In this month’s edition of the Geospatial Frequently Asked Question (G-FAQ), we turn our attention to the question of horizontal accuracy in high-resolution imagery products. While it seems like a topic that should be relatively straightforward, surprisingly it is wrought with complexities. Broadly speaking, orthorectification is the process of improving the horizontal accuracy of imagery; and as such, it will be the focus of this two-part series.
In this two-part G-FAQ series, we focus on this set of core questions:
What is the difference between georeferenced and orthorectified imagery? How is orthorectification completed and its accuracy tested? What are RPC files and how are they used in orthorectification?
Before we jump headlong into this two part series on orthorectification, let me frame the discussion a bit. First, my focus here is on high-resolution satellite imagery as that is where my knowledge base lies. While many of the principles discussed here would apply to aerial imagery as well, there are certainly topics relevant to aerial data and orthorectification that will not be covered. Second, my focus is on high resolution data and not on medium resolution data. This is done consciously as many medium resolution imagery products are simply too coarse to see features distinctly; and if you cannot see features (such as a sidewalk), then assessing the accuracy of an orthorectified dataset is impossible. And finally, the intent of this G-FAQ series is to address the basics of orthorectification so that our readers have an understanding of how an orthoimage is created and how its accuracy is assessed. I will not describe the mathematics behind orthorectification as admittedly this is well beyond my knowledge base.
Sources of Inaccuracies in Imagery
Many of our clients assume that high-resolution satellite imagery will be “accurate” to begin with, but this is not necessarily the case. To explore this topic, first we need to understand what the word, “accurate,” really means. In some cases, people use this term to refer to the alignment of the imagery they are purchasing with the spatial datasets they have in house. When this is the case, there is no way to assure accurate alignment without spatially registering the multiple layers to each other. Not only can the high resolution imagery we provide have accuracy errors, so too can the base layers a client has in house which can further accentuate the misalignment.
In other cases, the term accuracy can refer to how closely imagery aligns with its real position on the planet. While more modern satellites (such as GeoEye-1 and WorldView-2) have imagery that is closer to its actual position on the planet, the alignment is rarely, if ever, ‘perfect’ without additional processing of the data. These positional inaccuracies are caused by both internal and external factors. Internal factors are those caused by the imaging system itself, and include platform orientation and movement, optical and electrical aberrations and geometric characteristics. External factors will further compound internal ones and include the topography of the area imaged as well as atmospheric refraction. When taken together, internal and external factors can result in small or large positional inaccuracies; and further, accuracy will not be consistent from one image to the next even if it is collected by the same sensor, such as WorldView-2.
Georeferenced Versus Orthorectified Imagery
Now that we understand a bit about the sources of inaccurate data, let’s spend some time on three terms often used to describe satellite imagery: georeferenced, co-registered and orthorectified. Georeferencing is the process of projecting and thereby assigning coordinates (e.g. latitude and longitude) to an image. A georeferenced image has no accuracy guarantee; and while often times there is an accuracy statement for this type of data, it is a global average and not an assurance. In order to assure the horizontal accuracy of an image, it needs to be orthorectified.
Orthorectification is the process of removing internal and external distortions to assign more accurate coordinates to the final image. The goal of orthorectification is to create a final product whereby every pixel in the image is depicted as if it were viewed at nadir (or directly overhead), thereby removing the affects of hills, valleys, etc. on the data. Once data has been orthorectified, you are able to compute distances, areas and directions more accurately from the imagery. As a side note, it is never possible to remove all of the distortion in an image as after all, we display imagery on a flat surface (e.g. a computer monitor or paper) while the world is obviously curved; or put another way, no orthorectified image will be 100% true to its real world position.
Co-registration is the process of matching one (or more) spatial layer(s) to another so they align as closely as possible. It is possible to co-register two or more image files (i.e. raster to raster) to each other, multiple images to multiple vectors (i.e. raster to vector) and even multiple vectors to vectors (i.e. vector to vector). Unless one of the spatial layers used in co-registration has a known accuracy and its position does not shift, there is no way to assure the accuracy of the final product(s) created by this process.
