449
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
followed by the calculation of the broadband shortwave albedos
from the generated reflectance data. In order to avoid confusion
the terms MODIS s ; m and TM sim will be used in the following to
define the data that resulted from the spectral and spatial
processing described above.
Real Data
From the MODIS scenes, subsets were defined to cover the
relatively small study area. The “MODIS reprojection tool”
(MRT) was used to trim the data to the desirable extent and to
adapt the geodetic projection systems.
To allow a quantitative comparison of the albedo values
received from MODIS and TM, a systematic degradation of the
TM data was performed (Justice et al., 1989; Kavzoglu, 2004)
using the PSF of MODIS (derived from a series of prelaunch
measurements at system level; Barnes et ah, 1998).
The normalized PSF of each MODIS band was applied to the
corresponding TM band. As a final degradation step, TM data
were aggregated to a pixel size of 463m, which corresponds to
the MOD09 product. In the following, broadband albedos
calculated from MOD09 and the aggregated TM data (TM agg )
were compared for all acquisition dates.
Simulated Spectral Response Function MODIS Band 1
Figure 2. Example of using HyMap data to generate ideal
MODIS data: HyMap channels 13-17 (Gaussian response
functions) were weighted according to the MODIS spectral
response function to generate MODIS band 1
4. RESULTS & DISCUSSION
Simulated Data
The comparison of albedos derived from MODIS S i m and TM sim
showed an almost perfect linear relationship (Figure 3). Values
scattered marginally around the 1:1 -line at the highest range of
values, but in total, deviations from the 1:1-line were
negligible. Pearson’s r was 0.996, and the root mean squared
deviation (RMSD) equalled 0.006 (Figure 3). The conversion
formulae have been provided by Liang (2000) to cover a wide
range of natural surface types. The almost perfect match
between both datasets found here verified these formulae that
obviously can be applied consistently for surface characteristics
found in the central European study region with a hetero
geneous mixture of different cover types in one pixel. This
again suggests that a linear mixing principle is valid for
heterogeneous (mixed) surface types (Liang et al., 2002). As
the results of the applied processing scheme to degrade TM
data to broad MODIS-like pixels were consistent, this approach
was applied without modification to the real TM data.
Shortwave Albedo
Figure 3. Comparison of shortwave albedo derived from
MODIS sim and from TM sim after aggregation
Real Data
For the real data acquired at 28 May 2005, the albedos retrieved
from the degraded and aggregated TM data correlated highly
with those from the MOD09 product (Pearson's r = 0.88) and
grouped around the 1:1-line (Figure 4), although viewing
geometries differed significantly between both sensors (see
Figure 1 for MODIS, TM with near-nadir viewing geometry).
Daily albedos (MOD09) were also highly correlated with the
16-day composite, but with an offset of about 0.03 (MOD09 >
MCD43A3). Different from MOD09 and TM data, MCD43A3
reproduces the 16-day average condition. It also encloses a
correction of the surface BRDF effects.
Results were different for 19 July and 4 August 2003. Again,
linear relations between albedos from MOD09 and TM agg were
found, r equalled 0.94 and 0.86, respectively, but at both dates
albedos from TM agg had a clear offset towards MOD09 results
(TMagg > MOD09). One explanation for these results differing
from 2005 might be, that effects from BRDF changed distinctly
from July/August to May (mainly due to phenology). As for
example coniferous forests with relatively small phenological
variations were also included in the analysis, this yields not a
full explanation. MOD09 and MCD43A3 showed a very good
match at both dates in 2003 with low deviations around the 1:1-
line (19 July 2003: r = 0.9751; 4 August 2003: r = 0.9121). The
mismatch between TM albedos and MCD43A3 is not in line
with the findings of Liang et al. (2002), who stated a very good
agreement of total shortwave albedos retrieved from these data.
Nevertheless, Fang et al. (2004) also found some moderate
mimatches between albedos obtained from MODIS and Landsat
ETM+. For these deviations, uncertainties associated with the
atmospheric corrections (water vapor content and aerosol
corrections) may provide some explanation (Liang et al., 2002).
5. CONCLUSIONS
From our findings, the following main two conclusions may be
drawn:
- The formulae of Liang (2000) to calculate land surface
broadband albedos from TM and MODIS data were
successfully applied to specifically degraded HyMap data.
Conversion formulae were suitable for mixed surfaces of
different land cover types.
L