Production of a Landsat-7 ETM+ Orthoimage Coverage of Canada David B langer 1, Alain Benoit 1, Eric Loubier 1 and Mike Wulder 2 1 Centre for Topographic Information, Sherbrooke (Natural Resources Canada), 2144 King Street West, room 010, Sherbrooke, QC, J1J 2E8. Phone: (819) 564-5600; Fax: (819) 564-5698; Email: dbelange@rncan.gc.ca, abenoit@nrcan.gc.ca, eloubier@rncan.gc.ca. 2 Canadian Forest Service (Natural Resources Canada), 506 West Burnside Road, Victoria, BC V8Z 1M5. Phone: (250) 363-6090; Fax: (250) 363-0775; Email: mike.wulder@nrcan.gc.ca 1
Abstract The Centre for Topographic Information (CTI) of Natural Resources Canada is currently producing a complete set of cloud-free orthoimages covering the Canadian landmass using data from the Landsat-7 satellite (under a project called Ortho7). The project is being undertaken in partnership with GeoConnections, the Canada Centre for Remote Sensing (CCRS), provincial and territorial agencies as well as other federal-government departments. In addition to financial support, partners are providing topographic control data to assist in producing orthoimages of a higher accuracy. The creation of a national coverage with Landsat-7 ortho images will provide an up-to-date fundamental geospatial framework for Canada. These products will contribute as an excellent reference for map updating; their geometric integrity will facilitate data integration from other map and image sources; and finally the imagery s inherent information content can serve as a rich baseline for the Canadian landmass. Image acquisition for this initiative began in 1999 and will continue until a complete coverage of Canada is obtained (scheduled for completion in 2004). Of the estimated 750 scenes required to cover the Canadian landmass, 400 images have already been identified as suitable for production. The primary criteria is that the imagery must be cloud and haze free. The ortho-correction is being done in partnership with Canadian industry and is proceeding as scheduled. This article is intended to provide details about the Landsat-7 ortho-image data specifications and production process. Background The Centre for Topographic Information of Natural Resources Canada is responsible for the acquisition, management and dissemination of topographic information for the Canadian landmass. The CTI location in Sherbrooke (CTI-S), Quebec, is leading the project to create an orthographically correct Landsat image coverage for Canada. CTI will be using the ortho-images to update themes of the National Topographic Database (NTDB) such as the hydrographic network, built-up areas, vegetation, wetlands, among other aspects. In addition, this data, along with those acquired through partnership agreements and outsourcing (road network, rail network, etc.), will be used to populate CTI s new geospatial database (GDB). CTI will use the GDB to offer customers a new range of digital topographic products, which will be more accurate and up-to-date than the current National Topographic Database (NTDB). CTI has designed and implemented the production environment necessary for achieving the orthoimage coverage for Canada, and as well have designed the image acquisition, production, quality control, product distribution and overall project management for this initiative. Through GeoConnections, CTI was awarded funding to produce the ortho-image data for Canada. These funds largely support contracting out of the ortho production process. The various partners are providing funds, primarily for image data acquisition. Partners are also contributing groundcontrol data the provision of which facilitates the development of ortho-images with as high accuracy as currently achievable with the available resources. This project will also yield a metadata and control point database, appropriate for use in future ortho-image production projects. 2
The ortho-corrected Landsat-7 ETM+ data is being provided to all acquisition partners on compact disks (CD-ROM) as the data become available. Each CD contains information about the image ortho-rectification specifications, metadata on the ingest and output data, and includes the orthocorrected image data in a GeoTiff format. A web based distribution environment has also been established using the GeoGratis web site(http://geogratis.cgdi.gc.ca/frames.html) as well as the CTI On-line Purchasing web site (http://www.cits.rncan.gc.ca/). Thus, a broad distribution policy has been developed to ensure wide utilization of this particular product opportunity. In doing so, it is anticipated that the high quality data being provided will stimulate interest from less traditional users and thereby contribute to fostering greater interest in geomatics data and it s potential applications. CTI Landsat data collection consortium With the support of GeoConnections and the Canada Centre for Remote Sensing (CCRS), CTI-S established agreements with the larger stakeholders in the field of geomatics in Canada (federal, provincial, and territorial levels). Funding from these partners is primarily intended for data acquisition, since some 750 Landsat-7 scenes will be required to complete the project. Of the 10 Canadian provinces, 9 are full participants, and all three territories are participating. The participating federal government agencies, are Agriculture and Agri-food Canada, Canada Centre for Remote Sensing, Canadian Forest Service, Canadian Heritage (Parks), Canadian Transportation Agency, Department of Indian and Northern Development, Department of National Defense, Elections Canada, Environment Canada, Fisheries and Oceans Canada, and Statistics Canada. Image data description Landsat-7 ETM+ was launched April 15 th 1999 (Goward et al., 2001). Since then, CTI has been monitoring