Scale Factors ASTER L2 Surface Radiance TIR V004

[mirteriemens]

Hi,

I am using the ASTER L2 Surface Radiance TIR V004 to run my own TES algorithm as I want to see what happens in the different modules.
However, the final emissivity I get are not as similar as I would expect it to be to the ASTER emissivity. I believe this might be due to the scaling of the bands.

For all five bands there is are sir and sra tif rasters available.
I take the sra to be the surface leaving radiance, and found the corresponding scale factors for these on https://www.earthdata.nasa.gov/data/catalog/lpcloud-ast-09t-004#variables to be 0.006882, 0.006780, 0.006590, 0.005693, 0.005225 for the five different bands.

However, I cannot find believe the scale factors for the sir tif rasters, which I assumed to be the downwelling sky radiance. Could it be that the scale factor for the sir values are not the same? If so, what are / where can I find the correct values?

I tried to look for the scale factors in the metadata, but I cannot seem to find them in here. I am accessing the metadata from https://search.earthdata.nasa.gov/ , where I also downloaded the rest of the data from.

Thanks in advance

[LP DAAC - dgolon]

Hello @mirteriemens We're checking with our science team and will report back when we have any answer. Thanks --Danielle

[LP DAAC - jwilson]

@mirteriemens Our science team sent the following below on your inqury.
Hello,

Thanks for sharing the details of your workflow.

From your description, it looks like your handling of SRA (surface leaving radiance) is correct — the scale factors you found are appropriate for those bands. Your observation that the final emissivity differs from the LP DAAC ASTER product is understandable; these differences often arise from subtle differences in how inputs are interpreted or processed.

One point to clarify based on your workflow: the SIR bands are downwelling sky irradiance (units: W·m⁻²·µm⁻¹), not radiance like SRA. Because of this, they cannot be used in exactly the same way as SRA, and applying the same scaling approach or directly feeding them into TES as if they were radiance can lead to differences in emissivity. If you want to include SIR in your TES calculations, you would need to account for the unit difference (for example, converting irradiance to radiance under appropriate assumptions).

It’s also worth noting that ASTER’s official TES products incorporate internally derived atmospheric corrections and radiative transfer modeling via MODTRAN. So even with correctly scaled inputs, reproducing the exact emissivity results can be challenging.

The main takeaway is that the SIR values are in irradiance units, so differences in handling them are likely the source of the discrepancies you’re seeing.

If you have any other questions, please don't hesitate to reach out.

[mirteriemens]

Hello,

Thank you for the reply, it was very useful. I do still have one question:

To convert the SIR values I assumed a Lambertian sky and divided them by π. However, the values seem way too big (between 80 and 200 W m^−2 sr^−1 μm^−1) and shape of the curve also feels a bit unexpected to me. I attached an image of this sky downwelling radiance (SIR / pi) for the relevant pixels over the ASTER bands.

As such, this still feels like a scaling issue to me. Is there any scaling of the SIR bands that I am still missing?

Thanks for your help,
Mirte

[LP DAAC - jwilson]

@mirteriemens I forwarded you question to our science team. Once we have more information, I will let you know.
Janice

[LP DAAC - jwilson]

@mirteriemens Please find the response to your question from the science team.

Hi Mirte,

Thanks for sharing the plot — it's very helpful.

I took a look at a sample AST_09T V004 granule and inspected both the SRA and SIR GeoTIFFs. In that sample, the SIR bands appear to be stored as integer values without an additional scale or offset field in the file metadata. Based on both the product documentation and this file inspection, I don't see indication of a separate SIR scale factor equivalent to what is provided for the SRA bands.

I also checked the QA layers and they indicate that all required inputs were used in processing, with no indication of degraded or fallback conditions that would affect the SIR values.

Regarding your SIR/π values, both the magnitude and spectral shape look physically consistent. The smooth decrease from B10 to B14 is consistent with expected atmospheric emission behavior across the TIR bands, where radiance varies with wavelength due to absorption and emission features.

