X-B8, 2012 
discriminating 
and vegetation 
rounding land 
es were also 
1 glomerata, 
NDVI values 
etated wetland 
nber of pixels 
was computed 
ed wetland for 
\qua images. 
ween February 
d the period of 
v1 index were 
anges, as well 
rge (Ls? data 
nent Bureau of 
epartment of 
auge data are 
ousie Springs, 
area and the 
rng (DAAOIL) 
2002 and 2010 
nents are not 
neasures/year), 
Areawpyi Were 
ally, rainfall 
n Station, the 
1 west). 
  
jn of selected 
J: Upland; R: 
| are indicated 
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 
  
STRESS = 6.03 
  
  
  
  
Figure 2: Non-Metric Multidimensional Scaling (NMDS) ordination plots of the 21 ground points based on Bray-Curtis distance 
calculated with the NDVI values from 2002 to 2010. Pixel names as referenced on Figure 1 and 90™ percentile (P90) values 
calculated from the NDVI time series of the different cover type pixels are indicated in black: Pixel_Name, P90_value. The stress 
value close to 0 underlines the high quality of this NMDS representation. 
3. RESULTS 
3.1 Delineating wetland extent 
Figure 2 presents the NMDS plot based on Bray-Curtis 
distance calculated with the 8-year NDVI values (2002 to 
2010) extracted from the 21 respective pixels. The principal 
feature evident is the good grouping of sample sites 
representative of same wetland or land type (Figure 2). 
The vegetation communities indicative of the DSC are well 
differentiated from the surrounding non-spring fed land 
covers in the ordination. In terms of NDVI values, the 
wetland vegetation are characterized by P90 values 
systematically equal or greater than 0.2, with maximal NDVI 
values registered for the Phragmites australis sites (Figure 2). 
Surrounding non-spring-fed dryland responses exhibit lower 
NDVI values (0.09 < P90 < 0.17), excepted at the riverine 
sites (P90 > 0.21) (Figure 2). 
On the basis of these observations a threshold of 0.2 was 
established, above which NDVI values are considered to be 
indicative of wetland vegetation. As the riverine ecosystem 
pixels tested exhibit P90 values over 0.2 (Figure 2), a mask 
was applied to the MODIS images to confine analysis to the 
Dalhousie Spring discharge area of our study zone (Figure 
1c) and exclude surrounding riverine ecosystems. 
3.2 Temporal dynamics of wetland extent 
Figure 3 presents the temporal analysis of wetland vegetation 
areas (combined ephemeral, Melaleuca and Phragmites). 
Comparative analyses (not shown here) have been conducted 
to confirm compatibility between the AreaNDVI calculated 
from the Aqua and Terra NDVI products. All analyses were 
conducted on the whole DSC discharge area, except 
comparison with groundflow data, which was conducted for 
the individual spring DAAOI (Figure lc). Times series 
analysed are of 8 or 10 years. 
Multi-annual values of AreaNDVI implemented using the 0.2 
NDVI threshold are highly dynamic and characterized by an 
overall decreasing trend over the 2000-10 period, but with a 
slight return of wetland vegetation since 2010 (Figure 3). 
This index ranges from 5.77 to 107 km? with notable 
maxima recorded before 2002 by the MODIS-Terra satellite 
(Figure 3a). 
A relationship between groundwater spring flow and wetland 
area is observed for the spring DAA001 (Figure 1c and 3b). 
Multi-annual (2002-2010) vegetation areas and flow values 
are clearly correlated, except during years 2007 to 2009. 
Excluding these exceptional years, 79% of the AreaNDVI 
variance is explained by the multi-annual mean flow 
variability at DAA001 (Figure 3b). The lack of relationship 
observed between flow data and wetland vegetation from 
2007 to 2009 can be explained by an extensive fire started by 
a lightning strike in 2007 at spring DAAO01. This fire led to a 
marked decrease in the vegetation extent with a slow return 
of vegetation only in 2010 (T. Gotch, pers. comm.). 
At the multi-annual scale, changes in vegetation extent are 
also strongly correlated to magnitude of preceding rainfall 
(Figure 3c). Six-monthly smoothed rainfall and AreaNDVI 
values recorded between 2000 and 2010 show similar trends. 
Extended wetland areas recorded before 2002 respond to high 
rainfall totals whereas the minimum vegetation extent 
observed in 2008 is clearly linked to dry climatic conditions 
recorded this year (370 mm cumulative rain in 2000 vs. 129 
mm in 2008).