 36.0 GCPs, RPC order 0, 31.6 GCPs, RPC order 0, m 1.9m 5.2m 8.1m 9.8m 15.3m 1.9m 5.2m 8.1m 9.8m 15.3m The 1-D RMS errors based on the measurements of the 20 well-distributed ICPs on the final WV2 orthoimage after the single scene correction applying RPC-based model: - are sensitive to the DEM accuracy (RMSE_z) and increase with decreasing DEM accuracy. - meet the geometric specification of 2.5m for (a) at least 4 well-distributed GCPs and (b) a DEM of high accuracy in case of very inclined images, i.e. 3.5m if WV2 overall off-nadir angle exceeds 34 or 5.2m if the WV2 overall off-nadir angle is between 30 and 34.

5 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, GeoEye-1 Geometric Quality Analysis (1) Fig. 1-D RMSE_ICP [m] measured on the final GE1 orthoimages as a function of the GCPs number used during the single scene correction applying the first order PCI Geomatics or ERDAS LPS Rational Polynomial sensor model based on 0, 1, 2, 3, 4, 5, 6, 9 and 11 well-distributed GCPs and a DTM with 0.6m accuracy RMSE_E (RPC, PCI Geomatics) RMSE_E (RPC, ERDAS LPS) RMSE_N (RPC, PCI Geomatics) RMSE_N (RPC, ERDAS LPS) RMSE_E (RPC, PCI Geomatics) RMSE_E (RPC, ERDAS LPS) RMSE_N (RPC, PCI Geomatics) RMSE_N (RPC, ERDAS LPS) The 1-D RMS errors based on the measurements of the 13 well-distributed ICPs on the final GE1 orthoimage after the single scene correction applying RPC-based model: - are not sensitive to the number of GCPs used if they are well-defined, well-distributed and range between 3 and 11 (provided a DTM with 0.6m vertical accuracy); - The rigorous model requires a minimum 9 GCPs. 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, GeoEye-1 Geometric Quality Analysis (2) Fig. 1-D RMSE_ICP [m] measured on the final GE1 orthoimages as a function of the DEM accuracy (RMSE_z) after the single scene correction applying the first (left) and zero (right) order ERDAS LPS Rational Polynomial sensor model based on 4, 6 and 9 well-distributed GCPs. GCPs, RPC order 1, 35 GCPs, RPC order 1, 35 GCPs, RPC order 0, 35 GCPs, RPC order 0, m 5.2m 8.1m 9.8m 15.3m 1.9m 5.2m 8.1m 9.8m 15.3m The 1-D RMS errors based on the measurements of the 20 well-distributed ICPs on the final GE1 orthoimage after the single scene correction applying RPC-based model: - are not sensitive to the Rational Polynomial order (first or zero); - are sensitive to the DEM accuracy (RMSE_z) and increase with decreasing DEM accuracy; - are invariable with the GCPs number if points are well-distributed and range between 4 9.

6 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, GeoEye-1 Geometric Quality Analysis (3) Fig. 1-D RMSE_ICP [m] measured on the final GE1 orthoimages as a function of the DEM accuracy (RMSE_z) after the single scene correction applying the zero order ERDAS LPS Rational Polynomial sensor model based on 4, 6 and 9 well-distributed GCPs. On the left the GE1 overall off-nadir angle of 35, on the right off-nadir 28. GCPs, RPC order 0, 35 GCPs, RPC order 0, 35 GCPs, RPC order 0, 28 GCPs, RPC order 0, m 5.2m 8.1m 9.8m 15.3m 1.9m 5.2m 8.1m 9.8m 15.3m The 1-D RMS errors based on the measurements of the 20 well-distributed ICPs on the final GE1 orthoimage after the single scene correction applying RPC-based model: - are sensitive to the DEM accuracy (RMSE_z) and increase with decreasing DEM accuracy. - meet the geometric specification of 2.5m for (a) at least 4 well-distributed GCPs and (b) a DEM of high accuracy in case of very inclined images, i.e. 6m if GE1 overall off-nadir angle exceeds 34 or 7.2m if the GE1 overall off-nadir angle is between 30 and th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, m sensors: Kompsat-2 and THEOS K2 TH Orbit type Sun Synchronous Sun Synchronous Altitude km 822 km Inclination (Sun synchronous) 98 [deg] 98.7 [deg] Launch Data July 28, 2006 October 1, 2008 Spectral band PAN: 500nm 900nm Blue: 450nm 520nm Green: 520nm 600nm Red: 630nm 690nm Near-IR: nm PAN: 450nm 900nm Blue: 450nm 520nm Green: 530nm 600nm Red: 620nm 690nm Near-IR: 770nm 900nm Spatial resolution at nadir 1m (PAN); 4m (MS) 2m (PAN) and 15m (MS) Swath width 15km (PAN, MS) 22km (PAN), 90km (MS) Product level 1R, 1G 1A, 2A

