1

I have been reading about orthorectification and did some training on it but my question still remain unclear is how much RMSE should I aim to ?

I will be working on a lot of aerial photos that dated back from 1930's to 1950's. I will use ERDAS' Autosync to orthorectify aerial photos.

The training I had said to keep the RMSE lower as possible but it did not explain why or for any reason to keep them lower. It didn't say which imagery to use. It didnt tell why which referenced image had a better RMSE to work with.

8
  • ERDAS field guide is good reading hexagon-solutions.com.cn/libraries/misc_docs/….
    – user30184
    Commented Dec 4, 2017 at 17:18
  • @user30184 - Thanks but this ebook is just telling you how to use the software and what the toolbox are use for.
    – PROBERT
    Commented Dec 4, 2017 at 17:29
  • I guess that you did not read it.
    – user30184
    Commented Dec 4, 2017 at 17:35
  • Yes, I mostly read the ones that from the report from National Park that did the orthorecifiction and some graduate student thesis paper.
    – PROBERT
    Commented Dec 4, 2017 at 17:38
  • Didn't they tell what RMSE means in practice?
    – user30184
    Commented Dec 4, 2017 at 17:44

1 Answer 1

1

RMSE represents the square root of the average of the set of squared differences between elevation model coordinates and ground truth - that is, the control elevation model or raster to which you are orthorectifying your older aerial photos. What this means, essentially, is that the RMSE represents an averaged difference from your rectified imagery to where the points in the images actually are in the real world. RMSE is therefore a measure of accuracy.

There are a number of factors you want to consider in order to minimize inaccuracy in orthorectification - using a large number of Ground Control Points, evenly distributing those GCPs through each image, and ensuring that those points fall on fixed photo-identifiable features that are not elevated above ground too much are all important. This last point, selection of GCP locations, is especially important with vintage images, as many features may not exist yet or might have been eroded or destroyed.

What you want your RMSE to be limited to depends entirely on the applications - for example, mapping to certain scales should require a horizontal/radial RMSE of under a certain distance. Specific values can be found through the National Standard for Spatial Data Accuracy here.

As RMSE is a measure of accuracy, you want to remember that it is not an absolute measure of accuracy. The overall accuracy of the orthorectified imagery will depend on the variation in differences between the individual GCPs, as well as their distribution. If the GCPs show a good distribution and the error skews under a value of about + or - 0.5, you should be able to state with confidence that your RMSE represents 67% confidence, meaning that for the majority of the points sampled anywhere in the orthorectified imagery, the majority are closer to Ground Truth than the RMSE value.

Read the ERDAS doc indicated by @user30184, follow the instruction for a good distribution of GCPs and proper procedures, and pay attention to what you are intending to use your imagery for. If it is a mapping purpose that requires significant accuracy, you will want to bear in mind that any point in your imagery will likely be off by about the value indicated (in reference units) by your RMSE, on average.

6
  • One question I just discovered is that the aerial photo's map unit is in centimeters ? Would that made any difference ?
    – PROBERT
    Commented Dec 4, 2017 at 22:11
  • You should be able to specify projection as a part of the orthorectification process - typically it should be set to match the reference raster. But no, it won't affect the results, you just need to bear in mind that RMSE is measured in those same units - so if your cumulative RMS is 1999, for example, your mean RMSE is 1999 cm. If your units were meters, the same process would have RMSE of 1.999, assuming no change in your reference image or GCPs.
    – AlecZ
    Commented Dec 4, 2017 at 22:38
  • What I use the projection to referenced image as in NAIP 2015 it is in NAD 1983 UTM Zone 13N
    – PROBERT
    Commented Dec 4, 2017 at 22:51
  • One more thing came up my mind is that NAIP i have is in MrSID file .
    – PROBERT
    Commented Dec 4, 2017 at 23:10
  • Units will be in meters then for NAD83 UTM. MrSID format will work fine with the ERDAS toolset, but again check the instructional page.
    – AlecZ
    Commented Dec 4, 2017 at 23:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.