Home‎ > ‎Our Research‎ > ‎

Geomorphic Change Detection

Repeat surveys of streams can be immensely insightful in terms of understanding how channel form and associated habitat changes over the course of a single flood, or over annual or longer timescales. The River Feshie has been an instrumental site for the development and testing of tools used for geomorphic change detection (GCD). The procedure behind GCD is quite simple, and at its core involves a three-step process:
  1. Conduct a survey of a landscape and produce a digital representation of that landscape (called a digital elevation model, or DEM).
  2. Return to the site some time later, perhaps after a flood, or perhaps a year (or more) later; conduct another survey and produce a second DEM from the data.
  3. Subtract the old DEM from the new DEM to produce a map of changes that occurred on the landscape. This resultant dataset is termed a "DEM-of-Difference" or DoD.

In visual terms, the production of a DoD is simply the result of subtracting an older elevation surface from a newer one:

Given the multi-year survey record which exists for the River Feshie, and because of its dynamic nature as a braided stream, it provides an ideal test case for producing DoDs. Here we show a series of DoDs produced from surveys conducted annually between 2003 and 2007.


DoDs produced using DEMs from 2003 to 2007 on the River Feshie. The DEMs (top row) have been detrended (that is, they have had the down-valley slope component of elevation change removed) prior to DoD calculation. DoDs are shown in the bottom row, with areas undergoing erosion in red and those undergoing deposition in blue. Figure is from Wheaton et al. (2010), copyright Wiley and Sons.

While DoDs alone can be useful for examining channel changes over a variety of timescales, we must also consider the reliability of the data from which they are produced. That is, how much of the change that we see in the resulting DoD is the result of actual geomorphic shifts in the elevation of the landscape, versus how much of the change is due to inherent error or uncertainty (termed 'noise') in the surveys from which the DEMs (and eventually DoDs) are computed?

In a 2010 paper, members of our research team (led by Joe Wheaton) examined the influence of survey noise on DoDs. The figure below shows that depending on the reliability of the data from which DEMs are built, the amount of elevation change that we are able to detect in resultant DoDs can vary tremendously. In each DoD below, the 'minimum level of detection (minLoD)' - that is, the smallest elevation differences that can our instruments can reliably survey - is varied hypothetically, and the ultimate result on the DoD is shown. Note that with a higher minLoD (that is, hypothetically less reliable vertical resolution of DEMs), the volume of change that we see  in the corresponding DoD is altered.


The significance of minLOD on resultant DoDs and associated volumetric change estimates. DoD maps are shown on top and elevation change distributions (which are histograms of volume of change assigned to a given elevation change bin) are shown on bottom. The gross unthresholded DoD is shown on the far left, and moving toward the right progressively more conservative DoDs (i.e. higher minLoD) are shown. Figure is from Wheaton et al. (2010), copyright Wiley and Sons.

Knowing that the changes we can resolve in our DoDs are ultimately a function of the reliability of our topographic surveys from which DEMs are built, how can we incorporate survey uncertainty into the calculation of a DoD? There are several measurable quantities we can use to do this. In Wheaton et al. (2010), a combination of point density (higher density leads to a more accurate DEM), surface slope (steeper sloping areas result in higher uncertainty as they are more difficult to capture in surveys, leading to lower-quality DEMs), and GPS quality (lower point quality, as recorded by the instrument, leads to lower DEM quality) were used as inputs to a fuzzy inference system, and the combination of these three values yielded an estimate of DEM reliability at each point on these elevation surfaces. An example of the inputs to this fuzzy inference system are shown below.


Illustration of fuzzy inference system (FIS) construction and resulting probability map for 2006-2005 DoD. Inputs to the FIS include (a) survey point density, (b) surface slope, and (c) GPS point quality. Resulting map of elevation uncertainty for use in DoD calculation is shown at far right. Figure is from Wheaton et al. (2010), copyright Wiley and Sons.

When applied to the DEMs from the River Feshie, the use of the fuzzy inference system allows for removal of inherent survey noise by using measurable quantities derived from survey data. The result is a 'cleaner' DoD that in effect allows for much greater confidence that the geomorphic change we see is 'real' and not the result of survey errors.


Comparison of thresholded DoDs (at 95% confidence that changes seen are real and not due to survey noise) based on applying fuzzy inference system (FIS) are shown on top. Bottom row of DoDs also includes the application of a spatial coherence filter; that is, changes are more likely to be real (and not an artifact of survey noise) if they are surrounded, or spatially coherent, with other geomorphic change in a similar direction. Figure is from Wheaton et al. (2010), copyright Wiley and Sons.

The principles of geomorphic change detection developed on the Feshie have been instrumental in the development of software designed to allow users to produce DoDs using their own survey data. The Geomorphic Change Detection software is freely available here. You can find more information about the theory behind GCD in:

  • Wheaton JM, Brasington J, Darby SE, Sear DA. 2010. Accounting for uncertainty in DEMs from repeate topographic surveys: improved sediment budgets. Earth Surface Processes and Landforms 35: 136-156. DOI: 10.1002/esp.1886
  • Wheaton JM, Brasington J, Darby SE, Merz JE, Pasternack GB, Sear DA and Vericat D. 2010. Linking Geomorphic Changes to Salmonid Habitat at a Scale Relevant to Fish. River Research and Applications 26: 469-486. DOI: 10.1002/rra.1305.
  • Wheaton JM. 2008. Uncertainty in Morphological Sediment Budgeting of Rivers. Unpublished PhD Thesis, University of Southampton, Southampton, 412 pp.