Updating Maps of Kazakhstan Using Stitched SPOT Images
By Shouji Sakaino, Hideo Suzuki, Dr. Philip Cheng, and Dr. Thierry Toutin

Since the launch of the SPOT satellites, the data has been used successfully in many applications, such as updating topographic maps, digital elevation model (DEM) extraction, and land-use classification. SPOT has a 10m panchromatic sensor and a 20m multispectral sensor. The latest SPOT 4 satellite has added a SWIR sensor.
      This article will describe how SPOT images were used in updating maps of the Republic of Kazakhstan. A new technique to stitch and process SPOT scenes, taken from the same satellite path into a single strip, is also described. The technique reduces the number of ground control points (GCPs) required, decreases the amount of processing, and eliminates the mosaicking process. Consequently, the project time and costs have been reduced by a factor of two to three times.

Geographic background of The Republic of Kazakhstan
The Republic of Kazakhstan is situated in Central Asia. The territory of the Republic covers an area of approximately 2,717,300 square km, and stretches from the Caspian Sea in the west to the foothills of the Altai Mountains in the east, from the west Siberian lowland in the north to the Tien-Shan mountain range in the south. Since independence the capital was Almaty City, but it has recently been transferred to Astana City, which is located in the central part of the country. The population of the Republic is approximately 17 million, composed of 131 different ethnic groups. The national language is Kazak, but Russian is used equally as a national language. Wheat is the main agricultural product, occupying 20 percent of arable land and contributing more than 10 percent of wheat-share in the former USSR. Kazakhstan is rich in commercial minerals as well as coal and oil, where recently an oil field on the east coast of the Caspian Sea came into the limelight.

The need for updated maps
The Republic is covered by national base maps which were prepared during the period of the former USSR. These base maps had been corrected and revised every five to eight years. However, since the latter part of the 1980s, almost no revision had been conducted due to the financial stringency of the USSR. The situation after independence has not changed the state of the maps. More serious than this, however, is that under the agricultural policy enforced by the USSR from the 1950s, excessive development of arable land in the Syrdarya basin in the southern Republic resulted in a remarkable deterioration of agricultural productivity, and the deterioration of the rural environment. These environmental changes were brought about by civil works construction for irrigation, drying up and desertification of agricultural land, dropping of the level of ground water, and salination of the soil. These environmental concerns have received considerable attention from the World Bank, UNDP, and UNEP, etc., as an environmental problem of the Aral Sea. Numerous projects have been proposed, such as the replanning of agricultural land use, and planning for improvement of the actual environment by the riparian countries. However, it is virtually imposssible for these projects to be realized because of the lack of accurate, up-to-date base maps.
       In consideration of the serious deterioration of the environment, the Republic requested technical cooperation for the urgent revision of 1:200,000 topographic maps that cover approximately 150,000 square km of the Syrdarya, and 1:100,000 topographic maps covering the approximately 22,500 square km where most of the environmental change has occurred.
       In response to the request by the Government of Kazakstan, the Japan International Cooperation Agency (JICA) organized a "Study Team for the Urgent Establishment of National Basic Geographic Data" in the southern area of Kazakstan. Aero Asahi Corporation in Japan has conducted the three-year project as a consulting agent appointed and authorized by JICA.

New technique to stitch SPOT data
Satellite imagery has many advantages, such as observing a vast area in a short period, providing land cover classification, and the monitoring of environmental change. SPOT data was chosen for this project due to its high resolution, 8-bit image data, and the latest SWIR band on SPOT 4. Seventy-one panchromatic and 31 multiband scenes had been ordered for the area. In order to use the SPOT data for mapping, each scene must be orthorectified. This involves reading the data, collection of GCPs, rigorous satellite modeling, orthorectification, and mosaicking. Due to the large amount of data, it was necessary to reduce the time and cost in producing ortho-images. This was achieved by using a new technique which stitched and processed SPOT data, taken on the same satellite path, into a continuous strip.
     SPOT scenes are acquired in continuous strips and subset into smaller scenes (for example, 6000 pixels by 6000 lines for panchromatic) for sales. There are small overlaps between scenes along the same path. Therefore, it is possible to stitch scenes of the same satellite path back into one continuous strip. There are three ways to stitch SPOT scenes together. (1) Use a line correlation technique to find the overlap between scenes. This method is cumbersome and time-consuming. (2) Use the attitude record embedded in each line to match the attitude record with other images. This method is recommended for SPOT products recorded in the old format only. (3) Use the modelization record stored in the SPOT CAP format, which transforms line numbers between the small scene and the original strip. This method is recommended for the CAP format only.
      Since all the data are stored in CAP format, option 3 was chosen. Software was developed at PCI to stitch the SPOT data together. Up to a maximum of five SPOT scenes have been stitched together on the same path. As an example, five 6000-pixels by 6000-lines SPOT panchromatic scenes became a single 6000-pixels by 28,158-lines scene after stitching, and there is no geometric or radiometric discontinuity in the stitched strip.
      There are three advantages to using the stitched scenes. The first advantage is the reduction in the number of GCPs required. Instead of collecting a number of GCPs for each scene, the same number of GCPs is only required for the stitched scene. This reduces time and cost in collecting GCPs. The second advantage is that an ortho-image of an area can be generated from the stitched scene in one step instead of processing each individual scene separately. The third advantage is that, since scenes along the same path are stitched together before orthorectification, the resulting ortho-image is also continuous without geometric or radiometric discontinuity. The mosaicking process for scenes taken from the same path is thus eliminated.

