Hello sir, you're on the right track, but there are a few things to consider for a more accurate assessment. Evaluating the registration error between the datasets and using the roads as reference points is a reasonable approach, especially when no check points were collected. By comparing the vertical distances between points on the roads, you’re essentially evaluating the consistency of the georeferencing across your datasets.

However, the key challenge here is the assumption that the roads are perfectly stable and unaffected by erosion, which may not always hold true. Even small changes in road elevation over time could introduce errors into your analysis. Additionally, while an RMSE of 3 cm might suggest that your data is well-aligned, it doesn’t necessarily mean that you can confidently detect erosion smaller than this value. It’s important to consider the overall quality and resolution of your UAV imagery and the accuracy of your georeferencing process. Without check points, you might still have uncertainties in your absolute accuracy, so while your method is reasonable, be cautious in interpreting the results, especially if your goal is to detect small-scale erosion.

Yes, you can do this. Actually, its a broadly accepted approach for deformation monitoring using subpixel misregistration information and free data. Either you perform a moving window registration error assessment with some filtering or you utilize SIFT points and do the matching around. You can find a comparison in my paper list noted as 'Performance of correlation approaches...' If SNR is high, then subpixel misregistration assessment can be easily better than 1/1000 pixel.

Ben Pearse Yes, you can use registration error as an alternative approach to assess erosion monitoring in the absence of check points, but there are a few important considerations to keep in mind.

Imagine you are monitoring cracks in a tiled floor over time. If you have no exact measurements (check points) of each tile's position, you could compare how well the tiles align in two different surveys. If the edges of the tiles mostly match with small gaps (like 3 mm differences), you may assume no significant movement. However, if gaps larger than 6 mm appear, it may indicate noticeable shifts or damage.

Similarly, in your case, the roads at the top and bottom of the slope can serve as stable reference features (like the tiles). If you calculate the vertical differences between matched points on these roads across the two datasets and find a Root Mean Square Error (RMSE) of 3 cm, it means any changes smaller than 6 cm (double the RMSE for safety) might be within measurement noise, not actual erosion.

Your approach of evaluating registration error and assuming the roads as stable references is acceptable given the absence of check points. However, ensure that:

• The roads did not experience deformation.
• The RMSE calculation includes enough points to provide confidence in alignment accuracy.
• You treat any changes smaller than 6 cm with caution as they may be noise rather than true erosion.

This method, while not a perfect substitute for check points, is practical and reasonable for your erosion monitoring task.