and multi-location TIR imagery are still needed” (Weng, 2009,
p. 340) and the measurement of the urban heat island remains
difficult. In (Bechtel and Schmidt, 201 1) a large set of
predictors was compared with a long-term UHI dataset derived
from floristic proxy data and classical parameters like surface
temperature and NDVI were found useful.
In this study different (multitemporal) parameter sets were
tested for their potential to classify thermal LCZ (see Bechtel
and Daneke, 2012 for more detail) and derive empirical models
of the mean UHI from a mobile measurement campaign.
24. DATA
The city of Hamburg in Northern Germany was chosen as site
for the case study. The full domain covers 1132 km?
2.1 Multisensor and multitemporal features
Due to its high availability a strong emphasis was laid on
multitemporal Landsat TM and ETM+ data in this study. This
reflects on the different phenological conditions throughout the
year and the thermal response to different insulation conditions
which both reveal additional information. A disadvantage is
possible land cover change within the acquisition period which
may lead to classification errors. However, this is expected to
be acceptable since a high degree of persistence can be assumed
for the study area at the temporal scale of decades. The
multitemporal data were complemented with geometrical and
texture data from NEXTMap® Interferometric Synthetic
Aperture Radar (IFSAR). Overall seven feature sets were
compiled (see Table 1) and projected to a common 100 m grid
with SAGA (www.saga-gis.org). The parameters are also
named features (for classification) or predictors (for empirical
modelling) in the following.
The multitemporal multispectral (MS) parameters were derived
from 33 visually cloud-free scenes acquired between 1987 and
2010 (see Bechtel, 2011 for a detailed description of the
preprocessing). All spectral bands (1-5 & 7, ranging from blue
~485 nm to medium infrared 2.2 pm) were included in the
feature set. For each scene the Normalized Differenced
Vegetation Index (NDVI) was computed from the bands 3 and 4
as an additional band ratio. Atmospheric influence was
neglected, since the parameters were only used in trained
classifiers and models and no information about subscene
atmospheric conditions was available.
The multitemporal thermal infrared (TIR) data from TM and
ETM+ (Band 6, 10.4-12.5 um) was processed accordingly.
Digital numbers were used directly as features (without
atmospheric correction and calibration to radiance) for the same
reason. Further, surface temperatures were calculated for 22
scenes with National Centers for Environmental Prediction
(NCEP) atmospheric profiles available (Barsi et al, 2005,
Chander et al. 2009) in order to fit a simple model of the
annual cycle of temperature at acquisition time. The annual
cycle parameters (ACP) Yearly Amplitude of Surface
Temperature (YAST) and Mean Annual Surface Temperature
(MAST) contain information about the material specific thermal
surface properties (Bechtel, 2011; Bechtel, 20 12).
The geometric parameters were derived from a normalised
digital height model generated from NEXTMap Digital
Surface and Terrain Model on a 3 m grid. Besides simple statics
of the obstacle heights, further parameters were extracted by
Fourier techniques and morphologic filtering, in order to derive
spatial spectra and texture information and thus include spatial
information in the pixel-based classification approach (see
Bechtel and Daneke, 2012 for more details).
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
param category description number
ms mtms multispectral (TM/ETM- ) 198
ndvi | mtms ndvi 33
tir mt tir — thermal (TM/ETM- ) 33
acp mt tir ^ annual cycle parameters 2
shs geom simple height statistics 6
morph geom morphological profiles (texture) 22
fft geom bandpass & directional filters 190
overall 484
Table 1. Feature sets and number of features per set.
2.2 Training data — Local Climate Zones
The LCZ scheme is a local-scale landscape classification
system based on the thermal properties of urban structural types
(Stewart and Oke, 2009) and has high potential to become a
standard in urban climatology. The four basic landscapes series
(city, agricultural, natural and mixed) are each subdivided
according to their microscale (10s-100s of meters) surface
properties (more specifically: sky view factor, fraction of
impervious materials, Davenport roughness class, surface
thermal admittance and mean annual anthropogenic heat flux)
which affect the canopy-layer thermal climate. Since the
typology has a certain cultural bias towards Northern American
morphologies a compatible but slightly adopted scheme was
used for this study. For all classes representative reference areas
were digitized on a 100 m grid using map and high resolution
optical remote sensing data.
The urban series was subdivided into eleven categories. Urban
Core (urbcore) is representing the historic inner city with
massive buildings of uniform height with single spires like bell
towers. The compact morphologies were split into the classes
Urban Dense (urbdens) with perimeter block buildings of
uniform height with courtyards in the center and Terraced
Housing (terrace) with a regular pattern aligned in rows. Blocks
refer to clustered high-rise buildings in a uniform geometric
layout while Modern Core (modcore) comprises high rise
commercial buildings. Regular Housing (reghous) consists of
single family houses with a high proportion of greening
between the spaces and is typical for suburbs. The industrial
areas are divided into Industry (industr) with industrial or
commercial activities in low-rise buildings and Port (port)
which also contains container-arrays and storage facilities
beside similar structures. Rail tracks (rail), park and gardens
complement the urban series. From the natural and agricultural
series field, forest and water bodies were found relevant for the
area of interest.
2.3 UHI data
The UHI data was collected during a mobile measurement
campaign with public transportation buses in Hamburg during
the vegetation period from the 23™ of May until the 29" of
October 2011. Cooperation with the Hochbahn Hamburg
allowed for the collection of spatially dense air temperature
data in the inner city of Hamburg. Therefore, 15 buses were
equipped with temperature (and humidity) sensors of Driesen &
Kern. These sensors show a very fast responsiveness to
temperature changes which is necessary due to the fast
movement of the buses (up to 90 km/h). The DK311 loggers
were combined with CO-325 temperature sensors, RFT325
humidity sensors and a radiation protection shields. The
position was recorded with Qstarz BT-Q1000XT GPS-loggers
powered by the mobile power pack VT-PP-320 by Variotek.
The
mot
infl
tem
con
day
inte
Sin
the
coll
pos
inte
acc
dat:
min
stat
cen
Sub
Pos
stat
con
me:
revi
diff
-0.1
calc
the
Inst
like
cou
To
thar
sub
me:
The
can
