What makes a high resolution imagery?
When we access geospatial imagery (satellite, aerial, drone, street), we want to know what is visible in it and what is the reliability of the inferences drawn from it. Geospatial imagery has two key parameters that define its utility – resolution and positional accuracy. Resolution is the ability to see small objects. Positional accuracy is the ability to identify the actual location of an object from its location in the imagery. Of course, there are other parameters like spectral information, bit depth and level of processing but for the purpose of this post, we are going to focus on resolution and positional accuracy.
In case you want to understand geospatial imagery sources and how they are used in-depth, I recommend that you read our post ‘The Right Geospatial Imagery for your Project‘.
Understanding Image Resolution
Resolution is measured using GSD – Ground Spatial Distance – that is, the area on the ground measured by one pixel. A camera with a resolution of GSD 10 m (found on Sentinel 2 satellites) can take one reading for a block of 10m x 10m. This means the camera will average out bits of information smaller than 10m x 10m to capture a consolidated value. When you look at Sentinel 2 satellite image, you can not see objects smaller than 10m x 10m.
The following illustrations explain it. The objects that can be seen with the finer resolution are smaller.
Choosing the right image resolution
Resolution is tricky because you can see a plane at 3 m but you can’t identify it as an F-16 Falcon until you go down to 30 cm. At 30 cm, you can see shrubs in a landscape but you can’t extract usable measurements like the surface area until you go down to 7.5 cm. So resolution needs to be selected based on the insight that needs to be extracted from the imagery.
If you want to count new neighborhoods that have sprung up in the state of Arizona, Landsat 8 imagery is what you need. If you want to extract maps to build navigation applications, WorldView or Pleiades satellites are the best. If you want to measure outlines of landscape features, aerial imagery at 7.5 cm is provides the level of detail required.
A good source to refer to for further understanding is the National Image Interpretability Rating Scales developed by the Federation of American Scientists that gives a detailed description of how visibility increases at finer resolutions with a focus on military intelligence applications.
Let’s nerd out on positional accuracy
To get the positional accuracy of imagery, either of these two statistical parameters is used – RMSE (root mean squared error) and CE90 (Circular Error at 90% confidence). Both start with measuring the error distance between the position of an object on the ground and in the image is measured.
To calculate RMSE, the errors are squared, then their mean is taken and then the root of the mean is taken. It is a way of estimating error without considering the direction in which error is occurring. RMSE gives a mean error and so it is incomplete without the information on how much the error can vary.
To calculate CE90, the error values are plotted as shown above. The term 10 m CE90 means an identified point in an image will fall within a circle having a radius of 10 m, 90% of the time when compared to the true position on the Earth. CE90 is better than RMSE since it gives an idea of variance and directionality of error.
Choosing the right image for positional accuracy
The precision of maps needs to match the precision needed by the applications they serve. Architectural, constructional and landscape maintenance applications require a high degree of positional accuracy in site maps, better than 5 cm RMSE in most cases. This is because the construction of structures like buildings requires high-precision engineering. Navigation maps can work with a positional accuracy of 5 m CE90 since that is the best possible accuracy the GPS sensors in smartphones can achieve.
But maps are only as precise as the imagery used to derive them. Different image sources have different positional accuracy.
1. RapidEye satellites have a resolution of 5m and positional accuracy of 10 m RMSE throughout the globe.(1)https://resa.blackbridge.com/files/RapidEye_Image_Positional_Accuracy_Whitepaper_V1.0_ENG.pdf
2. Maxar Technologies satellite WorldView-2 has a resolution of 46 cm and positional accuracy of 5m CE90 and RMSE 3.3 m throughout the globe. (2) https://dg-cms-uploads-production.s3.amazonaws.com/uploads/document/file/38/DG_ACCURACY_WP_V3.pdf
3. Nearmap has two imagery products – 5.8 cm that has a positional accuracy of 25 cm and 7.5 cm that has a positional accuracy of 62 cm. (3) https://docs.nearmap.com/display/ND/Accuracy
4. Out-of-the-box drones like DJI Phantom 4 typically deliver positional accuracy of 1-meter. (4) https://blog.dronedeploy.com/accuracy-in-drone-mapping-what-you-need-to-know-10322d8512bb
Google Earth, the favorite free-to-use source for viewing imagery data has varying positional accuracy for different geographies. North America seems to do better with a study(5)https://www.tandfonline.com/doi/abs/10.3846/20296991.2017.1330767 stating that positional accuracy varies between 0.1m to 2.7m for Montreal, Canada. Another study(6)https://www.mdpi.com/1424-8220/8/12/7973/pdf done over the Gaza strip in West Asia shows positional accuracy varying between 0.4 m to 171.6 m.
Can I improve the positional accuracy of imagery?
Geospatial imagery has native positional accuracy that can be further improved it needed. Ground control points can be used to refactor imagery and improve the positional accuracy of the information available in it. Understand what ground control points are and how they are used will require another post in itself.(7)https://www.groundcontrolpoints.com/mapping-contour-lines-using-gcps
The maximum spatial accuracy that can be achieved is constrained by the resolution of the imagery since ground control points should be visible in the imagery.
The truth that you seek
Imagery is chosen depending on the size of the smallest object that you would like to see and the accuracy to which it needs to be localized in the real world. Choosing a high resolution imagery isn’t always the answer as such imagery also involve high costs. The resolution of the imagery is in fact defined by the application and processing that follows.
We, at Attentive AI, partner with our customers to select the right imagery and help them to navigate from imagery to digital map containing invaluable insights by extracting the right features from the imagery. We have been recognized by geospatial players and leaders and recently Feedspot decorated us within the top 35 blogs on GIS in 2020. Thus, if you want to map the world without leaving the room, reach out to us and let’s map it together.
References [ + ]