Applying DINEOF Algorithm On Cloudy Sea-Surface Temperature Satellite

Data Over The Eastern Mediterranean Sea

Andreas Nikolaidis*(a,b), Georgios Georgiou(b), Diofantos Hadjimitsis(a) and Evangelos Akylas(a)

a. Department of Civil Engineering and Geomatics, Cyprus University of Technology, 30
Archbishop Kyprianou Str.,3036, Limassol, Cyprus
b. Oceanography Centre ,University of Cyprus, PO BOX 20537,1678, Nicosia, Cyprus

ABSTRACT

Data Interpolating Empirical Orthogonal Functions (DINEOF) is a special technique which is based on Empirical
Orthogonal Functions (EOF) in order to reconstruct missing data from satellite images. It is an innovative method,
especially useful for filling in missing data from geophysical fields. Interesting examples could be clouds in sea-surface
temperature (SST). Past studies have shown that filtering the temporal covariance matrix allows to reduce spurious
variability and therefore a more realistic data reconstruction can be obtained. There is also provision in the estimation of
the error covariance of the reconstruction of the data. Moreover, the error fields can be obtained with some calculation
rearrangement. Error fields reflect the data-coverage structure and furthermore the covariance of the physical fields.
Successful experiments on the Western Mediterranean encouraged the extension of the application of the method
eastwards using similar experimental implementation. The present study summarizes the experimental work done, the
implementation of the method and its ability in reconstructing the sea-surface temperature field over the Eastern
Mediterranean basin, and specifically over Levantine sea and Cyprus.

Keywords: Remote Sensing, sea-surface temperature, DINEOF, data reconstruction

1. INTRODUCTION
Let us consider a basin filled with water whose surface is oscillating. To describe this surface wave we should note the
position of every water particle at every moment, which rapidly generates a huge amount of data. During the last decades
there has been effort to make possible to find simple (usually referred as ‘empirical’) functions, which allow such a
system to be described simply. The most common approach is of a spatial function which gives the form of the wave at a
given moment and a temporal function which characterizes the variation of the wave over time. The determining of these
functions then allows one to bypass an accumulation of data. In geophysics, a well known and used method (introduced
by Lorentz at 50’s) for this, is the geographically weighted Principal Component Analysis (PCA), which normally is
referred to, as empirical orthogonal functions (EOFs). Specifically, PCA is defined as an orthogonal linear
transformation that transforms the data to a new coordinate system such that the greatest variance by any projection of
the data comes to lie on the first coordinate (called the first principal component), the second greatest variance on the
second coordinate, and so on. EOFs can present different physical meanings, like coherent spatial patterns with
maximum variance, modes of energy, even just convenient mathematical abstractions, depending on the nature of the
problem. During 2003, Backers and Rixen demonstrated an innovative EOF method that does not needed any a priori
knowledge of a correlation function or correlation length, of the data. Due to the iterative nature of the algorithm, any
inhomogenity or non-isotropic behavior is automatically taken into account, generating an interpolation effect, hence the
name Data Interpolating Empirical Orthogonal Functions (DINEOF).An adaptation to handle the large data sets (typical
of satellite imagery) can be found in Alvera-Azcárate et al. (2005).

Filling the gaps, generated by the existence of clouds, rain, or simply due to incomplete track coverage of upper
atmospheric layers, is one of the most common problems, faced during the processing of satellite data. Many methods
have been tried out over the years for solving this problem, with different results, regarding the field of application and
the expertise of the scientists involved. Notable examples, are the Data-Interpolating Variational Analysis (DIVA) that

First International Conference on Remote Sensing and Geoinformation of the Environment (RSCy2013), edited by
Diofantos G. Hadjimitsis, Kyriacos Themistocleous, Silas Michaelides, George Papadavid, Proc. of SPIE
Vol. 8795, 87950L · © 2013 SPIE · CCC code: 0277-786X/13/$18 · doi: 10.1117/12.2029085

Proc. of SPIE Vol. 8795 87950L-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 allows the spatial interpolation of data (analysis) in an optimal way, and the optimal interpolation method (OI) (C.
Troupin et al., 2012). DINEOF method is simply faster.

