Does images have same spatial and spectral accuracy ?

Hilda

What are you using to reproject them into geo (Lat/Long)? When you reproject them do they line up prior to running the mosaic command? Have you tried gdal tools or fwtools? You could could also reproject the individual images into another projection , such as albers. I often do this with two different UTM images, such as Landsat scenes in two differing zones (e.g. 15 & 16). Are the images of the same resolution?

Are these Landsat images? I have found that I have to georeference them for my local analyses even though they may already be projected. In this case, I would try to find a common projection suitable for the region (an Albers projection maybe?), transform the images to the common projection, and georeference them against a higher resolution image, such as aerial photography, or ground control points, then mosaic them. There may be better ways to deal with this, but this is how I would approach it.

IMAGES should be of same spatial resolution, then it would be easy to do change detection. regarding projection, u can convert utm to geo-latlong, but there will not be any science in doing that. just u will get the pixel values which will overlay together. but if u really want the accuracy in doing this, to get longtitude of central meridian and latitude of origin of projection values of the location and then reproject the images to these values, with defined spatial resolution.

Usually you just pick one zone and reproject one of the images into that zone. Pick the image where more of the pixels of interest are nearer the zone boundary. Any reprojection software (e.g. the MATLAB mapping toolbox) should let you choose the zone.

Define your spatial reference perhaps if you are using ArcGIS.

http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#//009t00000029000000

Hi,

That seems to be a common issue.In ArcGIS, it will project on the fly and automatically align the images if working as a mosaic dataset. However, the projection of the mosaiced image will then depend on whatever your mosaic dataset projection was defined. For example, if you mosaic it as UTM14 then the zone 15 data will get re-projected on the fly to zone 14. However, when doing this the scale factor will increase and accuracy will degrade, when trying to measure distances the farther away you go from the true zone 14 boundary into the zone 15 data (now also projected at zone 14).For change detection purposes this may not be such an issue if only concerned with classifying pixels, etc. The other option is to convert both the zone 14 and 15 data into another coordinate system, such as state plane, that better captures the area or perhaps work in geographic.

BEWARE: using two different images will create problems and any published findings will be questioned and criticized. Though the study area is not changing, the direction and the fundamental radiometric properties are changing between the two scenes (think of how far apart the sensor is when one scene is captured and where it is when the second is captured). If you were just creating a map- it would be fairly simple to average the overlaying pixels and move on. But it is almost impossible to quantify the difference in the two images that can be attributed to difference scenes as opposed to your real question of the change in the study area at different times. If you are working with some index like vegetation, the range is usually so small that poor data can create very inaccurate results.

I have always been advised never to use differing scenes. Try to find another way.

Not sure what tools you are using. I carried out a change detection and had to MOSAIC four scenes together. The catch is that you need to adjust both the Range and Azimuth window parameter setting in the co-registration panel (which I did using the ENVI software). I would like to know some other details specifically about the tools you are using. ENVI works perfectly well with ArcGIS as well.

You can use gdalwarp (http://www.gdal.org/gdalwarp.html) to do this relatively easily.

Establish a "base" UTM zone (say zone 36 in your case). Use gdalwarp to reproject the image in UTM35 zone to UTM36. Once the images are in the same zone, use gdalwarp to mosaic them together. Remember, this only makes sense if both images are from the same date (i.e. paths are the same.)

1. The satellite data must have same resolution (pixel).

2. What tools you are using? I suggest to use ERDAS IMAGINE 9.X.

3. After you open program, in menu bar select “Data Preparation” and then click

“Reproject Images”.

4. In “Reproject Images” you can convert projection on satellite data.

Depend on what kind of change detection algorithm you run. If it is a pixel-by-pixel, then certainly I would not do any geometric transformation. Rather do my analysis on images from two zones separately. Combine the results in the end if that is necessary for visualization/presentation. While this is a common problem, I am not aware of any suitable technique to solve your problem. Remember - when you project your data, you have to live with the distortion. Not only your pixel locations will be shifted, but also the actual values of the pixel, because the new pixel values have to be interpolated by one of the methods such as nearest neighbor, cubic etc. In the process you change the pixel values - which I think is detrimental to any change detection you want to perform.

Hence, my suggestion - perform change detection on original images without any mosaic. Do it in the end when you are done with the analysis.

There are two options either you can georeference on you own else you can change the projection using envi software. Here i am giving steps. Go to map open the image and select the desired projection and set it. I hope this will help u

Both the images should be in the same projections otherwise it will effect your analysis. change the projection of either image using ArcGIS, Erdas Imagin or any other software.

Nice answer Bidhya. One thing to note is that reprojecting the image data from one UTM zone into a different UTM zone for an image that already has projection information stored for it, such as a geotiff, will not impact the image pixel values. This is because it will simply convert the header coordinate information over to the new coordinate system origin and then each pixel x,y location is based on its row,column offset and pixel size distance from origin. This process is different than georectification where you would be applying a geometric transformation to each pixel location to warp it to match control points in some other coordinate system. The image would then need to be resampled and hence the pixel values change which could mess up your change detection scheme.

Thanks alot for your suggestion, i will try all .

I think Michael Starek gave you a good suggestion, so you may reproject one of the images to the projection of the other. this will not affect pixel values. before you perform the reprojection to have to calibrate the images because they were acquired in different dates, as deduced from their location in different UTM zones. if you are not using all of the images, you have to reproject the images that covers small area to the projection of the big one.

I suggest some basic thing agreeing with Teeraswoung , Preferbly he has given suitable option for mosacing two scenes of images having different zone . Using ERDAS/ Imagine you reproject the Images and then mosacied it.

Make use of GRASS GIS by r.proj. Project all images in the same UTM zone and enjoy!

http://grass.osgeo.org/

http://tinyurl.com/tree-four-freedoms

http://grass.osgeo.org/grass64/manuals/r.proj.html

convert both of your mosaic image to conventional terresterial system(CTS)

and then do your opperation and finally take your information to every coordinate system that u want

If you do want to warp one or both of the images to register, you may get more consistent results if you apply a small amount of low-pass filtering before warping to get rid of information near the Nyquist frequency, which would appear or not in the warped image depending on the phase. As long as the changes you are looking for do not involve the Nyquist frequency only, you should be OK. You definitely don't want to use those frequencies for registration.

I would say work each image separately and combine after by re projecting it to one zone. Try to assign zone value which covers more area.