# Change detection using DEMs of difference and error quantification

Im currently working on a project in one of my seminars at university ("Advanced methods in natural hazard research"). We are modelling DEMs with Agisoft Photoscan using pictures from flume experiments (fluvial geomorphology) and are trying to detect changes between two different time steps. Now i got stuck with the error correction of the DEM of difference (DoD):

"For the interpretation of the result, a very fundamental step is still missing: as our SfM approach has errors, we need to incorporate these into our change detection analysis. Luckily, PhotoScan has a function that creates a report on the model that we calculated, which also has information on the errors of our model (File  generate report). Save this report and extract the total error of your model. As both of the models that you calculated have errors, in a change detection analysis, we need to propagate these uncertainties. While there are many ways to achieve this, we will use the standard “Gaussian error propagation” for this. Propagating errors in a DoD analysis gives us the so called “level of detection” (LoD). Values in our DoD that are below the value of the LoD cannot be interpreted as significant changes, as we cannot differentiate them from our analytical error. The LoD using Gaussian error propagation is calculated using the formula (1). Within this analysis, we will also incorporate a degree of confidence, based on a student’s t-distribution:

𝐿𝑜𝐷= 𝑡√(𝛿𝑍𝑇1)^2+(𝛿𝑍𝑇2)^2, (1); (square root after t!) --> vertical errors are known but i dont understand what t is supposed to be.

Where t is the critical value of a given confidence interval in a student’s t-distribution, 𝛿𝑍𝑇1 is the vertical error of the model at time step 1, and 𝛿𝑍𝑇2 is the vertical error of the model at time step 2. For the report (see below), create a figure showing the difference model you calculated, including the assessment of the LoD. Make sure that your colour scheme shows positive and negative changes in different colours and also the LoD is represented (by transparency or grey colours). Also calculate the magnitude of positive and negative changes and display them somehow, in a histogram, table, or some other form."

I understand that i have to delete every elevation that is <= LoD from the DEM of difference and create a new, corrected DEM of difference. Am I right?

• If you are looking geomorphological change detection you should have a look at this ArcGIS tool - gcd.riverscapes.xyz – Hornbydd Jan 30 at 10:52
• Thank you for your reply, although its not the change detection im looking for as this is rather straight forward. Im looking for how to incorporate the errors of my two DEMs into the change detection analysis. – Deibel Jan 30 at 11:42
• Ok i see what you mean. Thank you very much for your tip! – Deibel Jan 30 at 11:44
• It doesnt work in my case – Deibel Jan 30 at 12:55