as a
L. The
has a
ssified
on of
Lake
hallow
being
ion of
water
eters
| lake
ion of
ate of
à are
ability
iod of
J lake
urther
ncing
; and
their
JTE
1S-1C
vith a
scene
for a
panchromatic scene (5.8 m pixel size) it amounts to 70
km. For this study the VIS/NIR bands (see tab. 1) of the
LISS 3 sensor were used for 6 multispectral datasets
(see tab. 2) which were acquired under cloudfree
conditions. Additionally, a complementary panchro-
matic scene was acquired for June, 2, 1997.
Band 1 (VIS) 2 (VIS) 3 (NIR)
À in um 0.52-0.59 0.62-0.68 0.77-0.86
Tab.1: Spectral bands of IRS-1C LISS 3 sensor
During field campaigns secchi disk transparency was
measured for a varying number of lakes while the
lakes Gr. Wummsee, Gr. Zechliner See, and
Braminsee (fig. 2) were measured all the time.
Additionally, water samples were taken for
determination of chlorophyll-a content.
IRS-1C scene Field measurements
August/18/1996 -
February/2/1997 -
May/04/1997 May/15/1997
June/2/1997 (incl. PAN) June/10/1997
- July/23/1997
September/1/1997 September/2/1997
September/25/1997 September/25/1997
Tab.2: Acquisition dates of IRS-1C scenes and dates of
according field measurements
4.2 Preprocessing
Multitemporal analysis of IRS-1C scenes for
investigating spatial and seasonal variations of lake
properties requires a geometric and atmospheric
correction in order to carry out comparisons between
scenes.
Geometric correction of all datasets is based on a map
to image registration of the panchromatic scene by
determination of ground control points from 1:25000
topographic maps. All multispectral datasets were
rectified by an image to image registration with the
panchromatic scene being the master image. A RMS
error of less than 1 pixel shows the good spatial
correspondence between the geocoded scenes. For
resampling the nearest neighbor method was applied
in order to maintain the original gray values for
atmospheric correction.
For atmospheric correction (see fig.3) of the
multispectral datasets the ATCOR-2 model (Richter,
1990) was used. The correction is based on reference
spectra measured by the field spectrometer for the two
September scenes. With these spectra the calibration
coefficients (gain / offset) were estimated and
considered as stable for all scenes. The atmospheric
parameters (visibility, adjacency) were adjusted for
each scene using targets with lowest variability of
radiance. The result are reflectance values stretched
by standard coefficients to 256 digital numbers (8bit)
are obtained which allow a comparison between the six
acquisition dates of the multispectral scenes.
For the analysis of the lakes a water mask was applied
to the dataset. The mask represents the result of a
maximum likelihood classification of surface cover
types. In a next step, spectrally representative polygons
were determined within the lake areas by visual
analysis. The identification of these core lake areas
was performed in order to exclude mixed pixels in the
area of shore lines which are caused by vegetation and
the influence of the lake bottom in shallow areas. For
the purpose of avoiding any kind of ground signal in the
water spectra, the results of secchi disk transparency
measurements were analyzed in combination with
bathymetric information especially for the shallow lakes
to make sure that lake depth significantly exceeds
secchi disk transparency.
79 —e— Gr. Wummsee
original
60 —e-— Gr. Zechl. See
5 50 original
a
E ——k—— Braminsee
= 40 i
= original
S 30 --.--- Gr. Wummsee
©
S 20 FN atmos. corr.
10 ---@--- Gr. Zechl. See
0 atmos. corr.
2 2 Qa - A- - - Braminsee
atmos. corr.
IRS-1C Band
Fig. 3 Original and atmospheric corrected spectral
values of representative lakes for the
September, 1, 1997 scene
4.3 Approach for satellite-based lake water analysis
The three broad spectral bands of the IRS-1C sensor
within the VIS/NIR part of the spectrum offer a rather
limited range of possible band-related indices for
characterizing water properties, such as chlorophyll-a
content and secchi disk transparency. In this study
band to band relations in form of combinations of single
bands by various arithmetic expressions were
investigated and are called indices in the following. The
spectral properties of water for IRS-1C are
characterized by the peak in the visible green band
caused by scattering due to water constituents, by
chlorophyll-a absorption in the visible red band and by
absorption due to water in the near infrared (see fig.3).
The overall magnitude of spectra is dominated by
scattering of light by algae. For the broad IRS-bands
the strong chlorophyll absorption in the visible red is
superimposed by scattering effects for lakes with high
chlorophyll-à content. Based on an analysis of the
average reflectance values of the three bands for all
core lake areas, the mean value of band 1 (VIS green)
and band 2 (VIS red) and the difference between band
1 and band 3 (NIR) were chosen for further analysis. In
order to compare the results with single-point field
measurements and on a lake by lake basis within and
between scenes, standard statistical parameters were
calculated for each core lake area from the results of
the calculation of the indices. The values of the indices
are named as digital numbers (DN).
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 131
