Th e International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
2.2 Method: Time Series Long-Term Trend Analyses
The second analysis focused on the extraction of trends from
the 15 year time series. For a global context, we depict, which
areas have become wetter or drier between 1992 and today. A
period of 10-15 years is not enough to conclude on large scale
climatic change - however, no other soil moisture time series of
this length exists. It is furthermore the first time that the TU
Wien time series underwent a trend analyses. Some areas show
no change at all, while in several regions worldwide a clear
tendency towards drier or wetter soil conditions could be
observed. We furthermore investigated critically, which areas
of the global dataset are suitable for trend analyses. “Suitable”
means that, firstly, landcover conditions should favor unbiased
surface moisture extraction, and secondly, enough
measurements need to be available. Thus, we firstly masked the
data set to work with areas, which have favorable conditions for
highly accurate surface moisture extractions (see figure 2)
-90 0 90
50 60 70 80 90 100
Figure 2. The considered study area. Since soil moisture
derivation works best in unforested areas, a mask for the present
analyses has been generated from SPOT-based GLC-product
derived land-cover information. Only areas with at least 50%
grassland or Savannah were considered.
Examples of variation in the number of measurements per
month over the course of the time series are presented in figure
3. Due to sensor problems and data downlink capacities not
every spot on the earth is evenly covered with measurements.
Figure 3. Variation of data availability over time - from around
25 measurements per month to no data acquisition, a) 1994
December; b) 2000 July; c) 2001 February; d) 2007 January.
Overall the time-series between 1992 and January 2001 shows
optimal global coverage. From 2001 to mid-2003 sensor
problems lead to a gap in data acquisition. Since August 2003
data is only available within the visibility range of a limited
number of ground receiving stations.
The time series (only for the unmasked areas from figure 2 was
then analyzed in the following way:
1) Two arrays were created (for wet anomalies and dry
anomalies) with the dimension 15 x 12 (15 years time
series, 12 months per year).
2) For these arrays, the number of wet (respectively dry)
anomalies as a percentage of the total number of
measurements for that month has been calculated. A
wet anomaly is defined as a surface soil moisture
measurement, which is above the long-term mean for
that day of the year plus 10 times the noise level
associated to that long-term mean. A dry anomaly is
defined as a surface soil moisture measurement ,
which is below the long-term mean for that day of the
year minus 10 times the noise level associated to that
long-term mean
3) The yearly means of the percentage values in the
arrays were calculated by taking the arithmetical
mean of the 12 monthly values, for each year
resulting in 2 arrays of 15 years each.
4) Then a linear "a+bx" line was fit to each of the arrays
from Step 3. The 2 plots of figure 7 in the results
section for time series analyses show the "b" value of
this fit, hence its unit is "percentage per year".
3. RESULTS
In the following two subchapters we present 1) results of
anomaly analyses in the context of flood and drought
monitoring and 2) results for trend analyses of the complete
global time series.
3.1 Results of Anomaly Analyses
Among several example cases investigated the representation of
major floods and droughts, e.g. the 1999 drought in Southeast
China (figure 1) has been confirmed.
Furthermore, the potential of the 15 year time series for flood
forecasting has been analyzed. Figure 4 depicts the powerful
potential of the TU Wien time series for possible early warning
scenarios. A strong flood hit the UK in November 2000. 13
people died, 6000 inhabitants had to be dislocated, and the
estimated damage exceeded 3 Bio. USD. 22 rivers and an
overall area of 96950 km 2 were affected. The Dartmouth Flood
Observatory assigned Severity Class 3
The upper sequence of images in figure 4 depicts surface soil
moisture from January to December for the year 2000. Dark
blue areas indicate wettest conditions, light blue areas slightly
wet conditions and yellowish to dark brown areas drier to
completely dry conditions. The lower sequence depicts
anomalies. Grey areas indicate no outstanding / deviating
behavior while blue areas indicate “wetter than normal”
behavior and yellow-brownish areas “drier than normal”
conditions with respect to the 15 year monthly mean. It is
obvious that during the months January to March soil moisture
values in the year 2000 are high. However, the anomaly
sequence indicates that this is rather normal, except for some
slightly wetter occurrences in Ireland and Scotland in February.
During the summer, soil moisture is low and the anomaly