You are on page 1of 10

Numerical weather prediction models and SAR interferometry:

synergic use for meteorological and INSAR applications


Nazzareno Pierdicca *a, Fabio Rocca b, Daniele Perissin c, Rossella Ferretti d, Emanuela Pichelli d,
Bjorn Rommen e, Nico Cimini f
a
Univ. degli Studi di Roma La Sapienza, via Eudossiana 18, Rome, Italy 00141
b
Politecnico di Milano, Milan, Italy
c
Institute of Space and Earth Information Science, Hong Kong, China
d
Univ. degli Studi dell'Aguila, L'Aquila, Italy
e
European Space Research and Technology Ctr., The Netherlands
f
Istituto di Metodologie per l'Analisi Ambientale, Potenza, Italy

ABSTRACT

Spaceborne Interferometric Synthetic Aperture Radar (InSAR) is a well established technique useful in many land
applications, such as landslide monitoring and digital elevation model extraction. One of its major limitation is the
atmospheric effect, and in particular the high water vapour spatial and temporal variability which introduces an unknown
delay in the signal propagation. However, the sensitivity of SAR interferometric phase to atmospheric conditions could
in principle be exploited and InSAR could become in certain conditions a tool to monitor the atmosphere, as it happens
with GPS receiver networks. This paper describes a novel attempt to assimilate InSAR derived information on the
atmosphere, based on the Permanent Scatterer multipass technique, into a numerical weather forecast model. The
methodology is summarised and the very preliminary results regarding the forecast of a precipitation event in Central
Italy are analysed. The work was done in the framework of an ESA funded project devoted to the mapping of the water
vapour with the aim to mitigate its effect for InSAR applications.

Keywords: Synthetic Aperture Radar, Numerical Weather Prediction, water vapour, data assimilation, atmospheric path
delay

1. INTRODUCTION
SAR interferometry (InSAR) is based on the measurement of the difference in phase of the signal backscattered by each
land surface element observed from different points and/or at different times [1]. The atmosphere, particularly due to the
high water vapour spatial and temporal variability, introduces an unknown delay in the signal propagation. This effect
might be also exploited, so as InSAR could become a tool for high-resolution water vapour retrieval. The ingestion of the
latter into weather prediction models is very promising, since water vapour is one of the most significant constituents of
the atmosphere, and its state change is responsible for cloud and precipitation. The interaction of water vapour with
radiation is a crucial factor in climate change. Yet water vapour remains one of the most poorly characterized
meteorological parameters. Improving knowledge of the water vapour field is needed for a variety of atmospheric
applications and for studying the propagation of microwaves as well [2].
This paper is related to the ESA project METAWAVE (Mitigation of Electromagnetic Transmission errors induced by
Atmospheric WAter Vapour Effects), where the above mentioned problematic was deeply investigated by a large team
composed of SAR experts, meteorologists and atmospheric remote sensing experts. In the frame of such project the local
circulation in the urban area of Rome was studied using a high-resolution Mesoscale Model version 5 (MM5), a
microwave radiometer, and Global Positioning System (GPS) estimates of Integrated Water Vapour (IWV). Few
radioundings were also launched during a 15 days experiment and spaceborne estimates of IPWV were also collected for
comparison. The results of the multiplatform experiment are shortly summarized in the paper, together with a survey of
methods and tools we have exploited in order to predict maps of the water vapour field and related characteristics at

SAR Image Analysis, Modeling, and Techniques XI,


edited by Claudia Notarnicola, Simonetta Paloscia, Nazzareno Pierdicca, Proc. of SPIE Vol. 8179, 817912
© 2011 SPIE · CCC code: 0277-786X/11/$18 · doi: 10.1117/12.898756

Proc. of SPIE Vol. 8179 817912-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


resolution suitable for mitigating its effect on InSAR. However, the paper focus on a novel, still preliminary, attempt to
use the atmospheric signal into the InSAR phase for meteorological applications.
Such an attempt should face different problems, both on the SAR and meteorological side. First of all the interferometry
is intrinsically a differential technique, which provide phase difference at different SAR acquisition time. This makes it
difficult to inject this information into Numerical Weather Prediction (NWP) system which requires absolute geophysical
quantities. Moreover some other artefact can affect InSAR phase interferograms, such as linear trends due to orbit errors.
Then, the state of the art NWP software package are not always suitable to inject data at very high spatial resolution and
they are generally less sensitive to integrated information respect to meteorological quantity profiles.
This paper describes how we faced those problems in this preliminary exercise and the obtained results in predicting the
weather for a specific case study. The influence of the InSAR data assimilation was investigated and a comparison with
ground truth provided by a raingauge network was also performed to validate the effect.
*nazzareno.pierdicca@uniroma1.it; phone +39 06 44585411; fax +39 06 4742647; www.diet.uniroma1.it