Software and satellite modeling
PCI OrthoEngine software was used in this project. This software runs on most computer platforms and has the ability to orthorectify Landsat, IRS, SPOT, Radarsat, ERS and JERS data in different formats. The software uses a collinearity condition method developed by Dr. Thierry Toutin at the Canada Centre for Remote Sensing (CCRS), Natural Resources Canada. The algorithm uses principles related to orbitography, photogrammetry, geodesy and cartography. The algorithm and software have been successfully applied to VIR (Landsat, SPOT, IRS, etc.) and SAR (ERS, SIR-C, RADARSAT, etc) data. Since the method can adjust simultaneously, more than one strip can be used to improve the relative accuracy between strips. Based on good quality GCPs, the accuracy of the model was proven to be within 1/3 of a pixel for VIR satellite images, and one resolution cell for SAR images. The software has been tested at CCRS and the Canada Centre for Topographic Information, and it meets the specifications of Canada's National Topographic Database to update digital topographic data at 1:50000 scale using SPOT panchromatic data.

Ground control point collection and DEM
Two types of GCPs were used for the project. (1) Well-defined planimetric features on existing topographic maps, and those that are clearly interpretable on the satellite image, were adopted as "ground control point derived from map." (GCP-MAP) Coordinates of those points were measured on the reproduced printing plates. (2) On the occasion where remarkable planimetric features did not exist on the map, well-defined features, where accessible on the SPOT image, were selected as "ground control points derived from GPS." (GCP-GPS) GPS observations were carried out by placing a GPS antenna on the point with reference to a fixed GPS station placed on an existing first-order triangulation station, so that geo-referenced coordinates could be achieved. All GCP-MAP and GCP-GPS results were transformed to the WGS-84 system from the CK-42 coordinate system, and vice versa.
      To generate DEMs to be used for orthorectification, 1/100,000 and 1/200,000 reproduced contour-printing plates were digitized by scanner, and a raster to vector conversion process was used. The contour line vector data was compiled by adding elevation attributes, and the DEM was built from the vector data by OrthoEngine software.
      The area was divided into Zone A and Zone B. Zone A is mainly desert area. Zone B consists mainly of the Syrdarya basin and many artificial objects such as buildings, bridges, and rice fields. It was easier to collect GCPs in Zone B than it was in Zone A. Table 1 shows the number of scenes used before and after stitching.

Results
Table 2 shows two different results based on the source of GCPs. The first results are based on all the GCP-MAP and GCP-GPS GCPs, and the second results are based on all GCP-GPS GCPs only. The larger residuals in Zone A reflect the GCP cartographic coordinate errors due to the difficulty to acquire accurate GCPs in the extensive desert area. Table 2: Residual results (in meters) using different number of GCP-MAP and GCP-GPS GCPs. RMS is the root mean square residual.
      This is the final year of the three-year project, and the results of block adjustment of satellite images have been found acceptable for digital mapping. The quality of the block adjustment and the digital map was examined on the screen by overlaying satellite images with newly digitized maps and existing map layers. The most accurate terrain features on the existing maps are railways. They were plotted by polygonometric survey and not by photogrammetric means, and the digitized railways coincide within one or two pixels.

Summary
SPOT scenes taken from the same satellite path, using a new technique to stitch and process satellite data, was successfully applied in updating maps for large areas, thus reducing the time and cost for map updating by an approximate factor of two to three times.

Acknowledgements
The authors would like to thank PCI Enterprises and Aero Asahi Corporation for providing all the support for this work, and CCRS for providing the satellite modeling algorithms, technical assistance and the text editing.

About the Authors:
Mr. Shouji Sakaino is an engineer at Aero Asahi Corporation. He may be reached at [email protected]. Mr. Hideo Suzuki is an engineer at Aero Asahi Corporation. He may be reached at Hideo.[email protected]. Dr. Philip Cheng is a senior software engineer at PCI Enterprises, Richmond Hill, Ontario, Canada. He may be reached at [email protected]. Dr. Thierry Toutin is a senior research scientist at the Canada Centre for Remote Sensing, Natural Resources Canada. He may be reached at [email protected].

Back