Georeferencing with ArcGIS Pro
Georeferencing is the process of aligning spatial data from one coordinate system to another, typically linking data from a non-spatial source to spatial data. It involves assigning geographic coordinates to features in a dataset based on their location on the Earth's surface. Georeferencing allows for accurate spatial analysis, visualization, and integration of diverse datasets within a geographic framework.
Scenario: Imagine that you are working on a cultural preservation project aimed at documenting traditional gathering sites. As part of your research, you come across a valuable historical map or a hand-drawn sketch depicting the location of these sites. However, the map lacks geospatial information, making it challenging to precisely locate the gathering sites on the ground. To overcome this obstacle, you decide to georeference the map using modern GIS technology or/and satellite imagery. By aligning the features on the historical map with corresponding features in the geospatial dataset, you are able to overlay the historical information onto the current landscape. Through georeferencing, you bridge the gap between the past and the present, ensuring that traditional knowledge is preserved and accessible for future generations.
Software requirements:
ArcGIS Pro 3.x
1 - Data download
Follow this link to download the ArcGIS Pro package that will be used in this tutorial Georef_Lummi_Nation.ppkx. Save it in a meaningful location.
2 - Double-click on the Georef_Lummi_Nation.ppkx to open it in ArcGIS Pro.
Your display should be similar to the one below.
3 - Open your 'Catalog Pane' (View menu)
4 - Expand your 'Folders', 'Georef_Lummi_Nation', commondata, and the 'Georef_Lummi.gdb'.
The geodatabase has two rasters. One named 't1871'. This is your source image, and it is not georeferenced. One named 'WA_Bellingham_1975_geo.tif' that is georeferenced (reference map or target image).
5 - In your 'Contents Pane', right-click on 'WA_Bellingham_1975_geo.tif', select 'Properties', 'Source' and 'Spatial Reference' to take note of the coordinate system. Your target raster is projected (NAD 1983 StatePlane Washington North FIPS 4601 (US Feet)).
6 - In the 'Catalog Pane', select 't1871' and drag it in your data view. A pop-up window may ask you to calculate statistics. Select 'Yes'.
As you can see, your raster is added to your 'Contents pane' but not displayed in your data view. You may also see a notification telling you that the raster as an unknown coordinate system.
7 - Go to the 'Imagery" tab and click on 'Georeference' to open the toolbar.
The georeference ribbon r looks like this.
8 - In the Contents pane select your 't1871 layer', and in the 'Prepare' group select 'Fit to Display.'
Your source map should now be visible in your data view.
This map is part of the 19th century T-sheets created by the United States Coast & Geodetic Survey. It represents the topography of Gulf of Georgia, Village Point to Base of Sandy Point, 1888. Scale: 1:10,000; Surveyor: J.J. Gilbert, Puget Sound River History Project, University of Washington (see link in the References section below). It also shows the West part of the Lummi Nation (Treaty of Point Elliott, 1855).
9 - Make sure that your 't-1871 layer' is still selected and click on 'Move" in the 'Prepare' ribbon.
We want to move the 't1871' layer, so it displays side-by-side with the 1975 topographic map. The Lummi Nation is depicted in yellow on the topographic map. You can then click and drag 't-1871' to the side. SAVE your project!
10 - To switch back your cursor into selection mode. Go to the 'Map' menu and select 'Explore'. Go back to the 'Georeference' menu when you are finished. Now you are ready to start your georeferencing.
In the georeferencing process, a minimum of three ground control points is necessary to accurately align spatial data with the geographic coordinate system. These control points are identifiable features that are common to both the source dataset and the reference dataset (target), allowing for precise transformation.
11 - We will perform the georeferencing using a 1st order polynomial transformation (requires at least 3 control points). These control points are used to compute a linear transformation function that minimizes distortion between the two datasets. This transformation accounts for translation, rotation, and scaling difference, ensuring the spatial data is accurately aligned with the reference coordinate. By using the 1st order polynomial transformation, the georeferencing process achieves a basic level of spatial accuracy suitable for many mapping and spatial analysis applications. It is also the selected default transformation in ArcGIS Pro. Feel free to try the other transformations.
Suitable control points are identifiable features that are common to both datasets. Road intersections, river forks, prominent landmarks, building corners, and distinct geographic features such as mountain peaks or lake shores are all examples of suitable control points. Additionally, it is important to choose control points that are spread across the extent of the dataset to ensure comprehensive spatial alignment.
12 - Start by looking for features in your source image 't1871'. Sometimes it could be challenging to find features. We found three potential control points that we can see in both images.
13 - Make sure that your source image 't1871' is still selected in your contents pane. In the 'Adjust' group, select 'Add Control Points'.
The first control points should always be on your source image.
15 - Now your cursor displays a + sign with the label ' From point (source)'.
16 - Navigate to where the first point is located on your source map and left-click once to drop your point.
Now be careful because you have a string attached to your cursor.
17 - Navigate to where the same feature is located on your reference map (topographic map) and left-click once to drop your point.
If you need to pan your data view while the control point tool is activated, keep the C key pressed down on your keyboard.
Your source map will automatically shift to the left to align with your reference map.
SAVE your project!
18 - Now repeat the steps to add points #2 and #3. You can toggle the visibility of your source map on/off to drop the points on your reference map. Once again, make sure that you drop a control point on your source map 't1781' first, and then on your reference map.
19 - Save your project and save your georeferencing in the 'Save' group.
20 - Visually, you should be able to tell if your georeferencing is successful by adjusting the transparency of your source image (or using the swipe tool). Go to the 'Raster Layer' tab and adjust your source image's transparency.
The software can mathematically display the residual errors of each individual point as well as the complete Root Means Square (RMS) error.