Using SIR/π to approximate downwelling radiance under a Lambertian assumption is a reasonable first-order approach. However, the ASTER TES algorithm relies on internally derived atmospheric radiative transfer (via MODTRAN), so this simplified approach will not exactly reproduce the downwelling radiance used in the operational product. As a result, some differences in emissivity are expected even when inputs are otherwise handled correctly.

Taken together, this does not indicate a scaling issue. The differences are more likely related to how the downwelling term is incorporated in your TES implementation, rather than missing scale factors in the SIR data.

If you'd like, feel free to share the granule ID (or one of the file names) along with a brief description of how you are incorporating the SIR term in your TES workflow, and I'd be happy to take a closer look.

Best regards,
Science Team

[mirteriemens]

Dear Science Team,

Thank you for the reply and the time you are taking to help!

The file name of the data I am using are AST_09T_00403142018171852_20250826093212 and ASTER/AST_09T_00403142018171901_20250826093218. I am using both the sir and the sra data as input into my TES workflow. I now also included the plot of the surface leaving radiance (sra) for the pixels that I am using. This is the sra after applying the provided scaling (0.006882, 0.006780, 0.006590, 0.005693, 0.005225 for the different bands).

I am using the following workflow, I tried to summarise it:
Constants for Planck's Law:
c1 = 1.19104e8
c2 = 1.43877e4
max_iter = 12 # maximum amount of iterations
t2 = 0.05 # W m^-2 sr^-1 µm^-1
t1 = 0.05 # W m^-2 sr^-1 µm^-1

1. NEM
Assume a emissivity of 0.99 over all bands, compute the surface radiance as R = Ls − (1 − ε) · Ldown, where Ls is the surface leaving radiance (sra (scaled)) and Ldown is the sky downwelling radiance (sir / π).Then I invert the Planck function per band (c2 / wavelengths) / np.log((c1 * emis )/ (R * wavelengths**5) + 1) to derive the temperature and take the maximum band temperature as the NEM temperature, which is used to update the emissivity spectrum ( R/ planck R(T_NEM)). I check for convergence and divergence using the thresholds. Upon first convergence, the spectral variance of the emissivity is tested against against V1, and if it is lower I do the parabola fitting to attempt to refine the εmax (else I assume εmax to be 0.96). After a second convergence or exceedance of the max iterations, the NEM module is exited. After a divergence the process is aborted.
2. Ratio
Normalise the emissivity spectrum from the NEM module (β = N * emis / np.sum(emis)).
3. MMD
Calculate the spectral constrast (MMD = max(β) − min(β)). If MMD < 0.032, εmin is assumed to be 0.983. Else I do the following: MMD_corr = np.sqrt(np.abs(MMD_value**2 - c * (NEDe**2))), emis_min = a1 - a2 * (MMD_corr**a3). I scale the β using the εmin.
4. Final iteration
I recalculate the surface emitted radiance using the MMD emissivity. I take the max emissivity and use it to calculate the temperature. I use this temperature and the recalculated surface emitted radiance (R) to calculate the emissivity again. This emissivity is again an input to the ratio and the MMD module. The output from the MMD is are the final TES emissivities and corresponding temperature.

However, when I run this using the inputs (scaled sra and the sir / π) there is no outcome as everything diverges. If I multiply the sir / π by 0.01 I get the following outcome as included in the figure. The dotted lines are the emissivity as by the ASTER emissivity product and the full lines are the emissivity as by my model. Except for the red line with the crosses, they are all quite close. This seems to fit with your explanation on my non-inclusion of MODTRAN. This leads me to think that I am still doing something wrong with the scaling. The magnitude of the sra (scaled) (~8 W m^-2 sr^-1 µm^-1) and the sir / π (~ 80 - 200 W m^-2 sr^-1 µm^-1) are also pretty far from each other, which seems wrong.

I would really appreciate your thoughts and guidance on this matter.

Kind regards,
Mirte

[LP DAAC - jwilson]

@mirteriemens Your question was forwarded to the Science Team for further guidance.
Kind regards,
Janice