3. Types of DoD Uncertainty Analysis and Thresholding


Old Pathways


The above is an example of a Pathway 1 DoD Analysis with no uncertainty analysis in DoD 3.0. This is analgous to the Basic DoD Analysis in GCD 4.0.

Given the different ways of estimating errors in DEMs (e.g. spatially uniform, fuzzy inference systems, user-specified spatially variable, etc.), there will be lots of possible combinations for propagating those errors into the DoD. In addition there are at least three possibilities for what one does with the propagated errors into the DoD:
  1. Simple minimum level of detection thresholding
  2. Probabilistic thresholding (e.g. Lane et al. 2003; pp. 252-254)
  3. Weighted probabilistic thresholding (e.g. Lane et al. 2003; pp. 252-254)
If you use probabilistic thresholds, you can also use Bayes Theorem to update the probability that change is real with additional information by the following steps:
  1. Treat the probabilities estimated from the propagated errors as an a prior probablity
  2. Specify a conditional probability with additional information you may have (e.g. spatial coherence filter)
  3. Calculate the posterior probability with Bayes Theorem and use this to threshold the DoD
In the previous version of the software (DoD 3.0) a wizard guided the user through six potential pathways that represented different combinations of techniques for estimating errors, updating probabilities and thresholding DoDs (See flowchart at right). In GCD 4.0 more combinations of these techniques are allowed and we anticipate that in GCD 5.0 even more combinations will be allowed. The choice of which 'pathway' is best can not be prescribed and is something that a user needs to exercise judgment with based on their data and motivations for doing change detection in the first place. 

Further Reading

  • Chapter 4 of Wheaton JM. 2008. Uncertainty in Morphological Sediment Budgeting of Rivers. Unpublished PhD Thesis, University of Southampton, Southampton, 412 pp.
  • Lane SN, Westaway RM and Hicks DM. 2003. Estimation of erosion and deposition volumes in a large, gravel-bed, braided river using synoptic remote sensing. Earth Surface Processes and Landforms. 28(3): 249-271. DOI: 10.1002/esp.483.



Comments