Thermal Remote Sensing with Landsat

[anthonyw]

Hello All,

 

I am trying to estimate land surface temperature (LST) from Landsat 5 data. The study area is on the East coast in the US, and luckily it includes a temperature station with hourly data on the day the satellite passed. I've converted DN5 to radiance and used an inversion of Planck's to calculate brightness temperature-- this is all pretty standard stuff.

 

Here's my problem: The uncorrected at-satellite brightness temperature at the temperature station location is about 7 degrees F higher than the station measured at the temperature station. My general understanding is that the East coast is rather muggy, so I would suspect that atmospheric correction should cause LST to increase compared with the at-satellite brightness temperature, as there's probably a fair bit of moisture in the atmosphere absorbing radiance between the land surface and satellite. The fact that the uncorrected at-satellite temperature is higher than the measured temperature seems problematic. 

 

Indeed, using the Atmospheric Correction Parameter Calculator (http://atmcorr.gsfc.nasa.gov/) and a standard radiative transfer equation, my land surface radiance increases compared with the uncorrected at-satellite radiance, and the corrected temperature increases to nearly 12 degrees F above the temperature measured by the temperature station. 

 

I've checked and rechecked my radiance and Planck's inversion calculation multiple times. I'm confident these equations are fine. I've tried calculating radiance from both R/Grescale and LMIN/LAX equations, and I've even gone back to previous calibrations (e.g., Chander, 2003), but each of these approaches impacts temperature only slightly. 

 

Any ideas? I'm pretty much at a loss here. I would really appreciate any advice/suggestions. 

 

Thanks,

Anthony

 

 

ETA: Oh, I was wondering if this could be a time issue. My understanding is that Landsat image acquisition is listed as GMT. Is that correct?

Edited March 18, 2014 by anthonyw

[anthonyw]

Update:

 

I found another station in my study area. The at-satellite temperature there is also about 7 degrees higher than the instrumental LST.

 

I also just tried a Landsat image from a different date. I'm seeing more or less the same problem: at-satellite temperature is about 7 degrees too high at both stations. 

 

I'm perplexed here, and I'm quickly running out of ideas. Anyone have any ideas? 

 

Also, this is where I read that image acquisition times are in GMT: http://landsat.usgs.gov/acquisition_time_for_an_image.php

[anthonyw]

Update:

I was just reviewing some literature and noticed similar results (Yang and Wang, 2002; www.ltrs.uri.edu/research/LST_page/paper4.doc‎). Their uncorrected at-satellite temperatures were also significantly higher than station-based readings. Their research was conducted very close to my study area.

 

Am I fundamentally misunderstanding something about comparing satellite-estimated LST with station measurements? My recollection from literature is that multiple studies have done this and errors are usually rather small, and I don't recall reading that anything special need be done when comparing satellite estimated LST with station measured temperature. I know satellites are measuring the actual ground surface temperature based on its brightness temperature, and I know temperature stations are measuring air temperature and not ground surface temperature, but based on the literature I've reviewed I would have expected uncorrected temperatures to be a bit on the cool side based on atmospheric absorption between the land surface and satellite. Moreover, I would have expected uncorrected temperatures to be a bit closer than 7 degrees F! I don't know what the exact lapse rate in this area is, but I can't imagine it's high enough to account for this discrepancy!

[achoknyz]

Check the discussion below. You need to look at other parameters affecting temperature not just a black body. I hope this helps.

 

 

http://www.gisarea.com/topic/3547-calculation-of-emissivity-from-ndvi/?hl=calculating

[Arhanghelul]

Hello Anthony,

 

Did you applied the emissivity corection on your data ?

What algorithm did you used to estimate LST ?

Me, I know 3 major algorithms for Landsat thermal applications: the Radiative Transfer Equation, The Qin et al (2001) and Jimenez-Sobrino (2003; 2004; 2009) algorithms. User achoknyz posted above a link on my literature description.

 

Me, I have a problem with the the Atmospheric Correction Parameter Calculator (http://atmcorr.gsfc.nasa.gov/). At the Optional: Surface Conditions, the calculator requires the user to enter surface conditions. These surface conditions, you can take them from your local meteorological stations when the Landsat passes over the station. Ok ?

But what do you insert to "Altitude (km)" parameter ? What value ? In my opinion, it should be zero (0), because the local surface conditions are measured to no more than 2 meters above terrain surface (for air temperatures in meteo stations). Am I right for now ?

Please, explain me this problem....

 

 

P.S.: sorry about my english, is not perfect

Edited March 19, 2014 by Arhanghelul

[Arhanghelul]

Aaaa, and also I forgot to tell you.

