Fusion of Airborne Hyperspectral and LiDAR Remote Sensing Data to Study the Thermal Characteristics of Urban Environments - Computational Approaches for Urban Environments - page 274

11.5.1 Surface Material Map

By using the hybrid, three-stage classification approach presented in this study,
hyperspectral and LiDAR remote sensing data of Houston were fused and turned
into area-wide information on urban surface materials (Fig. 11.4 ). Unless extensive
field surveys are up for debate, the described mapping framework represents
an effective way to properly capture the high degree of spectral and spatial
heterogeneity found in the study area. At the same time, the obtained thematic map
is able to deliver an indirect overview of the land cover and land use elements of the
urban scenery, including industrial areas, single-/double-family homes and other
residential districts, roads, sports stadiums, running tracks, urban green spaces, and
water bodies. The mean user's and producer's accuracies of the surface material map
are 79 and 81 %, respectively. While very high accuracies ( 90 %) are obtained
for metal roofs, artificial turf, trees, and water bodies, high to medium accuracies
(between 90 and 70 %) are reported for concrete and tile roofs, grass, asphalt roads,
and bare soil. The lowest accuracies ( 70 %) are observed only for tar-bitumen
roofs and tartan (Table 11.2 ).
Misclassifications in the map are due to different reasons. To name a few, errors
in the building mask (partly originating from the simple calculation of the LiDAR
nDSM) compromised the differentiation of roof materials, the spectral similarity
between bare soil areas and sealed surfaces caused confusion (Yang et al. 2003 ;
Bauer et al. 2008 ;Weng 2008 ;Eschetal. 2009 ; Elmore and Guinn 2010 ;Luoand
Mountrakis 2010 ; Leinenkugel et al. 2011 ), the spatial resolution of the input data
hampered feature extraction due to the presence of different target classes within a
single pixel (mixed pixels) (Welch 1982 ; Woodcock and Strahler 1987 ;Jensenand
Cowen 1999 ; Ben-Dor et al. 2001 ;Small 2003 ), and the limited spectral range of the

Table 11.2 Confusion matrix for the surface material map of the study area

Reference

RTi RTa RCo RMe RAs BAR TA R WAT TRE GRA ART Tota l UA

Roof (tiles) 16 4 0 0 0 0 0 0 0 0 0 20 0.80

Roof (tar-bitumen) 4 12 0 2 1 0 0 0 0 1 0 20 0.60

Roof (concrete) 0 1 13 0 3 0 0 0 0 3 0 20 0.65

Roof (metal) 0 3 2 30 1 2 0 0 2 0 0 40 0.75

Road (asphalt) 0 1 0 0 18 0 0 0 0 1 0 20 0.90

Bare soil 0 0 0 0 2 13 0 0 0 5 0 20 0.65

Ta r ta n 4 0 0 0 7 1 15 0 2 10 1 40 0.38

Water 0 0 0 0 0 0 0 20 0 0 0 20 1.00

Trees 0 0 0 0 0 0 0 0 40 0 0 40 1.00

Grass 0 0 0 0 0 0 0 0 1 39 0 40 0.98

Artificial turf 0 0 0 0 0 0 0 0 0 1 19 20 0.95

Tota l 24 21 15 32 32 16 15 20 45 60 20 300 -

Producer's acc. 0.67 0.57 0.87 0.94 0.56 0.81 1.00 1.00 0.89 0.65 0.95 - -

Overall accuracy D 0.78 Kappa coefficient D 0.76