Geoscience Reference
In-Depth Information
SVMs refer to a supervised statistical learning and classification technique. They
are nonparametric, meaning that no assumption on the underlying data distribution
is made. During the training stage, the SVM algorithm aims to iteratively define
a so-called hyperplane in feature space. The hyperplane can be considered as
an optimal decision boundary differentiating a set of labeled input data into a
discrete, predefined number of classes that are consistent with the learning examples
(Mountrakis et al.
2011
). Once the hyperplane has been established, the SVM
classifier can be applied to the unlabeled data under consideration. Besides many
other advantages, some of the distinct amenities of SVMs in the context of
multidimensional data classification are their ability to incorporate hyperspectral
images without the need of any feature reduction procedure (Melgani and Bruzzone
2004
) and their relative insensitivity to training sample size and quality (Mountrakis
et al.
2011
). Considering that the areal percentages of some surface types (e.g.,
roofing materials and water bodies) are comparatively low in the study area,
SVM-based classifiers are well suited for the urban surface material mapping task
presented in this work.
In the third stage, the trained SVMs were applied to the input data. To assess
the accuracy of the resulting material map, use was made of several data sources
including 0.5 m spatial resolution aerial imagery provided by the NOAA digital
coast initiative (National Oceanic and Atmospheric Administration
2014
), Google
Street View, and Google Earth. A random sampling design was chosen comprising
at least 20 sample points per target class to assess overall accuracy, errors of
commission and omission, as well as the kappa index of agreement (Cohen
1960
;
Congalton and Green
2009
).
11.4.3
Microclimate Modeling
To simulate the urban microclimate, ENVI-met 4 (beta) was used (Bruse and Fleer
1998
; Huttner and Bruse
2009
). Based on the fundamental laws of fluid mechanics,
thermodynamics, and atmospheric physics (Bruse
2000
), ENVI-met is a 3D coupled
flow-energy balance model to predict the interactions between urban surfaces,
vegetation, and the atmosphere for a given test site and time interval (Bruse
1999
).
Version 4 of the model allows, for the first time, a complete 3D representation of all
land cover elements of the urban area considered, including the physical, thermal,
and hydrological properties of every house front and model building block (Huttner
and Bruse
2009
). Hence, the information provided by the urban surface material
map and the LiDAR nDSM (i.e., urban object heights) can be fully exploited for
urban microclimate modeling using ENVI-met 4. While the model enables the
calculation of several climatic parameters, the focus of this study was put on the
simulation of urban air temperature to analyze the thermal characteristics of two test
sites in the study area under varying conditions. To this end, a three-stage modeling
approach was employed (Fig.
11.3
).
Search WWH ::
Custom Search