IMAGE PROCESSING AND ANALYSIS

Ambient air temperature is not directly observed by investigating surface temperatures.  However, the near-surface atmosphere takes on the heat and moisture characteristics of the surface.  Digital satellite image data of thermal emittance were overlaid upon digitally mapped land use data to assess climatic heat loads characteristic of various land use categories.  An unsupervised multispectral classification (Figures 10 and 11) of reflected and emitted radiation was derived to develop a "climatically-driven" land use classification to be compared with more traditional methods of assessing land use.  [See Volume 3, Module 7.2 of the Remote Sensing Core Curriculum for a more detailed discussion of unsupervised image classification.]  The classification of this image scene employed the Landsat spectral bands of reflected solar energy and the thermal band.   The five clusters identified in this classified image were statistically derived according to surface reflectance and emittance patterns.  After comparing the unsupervised classified image with ground observation the classes were given titles according to predominant land use.  Mountainous areas were masked from the classification to avoid thermal effects of varying slope and exposure angles.  This was accomplished by creating a vector file of polygons covering mountainous areas of the image, and then masking these areas from analyzed image files using an overlay procedure.

 
[FIG 10]                     [FIG 11]

Most image processing and GIS software packages have the capability to extract numeric data from one image given the categories of another.  In this case, values of surface temperature (from Figure 8) were extracted for each unsupervised land cover category in Figure 10.  Table 1 presents seven significant temperature categories within the 6569 square kilometer region imaged over Phoenix, Arizona.  Heavily irrigated agricultural crops, parkland, and well-irrigated residential neighborhoods exhibited an average surface temperature which was 32.61^o Celsius.   This was just slightly higher than the observed surface temperature of open water surfaces (31.54 ^oC).  Xeriscaped residential neighborhoods, characterized by natural desert landscaping in addition to residential building materials, averaged 39.12 ^oC, which was nearly 0.8 degrees warmer than commercial areas.  Areas identified as primarily commercial land use, characterized by large areas of pavement and roofing materials with very little open water or vegetation, averaged 38.35 ^oC.  Natural desert landscape averaged a surface temperature of 42.08 ^o C.  Barren land, which was primarily comprised of construction areas and agricultural fields currently not covered with crops, exhibited a nearly identical temperature (42.07 ^o C).  These results are consistent with findings of Carnahan and Larson (1990) who, using Landsat thermal data, observed warm barren dry agricultural soils surrounding Indianapolis, Indiana as compared to a cooler urban environment.

It is no surprise that the desert and barren landscape exhibits a much warmer temperature than well-irrigated land.  However, it is interesting to note that the average surface temperature associated with much of the residential land use is nearly the same, and slightly warmer than commercial areas (See Table 1).  The land use categorization does not determine the physical climate of urban neighborhoods, but the land cover certainly does.  Well-irrigated residential neighborhoods, with lawn, shrubbery, and extensive stands of irrigated palm or citrus trees have a surface temperature as cool as that of irrigated golf courses and cropland.  However, residential neighborhoods which have little moisture available for evapotranspiration are as warm as commercial districts, having an average surface temperature which is 6.5^o C (12.7^o F) hotter than irrigated residential  areas.  Thus, traditional categories of land use may not correlate well with the thermal zones of the city, but the unsupervised cluster categories for remotely sensed imagery do represent a climatically derived categorization of the physical environment of the urban area.  The categories presented in table 1 are determined from patterns of solar reflectance and surface emittance. 

Most of the moisture available to the atmosphere in this region of study comes from the vegetation present at the ground.  There is very little area of open water surface.  In fact, a supervised classification of the 30 meter resolution thematic mapper data revealed that water surfaces comprised only 0.2 per cent of the total image scene.  To test the role of biomass in controlling observed surface temperature a regression analysis was performed comparing the thermal landsat image of Phoenix, AZ (Figure 8) with the normalized difference vegetation index image (Figure 9).  In this case the normalized difference vegetation index image (Figure 9) is the independent variable while the thermal image (Figure 8) is the dependent variable.  Figure 12 presents a graph of this regression analysis.  The R² is 0.66 (i.e., the coefficient of determination is 66.55 per cent).  This implies that approximately 66 per cent of the thermal variation in the imaged scene can be attributed to variations in the amount of biomass (and energy used for evapotranspiration) present in each pixel.  The R value of –0.8158 is negative because temperature declines with increasing biomass.  Each of the 307,719 pixels (120 m resolution) has been plotted on the graph in Figure 12.  In this case 120 meter pixels are used because the original thermal sensor has a pixel resolution of 120 meters. The data points in the lower left of the graph, showing cool temperatures associated with relatively low biomass, were found to be metal roofs with highly reflective metalized paints and materials exhibiting very low emissivity characteristics.

This regression analysis graph shows the relationship between biomass, as represented by the NDVI digital values (Figure 9) on the X-axis, and surface temperature (derived from Figure 8) on the Y-axis.  Over 300,000 sample pixel locations were used in the analysis.  The Y-intercept of 43.99 ^oC is almost exactly the same as observed surface temperatures of a large barren parking lot at the time of image acquisition.  Surface temperatures decline with increasing biomass.  The data points in the lower left of the graph, showing cool temperatures associated with relatively low biomass, were found to be metal roofs with highly reflective metalized paints and materials exhibiting very low emissivity characteristics.  The coefficient of determination for this analysis is 66.56 per cent, significant at the .05 level.

TABLE 1:

RADIATIVE SURFACE TEMPERATURES (10:00 AM) AND AREAL COVERAGE
June 24, 1992
PHOENIX, ARIZONA

Surface Type	Surface Temp Deg. C*	Min.Temp Deg. C*	Max.Temp Deg. C*	Standard Dev. Deg. C*	Area km2	% of scene**
Irrigated
Ag. & Res.	32.61	25	40	3.06	487	7
Dry Res.	39.12	34	44	2.10	1255	19
Commercial	38.35	41	49	1.29	409	6
Desert	42.08	37	46	1.92	1512	23
Barren	42.70	38	46	1.66	769	12
Unclassified	-----	---	---		2138	33

 

(masked mountain slopes)

* Surface Temperatures extracted from Landsat Thematic mapper band 6.
** Total imaged area of Metropolitan Phoenix equals 6,572 square kilometers.