2. MODELLING THE ATMOSPHERE


2.1 The role of NWP in the frame of the METAWAVE project
The primary providers of water vapour information potentially useful for InSAR data correction are the Numerical
Weather Prediction (NWP) systems. The increased computational power of computer machines allows for a
commensurate increase of the resolution of these models, which may become able to reproduce the physical phenomena
involved in water vapour formation and evolution. The fully compressible non-hydrostatic models allows for reaching
resolution in the order of 1 km or even higher. In this project, it has been used the PSU/NCAR mesoscale model (known
as MM5) that is a limited-area, nonhydrostatic, terrain-following sigma-coordinate model designed to simulate or predict
mesoscale atmospheric circulation.
Concerning NWP and InSAR, there are two open issues: the optimization of modelled high resolution water vapour to
correct InSAR interferograms and, eventually, the assimilation of InSAR path delay into the NWP model. For what
concerns the first problem, generally the water vapour produced by a high resolution NWP is a good approximation of
the real distribution and can be used to correct the radar interferogram, as was done in [3] using the UK Met Office
Unified Model. A limiting factor for high resolution NWP is the poor resolution of the initial condition. In this respect,
atmospheric Data Assimilation (DA) aims at incorporating observations into numerical weather prediction models with
maximum accuracy and efficiency and fills in the data gaps using physical, dynamical, and/or statistical information. In
the frame of the METAWAVE project, and with the objective to use NWP to correct InSAR, the envisaged approach
foresaw the assimilation of any observable (except InSAR observations) using the 3DVAR technique. We have followed
the approach used in [4]. The observations to be considered for this scope may include ground based networks, such as
GPS receiver slant-path delay or Zenith Total Delay (ZTD) estimates, or ground based microwave radiometers, as well
as spaceborne remote sensors, such as microwave or infrared radiometers.
Figures 1 and 2 compare the difference of integrated water vapour at two different days (February 3 and March 5, 1994)
predicted by MM5 (without any assimilation) and the APS map derived by InSAR acquisitions at the same days. The
most evident "signal" in the map is correlated to the topography of the area and the comparison looks fairly good in this
respect, as apparent for instance looking at the North East side of the image, covering the Apennine range. A less
successful comparison has been found when looking at the patterns associated to the atmospheric turbulent structures. A
relevant conclusion of the study indicates as NWP may provide reliable and useful information on the stratified
component of the path delay thought the atmosphere, which is dependent on the surface height and changes from time to
time. It is more difficult to correctly reproduce the turbulent component at the scale required for InSAR data correction.
As far as the possibility to use InSAR Atmospheric Phase Screen (APS) into NWP models is concerned, a major
difficulty is associated to the differential nature of the APS data (in time and space). APS's provide an insuperable high
resolution mapping of the atmospheric path delay changes with time over points over the earth surface which remain
steady, but they do not furnish absolute values. This is a relevant drawback if APS's are sampled at long time interval, as
in the case of the Envisat satellite, thus preventing for instance 4-D variational assimilation. It can be partially overcome,
and the high resolution APS information can be incorporated into the data used for initializing the NWP model. The way
we have faced this problem and the obtained results are described in the following sections.

Proc. of SPIE Vol. 8179 817912-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


Figure 1: Sample of MM5 integrated water vapour differences in the area of Rome (Italy) (February 3 and March 5, 1994 at 10:00
UTC) derived from ECMWF first guess. No assimilation of real observations was done.

Figure 2: APS from InSAR related to the acquisition of ERS-SAR on February 3, 1994 with respect to a master acquisition on March 5, 1994 at 9:55
UTC.

