1 Spatial Modelling of the Terrestrial Environment - Georeferencial

Near Real-Time Modelling of Regional Scale Soil Erosion 159
We applied the monitoring system outlined above for April 1998, the wettest month in
the 1998 wet season. This month was used because such wet periods are likely to produce
the most erosion. It is interesting to note that the region experienced extremely high rainfall
throughout the 1998 wet season that can be attributed to the El Niño conditions of that year,
thus rainfall was even higher than usual.
8.2 Methods
Methods of implementation of the three components of the sediment monitoring system
are discussed in turn below.
8.2.1 Soil Erosion Modelling
The following soil erosion model was applied to the Lake Tanganyika catchment at the
regional scale (1–8 km pixel size) on a decadal time step (Thornes 1985, 1989):
E = kOF 2 s 1.67 e −0.07v (1)
where E is the erosion (mm/decad), k is a soil erodibility coefficient, OF is the overland
flow (mm/decad), s is the slope(m/m) and v is the vegetation cover (%). The model predicts
soil detachment by rainsplash and the ability of overland flow to transport this material.
However, it does not consider where the sediment is transported to or any subsequent
deposition.
A problem is posed by the choice of the spatial scale of the model because soil erosion
predictions are sensitive to the spatial resolution of the model input parameters (Drake
et al., 1999). For example, as the scale of the digital elevation model (DEM) is reduced, the
slope derived from it is also reduced because the averaging inherent in the coarser resolution
data means that high altitudes become lower and low altitudes higher (Zhang et al., 1999).
The lower slopes result in lower predicted erosion. To overcome this problem we have
developed a way of calculating slope independent of scale using a fractal approach. Most
landscapes can be considered to be fractal and it can be shown (Zhang et al., 1999) that the
percentage slope (s) is related to s at any scale d by the equation:
s = αd 1−D (2)
where α is the fractal constant and D is the fractal dimension. Slope in this study was
estimated from the GTPO30 global 1 km digital elevation model (DEM) by calculating α
and D from the DEM and then using equation (1) with a d of 30 m in order to estimate the
slope at this scale. Zhang et al. (1999) have shown that the method is effective at the scales
being considered here. They tested the method using the same DEM, calculated slope for
a d of 200 m and 30 m, compared the slope estimates to two DEMs of this resolution, and
found RMS errors of 3.92% at 30 m and 8.63% at 200 m.
To estimate vegetation cover we used LAC AVHRR data obtained from the Kigoma
AVHRR receiving station. Imagery was calibrated, geocoded using information in the file
header, cloud masked, converted into NDVI and then calibrated to vegetation cover. The
cloud masking technique of Saunders and Kriebel (1988) was employed. The method uses