Geoscience Reference
In-Depth Information
In a second step, the standardized temperatures (T
az
) were included in a multiple
regression model as the dependent variable; the choice and the quantification of the
independent variables were the result of several experiments [ALC 06a]. They have
been computed within a 250 m grid around the measurement point. The use of mean
values within each “unit of 250×250 m” leads to a certain loss of information but it
also functions as a filter. Units of 100×100 m and 500×500 m were also tested and
have lead to worse results. As the relations between T
az
and the independent
variables are often not linear, some of the latter had to be subjected to exponential,
logarithmic, or other transformations [AND 03]. Multiple regressions of the
standardized temperatures and different predictors were carried out separately for
each thermal pattern. The coefficients of determination ranged between 0.68 and
0.92 [AND 03]. The most important predictors were: latitude, longitude, shorter
distance to the Tagus, vegetation index (using normalized difference vegetation
index (NDVI) factor, see Chapter 2), distance to the main axes, where tertiary
activities are concentrated, percentage of built-up area within the grid corresponding
to the measuring point and the product of the latter by mean building height of the
same grid area.
In the third step, the spatial interpolation of air temperature for the whole city
was carried out on a spatially-continuous base using a GIS [AND 98]. The values
computed through the regression model were filtered (the value of each unit was
replaced by the average of a group of 3×3 units (corresponding to a 750 m side
square [AND 03, p. 177]). Finally, the isothermals were drawn based on the
continuous thermal surfaces created. The three steps described have been repeated
for each thermal pattern.
One of the patterns occurring during cloudless nights with moderate or strong
north (N) and northwest (NW) wind (which is the prevailing wind in Lisbon) will be
described. The independent variables that were found to be significant were:
altitude, percentage of constructed areas, latitude, and distance to the Tagus
(equation in [AND 03, p. 183]). The highest standardized temperatures (2.5ºC) are
projected to occur in densely built-up city districts located in bottom of valleys
(Figure 5.5a). Highest temperatures extend to the N along the main streets. The
lowest estimated temperatures occur in Monsanto probably due to altitude (253 m),
vegetation, and absence of urban development. In this case, the maximum UHI
intensity is 4ºC. Topographic shelter is the main factor that influences the
temperature distribution, and this is enhanced by the effects of urban density:
altitude has the highest standardized regression coefficients (ȕ), followed by the
percentage of built-up area [AND 03]. In order to detect the urban effect, the model
was run excluding the predictors related to urban properties. The resultant thermal
pattern is shown in Figure 5.5b. In this case, the highest standardized temperatures
predicted are lower: the core of the UHI, which was located over the high-density
built up area and reached 2.5ºC (Figure 5.5a), now attains only 1.5ºC and has moved
to the riverside areas of south (S) and southwest (SW) Lisbon (Figure 5.5b), where
the topographic shelter effect is the highest. Thermal gradient in town is now only
3ºC.