2.2 MM5 model configuration


The fifth generation NCAR (National Center for Atmospheric Research) and Pennsylvania State University (PSU)
mesoscale model MM5 was used in this study; this is a non hydrostatic model at primitive equations with a terrain
following vertical coordinate and multiple nesting capabilities [5]. Four two-way nested domains are used (Figure 3) to
enhance the resolution over the urban area of Rome. The mother domain covers most of western Mediterranean area; it is
centred at 41.5° N, 10.0° E and it has a spatial resolution of 27 km (D01 in Figure 3). The nested domains cover Central
Italy with a spatial resolution from 9 km, for domain 2, to 1 km for the inner one (D04 in Figure 3); this last encompasses
the city area and its surroundings (Lazio region), greatly overlapping the ERS satellite swath.

Figure 3: MM5 domains configuration. Domain D1 has resolution of 27km; D2 has resolution of 9km; D3 has resolution of
3km; D4 has resolution of 1km.

Proc. of SPIE Vol. 8179 817912-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


Sensitivity tests and previous studies [6] allowed for tailoring the optimal combination of physical parameterizations for
the numerical experiments. Therefore, the following model configuration has been used:
• 33 unequally spaced vertical levels, from the surface up to 100 hPa, with a higher resolution in the PBL;
• the MRF scheme for the PBL. This scheme is based on Troen-Mahrt representation of counter-gradient term
and the eddy viscosity profile in the well mixed PBL [7];
• the CLOUD radiation scheme for radiative transfer processes. This scheme accounts for both shortwave and
long wave interactions with explicit cloud and clear air [8];
• the Kain-Fritsch 2 cumulus convection parameterization is applied to domains 1 and 2 [9] [10]; whereas no
cumulus scheme is used for domains 3 and 4;
• the Reisner 2 scheme for microphysics; based on mixed-phase scheme, graupel and ice number concentration
prediction equations [11].
The European Centre for Medium-Range Weather Forecast (ECMWF) analysis at 0.25° horizontal resolution for
temperature, wind speed, relative humidity, and geopotential height are interpolated to the MM5 horizontal grid and to
sigma levels to produce the model initial and boundary conditions.
2.3 Data assimilation approach
Atmospheric data assimilation aims at incorporating observations into numerical weather prediction models with
maximum accuracy and efficiency and fills data gaps using physical, dynamical, and/or statistical information. In other
words, physical consistency, spatial and temporal coherency, noise suppression are three major concerns for atmospheric
data assimilation.
Briefly, the variational method is an optimization problem: 3DVAR enables the optimal initial condition to be found that
fits an initial background field, called first-guess, with scattered observations (see [12] and [13]). The best fit is obtained
minimizing the so called cost function J, defined as:
1 1
J = Jb + Jo = (xb − x)T B−1(xb − x) + (yo − H(xb ))T (E + F)−1(yo − H(xb )) (1)
2 2
where xb is the background term, yo is the generic observation, H(xb)is the corresponding value evaluated by the operator
H used to transform the gridded analysis to the observation space. The solution of this equation x=xa is the a posteriori
maximum likelihood estimate of the true state of the atmosphere; B, E and F are the covariance error matrices for the
background, the observations and the operator H, respectively. The method has been used to assimilate InSAR data of
Integrated Water Vapor (IWV) to the aim of improving initial conditions used to initialize MM5. The adopted
assimilation scheme is schematically shown in Figure 4.
The absolute value of Integrated Water Vapour has been retrieved from the differential APS provided by InSAR, as
described in the next section. Then, as far as the H operator in eq. (1) is concerned, the one implemented for GPS data
assimilation has been adopted, considering that InSAR APS is a measure of the atmospheric path delay, similarly to data
provided by a network of GPS receivers.

Figure 4: MM5 3DVAR scheme


In order to test the approach, we have considered one of the days of the experiment carried out in the framework of the
METAWAVE project (see [14]), specifically October 3, 2008. The descending SAR overpasses were considered, at

Proc. of SPIE Vol. 8179 817912-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


