Dear Majid

It is an interesting investigation when you able to do simultaneous observation of satellite and in-situ measurement.

Here is my opinion regarding your findings;

1. The spectral region that able to penetrate the water mostly ranging from 450 to 560nm (approximately). This means that most probably, the NIR ability to penetrate the water was weak and really depends on the turbidity of the water.

Most satellite covered IR spectrum did not able to penetrate the water.

2. In the case of yours, there are SSC in the water means the penetration might be weaker.

3. If i were you, i would focus on the relationship found in the red spectrum because high SSC concentration would lead to high reflectance due to scattering from the particle size and grain size. The blue and green spectrum can be used as reference to determine the turbidity level of your study site.

4. The degree of relationship between SSC and red spectrum (~520-560nm) primarily varies within site to site due to numerous factors. Among those that are significance were types of water (sea, ocean, coastal, river), and types of sediments or particle. Most of the times, in many research the composition of multi-sediments within types of water was not clearly defined.

Do not hesitate to e-mail me if there are any further inquiries.

Good luck

I think in your case it would be better to follow through this method:

First of all unify your numbers (result) I mean to use DN digital number or reflectence or radiance,

Second try to find the correlation beteen each single band and SSC, and compare the result,

Third, Try to compare the result of NIR of Landsat and The NIR of Chinese HJ-1 A/B,

See the result and comeback to me later I will tell you what to do,

Regards,

Falah,

Dear Dr. Fakhri,

Thanks for your contribution and suggestions, I have followed your suggestions and results are attached.

I have unified the numbers in terms of reflectance.

The correlation between the reflectance of Landsat bands and SSC is as following 

Band1 = 0.33, Band2 = 0.53, Band3 = 0.59 and Band4 = 0.21

The correlation between the reflectance of Chinese HJ-1 A/B bands and SSC is as following

Band1 = 0.25, Band2 = 0.45, Band3 = 0.69 and Band4 = 0.10

Note: it should be remembered that the data for Landsat was used from years 200 to 2012, while for Chinese HJ-1 A/B was for years 2008 to 2012.

When combined both datasets i.e. Landsat and Chinese HJ-1 A/B, the correlations were as following;

Band1 = 0.28, Band2 = 0.45, Band3 = 0.60 and Band4 = 0.16

Thanks

Ok, I think keep the band 3 for both because it has higher correlation than the other,

(Is the band three the red one for the both satellites?)

Leave the combination for now to the later step,

Now, Try to compare for the same (band 3) for the convergent date as you have mentioned the 2012, and see the result,

Comeback later to tell you what to do,

Regards,

Falah,

Majid ... do also check your observations of SSC for sufficient variability. Do your observations have a large variation or not? Is the range in SSC is as large as that of others reporting high correlation? A larger range in SSC would (most probably) result in large correlation as it would be better resolved by remote sensing.

Dear Dr. Sehgal,

I think your observation is quite good. Yes, the SSC range in my study area is very variable. Mainly my study area (Hong Kong) has three main types of waters, the eastern side is more clear, western side is affected by the river flow from Pearl River while the central waters show a mixed type.

For my study I have combined the data of SSC from all water types.

Dear Dr. Fakhri,

Yes, the Band 3 is red spectral band for both satellite. 

Here are the results for the convergent dates (dates covering same time period for both satellites)

For Landsat

Band1 = 0.27, Band2 = 0.64, Band3 = 0.86 and Band4 = 0.15

For Chinese HJ-1 A/B

Band1 = 0.25, Band2 = 0.45, Band3 = 0.62 and Band4 = 0.10

Now, we could go back again to your question:

"Can someone explain why there is low correlation between NIR band and SSC for my study area?

Can this be attributed to the specific grain size and particle size distribution of the suspended solids found in this region?"

Now the higher correlation than the other is band 3 right Ok,

all other bands within the NIR band have approximately close result of low correlation, this could be or we can say might be attributed to the SSC properties as well as to the characteristics of the bands.

To be sure from this result follow through the following:

For band combination what to do now:

Select the bands with highest correlation and make the combination of each satellite separately and check the result.

