Georegistration Practices Using TerrSet - Tips for Accurate Spatial Alignment

Geometric correction for digital images for GIS applications

Georegistration Practices Using TerrSet  

Introduction  

In GIS applications, the data from multiple sources are usually being used. Thus, there are often a certain amount of inhomogeneities in such data. For example, old analog maps or satellite pictures. In order to make them suitable for further utilization in GIS projects a correct coordinate system needs to be assigned to them. The process of taking a digital image and adding geographic information to the image so that GIS or mapping software can ‘place’ the image in its appropriate real world location is called georeferencing. The accuracy of georeferencing is defined by RMSE - root mean square error. The importance of Georeferencing lies in its ability to turn non-spatial imagery into spatial raster data for use in a variety of circumstances.

The first step of this project was to carry out the georeferencing of satellite imagery (SPOT data). The map of the road network and the topographic map with correct reference system were used to assign a correct reference to the spot imagery. The main idea of georeferencing is to locate and recognize unique or significant features, where we expect lower degrees of uncertainty and that would reduce the total error. Therefore, crossroads or easily identifiable natural features (capes within coastlines, confluence of rivers) are being used. The points recognizable in both maps are called Ground Control Points (GCP). Ground control points are normally used to associate projection coordinates with locations on a raw (uncorrected) image.

Study Area and Dataset  

The municipality of Lund is located approximately between coordinates 380000 and

414000 (E-W) and 6153000 and 6185000 (S-N), therefore, it comprises the 1018.65 km2area of Lund kommun and its neighbouring area. The study area consists of built-up areas, agriculture zones, waterbodies and infrastructure buildings such as roads, rail tracks and an airport.

Satellite images from two different sources, Spot 5 and Sentinel 2, and the landmateriet dataset were provided by the department. Sentinel-2 Observations include 13 wavelength bands. Band 2, 3, 4 and 8 have 10-meter spatial resolution, band 5,6,7, 8a,11 and 12 have 20 m original spatial resolution and band 1, 9 and 10 have 60 m original spatial resolution. This dataset consists of three satellite observations for the dates 20150819, 20160502 and 20160505 from two different tiles(33UUB, 33UVB). On the other hand, the Spot 5 image, acquired on June 10, 2011, includes 4 bands. The image covers the southwest area of the Skane region and it is stored in GEOTIFF format with the spatial resolution of 10 meters.

Besides satellite images, there is landmateriet dataset provided in different formats. Höjddata consists of elevation values with two decimals in a raster grid with a resolution of 50 metres. The DEM raster is produced during 2009-2016 using airborne laser scanning and the final product is acquired by linear interpolation in a TIN(Triangulated Irregular Network) with 1m resolution. The resolution of The DEM data provided in tiff format is 50 meter resampled from the higher resolution by bilinear interpolation. Marktackedata is a land cover map converted to terrset .rst format. Terrangkartan file includes topographic information in tif file. Vagkartan dataset includes five shapefiles and they are considered the most useful layers for georeferencing. These five vector layers possess administrative borders, railways, urban areas, water, roads. The reference system of lantmäteriet data is SWEREF99 TM (epsg3006). Since TerrSet does not include the reference file (.ref) to register SWEREF99 TM, this reference file and the symbology files (.smp) for Terrängkartan and Marktackedata dataset is also provided.

Table 1: Metadata of datasets used in the project.

Methodology  

First the .tif-file imagery was imported into TerrSet, and converted to .rst format that includes four different bands by using the tool called GDAL Raster Conversion Utility. The topographic reference image was also converted from .tif to .rst format. However, in the metadata of the topographic map, the reference system had to be defined, by choosing the sweref99tm.ref file that was provided.

The function WINDOW was used in the topographic map to crop out the area where the municipality of Lund is located. That cropped topographical raster was used as the reference input image for the georeferencing.

The IDRISI tool for georeferencing is called RESAMPLE. The window of the tool provides a table of GCP and preview of both images, which makes it a bit effortless to look for identical features. However, because of the quality and the resolution of the input data it wasn’t always easy to find the point locations. The easiest to locate were man-made infrastructure, basically roundabouts, crossroads and unique natural features like islands in lakes. Sometimes Google earth was a useful tool to compare the images with reality. 16 ground control points (GCPs) were utilized.

Figure 1: Depicts the GCP selections

Results  

After georegistration to the coordinate system SWEREF99™, the form of the raster is oriented 16 degrees in the clockwise direction. The background value 0 is filled out to the area where the raster cells do not exist. The total root mean square error (RMS) for the 16 GCPS is 0.55.(figure below) According to the literature, there is a threshold that should be below 5 which in this case is accurate enough.

Figure 2: Shows the total RMS

Figure 3: Georeferenced Spot-5 Band-3.

Discussion  

The accuracy of the method used to locate the GCP points is highly influenced by the resolution of the reference rasters. Other factors that can cause lower accuracy are a map production method with an error already included in the reference map, a spatial distribution of GCPs and the number of GCPs. Furthermore, a subjective factor like user experience has to be taken into account. The year of the topographic map produced is unknown. The input reference image Spot-5 acquired in 2011, thus these two sources could be created in different periods of year. The assignment of GCPs may be misled by the replaced object we used to match our GCPs. Moreover, the georegistration was performed by only one process although the consecutive iterative operations may increase the quality of the result. However, given all the limitations described above the total RMS of 0.55 indicates that the quality and precision of input data were sufficient and the identification of GCPs was done properly.