How to Create an Orthorectified Image
Generally speaking, an orthorectified image starts georeferenced. To orthorectify an image, you need at minimum these two items:
- An elevation model – this is a spatial layer that shows the topography of the ground under the imagery you plan to orthorectify. The preferred format of the elevation layer is a digital terrain model (DTM) whereby human-made features as well as vegetation and other surface features have been removed; accordingly, this is sometimes called a bare earth model.
- A camera model or rational polynomial coefficients (RPCs) – we will describe both of these items in more detail shortly, but simply stated, the camera model or RPCs describe the positional relationship of the imagery to the ground below when it was collected by the sensor.
In order to have a guaranteed accuracy for orthorectified imagery, ground control points (GCPs) are required. While we will discuss ground control in more detail in upcoming G-FAQs, for now let’s define them as known points on the surface of the planet that can be found in the imagery you plan to orthorectify. When an elevation model, a camera model/RPCs and ground control are taken together, orthorectified data can be extremely accurate – the higher the quality of the DEM and/or GCPs you use, the more accurate the final product can be.
For our visual-learning readers, orthorectification can be pictured in this fashion. High-resolution imagery is akin to a balloon that is stretched over the topography of the planet (i.e. the elevation model) and is tied down at multiple locations with known coordinates (i.e. GCPs). Once this stretching and tying-down process is completed, the balloon is now positioned accurately on the surface of our planet, or it has been orthorectified.
From the graphic and description above, it might be apparent that removing the influence of topography on imagery is one of the most important steps of orthorectification. The effect of topography is to tilt features in an image away from the center of the camera, and this is counter-acted by employing an elevation model in the orthorectification process. Accordingly, orthorectification has a more drastic impact on accuracy in areas of high relief. As a general rule of thumb, when orthorectifying high-resolution imagery, you should use a 10-meter elevation model or better; and in areas with extremely high relief, you might consider a higher resolution elevation model than this. Obviously, this is a general rule of thumb as it can be nearly impossible to find free 10-meter elevation models outside the USA; and creating them can be very costly and take weeks to months to complete.
As mentioned at the outset of this article, the mathematical models used to orthorectify imagery are beyond the scope of this G-FAQ. That said, it is useful to describe two of the more common methods employed in the orthorectification process from a high-level. The most common method is the black-box (or the analytical) model. This method does not look at the specifics of a sensor and/or the collection geometry of the imagery, rather it relies on RPC files (which we will address in more detail in the second part of this G-FAQ) for orthorectification. The advantage of the black-box method is that it is easy to implement as software developers do not have to gather proprietary information (i.e. the camera model) for each sensor.
The second methods for orthorectification are physical-based models which take into account a wide variety of factors influencing the acquisition of imagery. One of the most popular physical-based methods is referred to as the rigorous model or camera model. The rigorous or camera model is based on proprietary information about each sensor that can only be obtained from its owner; as such, many commercial software packages do not include this method. This model takes into account multiple factors not considered in the black-box model including the exact position of the satellite in space when the image was acquired, the sensor’s electronic and optical characteristics as well as atmospheric effects. The advantage of the rigorous model is improved accuracy of the orthorectified image, though my review of the available research suggests the impact is minimal. For example, Dial and Grodecki found less than a 0.1 pixel residual error difference between IKONOS imagery orthorectified by the black-box and rigorous models.
In the second part of this G-FAQ series, we will continue the discussion on orthorectification with a focus on accuracy testing and RPC files.
Do you have an idea for a future G-FAQ? If so, let me know by email at email@example.com.
Find Out More About This Topic Here
- Center of Geotechnologies, Riccardo Slvini – QuickBird Stereophotogrammetry for Geological Mapping
- DIIAR, Maria Antonia Brovelli – Accuracy Assessment of High Resolution Satellite Imagery
- Michigan Tech University – Photogrammetric Control Surveying
- Penn State University – Orthorectification
- Rutgers University – Geometric Correction of Imagery
- University of Oregon – Geometric Rectification
Brock Adam McCarty