20 - Locate the 'Review' group and open the 'Control Point Table'. Your table may look different than ours.
The control point table recorded a list of your identified control points, including their coordinates and corresponding locations on both the source dataset and the reference map. Each entry in the table provides essential information for aligning the spatial data accurately with the reference coordinate system.
You can notice that our residuals are not displayed. Residuals refer to the difference between the observed values (in our case, the locations of control points in the source dataset) and the predicted or expected values (the corresponding locations in the reference map) after performing a transformation. Residuals can be calculated for each control point and for each dimension (X and Y).
When only three control points are used in the georeferencing process, the residuals are not calculated because they are not as informative or meaningful compared to cases where more control points are used. With only three points, there isn't enough data to provide a comprehensive understanding of the accuracy of the transformation.
While some software may calculate residuals for three control points (not with ArcGIS Pro), it is important to interpret then cautiously and recognize their limitations due to the small sample size.
Additional control points provide more data points for the transformation algorithm to optimize the spatial alignment and calculate the residuals, leading to a more robust georeferencing.
21 - Add a fourth control point. The image below displays potential choices.
We chose the Lummi River forks intersection.
22 - Now take a look at your residuals.
The goal in georeferencing is to minimize residuals as much as possible. Lower residuals indicate that the georeferencing transformation is accurately aligning the spatial data with the reference map, resulting in less discrepancy between observed and predicted locations.
Regarding the units of residuals, if your reference map's linear unit is in feet (our case here) and you have a residual of 373.645112 (our point #1), it means that on average, the observed locations of your control points are 373.645112 feet away from their predicted locations after the georeferencing transformation. This provides a measure of the average error or discrepancy in the spatial alignment between the source dataset and the reference map.
When the residuals are close to 0, it indicates that the observed locations of the control points closely match their predicted locations after the georeferencing transformation. Residuals close to 0 are desirable and typically indicate that the transformation has been successful in accurately registering the spatial data to the reference coordinate system.
23 - Edit a few points by choosing different features to see if you can improve your residuals. You should usually remove the points that have the highest residuals first. However, our point #4 has the highest residual but it seems to be a better feature than our point #1.
24 - In your 'Control Point Table', select the point that you would like to delete and click on the 'Delete Selected" tool. You need to select the entire row.
25 - Choose other features on both maps.
26 - Our residuals are way lower.
When analyzing residuals, it is essential to consider both their direction and their magnitude to understand their implications for the accuracy of the transformation.
If the Y residual (vertical direction) is negative, it suggests that the corresponding points in the Y direction are shifted downwards from their expected locations (the opposite for the positive ones).
If the X residuals (horizontal direction) is negative, it suggests that the corresponding points in the X direction are shifted to the left from their expected locations (the opposite for the positive ones).
If the residual (overall residual) is positive, it indicates that the overall discrepancy between the source points and their corresponding locations in the reference map is positive.
While each individual residual provides information about the direction of displacement in its respective dimension, the overall residual (positive in our table above) gives an indication of the overall discrepancy or error in the georeferencing transformation. Therefore, the overall residual is typically the most important value to consider when evaluating the accuracy of the transformation.
27 - You can also choose a different transformation if you decide to add more than 4 points to see if you can improve your georeferencing.
You can select the transformation directly in your 'Control Point Table' or by choosing 'Transformation' in the 'Adjust' ribbon.
28 - When you are finished, save your project and save your georeferencing.
29 - You can generate a report of your georeferencing for your records and/or to share with others. Locate 'General Report' in the 'Save' group.
The georeferencing report provides three different Root Means Square error (RMS) values:
The total RMS error (forward) of 43.765870 indicates the average discrepancy between the observed locations of control points in the source dataset and their corresponding locations in the reference map after performing the forward georeferencing transformation. This value suggests a relatively high level of error in the forward transformation.
The total RMS error (inverse) of 0.006571 represents the average discrepancy between the observed locations of control points in the reference map and their corresponding locations in the source dataset after performing the inverse georeferencing transformation. A low value like this indicates a high level of accuracy in the inverse transformation, meaning that the reference map aligns well with the source dataset.
The total RMS error (forward-inverse) of 0.000000 suggest that there is no discrepancy between the forward and inverse transformation, indicating that the georeferencing process is consistent and reversible.
Overall, while the forward transformation has a relatively high RMS error, the inverse transformation is highly accurate, and both transformations are consistent with each other. Additionally, while the forward transformation error may be relatively high, the low inverse transformation (0.006571) suggests that the georeferencing process aligns well with the reference map, which is a positive indication of accuracy.
Given that the source map is hand-drawn, achieving a total RMS error (forward) of 43.765870 may be considered acceptable. That would also depend on the specific requirements and accuracy standards of your project. 19th century hand-drawn maps inherently contain more uncertainties and inaccuracies compared to modern digital or professionally surveyed maps, so some level of error is expected.
With hand-drawn maps, your georeferencing may be challenging but you can get pretty close. Sometimes, 'pretty close' is enough!
30 - SAVE your project and close the georeferencing tool.
We hope that this article has been helpful! If you have any feedback or questions, please feel free to send us an email or connect with us for a chat. The NTGISC team is here to assist you further!
References
Original scanned image T-1871, United States Coast & Geodetic Survey Topographic Sheet,
Topography of Gulf of Georgia, W T, Village Point to Base of Sandy Point, 1888
Scale: 1:10,000; Surveyor: J.J. Gilbert, Puget Sound River History Project, University of Washington: https://riverhistory.ess.washington.edu/
USGS TopoView, Topographic Map, Bellingham (WA), 1975, Scale: 1:100,000 https://ngmdb.usgs.gov/topoview/viewer/#4/40.00/-100.00