Optimal Interpolation is the best known and the most classical method. The main problem is the demanding calculating
time. For a typical ensemble of data, DINEOF is an innovative method, 30 times more rapid than OI. This acceleration is
a direct consequence of the different statistical methodology between the two different approaches. In particular
DINEOF constitutes a procedure that fills gaps by iteratively decomposing the data field via Singular Value
Decomposition (SVD) until a best solution is found as compared to a subset of reference values (non-gaps). This is done
by progressively including more EOFs in the reconstruction of the guppy locations until minimization of error
converges.

In this paper, we apply the DINEOF method in order to reconstruct a fully filled satellite derived sea surface temperature
field, for the area of Eastern Mediterranean, Levantine sea and Cyprus coasts. SST is a physical parameter commonly
used in most oceanographic and meteorological applications and among others is a principal factor for relevant
arithmetic model calibration, assimilation and initialization. DINEOF is applied for the first time over the specific area of
Eastern Mediterranean Sea, a step which is crucial, both for the research as for the operational implementation of the
method.
The structure of this work is as follows. In section 2 a description of the DINEOF method and of the corresponding
algorithm is given. Section 3 describes the application of the method in Eastern Mediterranean Sea and in section 4 the
results of the implantation are analyzed and discussed.

2. DESCRIPTION
By using EOF method one is able to identify a set of orthogonal spatial modes, such that, when ordered, each successive
eigenvector explains the maximum amount possible of the remaining variance in the data. Each eigenvector pattern is
associated with a series of time coefficients that describe the time evolution of the particular spatial mode. The
eigenvector patterns that account for a large fraction of the variance are, in general, considered to be physically
meaningful and connected with important ‘centers of action’.

The EOFs can be regarded as a kind of eigenvectors, which are aligned so that the leading EOFs describe the spatially
coherent pattern that maximizes its variance. EOFs are often used as functional basis (a new set of axes or reference
frame), providing a convenient method for studying the spatial and temporal variability of long time series of data, over
large areas. The method splits the temporal variance of the data into orthogonal spatial patterns called empirical
eigenvectors. The EOF analysis may be thought of as being analogous to data reconstruction based on Fourier transforms
(FT), in the sense that both produce series (vectors) which form an orthogonal basis. In the following, we briefly
summarize the mathematical principles of the algorithm.

2.1 The algorithm

Recalling the description given by A. Alvera-Azcarate et al.,(2009), let us consider a matrix X with size M×N, with M
the spatial size and N the temporal size. Anomalies are computed and the missing data are initialized to the mean (i.e. to
a zero anomaly). Initially the most dominant EOF mode of this matrix is obtained, and the missing data are calculated by

T
X m,m =USV (1)

where m=1...M and n=1...N. U represents the spatial EOF modes, with dimension m x r, VT are the temporal modes, with
dimension n x r and S are the singular values, with dimension r x r. The value r <= min(m,n) is the rank of the X matrix.
For the reconstruction of X, only the most significant spatial and temporal EOF is used. The new estimation (of X) for
the missing data is reintroduced in the data matrix, and the EOF mode calculation is repeated. This process is continued
in a successive way until the convergence of the missing values, and then the EOF modes calculated are increased to
two, then to three, etc. The EOF mode calculation is succeeded using a Lanczos solver provided by the ARPACK free
software (Lehoucq et al., 1997).A major improvement of the latest DINEOF version, is the ability of outlier detection,
provided by the ratio between the analysis residuals and their expected standard deviation,

Proc. of SPIE Vol. 8795 87950L-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 X ia − X io
Oi = , for non-missing X iO (2)
Δi

where i=1...m is the spatial index, X ia is the new value, as created by DINEOF, X io is the original given value and Δi is
the expected misfit, calculated by

2
Δi = μeff − ∑ Ei,k2 (3)

2
In the above, N is the number of EOFs used from the algorithm for the reconstruction, while k=1..N, Parameter μeff is an
estimation of the average noise of the initial field, obtained as a cross-validation error. The expected error Ei,k of each i-
position is calculated by

E = Lp Sc with C = Sc ScT (4)

where the Lp is constructed by m x N columns of the spatial EOFs multiplied by the corresponding singular values and Sc
is a N x N factorization (Cholesky factorization) of C. More details on the mathematical procedure can be found on
A.Alverta et. al. (2012). In Figure 1 (A.Alverta et. al, 2010), the outline of the implementation of the DINEOF algorithm
is presented.