Comeback to me to tell you what to do later,

Regards,

Falah,

Dear Majid,

I think your results are pretty well, but I can give you some clues that perhaps help you to find a some theoretical justification of your problem. First of all, a spectral reflectance is a function of inherent optical properties (IOPs), like absorption, scattering, backscattering coefficients, asymmetry parameter, etc. Also, this parameter depends in some degree on atmospheric conditions (i.e., wind speed and atmospheric optical depth), solar and viewing zenith and azimuthal angles. In turn, IOPs depend on particles sizes, their forms, structure and orientation, as well as on complex refractive index of particles and wavelength. Intuitively it should be clear that such a variety of parameters yields a wide variety of the spectral reflectance dependencies on the SSC. And a literature analysis shows that a biggest correlation between SSC & reflectance not necessary happen at the NIR. Look, for instance, Fig. 1 in "Inversion of reflectance spectra of nonchlorophyllous turbid coastal waters" by Forget et al. (1999) or Fig. 1 in "Colour classification of coastal waters of the Ebro river plume from spectral reflectances" by Lahet et al. (2001) or Fig. 2 in "A reflectance band ratio used to estimate suspended matter concentrations in sediment-dominated coastal waters" by Doxaran et al. (2002) - in all these figures the best reflectance resolution (and, hence, the best correlation) is achieved in the red spectral range (~650-700 nm), while for the larger wavelengths this resolution is deteriorates. In the other publications [i.e., Fig. 6 in "A reflectance band ratio used to estimate suspended matter concentrations in sediment-dominated coastal waters" by Doxaran et al. (2002) or Fig. 6 in "Spectral signature characters based on measurement in situ of high turbid water in Yangtze River estuary" by Shen et al. (2006)] a NIR range gives good resolution.

Sincerely,

Leonid

Dear Prof. Leonid,

Many thanks for sharing your views about this problem, I found this explaination very useful.

Thanks

You are welcome!

Dear Majid

It is an interesting investigation when you able to do simultaneous observation of satellite and in-situ measurement.

Here is my opinion regarding your findings;

1. The spectral region that able to penetrate the water mostly ranging from 450 to 560nm (approximately). This means that most probably, the NIR ability to penetrate the water was weak and really depends on the turbidity of the water.

Most satellite covered IR spectrum did not able to penetrate the water.

2. In the case of yours, there are SSC in the water means the penetration might be weaker.

3. If i were you, i would focus on the relationship found in the red spectrum because high SSC concentration would lead to high reflectance due to scattering from the particle size and grain size. The blue and green spectrum can be used as reference to determine the turbidity level of your study site.

4. The degree of relationship between SSC and red spectrum (~520-560nm) primarily varies within site to site due to numerous factors. Among those that are significance were types of water (sea, ocean, coastal, river), and types of sediments or particle. Most of the times, in many research the composition of multi-sediments within types of water was not clearly defined.

Do not hesitate to e-mail me if there are any further inquiries.

Good luck

Dear Rizaludin,

Thank you so much for such a good explanation its very helpful for me to interpret the results. 

Dear Rizaludin,

I agree with your general approach and like it. But I would like also to do several additional remarks for the best understanding a situation. As I understand Mr. Majid Nazeer, his task is to develop remote-sensing algorithm for SSC estimation. It is obvious that this task may be solved for such spectral ranges were a clear (significant) dependence of reflectance (R or Rrs) on SSC is exist. If we use a spectral ratio, then again one wavelength should yield significant dependence R on SSC, while the second wavelength should be selected so that yield minimally dependence R on SSC. In turn, reflectance is primarily a combination of two IOPs: absorption and back scattering. Absorption has spectral minimum dependent on water quality; for very clear waters  a minimum achieved at blue range, for very turbid waters a minimum shifted to the red or even NIR range. Backscattering, however, clear decreases from blue to NIR for any type of natural waters. Note, however, that variations of backscattering with wavelength generally much weaker than the same variations for absorption. Thus, the last property is probably the main contributor to the reflectance. Next, taking into account that in the first approximation, R or Rrs are proportional to the b_b/(a+b_b), where a and b_b  are the total absorption and backscattering coefficients, respectively, we get that the spectral reflectance has maximum depending on turbidity: in the blue for very clear oceanic type waters and in the red or even NIR in the extremely turbid waters. However, spectral R or Rrs dependences are more complicated than for a and b_b, because the first are complicated functions on IOPs, solar and observation geometries, sky conditions etc. This is a reason of a large discrepancies in dependencies of R or Rrs on wavelength and SSC in different publications. However, the common between them is that the spectral range, where a significant dependence of reflectance on SSC exist is the green-red or green-NIR. For the Majid (Hong Kong) case is probably the red range. It is also close to my current waters (east coast of China), for which an optimal method was  developed as a spectral ratio of red (645.5 nm) to green (553.6 nm). I am attaching the last our publication on this issue. 