0930UTC in the morning, therefore to assimilate the data a background analyses (First Guess) is necessary at the same
time. Standard ECMWF analyses used to initialize model simulations are produced at synoptic times, only every 6 hours
(e.g., 0600 or 1200 UTC), which makes the time of the SAR passage too far. To solve this problem an ad hoc first guess
has been produced to allow us to assimilate APS data available at ENVISAT overpass time.
A technique similar to the ‘warm start’ approach has been applied to this purpose. A short term MM5 simulation starting
at 0600UTC of October 3 and ending at 0900UTC of the same day is performed. The output at 0900UTC of this
simulation is fed back to produce the input file to be used as first guess for the 3DVAR. This procedure allows for
having Initial Conditions (IC) at 0900UTC to initialize the forecast (MM5_VAR). Similarly, IC without assimilation of
InSAR data are produced to perform a control run (MM5_NOVAR). The assumption of a frozen atmosphere between
0900UTC and 0930UTC is applied for using the ENVISAT passage at 0930UTC at the simulation start time.
To assimilate the InSAR data, a background (B) and an observation error matrix (E) have to be defined as equation (1)
shows. The B matrix is related to the climatology of the event. To compute the B matrix the "NMC method" is
commonly used [15]. According to it, forecast error covariance is approximated using forecast difference statistics (e.g.,
differences between forecasts at T+48 and at T+24). The E matrix is build based on the assumption of a constant error
estimated into 0.05cm for IWV.

3. APS ASSIMILATION INTO NWP: METHODS AND RESULTS


3.1 Integrated water vapour maps from APS
The first problem to be faced when trying to assimilate InSAR Atmosphere Phase Screen (APS) data is translating the
differential information provided by InSAR into an absolute value of the Integrated Water Vapour (IWV). Since an
interferogram from SAR represents the path delay difference in both time and space and the InSAR APS at any given
time is obtained by removing noise and phase ramp, it is not trivial to derive absolute IWV from InSAR APS images.
Beside conversions from path delay into IWV, which can be assumed roughly a proportional factor, the basic idea is to
estimate the mean (average in time) distribution of IWV in a given area and within a time frame by relying on an
external source, such as IWV maps from Earth Observation missions.
In more details, consider that the interferometric phase represents a difference between two SAR overpasses (at time i
and j, respectively) referred to a single point x0 in the image. It contains a displacement term (subscript DISPL) and a
term due to a variable atmospheric path delay (subscript ATMO). Indicating by Δij the differentiation in time, by Φ the
phase and by L the path delay, we can write for a SAR interferogram:

Δ ij Φ(x) − Δ ij Φ(x0 ) = Δ ij Φ DISPL (x) − Δ ij L ATMO (x) − Δ ij Φ(x0 ) (2)
λ
where the atmospheric one-way path delay L and phase are simply related by two times the wave number k=2π/λ, being
λ the wavelength and considering the radar signal round trip. In clear sky conditions the path delay L has an hydrostatic
dry term and a term which is almost proportional to IWV through a factor Π (i.e., L=IWV/Π). It can be demonstrated
that this factor is in the order of 0.15, so that for instance 6 mm of path delay due to the water vapor is equivalent to
about 1 mm of IWV. For a steady earth surface (or a surface whose deformation can be modelled), from a sequence of
SAR images and differentiating with respect to a unique master at time j=M, the atmospheric delay (or Atmospheric
Phase Screen, APS) can be produced in each point x, with arbitrary unknown constant (consti) corresponding to the
phase difference of the reference point x0, i.e., consti =ΔΦiM(x0):
APS i (x) = Li (x) − LM (x) − consti
(3)
From equation (3) it is apparent that having the APS, in order to derive an absolute value of the path delay at a given
time i, which can be associated to the atmosphere an in particular to IWV (we assume here that the hydrostatic
component is constant within a typical SAR image, so that it is included into consti), one should know the atmosphere
when the master image was acquired (that is LM), still remaining a mean value ambiguity of consti. LM could be provided
by other sources, like for exampled Earth Observation satellites carrying aboard a radiometer sensitive to water vapour
content, either in the infrared or microwave spectral band. NWP outputs could be another source to provide LM and thus
estimate Li, but they cannot be considered when attempting to assimilate the APS since they would introduce statistically
dependent observations into the assimilation process. Anyhow, the error variance associated to the water vapour content

Proc. of SPIE Vol. 8179 817912-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


of the master acquisition provided by the external source (σ2EXT) can be very significant and it would directly reflect onto
the error variance of the estimated IWV (σ2IWV) at required time i and the APS intrinsic error (σ2APS) according to the
following:

σ IWV
2
= σ EXT
i
2
+ σ APS
2
i
(4)

If we have a suitable sequence of external data, in parallel to the sequence of SAR images used to estimate the APS, we
can adopt another strategy. By averaging many APS’s and relying on the external source to estimate the expected value
of the atmosphere signal instead of the master, one can reliably estimate the absolute atmospheric delay from APS:

Li (x) = APS i (x) − Mean[APS i (x)] + Mean[LEXT


i (x)] + const iEXT (5)

