add as much as GCP points till you get accuracy.

Hi Balan,

Introduce more GCPs, wherever you find more mismatches.

The distribution of GCPs should cover all over the image.

Reduce your RMSE value as much as possible.

Then you can improve your quality with DGPS Survey.

One more suggestion,

Instead of using toposheet, try to use IRS P6 LISS III data or geocoded Landsat data which is available in GLCF website for free download to rectify your area.

You can select GCPs very clearly when you using image to image rectification method.

Thank you

Hi Ramesh,

Thanks for the suggestion.

But i want to know whether can we obtain high accuracy (RMSE < 0.5) using toposheets as references..

Oops, adding more GCPs could be a solution if only the projections are the same!

My suggestion is to check the map projection of the satellite image. Usually it is reprojected to UTM or other 'standard' projection with WGS84 geodetic datum. (If the satellite image is in its native format, it is another piece of cake - if it is indeed your case, let's return to it after the initial check). However, if it is in UTM, *the coordinates of your local control points from your toposheets should be given in UTM, too!* Using any other projection results an error in the range of the distortion between the two projections (the ones of the satimg and the toposheets), which is around the value you mentioned.

I hope this helps.

Liss 4 image is of resolution 5m and you have not mentioned the scale of toposheet which you are using as reference.,. you need to use 1:25000 scale toposheet for referencing then it will be more accurate.

Thank you for your suggestions,

The satellite image was in UTM projection and I reprojected it to Geographic(lat/long), which is our native one. Georeferencing was done after reprojection.

Bharath, thanks for pointing out the scale of the toposheet , I use a toposheet with a scale 1:50000 and there might be the problem.

Can the mismatch between toposheets also be a factor for high RMSE as i have used three toposheets simultaneously as reference images

Yes of course...

And what is the projection of your toposheets?

Moreover, if the satellite image is in UTM... then it is already geo-referred somehow. Perhaps it is only a preliminary georeference, with poor accuracy but it can be corrected easily. Pick just one point in the terrain (that is identified also in your satellite image) by GPS (location error is in range of your pixel size), read it in UTM and shift your satellite image in the UTM plane, using only this one point, keeping the image orientation. If the terrain is not accessible, you can try also Google Earth but keep in mind that its horizontal accuracy is not so good (errors up to 30 meters).

So your problem might be with geo-referring your toposheets.

The toposheets projection is Geo(lat/long)

take polynomial order 2 or 3,after that apply whole process, check projection should be same, and take maxi. point on intersection as two road join, and at temple, etc...

I hope you will got your required accuracy

Srijith,

Your 1:50,000 Topo sheet is most certainly NOT Geo(lat/long). Chances are, it is Lambert Conformal Conic projection (one of the "India Zones") developed by the Survey of India in the early 1900's. All Topo sheets are on some type of projection in order for them to be printed onto a flat sheet of paper! Difficulties in geo-referencing satellite imagery to old Topos can be a result of the old Topo sheet being of poor accuracy to begin with. Common in Africa, less so in parts of Asia, see my publication "Maps in the Middle of Nowhere" at www.ASPRS.org. Look under publications, Photogrammetric Engineering and Remote Sensing, Volume 78, Number 10, October 2012, Page 1015. It can be downloaded as a pdf file at no charge.

Thank you, I'll surely look into it..

Dear Srijith (and Clifford): the 1:50k topos are in Polyconic projection, not LCC, but more importantly, they are NOT in WGS84 spheroid-datum. They are in India-Nepal Datum with Everest 1956 spheroid (if they are post-1970). Given this, you have two options:

Option1: just ditch topos, and go to the field, get GPS readings for 10-20 well-spread-out control points that are clearly visible in your sat-image (road-road, road-river intersections, dam corners, etc), and (making sure the readings are in WGS84 setting) geo-rectify the image (after removing its original pseudo-coordinates). Note that you can never reduce RMSE much below 1 pixel (although you can get a pseudo RMSE that is lower), because that is the least error in your locating your GCP in your image.

Option 2: (if you cannot go to the field): georectify the topos properly. this requires taking into consideration the different datum and spheroid I mentioned above. I attach a file outlining the method (we have created this as an internal procedure file).

There is also a third option: if you cannot go to the field, you CAN use Google Earth as a source of GCPs: often, the accuracy is better than the topos, because the GE image is recent and so has the most recent road locations (which do shift around over the 30-40 year period since the topos were made). My experience with GE has been that it generally matches GPS readings down to 5-10m (which is the error in GPS anyway).

Just to add: I read and enjoyed Clifford's short piece on 'middle of nowhere' and basically agree that in addition to systematic issues listed above, there are also going to be basic inaccuracies in the way the topos were prepared that cannot really be dealt with. Here is another piece on the errors in Indian context.

It couldn't have come at a better time.

Thank you very much

hi srijith, as our senior's told is correct one, you can consider all those suggestions, and one more thing, i think due to plate techtonic theory, all the plates are moving on the earth surface, this could also one of the reason u try to consider and discuss with other geographer and Erosion and accretion if coastal area.