4. POST-KATRINA URBAN SURFACE TEMPERATURE PATTERN AND INTENSITY
4.2 Analyzing the Land Surface Temperature Within the Study Area
LST retrieval for the AOI required several processing steps in order to extract the data. Atmospherically corrected LST (AC-LST) was derived from Landsat TM imagery using a
single-channelalgorithm (VII) given by Brunsell et al. (2002):
LST = L
sens+ t + (1 – ε)L
d+ L
up (VII)Before this equation could be solved, several pieces of data from different sources needed to be collected in order to populate the formula. Figure 4.1 below shows a general flowchart of the steps of this process. During the image comparison phase, additional LULC classified images, based on the non-thermal bands, were used in order to analyze and verify the results.
Figure 4.1 – LST change detection processing flowchart.
4.2.1 Normalization of Satellite-derived LST
Although not always necessary, atmospheric corrections can be applied to at-sensor temperature to give a more accurate reading of surface temperatures. This process is difficult and tedious, as it has to be done for each individual image with different constants, such as: Sun elevation, Earth-Sun distance, and sensor calibration constants. These are necessary in order to convert the DNs to at-sensor radiance. In order to derive top-of-the-atmosphere (TOA)
temperature, which is the temperature corresponding to a blackbody radiator emitting the same radiance, also called the brightness temperature (BT), one must use the equations found in the Chapter 4. The data that was needed for TOA–LST to be retrieved is listed below. The example given is data for the October 9th, 2005 image:
• Date acquired = 2005-10-09
• Time acquired = 11:20am 16:20 GMT • 282nd = day of the year
• Earth-Sun distance = 0.99890 • Sub elevation = 47.6210979 • Lmax Band 6 = 15.303 • Lmin Band 6 = 1.238 • Calibration constant 1 = 1260.56 • Calibration constant 2 = 607.76
In order to add the atmospheric correction, the following information must be obtained in order to use NASA's online APCC, which utilizes a MODTRAN algorithm
(http://atmcorr.gsfc.nasa.gov). The algorithm was used to compute downwelling irradiance, atmospheric transmittance and upwelling radiances, parameters needed to compute surface leaving radiance.
5. Weather Reported for 11am, Oct. 9th, 2005:
6. Station: 722310 MSY N.O. International Airport 7. Temperature: 24.4° C 8. Relative Humidity = 56% 9. Air Pressure: 1010.494mb 10. Wind: 6 MPH 11. Alt: +0.0012192 km (1.2192m) 12. Latitude: +29.993 13. Longitude: -090.251
The next step in the AC-LST retrieval procedure was to obtain surface emissivity of the AOI, acquired from: http://speclib.jpl.nasa.gov, Version 2.0 of which has a library of over 2300 spectra. Chapter 4 already described the results of these corrections. What was not mentioned was the fact that the APCC only goes back to the year 2000. This made deriving AC-LST for the years prior impossible using these methods and current software set-up. Extra software tools could have been purchased, including a fast and easy way to atmospherically correct all of the images, however part of the goal of this entire project was to learn how to do my own
calculations whenever possible or practical. Other methods were investigated but took too long to implement.
A solution was found, detailed in Table 4.2 below, where the atmospheric conditions for dates prior to the year 2000 could be estimated by using dates with similar weather conditions post-2000, and plugging them into appropriate dates for the APCC. This was a very time consuming process, that was abandoned for non-Fall dates. The method worked, and the
preliminary results were very similar to post-2000 dates. The results were considered 'good enough', considering the MODTRAN routine is itself a model of the atmosphere based on generalized parameters, and not reality. In order to obtain 'real' atmospheric conditions, one must use a radiosonde or aircraft sensors to measure real-time conditions.
Year T °C RH% W kp/h Pr mb t Lu Ld Tp °C Climate Conditions 2000-2012
1999 19.444 47 9.656 1025.056 --- --- --- 25.607toa Mostly clear sky 2000* 25.556 58 19.312 1020.992 0.77 1.63 2.68 27.972
2001 22.778 37 25.750 1019.976 0.78 1.58 2.59 27.928 2002 25.556 79 14.484 1014.219 0.54 3.08 4.72 29.280 2003 26.667 71 14.484 1019.299 0.75 1.80 2.92 28.133 2004* 29.444 63 19.312 1013.542 0.63 2.65 4.11 28.914
2005 17.778 40 22.530 1019.638 0.84 1.16 1.95 27.541 Clear sky - Better 2006 21.111 53 12.875 1020.315 0.55 3.12 4.71 29.330
2007 16.667 54 4.828 1012.865 0.75 1.76 2.86 28.090
2008* 18.889 84 14.484 1015.574 0.85 1.16 1.91 27.549 Partly Cloudy - Good 2009 20 55 14.484 1016.928 0.79 1.53 2.49 27.884
2010 30 61 25.750 1009.478 0.76 1.76 2.83 28.099
2011 26.667 42 9.656 1020.992 0.84 1.20 1.96 27.581 Clear - Best – LA climate similar 2012* 26.667 67 16.093 1016.590 0.83 1.32 2.15 27.699
1999† 19.444 47 9.656 1025.056 0.843 1.173 1.94 27.556 0.01, 0.01, and 0.03 difference °C Table 4.2 – Simulated climate of images before year 2000 for Oct. 25th 1999 .* = Leap year; † = Averaged year; T = Air temperature degrees Celsius (°C) at 17:00 GMT KMSY station; RH = % Relative humidity; W = Wind speed (kp/h); Pr = Air pressure millibars (mb); Lu = Upwelling radiance W/m2/sr/μm; Ld = downwelling radiance W/m2/sr/μm; Tp = Random pixel temperature °C.
The other reason this method was abandoned was because in order to account for seasonal variations and weather conditions, the LST data needed to be normalized. One way to do this was to linearize the data for use in a multi-date, time series, as well as rescale it for comparison to other bio-physical features, and socio-economic patterns. The atmospherically corrected LST images were rescaled using the Normalized Bright Temperature (NBT) method:
T
NBT= (T
i-T
min)/(T
max-T
min)
(VIII)
Where,
TNBT = the normalized LST image
Ti = the input LST image
Tmin = the minimum temperature of the reference image
In Figure 4.2 below, the Landsat TM band 6 LST pixel is at the exact location (WGS84 Latitude: 29° 59' 48.0691 N; Longitude: 90° 15' 17.4600 W) of the KMSY station (4ft or
1.2192m above sea level). The AC-LST pixel is 30.829° C at 11:22am over a grassy surface, and the TOA-LST pixel is 27.843° C, a difference of 2.986° C, which demonstrates an outcome of using atmospheric correction. The air temperature recorded was 24.4° C at 11:53am, October 9th,
2005, the wind was 6mph, and RH 56%. Grass emits higher temperatures than air, depending on the time of day. The KMSY weather station has the longest and most reliable weather and climate data in the AOI. When performing tests during the day and at night, it was the only weather station that had results similar to the mobile weather equipment used for this study.
Figure 4.2 – The LST pixel in the cross-hairs is 29.9° C at KMSY station location
256 random points were created to compare the difference between the TOA-LST and the AC-LST pixels. Several of the points were removed because they were placed outside the AOI.
All of the pixels within the AOI had a difference of 2.986° C for October 9th, 2005. This process
was repeated for all of the Fall images. Results are listed in Table 4.3:
Year *Air Temp °C TOA °C LST °C Difference °C
1987 21.7 26.2 **28.1 1.9 1992 25 24.9 **27.1 2.2 1997 26.7 30.3 **32.5 2.2 2002 20 23.6 25.9 2.3 2003 25.6 27.4 30.8 3.4 2005 22.8 27 29.9 2.9 2006 24.2 29.5 31.9 2.4 2008 24.4 29.9 32.4 2.5 2010 25 31.9 33.8 1.9 Average 23.9 27.9 30.3 2.4
Table 4.3 – Averaged pixel temperature results for two methods of LST retrieval. *Air temperature °C sampled at KMSY station; **based on simulated APCC profiles.
The TOA-LST retrieval method alone underestimated the surface temperature more than the AC-LST, where the latter can also underestimate actual temperatures (Qin et al. 2001). the contributions of the emissivity of surface features and the atmosphere, returning a lower
temperature. This finding agrees with the results of several other studies, some of which used in situ instruments to test the local conditions (Sobrino et al. 2008). While experimenting with the input variables of the AC-LST model, I found upwelling to have the most significant
contribution. Changing it just slightly gave drastically different results, whereas the other variables changed the results very little by comparison.