1 Spatial Modelling of the Terrestrial Environment - Georeferencial

Near Real-Time Modelling of Regional Scale Soil Erosion 165
the ATSR-2 sensor on-board the ESA ERS-2 satellite was also used and two largely cloudfree
images of the lake were found in the archive. Both sensors have a spatial resolution
of approximately 1 km, however, ATSR-2 images have a much smaller scene size than
AVHRR (512 × 512 km compared to 2700 km × 2700 km), and therefore never cover the
whole of Lake Tanganyika in one image. AVHRR images are large enough to cover the
entire lake. Furthermore, ATSR-2 provides a return period of roughly 3 days while AVHRR
acquires imagery on a daily basis.
AVHRR imagery was automatically geocoded using information in the file header.
ATSR-2 images were manually geocorrected using a DEM of the area as the base image.
Ground control points were located around the lake shore and RMS errors were kept below
0.55 of a pixel for this area. Both ATSR-2 and AVHRR imagery were calibrated to radiance
or brightness temperature and a cosine correction was applied to correct for the effect of
the solar zenith angle.
The main problem that must be faced when analysing remotely sensed data of water
bodies is that of the atmosphere. Only about 20% of the radiant energy of water bodies
arriving at the AVHRR sensor channel 1 (Red) is from the Earth’s surface, the majority is
due to atmospheric effects, and in AVHRR channel 2 (NIR) over 95% of the signal is due to
the atmosphere. It is especially important in remote sensing studies of suspended sediments
to obtain accurate estimates of the reflectance of the lake water since small deviations in
reflectance caused by atmospheric effects will translate into erroneous identification of
significant concentrations of suspended matter. Thus, atmospheric correction and cloud
masking are essential. The atmospheric correction of Stumpf (1992) was applied. This
technique includes processing to remove the effects of changes in the down-welling solar
irradiance caused by aerosols, Rayleigh scattering and glint (Stumpf, 1992). Cloud masking
was implemented in the same way as that used in the vegetation cover estimation; however,
a second step that uses a ratio between NIR and Red reflectance was introduced in order to
mask out the land. A simple histogram analysis of the ratio was used to identify cloud-free
peaks over land (greater than unity) and water (less than unity). Since we are interested
only in removing clouds from a large water body, a threshold below unity was set so that
all pixels other than those representing water are masked.
Images of the lake processed in the manner outlined above will exhibit changes in
reflectance due to changing suspended sediment or chlorophyll concentrations in the near
surface waters. Because Lake Tanganyika is a nutrient-limited system chlorophyll is of
little importance and changes in reflectance will be due to near surface sediments. Under
these conditions it is possible to calibrate water reflectance to suspended sediment concentration
using a comparison of water reflectance and measured suspended sediment
concentrations. This final step was not employed due to the lack of field data. Instead we
use an increase in reflectance of water at red wavelengths (i.e., channel 2 of AVHRR and
ATSR-2) as a relative indicator of near surface sediment concentration.
8.3 Results
There was a considerable amount of rainfall throughout much of the region in April 1998
and this produced much erosion (Figure 8.2). Most of the erosion occurred in regions outside
the Lake Tanganyika catchment for much of the time, only in the second decad was there