There is no exact method to validate the atmosferic and emissivity corrected LST derived from Landsat thermal images.

If you rewiev the literature, you will see that there are still unsolved problems with the LST ground validation.

So, don't expect to obtain a real LST, just a relative LST is more correct to say.

Edited March 19, 2014 by Arhanghelul

[anthonyw]

Hello Anthony,

 

Did you applied the emissivity corection on your data ?

What algorithm did you used to estimate LST ?

Me, I know 3 major algorithms for Landsat thermal applications: the Radiative Transfer Equation, The Qin et al (2001) and Jimenez-Sobrino (2003; 2004; 2009) algorithms. User achoknyz posted above a link on my literature description.

 

Me, I have a problem with the the Atmospheric Correction Parameter Calculator (http://atmcorr.gsfc.nasa.gov/). At the Optional: Surface Conditions, the calculator requires the user to enter surface conditions. These surface conditions, you can take them from your local meteorological stations when the Landsat passes over the station. Ok ?

But what do you insert to "Altitude (km)" parameter ? What value ? In my opinion, it should be zero (0), because the local surface conditions are measured to no more than 2 meters above terrain surface (for air temperatures in meteo stations). Am I right for now ?

Please, explain me this problem....

 

 

P.S.: sorry about my english, is not perfect

 

Thanks for the response.

 

I have calculated LST three ways:

1. Uncorrected at-satellite

2. Corrected temperature via Atmospheric Correction Parameter Calculator and radiative transfer equation

3. The Jimenez-Sobrino (2003) single-channel method.

 

I have estimated emissivity using the NDVI Thresholds method as per Sobrino et al. 2004. 

 

For the Atmospheric Correction Parameter Calculator, I tried a number of different options to see how upwelling and downwelling radiance and atmospheric transmittance would be affected. I tried it with and without surface conditions, with and without interpolation, with summer and winter atmospheres, and with an altitude of 0 to 0.1 km (the weather station in my study area is on top of a tall building, at an elevation of about 300 ft.). Overall, these changes didn't have much impact on upwelling and downwelling radiance and atmospheric transmittance. 

 

I'm just a bit confused as to why uncorrected at-satellite temperatures are so much hotter than weather station temperatures-- especially in this area. I would expect ground surface to be a bit hotter than air temperature, but in a muggy area with lots of water vapor, I would expect the thermal radiance sensed by the satellite to be much lower than that measured on the ground, as water vapor between the ground and satellite should absorb radiance. I thought the "cooling" impact of water vapor would exceed the "warming" impact of LST vs air temperature. I know Band 6 is in the atmospheric window, but I still thought water vapor would be a big enough deal to absorb enough radiance to make uncorrected at-satellite temperatures appear cooler than weather station temperatures. I guess I was just wrong about that?

One last question: When you have used Jimenez-Sobrino, how did you estimate total atmospheric water vapor content?

 

Thanks,

Anthony

 

Edited March 19, 2014 by anthonyw

[Arhanghelul]

You have a very interesting situation there, my friend. When dealing with satellites having just 1 thermal band, it produces confusing results.....

I can't explain you why you get those results, I don't have the necessary knowledge

 

But have you tried to use ASTER thermal images ? I know that, for the USA territory, you can download ASTER images for free, from Glovis.

Me, from east Europe, I don't have acces to free ASTER.

You could use a split-window algoritm to calculate LST from ASTER's thermal bands (it has 5 thermal bands from wich the last 2 bands cover a similar thermal window to LANDSAT). Also, ASTER has a good resolution of 90 m, it registers day and night images. You have more advantages using ASTER.

And the method to calculate LST is more precise than the mono-window algorithms for LANDSAT.

You could also compare the daylight LST derived from ASTER to that produced from LANDSAT, if you could find 2 images having a good time overlapping.

Do you understand me ?

 

Sorry about my english......

 

Search more articles about this.

 

Here, more info about ASTER: http://www.yale.edu/ceo/Documentation/ASTER.pdf

Edited March 19, 2014 by Arhanghelul

[Arhanghelul]

And here, an very useful article about using LANDSAT 5 TM and ASTER images :

http://www.mdpi.com/2072-4292/3/7/1535

[Arhanghelul]

citation:

 

"Lack of atmospheric correction may introduce a temperature error of 4 to 7°C for the midlatitude summer atmosphere (Voogt and Oke, 1998). The magnitude of atmospheric correction depends on the image bands used, as well as atmospheric conditions and the height of observation. Errors owing to urban effective anisotropy depend on surface structure and relative sensor position and can yield a temperature difference of up to 6 K or higher in downtown areas (Voogt and Oke, 1998)."

[terefice]