International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
that, since the two images acquired from sensors with the same
orbital properties, a change of building heights between the two
dates will be reflected in a change of shade fractions. Likewise,
dense development of urban neighborhoods and addition of
tress should be reflected in an increase in shade fractions, while
deforestation and removal of tress should be reflected in a
decrease in shade fractions. The final two sets of endmembers
selected in the image were linked to their respective image to
determine their physical correspondence in the urban scene and
included the following endmembers (for each date):
|. One endmember for the water (the ocean, lakes) and shade
category (shd).
2. Two endmembers for the green vegetation category (vegl
& veg2). Vegl corresponded to urban vegetation found in
residential lawns, gardens, parks, golf courses, cemeteries, and
shrublands, while Veg2 was used for natural vegetation located
in the coastal sage and chaparral occupying the lower elevations
of the Santa Monica and San Gabriel mountains, in addition to
the oak-grass woodland located in the eastern portion of the
image. A slight difference was observed between the spectral
profiles of these two vegetation endmembers due to the level of
water content, which is higher in natural green vegetation than
in urban vegetation.
3. Two endmembers for the impervious surface category
(impl & imp2). The first was used as an endmember for
parking lots and dark gray roads, while the second
corresponded to red tile roofs and wood shingle roofs.
4. Two endmembers for the soil category (soill & soil2). The
first corresponded to bare soil in the urban scene, while the
latter corresponded to sparsely vegetated soils. Observed
differences in the spectral profiles of these two endmembers
result from variations in the organic matter and mineral
compositions.
Figure 1 shows the multi-temporal endmember fractions of
vegetation, impervious surface, soil and water/shade for the
northwestern (NW) portion of Los Angeles near San Fernando
Valley. The fractions were produced independently from
applying MESMA to the two multispectral images. Brighter
areas indicate a higher fractional abundance, while darker areas
indicate lower abundance. The fractions provide a measure of
the physical properties of the dominant land cover classes in the
scene at two different dates, thus helping to reveal the
morphological patterns of neighborhoods in this part of Los
Angeles at two different snapshots in time. Obvious changes
between 1990 and 2000 are readily observed in Figure 1 with
respect to an increase in the brightness of vegetation and shade
fractions, and a decrease in the brightness of soil and
impervious surface fractions. Nevertheless, the spatial patterns
of the fractional abundance of all land cover classes in the two
dates are very similar. A similar interpretation can be drawn for
all of Los Angeles County (not shown). This suggests that
patterns of morphological change in Los Angeles are primarily
within land cover classes rather than between land cover
classes. In other words, change in land cover is still taking place
at the sub-pixel level but not so much at the pixel level. Thus, a
crisp classification would likely result in a misleading
conclusion that no change is taking place in Los Angeles.
MESMA, however, shows that urban morphology is Los
Angeles is actually undergoing continuous changes.
Imp
Soil
Shd
M
» MS rift
Hs
Figure 1: MESMA Land Cover Fractions for the NW
portion of Los Angeles
The accuracy of MESMA fractions was assessed by comparing
the accumulated fraction estimates in areas with relatively
homogeneous land cover that did not experience change in
fractions between the two dates with areal estimates derived
from the higher resolution aerial photos (Table 1)
Table2: Results of the Accuracy Assessment of MESMA
Fractions. Calculation method can be found in Peddle et al
(1999) and Rashed et al. (2003). Areas measures in M2.
1990 2000
Area of Area Estimated Area Estimated
Reference |From accumulated e From accumulated rea.
Data fraction fraction
Vegetation i £y dd
Site1 242,500 235,889 0.052 238,624 0.051
Site2 211,250 222,629 234,688
Site3 325,000 310,365 (TE ane
Site4 1,034,375 1,120,759 1,117,548
Impervious Surface
Site1 101,875 103,411 0.088 101,934 0.063
Site2 1,743,125 1,829,180 1,815,604
Site3 25,000 28,539 27,037
Site4 718,750 822,654 835,140
Soil
Site1 128,125 137,078 0.046 142,694 0.054
Site2 125,000 132,115 136,527
site3 302,500 sas 1 02
Site4 1,321,250 1,288,430 1,315,549
Shade a A ME
Site1 310,625 258210. |. 0.100. ... 326,800 5 0.077
Site2 ____ 298,125 _325,682 214758. 5
Sie3:— : 279375 |. 12682200 — 00 15 298121. Ls
Site4 450,000 495,366 514,364 e
This approach was deemed sufficient because the 1993 aerial
photos were acquired between the image acquisition dates
(1990 & 2000). In addition, for most applications, one is
interested in the aggregation of fraction measures over well-
defined regions (e.g. census tracts, ecological fields) rather
than individual pixels. The results shown in Table 2 indicate
that there is good agreement between MESMA measures at
both dates and the aerial photo-derived estimates for all four
land cover classes. Both the vegetation and soil fractions have
506