Note that for each time there is still an unknown constant which can be estimated from the external source in that
specific time, so that we have introduced the term constEXTi. The advantage to derive from the external source only the
mean (i.e., climatological) value is the less influence of the errors of the external source, since the variance of the
estimation of the mean is reduces by a factor equal to the number of available observations:

σ IWV
2
= σ MeanEXT
i
2
+ σ MeanAPS
2
+ σ APS
2
i
(6)

In our experiment, described in the sequel, we have adopted as external source the estimates of IWV provided by MERIS
aboard of Envisat, that are contemporaneous to the Advanced SAR (ASAR) acquisitions when considering the satellite
descending passes (daily passes). Some problems were associated to the presence of cloud cover which affect MERIS
estimations and to the presence of the orbital phase ramp into the interferograms, which has to be estimated for a better
evaluation of the bias term.
From the absolute path delay, the IWV can be derived and assimilated into NWP by the following:
IWV = ΠL = Π(λ / 4π )Φ (6)

Note that the APS's bring into the assimilation process only the space variable term of the require quantities, whereas
they still have an undetermined bias. In other word, they provide the high spatial frequencies of the path delay or IWV
horizontal field, being the low frequency components still furnished by the traditional sources of information (like for
instance ECMWF analysis or EO products).
3.2 Results of APS assimilation
A preliminary assessment of the impact of InSAR assimilation is performed by comparing the vertical distribution of
water vapour at the lower atmospheric layers with and without assimilating the InSAR data. A cross-section on the MM5
domain 4 (Figure 5) is taken crossing the urban area of Rome.

Figure 5: Left panel shows MM5 domain 4, the red line indicates the cross section to which the other panels refer. Central
and right panels a) and b) show water vapour vertical distribution along the cross section from the surface to 900hPa at
1100UTC on Oct. the 3rd. a) MM5 simulation without InSAR assimilation (MM5_NOVAR); b) MM5 simulation with
InSAR assimilation (MM5_VAR).

Proc. of SPIE Vol. 8179 817912-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


The vertical distribution of water vapour clearly shows a different distribution also two hours after the start time; panel
a) of Figure 5 shows the water vapour for the MM5 simulation without InSAR assimilation (MM5_NOVAR); the
comparison with panel b) considering InSAR assimilation (MM5_VAR) displays a fairly significant change in the water
vapour vertical structure. Indeed, a decrease of the water vapour content is produced by MM5_VAR at 950hPa at 10km
from the western boundary of the cross section. On the other hand, an increase is produced at approximately the same
level but at 15km from the western boundary.
A further comparison with radiosounding observations (RAOB) has been performed to understand the InSAR data
impact on temperature (T), humidity (RH) and wind vertical (WSP for speed and WDR for direction) profiles; soundings
launched from a site in the centre of Rome and from Pratica di Mare (41.65N, 12.43E, on the west cost of Lazio region)
have been taken into account. No evident influence of assimilation on the profiles extracted over Pratica di Mare has
been detected (not shown) allowing to asses an impact of the assimilation strictly around the area where data are
available (the Pratica di Mare site is out of the ENVISAT swath).
Two soundings were, instead, available over Rome at 1000UTC and 1230UTC; only small changes are shown in the
comparison between the MM5_VAR simulation and the RAOB at 1000UTC, with an increase of the bias1 for the
assimilated results both for relative humidity and for temperature at lower levels (not shown). The MM5_VAR increases
the overestimation of RH, such as the underestimation of T at low levels respect to the MM5_NOVAR simulation.
Biases in this case increase less than 10% for humidity and than 0.7 °C for temperature.

Figure 6: Comparison between radiosounding over Rome on October the 3rd 2008 at 1230UTC (red line) and MM5 profiles
with and without InSAR assimilation (yellow and blue solid lines respectively) at the same time and coordinates.
Dashed lines show biases between observed and simulated data. The three panels represent: relative humidity (top left),
dew point temperature (top right) and temperature (bottom); on each panel the minimum and maximum bias along the
whole profile is indicated.

