Random Points: Just a Little Bit Tighter, now Baby…

A 341Kb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

I was recently engaged in a very interesting conversation with a company who is performing a number of survey related operations for mine site management. This conversation led me to think, once again, about the overall issues of accuracy. The end purpose of this month’s column is to implore the service and equipment providers to clearly communicate to clients the capabilities of systems in a vernacular that is understandable by the client.

A few years ago I discussed the parameters of accuracy in this column. Just a brief refresher:
Resolution–the "fineness" to which a device can measure (think marks on a ruler)
Precision–the repeatability of a series of the same measurement (the "spread" when making repeated measurements)
Local Accuracy–how closely point to point length measurements come to "truth" (e.g. measuring the height of a door)
Network Accuracy–how closely measurements of position match an external reference system (e.g. how closely a survey mark matches an independent RTK reading)

It is also critically important to appreciate the distinction between random ("stochastic") and systematic errors. Random errors (often, right or wrong, referred to as "noise") are errors that can only be modeled using statistical analysis and generally cannot be removed. A good example of a random error is the variability of range timing in a LIDAR system. Systematic errors (again, right or wrong, often termed "bias") are errors that are present system-wide and can, if properly modeled, be removed or compensated. A good example of a systematic error is the change in focal length of a metric camera caused by temperature or incorrectly setting antenna height in a base station computation. Note that systematic errors are not necessarily just simple offsets. In Figure 1 is illustrated a possible vertical bias in a LIDAR data set.

Users of data and data services are often very clear about their geospatial business needs but not so clear about what is needed, accuracy speaking, to achieve these needs. There is also a bit of misplaced thought that is occurring due to the high resolution of data now being collected.

One of the most common misunderstandings of clients is resolution. For example, very high resolution orthos and LAS point clouds are being produced by mobile laser scanners (MLS), helicopterborne LIDAR systems and dense image matching (DIM) from cameras on small unmanned aerial systems (sUAS). It is quite common to deliver an sUAS-derived ortho with a ground resolution of 2 cm. Most customers have a sort of intuitive expectation that this 2 cm resolution ortho will fit a survey coordinate system to within 2 cm (network accuracy) and that any measurements they make from the orthos will have local accuracies also on the order of 2 cm. Of course, both assumptions are very wrong. However, I have found it difficult to explain to clients that high resolution means nothing other than high resolution.

A sometimes more obscure topic to present to clients is the precision of the measurement. Again, with a stated vertical resolution of a centimeter or so, there is (again wrong) an expectation of relative or absolute vertical accuracies in this same range.

If you think about it, this was not such a big problem when we were collecting LIDAR data with spatial resolutions of 2 m and precision on the order of 12 cm or so. Our systems could typically collect with horizontal and vertical local accuracies within this same 2 m envelope and the instrument precision was about an order of magnitude better. If we tied in to ground control, we could also achieve network accuracies in this same ballpark.

Interestingly, the resolution capabilities of our measuring systems have improved much faster than the accuracies of the associated positioning equipment. For moving sensors these positioning systems are usually Real Time Kinematic (RTK) Global Navigation Satellite Systems (GNSS) for X, Y, Z and an Inertial Measurement Unit (IMU) for linear and angular accelerations. Of course in the case of structure from motion (SfM) as is commonly used in sUAS, the positioning is often inferred from ground control. We have been doing a lot of experiments within GeoCue Group to characterize the accuracies we can expect using these various techniques. Environmental parameters have a big impact so the characterization is not straightforward. More on this in future columns. I would say, however, that 4 cm vertical and 2 cm horizontal (Root Mean Square relative to independently measured control) is pushing the envelope.

One final big issue that I have seen in the sUAS services market is a misunderstanding of network accuracy. Sure, you can compute volumes using only scale. But the stockpile toes used in this computation cannot be accurately placed on a map and, even more limiting, cannot be used in time series studies (how does my site this month compare to last month?). Of course, any contours derived from the data set are of very limited and specialized use. I think anyone selling data that is not referenced to a network with decent accuracy metrics is doing their customer a big disservice; these data have little integration value.

I will explore these topics in more detail in a future column (or perhaps a dedicated article). In the meantime, however, we all need to be a lot more careful in addressing these issues with our clients. It is an education process that requires us to be teachers!

Lewis Graham is the President and CTO of GeoCue Corporation. GeoCue is North America’s largest supplier of LIDAR production and workflow tools and consulting services for airborne and mobile laser scanning.

A 341Kb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE