I am analysing urban heat island using thermal band 6 of landsat TM-5 sensor with respect to Normalized difference vegetation index that derived from the Band 3(Red) and Band 4(NIR). Lansat TM years 1992, 1998, 2002, 2009, and 2011 images were downloaded from Glovis.
In order to reduce the noises caused by water vapor, aerosol optical depth and also the gains of satellite launch parameters, the original DN needs to be calibrated.
The study area were clipped by using the shape file using clipper tool in QGIS. The DN of thermal band 6 was converted to top-of-atmospheric (TOA) radiance and then at-satellite brightness temperature. These two-fold process of conversions were performed in semi-automatic classification plugin using QGIS. Data type of converted band was 64 bit tiff raster.
At-satellite brightness temperature Band 3, Band 4 and thermal band 6 statistic minimum, statistical maximum and statistics mean values ranges that were noted from metadata tab of layer properties using QGIS are as follows:
At-satellite brightness temperatures min, max and mean respectively:
Band 3 = 0, 0.29, 0.08 Band 4 = 0, 0.39, 0.12 Band 6 = -71.28, 41.36, and mean is -19.03 (Celcius)
At-satellite brightness temperature was used to analyse the thermal environment of land surface but it is not real surface temperature due to the influence of atmosphere. So the atmospheric correction is required. Single channel atmospheric correction (Sobrino et al. 2004) method was used to eliminate the atmospheric effects by retrieving the land surface emissivity of the corresponding land cover category.
Hu and Jia 2010 Influence of land use change on UHI derived from multi-sensor data adopted the same method as Sobrino et al (2004) did by threshold NDVI values and can be assumed the emissivity values as follow: The emissivity was assigned to be 0.97 at NDVI <0.2, and 0.99 at NDVI >0.5. When 0.2 ≤ NDVI ≤ 0.5, the emissivity was calculated by the following formula:
e = 0.004 [(NDVI - NDVImin) / (NDVImax - NDVImin)]2 + 0.986 ------Equation 1
Here in equation NDVImax and NDVImin were taken as 0.50 and 0.2 respectively and NDVI derived for his study area (I don't know what is the dynamic range of possible values that might would come).
NDVI Calculation for my study work and dynamic range are as follows:
NDVI = (Band 4 - Band 3) / (Band 4 + Band 3) -----------Equation 2 Here Band 3 and Band 4 are at satellite reflectance values (conversion method described above) dynamic range = -0.32313865423203, 0.69899702072144, mean = 0.20858884798409
NDVI threshold was done in raster calculator of ArcGIS 10.1 and syntax of conditional statement is as follows:
Con("NDVI.tif" < 0.20, 0.97, Con( ("NDVI.tif >= 0.2) & ("NDVI.tif" <= 0.50), 0.004* (("NDVI.tif" - 0.20)/(0.50-0.20))^2 + 0.986, 0.99 ) ) NDVI.tif is output main raster contained dynamic values as mentioned in equation 2.
My questions here is that,
Can I use this same formula to derive emissivity for my study area? Explain in a precise way to threshold the values by using either syntax for conditional statement in arcGIS, QGIS or ERDAS IMAGINE model maker? (share if any tutorial you have)? and Is this conditional statement syntax that used for my work true. I was actually not sure about the last parameter that is 0.99, will it be applied where NDVI > 0.5. what about the other background or abnormal values if it has in. My emissivity raster gives me the range of minimu 0, maximum 0.99000000953674, and mean as 0.58483839008459.
Does it matter if the product is 64 bit type being made by QGIS 2.4.0 Chugiak? What impact can be made while doing analysis if raster data type is 32 or 64 bit type?
The tools that I can use are arcGIS, QGIS, ERDAS IMAGINE, R statistics, not a expert but medium level.
Calculation of Green vegetation fraction was calculated by using the equation:
GVF = (NDVI - NDVImin) / (NDVImax - NDVImin) ------------ Equation 3
I took the same values ranges as mentioned above for the main NDVI raster as dynamic range and the GVF values comes ranges from 0 - 1.0 that is actually I need to draw the colour map as Hu and Jia has made in figure 4 with 0-0.20, 0.20-0.40, 0.40-0.60, 0.60-0.80, and 0.80-1.0 legend. But my whole study area comprises six subdivisions and I want to display GVF for all my subdivision boundaries. As well as with whole dynamic range of study area concerns, it is nicely computed with full range of 0-1 but when i chop 6 subdivisions from the main raster that comes from GVF this ranges would gone smaller like minimum and maximum of one subdivision that I chop from main GVF raster are 0.6524104475975, 0.099329931638093 respectively and could not make the equal colour map of full range as I did with the main raster. Is there any way out to cope with such issue or anyone explain about the method that follows Hu and Jia.