1
The bias is calculated as a difference between the observed data and the simulated ones.

Proc. of SPIE Vol. 8179 817912-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


The comparison with radiosounding at 1230UTC (Figure 6, red lines), i.e. 3 hours and half after start time, shows an
evident impact of InSAR assimilation on vertical profiles. The most important changes are detected below 5 km of
height (Figure 6): InSAR data assimilation decreases the overestimation of relative humidity (Figure 6, top left panel,
yellow line) produced by MM5_NOVAR simulation (blue line). Between the layers 1500 and 3500 m the model (blue
and yellow solid lines) looses most of the variability measured by the RAOB (red line), with the MM5_VAR simulation
reproducing a sort of mean of the observed profile (yellow solid line) and reducing the error respect to observation. The
correction is indirectly induced also on dew point temperature profile as evident on top right panel of Figure 6. Changes
on temperature profiles (Figure 6, bottom panel) are less evident between MM5_NOVAR and MM5_VAR: anyway a
tendency to reduce errors in the simulated vertical profiles is found if InSAR data are assimilated into initial conditions
data.
The further comparison on Figure 7 between observed (red line) and simulated wind profiles (blue and yellow solid
lines) shows as the InSAR assimilation has an impact also on dynamical variables; the IWV assimilation (Figure 7,
yellow lines) improves both the speed and the direction wind simulation respect to MM5_NOVAR (blue line).

Figure 7 Comparison between radiosounding over Rome on Oct. the 3rd 2008 at 1230UTC (red line) and MM5 profiles with
and without InSAR assimilation (yellow and blue solid lines respectively) at the same time and coordinates. Dashed
lines show biases between observed and simulated data. The two panels represent wind speed (left panel) and direction
(right panel); on each panel the minimum and maximum bias along the whole profile is indicated.

Finally, to assess the impact of the InSAR data assimilation on the forecast, the most difficult field to predict is now
analyzed, that is the hourly precipitation. A comparison between the model hourly precipitation, 8 hours after the start
time with (MM5_SAR) and without (MM5_NOSAR) InSAR data assimilation is performed. Furthermore, the outputs
are compared with observed rainfall to better verify the impact.
The MM5_NOSAR clearly shows a precipitating cell in the south east corner of the domain, producing more than
18mm/h (Figure 8A); a similar structure is produced by MM5_SAR (Figure 8B), but a reduction of the rainfall is clearly
produced by the InSAR assimilation mitigating the model error. The comparison with the observation (Figure 8C)
clearly suggests that the cell position is well reproduced by both MM5 simulations, but a 3 hour anticipation is found.
Moreover, an overestimation by MM5_NOSAR is produced: the maximum observed precipitation is approximately
16mm/h (Figure 8C). Moreover, both MM5 simulations miss the precipitation in the eastern part of the central domain.
The time shift is a well known MM5 structural problem that barely would be corrected by the InSAR assimilation; on the
contrary, the IWV assimilation clearly reduces the rainfall overestimation by lowering it to 16mm/h (Figure 8B).

Proc. of SPIE Vol. 8179 817912-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


Figure 8: One hour accumulated precipitation on Oct. the 3rd for: A) MM5 without InSAR assimilation between 1600UTC
and 1700UTC; B) MM5 with InSAR assimilation for the same hours as for a; C) observed between 1900Uct and
2000UTC.

4. CONCLUSIONS
The paper gave a short overview of the contribution of Numerical Weather Prediction systems in the frame of SAR
interferometry, as exploited in the METAWAVE project. NWP can help to mitigate the disturbing effect of the
atmosphere, and especially water vapour changes, on InSAR. The main focus of the paper was however a novel but still
very preliminary investigation on the possibility to use the Atmospheric Phase Screen produced by multi-pass InSAR
technique, and specifically the Permanent Scatterer technique, to provide useful information on the atmosphere to be
assimilated into NWP systems. The problem of the APS differential nature, hampering a direct assimilation by 3DVAR
techniques, has been faced and solved by relying on temporal series of water vapour maps from external sources. It has
finally demonstrated that by using APS the NWP forecasts can be improved, albeit slightly. This has been demonstrated
at the moment for a single case study and more must be done to draw final conclusions. However, we also expect that the
APS assimilation approach can be further improved, thus exploiting the high spatial resolution of the APS even better. In
particular the following could be further investigated:
ƒ 3DVAR adjustment of the meteorological fields to the large scale may remove InSAR high resolved features. More
effective techniques have to be considered, including Kalman filter.
ƒ A higher resolution model would allow one to better retain the high frequency structures of the InSAR data to
correct the forecast.
ƒ The warm start used for initializing the forecast at the time of the Envisat passage should envisage the use of local
observations to ensure a reliable Initial Condition.