the quality of images collected over Canada and purchasing those that meet seasonal and atmospheric criteria. The Landsat-7 imagery is collected from May to October. Only those images without cloud or haze are purchased. A brief description of the Landsat-7 Orthoimage production plan resides on a web site at CTI 1. The Ortho7 web site is comprised of three sections: The first section provides production progress on a per province or region basis showing the orthoimages produced, those currently in production and the image availability for ortho-correction (cloud free). A list of the selected images is available on-line. The second section provides the theoretical planning for Landsat-7 coverage of Canada. The Worldwide Referencing System (WRS) is used to define the acquisition and collection of Landsat imagery. A summary of the WRS, and specifically how it relates to Canada, can be found in Wulder and Seemann (2001). The overlap of Landsat frames is approximately 40% at Canada s southern border, increasing to greater than 80% in 1 http://www.cits.rncan.gc.ca/sitecits/servlet/cits?site_id=01&sessionid=&page_id=1-011-005 3
northern Canada. Using thinning rules to maintain all images at <60, every second image from 60 to 68, and every third image above 68 results in a theoretical coverage of 650 images. Prior to thinning of the WRS, over 1200 images cover Canada s land mass. The last section shows graphically the progress of the project. Cloud free images acquired May to October are eligible for purchase for the orthoimage. To date, over 140 orthoimages have been produced and distributed to partners. Another 60 orthos are currently in production and should be available for distribution by fall 2002. The production schedule is established according to availability of cloud-free imagery, (where possible with contiguous scenes), availability of sources for control, and stakeholders priorities. The data is geocoded to the UTM format in the NAD83 datum. The image data are resampled using a cubic convolution algorithm. The image radiometry is altered using the gain and offset values appropriate for the date of image acquisition. The metadata for the imagery contains information on scene centre, sun elevation, sun azimuth, and time to allow for additional radiometric adjustments. A National Context for Orthoimages Production Accuracy is one of the unavoidable criteria for this production project. In order to obtain the highest image accuracy, a rigorous mathematical model and uniform methodology in selecting the control points are used, in addition to accurate control data. Sources of Control Control from various sources is used for image correction, the most accurate being given the priority. The usual priority ranking is: GPS source for the National Road Network, provincial vector data, accurate NTDB data (10 m to 90 %), federal-government aerotriangulation data, and other sources. Mathematical Model The decision to use the mathematical model developed by Dr. Thierry Toutin was based on the fact that it provides a rigorous geometric correction method that takes into account all distortions. The model is based on principles such as orbitography, photogrammetry, geodetics, and cartography (Cheng et al., 2000). Methodology for Selecting Ground Control Points Giving the national scope of this project and Canada s geographic situation, a rigorous methodology that could be used for the entire country had to be developed. A series of test lead to the definition of criteria for the number of control points to be used. A minimum of six points is needed per NTS 50k sheet. Some additional points are needed for maximum and minimum 4
elevations. Points located in the overlap between images are required to provide smooth transition and consistent elevation values. Testing has also demonstrated that the centres of mass of lakes/islands and road intersections are the two types of control points that yield the best results. These criteria served as the basis for outsourcing specifications and for developing a uniform methodology for producing orthoimages for the entire country. 1.1 Awarding of Contracts to Geomatics Firms The work for delimiting the entities used for image orthorectification is carried out by Canadian geomatics firms. Approximately 700 Landsat 7 are contracted out through requests for proposals (RFP) where bidders must demonstrate that their processes can meet the technical specifications developed and issued by CTIS. Two types of contracts are open to firms depending on whether the control-data source is vector or federal-government aerotriangulation data. CTIS, for its part, is responsible for controlling data capture quality, calculating road intersections and the centres of mass for lakes and islands, final selection of ground control points, and batch processing the geometric correction of the images. Quality Control In order to minimise the time and resources associated with quality control, CTIS has put in place a computerised quality-control process that consists of automated and semi-automated data processing (see Figure 1) to verify the structure of the entities provided and their compliance with the related metadata. A bonus/penalty system provides incentive to contractors to strive for quality. Upon delivery of data, once the structure has been accepted, a semi-automated process is initiated. This process verifies if the choice of entities and if the data capture quality with respect to the source, complies with specifications. A 3-person team participates to the quality control of 250 orthoimages/years. 5