DINEOF
1st: Demeaned matrix missing data flagged and set to zero
Some data are set aside for cross -validation

2nd: EOF decomposition with N =1 EOF \

Calculate missing vakies

=E0,(u,),1,,
Improved guess for missing values r-1

Convergence ( best value for missing data wrth 1 EOF

ctoss validation: error

EOF decomposition with N =2 EOFs

Calculate missing values

Improved guess for missing values

Then we repeat with N= 3 EOFs

N EOFs and so on...
Figure 1. DINEOF algorithm iterative scheme (A. Alvera-Azcárate, A. Barth, J.-M. Beckers,GHER – University de Liège, FNRS –
National Fund for the Scientific Research, Belgium, 2010).

The optimal number of EOFs needed to calculate the values at the missing locations is determined by cross-validation: a
small percentage of valid data (typically 1% of the total data) are initially set apart and flagged as missing. Once
convergence is reached for a given number of EOF modes, a root mean square error is calculated between the newly
obtained estimate and the initial data set. The number of modes that minimizes this error is considered as optimal. It is
notable that not all modes need to be calculated, as one can observe that when the error increases steadily for 3
consecutive modes, a minimum has been reached (A. Alvera-Azcárate et al., 2012). The optimal number of EOF modes
retained are calculated by cross-validation (i.e., a few valid data are set aside and the error of the reconstruction is

Proc. of SPIE Vol. 8795 87950L-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 Z (level) : 0
TIME : 09 -JAN -2013 10:00 DATA SET: InputFile TIME : 09 -JAN -2013 10:00 DATA SET: DineofOutput

21
20 20
39.0°N -
19 19
18 18
17 37.0°N - 17
18 16
15 15
14 14
g 35.0°N
13
12 12
33.0N - tt

10 'C

31.0N -
22 °E 26't 309E 34°E 22t 26 °E 309E 34°
9a11aue LONGITUDE CONTOUR LONGITUDE
cONraR oON10uN:

Initial data SST (oC) Result data SST (oC)

Z (level) : 0
MAE : 09-JAN-2013 20:00 DATA SET: InputFile TIME : 09 -JAN -2013 20:00 DATA SET: DineofOutput

21
20 20
39.0°N -
19 19
18 - IS
17 37.0°N - 17
16 18
15
14 F 35.0°N - 14
13 5 13
12 12
11 33.0 °N - 11
10 10
9
31.0 °N -
8
22.E 28 °E 30 °E 34.E
COKISING LONGITUDE
canoun:
Initial data SST (oC) Result data SST (oC)

Z (level) : O
TIME 10 -JAN -2013 10:00 DATA SET: Inputfile TIME : 10-JAN-2013 10:00 DATA SET: DineofOutput

21 21
20 20
19
le
17
le
15
14
13
12

10

22E 28E 30E 22E 26 E 30E 3E

OONTOING LONGRUDE LONGITUDE
CONTOUPt

Initial data SST (oC) Result data SST (oC)

DIME : 10- JAN -2013 20:00 DATA SEI: Inpulfile TILE : 10-UW -2013 20:00 DATA SET: DinaofOutput

39.01 39.0N

37.0,1 37.0N

3s 35.0N

33.0N
10
9
31.0, 31.0N 8
34 2eE
LONGITUDE LONGITUDE
CONTOUIO eeloloMe
Initial data SST (oC) Result data SST (oC)
Figure 3. Original and reconstructed, using DINEOF, SST data for the Eastern Mediterranean, from 9 - 12 /1/2013.