5. ACKNOWLEDGMENT
The project was funded by the European Space Agency under contract N. ESTEC-21207/07/NL/HE. We would like to
thank all the METAWAVE team for their valuable contributions and discussions.

Proc. of SPIE Vol. 8179 817912-9

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms


REFERENCES

[1] Rocca, F., “Modeling Interferogram Stacks”, IEEE Trans. Geosci. Rem. Sens, 2007, in Vol. 45, Issue 10, pp. 3289–
3299, October 2007.
[2] Peixoto J.P., A.H. Oort, [Physics of climate], American Institute of Physics, 1993
[3] Wadge G., Zhu M., Holley R. “Forward atmospheric modelling to correct for water vapour delays to InSAR ground
deformation measurements over mountains”, Geophysical Research Abstracts, Vol. 8, 07384, 2006
[4] Faccani C., D. Cimini, R. Ferretti, F.S. Marzano, and A.C. Taramasso, “3DVAR assimilation of SSM/I data over
the sea for the IOP2 MAP case”, Adv. in Geosci., vol. 2, p. 229-235, 2005.
[5] Dudhia, J., “A nonhydrostatic version of the Penn State-NCAR Mesoscale Model: validation tests and simulation of
an atlantic cyclone and cold front”, Mon. Wea. Rev., 129, 1493-1513 (1993).
[6] Ferretti, R., Mastrantonio, G., Argentini, S., Santoleri, L., and Viola, A., “A model-aided investigation of winter
thermally driven circulation in the Italian Tyrrhenian coast for a case study”, J. Geophys. Res., 108 (D24), 4777-4792
(2003).
[7] Hong, S.-Y., Pan, H.-L., “Nonlocal boundary layer vertical diffusion in medium-range forecast model”, Mon. Wea.
Rev., 124, 2322-2339 (1996).
[8] Stephens, G. L., “The parametrization of radiation for numerical Weather prediction and climate models”, Mon.
Wea. Rev., 112, 826-867 (1984).
[9] Kain, J.S., Fritsch, J.M., “Convective parameterization for mesoscale models: the Kain-Fritsch scheme”, Meteorol.
Monogr., 24 (46), 165–170 (1993).
[10] Kain, J.S., “The Kain-Fritsch convective parameterization: an update”, J. Appl. Meteorol., 43 (1), 170–181, (2004).
[11] Reisner, J., Rasmussen, R., Bruintjes, R., “Explicit forecasting of supercooled liquid water in winter storms using
the MM5 mesoscale model”, Q. J. R. Meteorol. Soc., 124, 1071–1107, (1998).
[12] Courtier P., Anderson E., Heckley W., Pailleux J., Vasiljeric D., Hamrud M., Hollingsworth A., Rabier F., and
Fisher M., "The ECMWF implementation of three-dimensional variational assimilation (3DVAR). I: Formulation", Q. J.
R. Meteorol. Soc., 124, 1783-1807, 1998
[13] Lorenc, A.C., Ballard S.P., Bell R.S., Ingleby N.B., Andrews P.L.F., Barker D.M., Bray J.R., Clayton A.M., Dalby
T., Li D., Payne T.J. and Saunders F.W., "The Met Office global three-dimensional variational data assimilation
scheme", Q. J. R. Meteorol. Soc., 126, 2991-3012, 2000.
[14] Pierdicca N., F. Rocca, B. Rommen, P. Basili, S. Bonafoni, D. Cimini, P. Ciotti, F. Consalvi, R. Ferretti, W. Foster,
F.S. Marzano, V. Mattioli, A. Mazzoni, M. Montopoli, R. Notarpietro, S. Padmanabhan, D. Perissin, E. Pichelli, S.
Reising, S. Sahoo, and G. Venuti, “Atmospheric water vapor effects on spaceborne interferometric SAR imaging:
comparison with groundbased measurements and meteorological model simulations at different scales”, Proc. IGARSS-
2009, CapeTown (South Africa), July 13-17, 2009.
[15] Parrish, D. F., and J. C. Derber, “The National Meteorological Center’s Spectral Statistical Interpolation analysis
system”, Mon. Wea. Rev., 120, 1747–1763 (1992).

Proc. of SPIE Vol. 8179 817912-10

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 11/21/2013 Terms of Use: http://spiedl.org/terms

You might also like