Figure 1 Quality-control and production system for Landsat 7 orthoimages. This figure illustrates the main steps involved in correcting Landsat 7 images. The grey rectangles represent steps that have been automated. 1.2 Generating Orthoimages Another completely automated process is executed to complete the QC. It calculates the final mathematical model for geometric correction, generates a digital elevation model (merge of various DEMs) for the complete satellite scene, generates the orthoimage metadata, and produces the orthoimages in the projections requested by partners (see Figure 1). The control entities and image entities used to correct the images are stored in a spatial database (Oracle) for future use. Many images have already been processed in this way and delivered to partners (see Figure 2). In addition, many contracts are now underway. Our target is to have complete coverage of Canada with Landsat 7 orthoimages in 2004. 6
Figure 2 Summary of the production of Landsat 7 orthoimages in Canada as of March 26, 2002. This figure shows the images that have been orthorectified (completed images), those being processed by contractors (in progress), and those that have been selected because they have little or no cloud coverage (cloud-free). The latter are considered as potential images for upcoming contracts. 2. Orthoimage Accuracy A number of parameters are analysed in order to obtain orthoimage accuracy with an acceptable level of confidence. These parameters include the mean square error (MSE) along the x-axis and y-axis of the ground control points after the correction model has been generated, the accuracy of the image control data, and the altimetric and planimetric accuracy of the digital elevation model (DEMs) covering the satellites scene. In order to determine the impact of the DEM on orthoimage accuracy, a slope analysis is carried out for the entire satellite scene using Canadian Digital Elevation Data (CDED) at the 1:250 000 scale. The assessment of this impact takes into account the maximum slope of the terrain and the maximum angle of incidence of the sensor. Once all the errors have been combined, the accuracy is calculated to a level of confidence of 90% (Circular Map Accuracy Standard (CMAS) (CCSM, 1984)). More than 93% of the orthoimages generated to date have an accuracy better than 20 m (see Figure 3 and Table 1) with a level of confidence of 90% (Circular Map Accuracy Standard 7
(CMAS)). The production of Landsat 7 orthoimages will make it possible to have the most accurate and consistent source for complete coverage of the country. Figure 3 Position and accuracy (to 90%) of images orthorectified as of March 26, 2002. Figure 3 illustrates the accuracy of corrected images with an interval of 5 m, each corresponding to a different color (hue). Table 1 Accuracy to the nearest metre of orthoimages by province or territory according to level of confidence as of March 26, 2002. Province or Territory Number of Average MSE Average CMAS (metres) Images (metres) Newfoundland 0 --- --- Nova Scotia 1 12 18 Prince Edward Island 1 12 18 New Brunswick 2 14 21 Quebec 5 9 13 Ontario 41 10 15 Manitoba 22 11 16 Saskatchewan 26 10 15 Alberta 25 10 15 British Columbia 3 14 21 Yukon 0 --- --- Northwest Territories 22 9 14 Nunavut 43 9 14 8
Mean square error (MSE) combined in X and Y corresponds to accuracy with a level of confidence of 63.21%. The CMAS is more used in cartography. It corresponds to the MSE circular multiplied by a factor of 1.5174 (CCSM, 1984). This represents a circular accuracy with a level of confidence of 90% and is used in assessing the orthoimage accuracy. An image may be located in more than one province or territory. Conclusions This national Landsat-7 coverage of Canada, enabled through GeoConnections funding, CTI and CCRS leadership is providing invaluable resource characterizing the state of the Canadian landmass. The sharing of this data with partners and availability via the web provides an unprecedented opportunity for the characterization and assessment of the Canadian landmass and in so doing providing baseline documentation for the status of our terrain circa 2000. The integrity of the ortho-corrected imagery allows for a common geometry between the Landsat- 7 scenes and other geographic data of Canada. The successful partnerships brought to bear in this initiative are an excellent example of a working framework for future opportunities. 9
Acknowledgments The support and partnership of the provinces, territories, and federal agencies (Agriculture and Agri-food Canada, Canada Centre for Remote Sensing, Canadian Forest Service, Canadian Heritage (Parks), Canadian Transportation Agency, Department of Indian and Northern Development, Department of National Defense, Elections Canada, Environment Canada, Fisheries and Oceans Canada, and Statistics Canada) are gratefully acknowledged. GeoConnections is thanked for the provision of funding to aid in the implementation of this project. Yvon Boucher and Pierre-Yves boisvert, of the Centre for Topographic Information, are thanked for their valuable insights and communications. The support and scientific advice of Dianne Richardson, Tom Feehan and Thierry Toutin of the Canadian Centre for Remote Sensing are gratefully acknowledged. 10
References CCSM (1984) National standards for the exchange of digital topographic data, Data Classification, Quality Evaluation and EDP File Format, Volume 1, E.M.R. Canada: 153,157 Cheng P, Toutin Th, Tom V (2000) Unlocking the Potential for Landsat 7 Data, EOM, 28-31. Goward, S., J. Masek, D. Williams, J. Irons, and R. Thompson, 2001; The Landsat 7 mission: Terrestrial research and applications for the 21 st century, Remote Sensing of Environment, Vol. 78, pp. 3-12. Wulder, M., and D. Seemann, 2001; Spatially partitioning Canada with the Landsat Worldwide Referencing System, Canadian Journal of Remote Sensing, Vol. 27, No. 3, pp. 225-231. 11