1X-B8, 2012 
surprising that no 
til this date - has 
sensing imagery to 
ling other than by 
294; Lu, Prosser et 
ever, in open plant 
annas of the study 
shown to perform 
able TVC and its 
s modelling (van 
or or inclusion of 
he TVC in erosion 
rosser et al. 2003; 
vegetative cover 
mpeding soil loss, 
| cover subfactors 
two major vertical 
r (as for example 
(1997). As most 
ictive upper and a 
separation of the 
is considered to 
that cover most of 
Cf and subfactor 
hmeier and Smith 
hed in Rosewell 
(I) 
r and SurfCov 
f vegetative cover 
Revised Universal 
sewell (1997) for 
3 to determine the 
ven as follows: 
Q) 
©) 
) (3a) 
(3b) 
sewell (1997), CC 
ht, Ay is height to 
op of the canopy, 
concentration of 
008), and is CH is 
y been determined 
le and Ellis 2009) 
SDA (2008) and 
idy were designed 
3 of more recent 
F2 model) or 
, to date used in 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
Australia (SOILLOSS model) and variations of Rosewell's 
method as applied e.g. by Searle and Ellis (2009). This study 
calculated three vCf for the VSC representing the study area 
based on eq. 2 and eq. 1 using eq. 3a or 3b. A regionally 
developed Landsat TM and ETM+ overstorey’s woody extent 
product (woody Foliage projective cover (WFPC)) from 
Queensland Department of Environment and Resource 
Management (QDERM) had been used to stratify the study area 
into VSC (Schoettker, Phinn et al. 2010). 
The relevant variables for the subfactors were determined as 
follows: MODIS FPAR was used to approximate the GC 
variable in eq. 2 to establish its potential suitability for erosion 
modelling. To calculate the SurfCov we differentiated between 
more grassy and herbaceous VSC applying different coefficients 
a, b, c, and d for eq. (2) (Rosewell 1997). For eq. 3, averages of 
CC per VSC were derived from a relationship of wFPC to CC 
by Scarth, Armston et al. (2008). Median CH (incl. ^, and h,) 
was derived from ICESat for each VSC. The ICESat data had 
been processed byScarth, Armston et al. (2010). For eq. (3a), A; 
and h, were calculated from median ICESat pulses representing 
the upper and lower bounds of the canopy. The coefficients a; 
and a, in eq. (3a) were taken from (USDA 2008). For eq. (3b) 
CH was calculated as one third of the ICESat median centroid 
canopy height of each VSC as describe in Schoettker, Scarth et 
al. (2010). High temporal trajectories of vCf predictions for 
each of the three schemes to calculate the vCf (using eq. 2, and 
eq. 1 with 3a or 3b) were calculated and then extracted for 
representative and homogeneous regions of interest per VSC for 
the time span from 2000 to 2006. The ROIs were widely 
distributed over the study area (sizes of ROIs varied between 3 
and 10 km”. 
2.2 Modelling soil loss 
The (R)USLE is commonly known in the following form and 
soil loss is predicted as the product of six factors (Renard, 
Smith et al. 1997; Rosewell 1997): 
A mo R*KFL*S*C*P (4) 
, Where A is average soil loss (t/ha/yr), R is rainfall erosivity 
(MJ mm ha/hr/yr), K is soil erodibility, L is slope length factor, 
S is slope steepness factor, C is crop and cover management 
factor (here vegetative cover factor (vCf)), and P is (due to lack 
of data usually assumed to be 1 (Searle and Ellis 2009)). 
A temporarily and spatially explicit implementation of a 
modified version of the USLE was undertaken using a purpose 
written piece of C+ code as described in Searle et al. (2009). 
The variable vegetative cover model of the USLE was based on 
the spatial and temporal processing of raster surfaces 
representing the components of the Revised USLE (RUSLE) 
(Renard, Smith et al. 1997). Daily soil loss predictions were 
made for pixels of 25m, that is, the MODIS FPAR time series 
had been resampled to 25m (nearest neighbour resampling 
technique). 
