Environmental Engineering Reference
In-Depth Information
Model 7
Model 6
Model 5
Model 4
Model 3
Model 2
Model 1
0.0
0.1
0.2
0.3
0.4
0.5
R
2
0.6
0.7
0.8
0.9
1.0
FIGURE 13.3
Goodness of fit of seven population estimation models.
building height was influenced by trees partially (or even totally)
occluding the building, especially for SF residential buildings.
The inaccurate height information negatively impacted the pop-
ulation estimations from building volumes. Presumably, this also
explains why Model 7 was no longer the best performing model.
Nevertheless, the differential improvement by the incorporation
of the finest land use information seemed to hold in the validation
sample (compare Model 1 with Model 2, Model 3 with Model 4,
and Model 5 with Model 6).
13.4.3.3
Validation of models
In order to confirm the fitting trends observed during the cali-
bration stage, an independent set of samples was used to validate
each model. The
R
2
and median absolute error with interquartile
ranges are plotted in Fig. 13.4 for a visual comparison. Con-
trary to expectation, Models 2-4 performed comparably to, or
even better than, models that incorporate volume information
(Models 5-7). This may be due to the fact that the estimated
Model 7
Model 6
Model 5
Model 4
Model 3
Model 2
Model 1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
R
2
(a)
128
64
32
16
8
4
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
(b)
FIGURE 13.4
R
2
statistics (a) and median absolute error (b) with interquartile rage (vertical bars) based on validation samples.
Note the vertical axis is in a logarithmic scale in (b).
Search WWH ::
Custom Search