Dear Srishti,

This is a common problem in satellite remote sensing. Indeed, since your two data sets have been acquired a month apart, the retrieved temperature fields are inhomogeneous because, among other possible factors:

- Environmental effects (normal seasonal evolution of the temperature field):

* the state of the atmosphere has changed (atmospheric 'corrections' are never perfect)

* the vegetation has grown (or decayed)

- Observational effects:

* the Sun is located at different zenith and azimuth angles, thereby generating different heating rates, especially over sloping terrain

* the satellite may (or may not) be at a different position relative to the target and the thermal radiation field may be anisotropic

Please note that the environmental causes to temperature changes are real and significant changes: they should probably be taken into account in your analysis, instead of being masked or deleted by some statistical procedure. However, if the main goal is to provide a homogeneous pattern of temperature variability (around an arbitrary mean), then the problem is to adjust the images to remove or at least reduce the obvious spatial contrast present in the straight composite of the two raw images.

There are two aspects to this:

A1. What to do where you have a value in both images (i.e., where they overlap)?

A2. What to do where you have data from only one image?

One simple (not necessarily optimal, but practical) suggestion might be as follows. Let T1 and T2 be the land surface temperatures at each pixel in images 1 and 2:

Step 1. Compute the average of the whole temperature field of the first image:

Tav1 = sum(T1)/n(T1)

where n(T1) is the number of 'valid pixels' (i.e., outside of the cloudy areas, water bodies, etc., if any)

Step 2. Compute the average of the whole temperature field of the second image:

Tav2 = sum(T2)/n(T2)

where n(T2) is the number of 'valid pixels' (i.e., outside of the cloudy areas, water bodies, etc., if any)

Step 3. Compute the mean of those two values: Tmean = (Tav1 + Tav2)/2

Step 4. Create a new combined image as follows:

4.1 for those pixels where no data is available on either of the original images, specify an arbitrary missing value code, clearly different from any usual value (for instance -99.9)

4.2 for those pixels where only one observation is available,

- if that observation is from image 1, compute Tnew1 = T1 - Tav1 + Tmean

- if that observation is from image 2, compute Tnew2 = T2 - Tav2 + Tmean

4.3 for those pixels where observed values are available from both images, compute

Tnew = (T1 + T2)/2

This procedure should preserve the spatial patterns of each original image while adjusting the means of those images to a common value, and thus remove (or at least reduce) the inhomogeneity in the new composite image.

Again, remember that now you will see a nicer image, but the temperature distribution will not be representative of any particular situation in the environment: it may well be that your target never experienced this particular temperature pattern on any given day within that month! All that you can do with this product is discuss relative spatial changes, assuming that those did not vary between the two dates of acquisition. Be sure to explain in detail what you have done and refrain from drawing bold conclusions that may not be supported by that evidence.

Lastly, this procedure is not specific to the temperature field: it could be applied to any biogeophysical field of interest. However, if you are interested in very accurate computations, you should be prepared to take other effects into account, such as the anisotropy of the radiation field, or computing the average fields on the fourth power of the temperature fields (Stefan-Boltzman: https://en.wikipedia.org/wiki/Stefan–Boltzmann_law).

Good luck. Michel.

There is not much to add after the insightful comments by Prof. Michel. Just want to add, there are seamless mosaic tool options available in ENVI, ArcGIS, QGIS etc. I've attached one of the ENVI links here. Seamlines in mosaic may also occur 'when two scenes have different gain states, they have different dynamic ranges, and there will be a shift in Digital Number (DN) values from one scene to the next'. It is definitely a good idea to review the metadata for each of the two scenes used in the mosaic, especially because they are from two different time periods.

The question that I have in mind is, what is the reason for creating a seamless mosaic of two Landsat LST scenes from two different time periods other than for display or presentation purposes? I would be wary of making any quantitative analysis in and around the region where the two scenes are joined. What do you think Sir?

https://www.harrisgeospatial.com/docs/MosaicSeamless.html

https://landsat.usgs.gov/what-could-be-causing-seamline-mosaic-multiple-images

Temperature is an environmental variable. The simultaneous observation is very important. It is a really difficult problem in the area which cover more than one image.

First Mosaic and then by applying upper and lower limit of temperature bring in assets range to the image.

Question:

What could be causing the "seamline" in a mosaic of multiple images?

Answer:

In order to obtain a "seamless" mosaic, radiance (and preferably reflectance) may need to be calculated before performing the mosaic, because gain changes may occur in one or more of the scenes. When two scenes have different gain states, they have different dynamic ranges, and there will be a shift in Digital Number (DN) values from one scene to the next. This can be determined by reviewing the included metadata file for each scene.

Some imaging software packages have a utility that can automatically create the radiance image, based on the gain/bias in the header file. If this is not available, a manual "band math function" may be necessary.

Details about conversion to radiance and converting radiance to reflectance are provided in the Landsat Data Users Handbook.