Proc. of SPIE Vol. 8795 87950L-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 Z (level) : 0
TIME : 09 -JAN -2013 10:00 DATA SET: InputFile TIME : 09 -JAN -2013 10:00 DATA SET: DineofOutput

21
20 20
39.0°N -
19 19
18 18
17 37.0°N - 17
18 16
15 15
14 14
g 35.0°N
13
12 12
33.0N - tt

10 'C

31.0N -
22 °E 26't 309E 34°E 22t 26 °E 309E 34°
9a11aue LONGITUDE CONTOUR LONGITUDE
cONraR oON10uN:

Initial data SST (oC) Result data SST (oC)

Z (level) : 0
MAE : 09-JAN-2013 20:00 DATA SET: InputFile TIME : 09 -JAN -2013 20:00 DATA SET: DineofOutput

21
20 20
39.0°N -
19 19
18 - IS
17 37.0°N - 17
16 18
15
14 F 35.0°N - 14
13 5 13
12 12
11 33.0 °N - 11
10 10
9
31.0 °N -
8
22.E 28 °E 30 °E 34.E
COKISING LONGITUDE
canoun:
Initial data SST (oC) Result data SST (oC)

Z (level) : O
TIME 10 -JAN -2013 10:00 DATA SET: Inputfile TIME : 10-JAN-2013 10:00 DATA SET: DineofOutput

21 21
20 20
19
le
17
le
15
14
13
12

10

22E 28E 30E 22E 26 E 30E 3E

OONTOING LONGRUDE LONGITUDE
CONTOUPt

Initial data SST (oC) Result data SST (oC)

DIME : 10- JAN -2013 20:00 DATA SEI: Inpulfile TILE : 10-UW -2013 20:00 DATA SET: DinaofOutput

39.01 39.0N

37.0,1 37.0N

3s 35.0N

33.0N
10
9
31.0, 31.0N 8
34 2eE
LONGITUDE LONGITUDE
CONTOUIO eeloloMe
Initial data SST (oC) Result data SST (oC)
Figure 3. Original and reconstructed, using DINEOF, SST data for the Eastern Mediterranean, from 9 - 12 /1/2013.

Proc. of SPIE Vol. 8795 87950L-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 Z (level) : o
TIME : 11 -JAN -2013 10:00 DATA SET: InputFile TIME : 11 -JAN -2013 10:00 DATA SET: DineofOutput

22E 264 34 °E 2e °E 30, 344

LONGRUDE LONGITUDE
COMOUlt
Initial data SST (oC) Result data SST (oC)
Z (level) : O
TIME 11 -JAN -2013 20:00 DATA SET: InpuW ile TIME : 11 -JAN -2013 20:00 DATA SET: DineofOutput

28S 30E 26°E E4

LONGITUDE LONGITUDE

Initial data SST (oC) Result data SST (oC)

Z (level) : 0
TIME : 12 -JAN -2013 10:00 DATA SET: Inputfile TIME : 12 -JAN -2013 10:00 DATA SET: DlneofOutput

21 21
20 20
19 19
18 18
17 17
16 16
15 15
14 14
13
12
11 11
10 10
9 9
8
2 °E 20 E 302 342 22°E 262 30°E 34
COMMA: LONGITUDE 001110e1í LONGITUDE

Initial data SST (oC) Result data SST (oC)

Z (level) : 0
T1ME : 12-JAN-2013 20:00 DATA SET: Input'Ile TIME : 12 -JAN -2013 20:00 DATA SET: DineofOutput

21
20
39.0°4 -
19
18

37.0°N - 17
16
-
15
35 0N 14
13
12
33.0°4 - 11

10
9
31.0~1 -
e
22 °E 262 30 °E 342 222 26 °E 30 °E 34
LONGITUDE LONGITUDE
00100914

Initial data SST (oC) Result data SST (oC)

Figure 3. continue

Proc. of SPIE Vol. 8795 87950L-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 4. VALIDATION OF RESULTS
In order to estimate and test the successful implementation of the presented application, a comparison of the recovered
results against the outputs of oceanographic models providing regional now-casting real time information is further
performed. Two of the most important relevant models, are:

(1) the CYCOFOS, the Cyprus Operational Oceanographic System, providing very accurate data in high-resolution
around Cyprus and Levantine sea, and
(2) the U.S. Navy Coastal Oceanographic Model (NCOM), which covers a wider area, in less detail.

