Validation of SMOS satellite data over Ghana and Burkina Faso
Testing the SMOS algorithm over Guinea and Sudan savanna vegetation
by Anke Poelstra
Soil moisture (SM) is an important parameter in weather, hydrologic, climatologic and
atmospheric models because of its influence on evaporation, infiltration, runoff and uptake of
water by vegetation. Because of its variability in time and place, however, creating a database of
the entire earth based on field measurements would be extremely time- and labor-intensive. A
solution to this is satellite imagery. Satellite imagery can give information on SM on a global
scale, at intervals of at most a couple of days.
For this purpose, the European Space Agency (ESA) has developed the Soil Moisture and Ocean
Salinity (SMOS) satellite. This satellite measures brightness temperature (TB) at L-band. From
these measurements, SM in the top five centimeters of the soil can be determined. After
resampling, the data resolution is approximately 15 by 15 km. One such cell of 15 by 15 km is
called a node. The satellite covers a specific node on average every three days. ESA's objective is
to measure SM within 4% of its real value.
The SMOS satellite has been in operation since November 2009. Decades of research preceded
this launch. Both a suitable measuring device and an algorithm had to be developed. TB that is
measured by a satellite is not only dependent on SM, but also on vegetation, topography, snow
conditions et cetera. Different parameters model these conditions in the algorithm.
Many parameters have been tested in research on plot sized areas, in the order of square meters
instead of kilometers. Now that the satellite is launched, the algorithm has to be validated and
calibrated for different areas in the world. This report focuses on the validation of the algorithm
for West-Africa. The first step in this process is a sensitivity analysis of the parameters in the
algorithm. This is used to determine which parameters are most influential. This knowledge can
be used in further validation and calibration efforts and the design of future field experiments.
Validating data was done by comparing SMOS SM values to SM values from field research and
other satellite data.
The sensitivity analysis showed that the parameter that models the scattering of radiation
through vegetation (scattering albedo ω), roughness of the area (modeled with the dimensionless
parameter HR), temperature T and litter properties (modeled with its effect on the optical depth
of vegetation τL) are the most influential parameters. Default values as found in the L2PP
processor used for this analysis did not always seem to represent the research area best. E.g.
litter was put on a default value of 0, while it is known that at certain periods in time, litter is
found in the area. These differences cause the largest changes in SM. Looking solely at the
influence of parameters when changing them within a range that is thought to represent the
research area best, HR seems to have the largest impact on SM. All results should be met with
caution. The functional form of the dependences showed that this is dependent on the measured
TB, the original value for the parameter and the amount of change the parameter undergoes.
Therefore, results cannot simply be generalized for other areas, even when the range of the
parameter is the same.
The second step was to compare SMOS SM data with other SM datasets. In order to obtain field
data a small field work was executed in northern Ghana, near Tamale. This was done in April and
May, at the end of the dry season,. Its objective was to determine the average SM content of a
SMOS node which could be compared to the satellite measurement of the area in that period. In
order to compare field measurements with satellites measurements, an average that represented
the SM in the complete pixel had to be found. This average is called a well defined average. In
order to determine this average value, the area was divided into so-called hydrotopes on the
basis of hydrologic features. Hydrotopes have two important characteristics. The SM values
measured in a hydrotope show a distinct distribution and, in the wet season, the difference
between the means of two hydrotopes is larger than the variances of the hydrotopes. The
average values of the different hydrotopes should thus be very distinct and the overlap of their
distributions is limited. It turned out however that a better distinct distribution for the top five
centimeters of soil was obtained by dividing the area based on vegetation type. Comparison of
the well defined average of the area with SMOS satellite data showed a relatively low value for
the SMOS satellite, that nevertheless falls within 4% of the value measured in the field.
Other comparisons were done with measurements in Burkina Faso, SM data from other satellites
and precipitation data. The field measurements in Burkina Faso are located near a stream and
cover a line of a couple of kilometers, rather than a complete pixel. Moreover, the field
measurements are done at a different depth than satellites' measurements. By comparing values
for different locations, similarities of and differences between the patterns of the different
satellites over time become apparent. The comparison with precipitation data shows whether the
SM data reacts to rainfall. SMOS showed a very poor correlation with the field measurements in
Burkina Faso. It also showed a small range compared to two of the three other satellite datasets
and its absolute values were far lower than the field measurements. SMOS did show good
correlation with other satellite data for areas with little precipitation. This latter data has in earlier
research been found to have an error of 5.4%. SMOS also showed a good correlation with
rainfall; its value goes up with precipitation and down in dry periods. The correlation with other
satellite datasets for wet areas is however poor. The best correlations are found in the northern
part of the research area, with less vegetation and rainfall. Some pixels more to the south,
receiving very little rainfall, also showed a good correlation. Remarkable is the difference
between SM values from SMOS morning and evening overpasses. Comparisons with satellite data
did not show which overpass gives the more realistic value.
Concluding, it seems that the SMOS SM values are relatively low and although SMOS data
responds to precipitation as expected, the amount of change is too small to ensure good
correlation for wet areas. The correlation for dry areas is usually acceptable. Ensuring the right
values for certain important parameters will probably improve the correlation values. It is
expected that then the 4% boundary will be met for dry areas. Whether this will also be the case
for wet areas will have to be investigated further.