Dear Majid,

Water tends to absorb strongly in the NIR region, even if it has solids in suspension, so may be that is why you find a low correlation between NIR and SSC.  Depending on the type of solids (soil sediments, algae), it is more likely to find correlations in the red and green regions.

Hello Everyone,

I have used remote sensing to investigate ways to detect and monitor suspended sediment information at several locations including San Francisco Bay (using Landsat TM images), coastal regions in Hawaii (using Landsat TM, aerial imaging, and field based digital camera systems), and the Colorado River at the bottom of the Grand Canyon (mounted a spectral radiometer with Landsat TM bands on the shore line and collected spectral measurements every 15 minutes for over a year).  The correlation between the remotely sensed spectral information were correlated with field samples that were collected at various times during the life of the projects.

A couple of things that are important to keep in mind when investigating the use of spectral data to help monitor water surfaces are:

1. Sun glint and wave action can (and will) influence the spectral measurements that are collected for some of the pixels/areas being monitored.  Depending on the sun elevation and view angle this can be a minor or major issue to deal with.  The higher the spatial resolution the higher the influence will be on an individual pixel.

2. The correlation between the spectral measurements (and various ratios) can (and will) depend on the range of the concentration levels of the suspended sediments in the water.  In the case of coastal waters often the concentration levels  are relatively low to moderate.  In this case the visible bands will be the most useful, however, when the concentration levels get high (as can on coastal waters during flood runoff or rivers during flooding --- as it did on the Colorado River in the Grand Canyon) the NIR band can become quite important.

The link below will show you an example of the use of Landsat TM in the San Francisco Bay and the attached file has a graph showing the relationship between the silts/clays and the ratio of the Nir2Red radiance values collected using a spectral radiometer with Landsat TM bands.  Note that once the concentration levels get moderate to high within the river that the correlation with the ratio that includes the NIR band becomes high.

    http://water.usgs.gov/osw/techniques/workshop/chavez.html

Good luck with your research.

Pat

I have been glad to read remarks by Dr. Chavez, well-known expert in the atmospheric & ocean optics! And I completely agree with these remarks. Just now we are preparing a paper about SSC vs reflectance relationships in the East China coast. In the frame of this study we performed an analysis of the world literature, and see that different reflectance indices may yield the best results in different water basins. If to limit ourselves only by turbid Case 2 waters, then such indices use generally the following spectral combinations: Red/Green, NIR, NIR/Green, NIR/Red. We going to check these indices for Chinese waters. 

Dear Dr. Pat,

Many thanks for your valuable suggestions and sharing your experience with us about the relationships of satellite and in situ measured reflectances and their effects on waters.

Dear Prof. Sokoletsky,

Thanks for your comments and suggestions, also it was great to talk to you during the Ocean Optics conference in Maine.

Thanks

Dear Majid,

You are very welcome!

Could you, please, say that is your final aim? I believe that an establishment of correlations between particles and reflectances is only intermediate aim, not the final one. And solution for in situ and remote sensing SSC assessment may be different.

Best regards,

Leonid

SSC usually has a high correlation with Red band rather than NIR band. You might try Red band to see what happen. A low correlation could also be caused by the mismatch between image pixel and in-situ sample point.