Due to the lack of field observations of high temporal frequency 
soil loss, only a relative validation of predicted soil losses from 
the study area could be achieved by exploring the relationship 
of the soil loss predictions to daily rainfall observations and to 
measurements of daily total suspended sediment TSS and daily 
streamflow at the outlet of the catchment and study area. 
To evaluate the effect of the integration of the high-temporal 
frequency vCf predictions into the USLE soil loss predictions, 
comparisons to formerly made soil loss predictions were made. 
The relationship between average daily rainfall and (a) the to 
date commonly used, annual vCf (BGI_vCf based on Landsat 
imagery from QDERM) predictions and (b) this study's high 
temporal frequency vCf predictions using MODIS FPAR (eq. 2) 
(MODIS FPAR vCf) was determined over the whole time span 
of seven years for the areas of wFPC below 30%. Those areas 
were chosen, since Searle and Ellis (2009) had applied their 
modified model previously to those areas only. 
2.3 Data used 
2.3.1 Remotely sensed data 
A time series of the global MODIS FPAR (collection 4) data 
from 2000 to 2006 had been quality controlled and analysed for 
its sensitivity to regionally validate Landsat and MODIS based 
products in an earlier study (Schoettker, Phinn et al. 2010). The 
Landsat wFPC product used here to derive the VSC was based 
on a standardised Landsat TM and ETM+ time series developed 
at the QDERM (Danaher, Scarth et al. 2010). ICESat canopy 
height information was derived through waveform aggregation 
methodology and provided to the author by Scarth, Armston et 
al. (2010). 
2.3.2 In situ measurements and rainfall data 
In situ measurements of total suspended solid (TSS) (mg/l) for 
the wet seasons 2003/2004, 2004/2005, and 2005/2006 at the 
Myuna station, the station furthest downstream in the catchment 
of the study area, were provided by David Post, CSIRO. The 
data were recorded in hourly to minute-intervals and were here 
aggregated to average daily and cumulative TSS measures for 
comparison to daily predicted soil loss from the USLE model. 
Water quality and streamflow data were collected from David 
Post, CSIRO Land and Water Canberra and the only data 
recorded between 2000 and 2006. Daily rainfall surfaces and 
streamflow data (cumecs) were also provided by QDERM 
(http://watermonitoring derm.gld.gov.au/host htm). 
3. RESULTS AND DISCUSSION 
3.1 High temporal frequency, remotely sensed vegetative 
cover factor estimates 
The time series of high temporal resolution vCf, resulting from 
eq. 2, 3a, and 3b by assuming the SurfCov can be approximated 
as a function of the time series of MODIS FPAR, for four VSC 
in the study area are shown in Figure 1 (Scheme I, IIa, and IIb 
respectively). MODIS FPAR was used in that way because of 
its statistically significant sensitivity to green and non-green 
ground cover fractions, and despite the original eq. (2) being 
developed for ground cover products only. The vCf trajectories 
for the first four VSC are shown exemplarily; average wFPC for 
those four VSC are 0-296, 3-1096, 11-30%, and 31-50%, 
respectively. Note, the higher the vCf value, the lower is the 
estimated protective function of the TVC. Overall higher vCf 
estimates and larger annual amplitudes can be found with 
decreasing wFPC percentages (lower VSC classes), suggesting 
a vegetative cover in those VSC classes with high biophysical 
variability. 
A clear seasonality and distinct annual differences, such as the 
dry period peaking at the end of the year 2002, are visible for all 
three vCf schemes in all VSC (Figure 1). Average maximum 
vCf values in Figure 1 lie at 0.12, minimum values at 0.05. 
Differences between the vCf trajectories are generally very low 
but appear most prominent at the end of the dry seasons and 
more so in the higher wFPC classes (local maxima (for just one 
pixel) in differences of average vCf predictions in the study area 
lie at 0.04 and 0.048 for the dry season 2000 and 2004 
respectively). It was expected, that the CanCov subfactor from 
eq. (3a and b) affects the final vCf estimates more in VSC with 
denser wFPC or CC. Cover factor estimates of scheme (3b) are 
generally the lowest, which can be attributed to the fact that hf