7 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Kompsat-2 Geometric Quality Analysis (1) Fig. 1-D RMSE_ICP [m] measured on the final K2 orthoimages as a function of the overall off-nadir angle after the single scene correction applying the first (left) and zero (right) order ERDAS LPS Rational Polynomial sensor model based on 6, 9, 12 and 15 well-distributed GCPs and a DTM with 0.6m accuracy. RMSE_E (v_6, ERDAS, RPC 1) RMSE_N (v_6, ERDAS, RPC 1) RMSE_E (v_9, ERDAS, RPC 1) RMSE_N (v_9, ERDAS, RPC 1) RMSE_E (v_12, ERDAS, RPC 1) RMSE_N (v_12, ERDAS, RPC 1) RMSE_E (v_6, ERDAS, RPC 0) RMSE_N (v_6, ERDAS, RPC 0) RMSE_E (v_9, ERDAS, RPC 0) RMSE_N (v_9, ERDAS, RPC 0) RMSE_E (v_12, ERDAS, RPC 0) RMSE_N (v_12, ERDAS, RPC 0) RMSE_E (v_15, ERDAS, RPC 1) RMSE_E (v_15, ERDAS, RPC 0) 0 deg 13 deg 14 deg 30 deg RMSE_N (v_15, ERDAS, RPC 1) 0 deg 13 deg 14 deg 30 deg RMSE_N (v_15, ERDAS, RPC 0) The 1-D RMS errors based on the measurements of the well-distributed ICPs on the final K2 orthoimage after the single scene correction applying either the PCI rigorous model or the PCI RPC-based or the ERDAS RPC-based model: - are not sensitive to the number of GCPs used if they are well-distributed and range between 9 and 15 (provided a DTM with 0.6m vertical accuracy); - are sensitive to the overall off-nadir angle and increase with increasing off-nadir angle. 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Kompsat-2 Geometric Quality Analysis (2) Fig. The average 1-D RMSE_ICP [m] measured on the final K2 orthoimage as a function of the overall off-nadir angle after the single scene correction applying first (left) and zero (right) order Rational Polynomial using ERDAS LPS (in blue) and PCI Geomatics (in red) or PCI rigorous model (in green). mean RMSE_E (PCI, RPC 1) mean RMSE_E (PCI, RPC 0) mean RMSE_N (PCI, RPC 1) mean RMSE_N (PCI, RPC 0) mean RMSE_E (PCI, rigorous) mean RMSE_E (PCI, rigorous) mean RMSE_N (PCI, rigorous) mean RMSE_E (ERDAS, RPC 1) mean RMSE_N (PCI, rigorous) mean RMSE_E (ERDAS, RPC 0) 0 deg 14 deg 30 deg mean RMSE_N (ERDAS, RPC 1) 0 deg 14 deg 30 deg mean RMSE_N (ERDAS, RPC 0) The overall average 1-D RMSEs are 2.1m and 4m (the maximum 1-D RMSE values are 3.2m and 6.2m) of Easting and Northing direction respectively, provided a DTM with 0.6m vertical accuracy, and the GCPs with 0.05m < RMSEx, RMSEy < 0.90m are used. - These values are in accordance with SAUNIER s (2008) Kompsat-2 Mission Quality Assessment; and are superior to the nominal horizontal geo-location accuracy of Kompsat-2 reported by SEO (2008) from Korean Aerospace Research Institute, the K2 image provider.

8 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, THEOS Geometric Quality Analysis (1) Fig. The 1-D RMSE_ICP [m] measured on the THEOS 2A PAN products as a function of the overall off-nadir angle before (left) and after (right) systematic error elimination. The 1-D RMS errors measured on the THEOS 2A PAN product are sensitive to the overall off-nadir angle and increase with increasing off-nadir angle, especially when the angle exceeds 22 degrees. The average absolute 1-D RMSE for the THEOS 2A PAN product are 5m and 6m for Northing and Easting direction respectively, provided systematic error elimination (otherwise the values are as big as 87m and 137m). 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, THEOS Geometric Quality Analysis (2) Fig. The 1-D RMSE_ICP [m] measured on the orthorectified THEOS 1A PAN image as a function of the overall off-nadir angle after the single scene correction applying the Keystone Spacemetric SIP/Ortho rigorous model based on the 9 GCPs collected from 0.5m ADS40 orthorectified image and 0.6m accurate DEM (image matching techniques applied to GCP/ICP measurement). The 1-D RMS errors measured on the orthorectified THEOS 1A PAN image are sensitive to the overall off-nadir angle and increase with increasing off-nadir angle, especially when the angle exceeds 22 degrees. The average absolute 1-D RMSE for the orthorectified THEOS 1A PAN product are 2m and 2m for Northing and Easting direction respectively (1 pixel accuracy) provided a DTM with 0.6m vertical accuracy and 9 GCPs characterised by max RMSEx,y of 0.90m.

9 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, m sensor: DMCII and THEOS MS UK-DMCII THEOS (MS) Orbit type Sun synchronous Sun Synchronous Altitude 686 km 822 km Inclination (Sun synchronous) 98.2 [deg] 98.7 [deg] Launch Data July 29, 2010 October 1, 2008 Spectral band PAN: 450nm 900nm MS1 (Green): 520nm 600nm Blue: 450nm 520nm MS2 (Red): 630nm 690nm Green: 530nm 600nm MS3 (NIR): 760nm 900nm Red: 620nm 690nm Near-IR: 770nm 900nm Spatial resolution at nadir 22 m (MS) 15m (MS) Swath width 600 km 90km (MS) Product level L0R, L1R, L1T 1A, 2A 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, THEOS MS Geometric Quality Analysis (1) Fig. The 1-D RMSE_ICP [m] measured on the THEOS 2A MS products as a function of the overall off-nadir angle before (left) and after (right) systematic error elimination. The 1-D RMS errors measured on the THEOS 2A MS product are insignificantly sensitive to the overall off-nadir angle. The average absolute 1-D RMSE for the THEOS 2A MS product are 11.5m and 8.6m for Northing and Easting direction respectively, provided systematic error elimination (otherwise the values are as big as 119m and 90m).

10 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, THEOS MS Geometric Quality Analysis (2) Fig. The 1-D RMSE_ICP [m] measured on the orthorectified THEOS 1A MS image as a function of the overall offnadir angle angle after the single scene correction applying the Keystone Spacemetric SIP/Ortho rigorous model based on the 9 GCPs collected from orthorectified Landsat7 ETM PAN data of 15m accuracy and SRTM DEM CGIAR version 4 of 90 m grid interval (image matching techniques applied to GCP/ICP measurement). The 1-D RMS errors measured on the orthorectified THEOS 1A MS images are insignificantly sensitive to the overall off-nadir angle. The average absolute 1-D RMSE for the orthorectified THEOS 1A MS product are 6.8m and 6.8m for Northing and Easting direction respectively (1/2 pixel accuracy), provided systematic error elimination (otherwise the values are as big as 25m and 8m). 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, DMCII Geometric Quality Analysis Fig. 1-D RMSE [m] for 35 ICPs measured on the orthorectified DMC2 1T image as a function of the overall offnadir angle after the single scene correction applying the Keystone Spacemetric SIP/Ortho rigorous model based on the 20 or 50 GCPs collected from Landsat ETM+ data and SRTM DEM V31 data with 16m vertical accuracy EQC RMSE_X EQC RMSE_Y EQC RMSE_X incl. point ident. Error EQC RMSE_Y incl. point ident. Error 0 6 deg 19 deg 1-D RMSE measured on the orthorectified DMC2 1T image are slightly sensitive to the overall off-nadir angle. The average 1-D RMSE are 22m and 21m of Easting and Northing direction respectively (1 pixel accuracy), or 25m and 24m if DMCII point identification error is included. Promising results BUT based on limited DMCII samples - further testing is required!

11 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Key Issues (1) Based on the limited GE1 and WV2 sample images that were made available to us: - the GeoEye-1 orthoimage accuracy meets the geometric specification of 2.5m (1D) RMSE corresponding to EU technical requirements for: (a) at least 4 well-defined, well-distributed ground control points in case the Rational Functions mathematical model is used (the rigorous model requires a minimum 9 GCPs) (b) DEM of high accuracy in case of very inclined images, i.e. RMSEz = 6m if GE1 overall off-nadir angle exceeds 34 or 7.2m if the GE1 overall off-nadir angle is the WorldView-2 orthoimage accuracy meets the geometric specification of 2.5m (1D) RMSE corresponding to EU technical requirements for: (a) at least 4 well-defined, well-distributed ground control points in case the Rational Functions mathematical model is used (the rigorous model requires a minimum 9 GCPs) (b) DEM of very high accuracy in case of very inclined images, i.e. 3.5m if WV2 overall off-nadir angle exceeds 34 or 5.2m if the WV2 overall off-nadir angle is Accuracy requirements for auxiliary date, i.e. ground control points and DEM, given in Guidelines for Best Practice and Quality Checking of Ortho Imagery should be strictly followed 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Key Issues (2) Based on the limited K2, THEOS and DMCII sample images that were made available to us: - the Kompsat-2 orthoimage accuracy meets the geometric specification of 2.5m (1D) RMSE if the imagery is characterised by an overall off-nadir angle close to zero degrees and the rigorous model or first order Rational Polynomial sensor model for at least 9 GCPs is applied; - the THEOS PAN orthoimage can reach 2m accuracy provided that a dedicated rigorous model based on at least 9 well-distributed GCPs of high accuracy (i.e. RMSEx,y < 0.90m) is applied. - the orthorectified THEOS 1B MS product accuracy can reach 6.8m, provided that a dedicated rigorous model based on at least 9 well-distributed GCPs of appropriate accuracy is applied. - The orthorectified DMCII image accuracy can reach 1 pixel accuracy (22m), provided that a dedicated rigorous model based on at least 20 well-distributed GCPs and a DEM of appropriate accuracy are applied. PCI Geomatics and ERDAS IMAGINE have not implemented a THEOS dedicated model yet. Accuracy requirements for auxiliary date, i.e. ground control points and DEM, given in Guidelines for Best Practice and Quality Checking of Ortho Imagery should be strictly followed

12 16th Conference 'Geomatics in support of the CAP" Bergamo, Italy, November 24-26, Thank you! Satellite Image Quality Research Cooperation VERY WELCOME