Both models are used in an operational basis by the Cyprus Government and Research Institutes around the world, as
well as the US Navy. In Figure 4, we outline the testing procedure. The initially cloudy image (a) is regenerated with
DINEOF (b), as explained, the respective model’s output for the same time is selected (c) and the differences between
(b) and (c) at all points are computed (d). The same procedure is followed for each time step for both models.(Figure 4)
Z (level) : 0
TIME : 09-JAN-2013 20:00 DATA SET: InputNle TIME : 09-JAN -2013 20:00 DATA SET: Dlneof0utput

20.5
20
19.5
19

18.5
18
7.5
7
'6.5
'6
55

31.0E 32.0E 33.0 °E 34.00E 35.0E 36.0E 31.0°E 32.0 °E 33.0Æ 34.0 35.0°E 36.0 °F
COMA LONGITUDE 34WTOOR: LONGITUDE
CONTOUR: O011TWR:

Initial data SST (oC) Resulted data (oC)

36.0°N -
z:0
TIME : 09 -JAN -2013 20:00

21

20.5
20
36.01N -
z (leel)
11Á(E :
v 69-0

lllllll
JAf1-2013 20:00

3
19.5
2
19
1
18.5
,2 35.0M7 - 18 35.0°N - 0
17.5
g
g
n 2
34.0N - 16.5 34.0°N -
3

1TÌ
4
15.5
33.0°N - 1
1 1 1
1 1 1
15 33.0°N 5
31.0 32.0°E 33.0 °E 34.011 35.0 36.011 31.08E 32.0 11 33.0 34.0 35.01E 313.0°E
cONT01Rí LONGITUDE CONTOUR: LONGITUDE
CONTOUR:

CYCOFOS Model SST (oC) Positive /Negative differences (oC)

Figure 4. Outline of the validation procedure for an arbitrary time-step of Eastern Mediterranean SST data.

Firstly we present the comparison against the less detailed model US NCOM, in Figure 5. The results are for the
difference between the model and the recovered SST values. For all the time-steps, it is evident that the recovered data
are partially overestimated compared to the model. Although over the most of the studied area the differences remained
relatively small, (less than 1 oC degree), near the Eastern Mediterranean coastal area the difference peaks around 4 oC
degrees. This difference could be wrongly attributed to the fact that the specific area lucks original SST information as
shown in Figure 3. In other words this difference could be due to DINEOF’s inaccuracies. The truth behind is completely
at the opposite direction.
The US NCOM model, covers the specific area in low resolution resulting in less accurate results. In fact figure 6, which
focuses in the vicinity of Cyprus through CYCOFOS operational model, proves the high level of agreement between the
model results and the DINEOF outcomes. As mentioned, CYCOFOS is a source of very accurate data, given in high-
resolution for Cyprus coasts and Levantine sea. The observed agreement, with absolute differences of less than 1 oC
degree, is further certified by respective in-situ measurements from Paphos tide-gauge station (data available from
CYCOFOS web site), revealing the same behavior. The overall success suggests the existence of a strong basis available
for further testing and application, including more sophisticated procedures, in the direction of successfully
implementing a continuous recovery of SST data from satellite images over Eastern Mediterranean. It is expected, that in
a future step such results could serve as input for assimilation processes and model calibrations.

Proc. of SPIE Vol. 8795 87950L-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 Z (level) : 0 Z (level) : 0
TIME : 09 -JAN -2013 10:00 TIME : 09-JAN -2013 20:00

39.01% 39.0.4

33.09

31.094

229E 281E 309E 3414 229E 28% 3014 3414

LONGITUDE OONNIIR: LONGITUDE
Positive/Negative differences with US NCOM model Posltive/Negntive differences with US NCOM model

Z 2 level) : 0
TIA(E:1Ó-J11-2013 10:00 li E : 10 -JAN -2013 20:00

3
39.0°4 4 39.04
3
37.011 2 37.0°N

E
0
g35.019
5 35.0'N 1

2
33.011 33.0°N
3

31.014 31.0°N
5
221E 2814 30E 3414 22 2814 3014 3414
COMM LONGITUDE COteWR: LONGITUDE
Positive /Negative differences with US NCOM model Positive/Negative differences with US NCOM model

Z (level) : O Z (level) : 0
TIME : 11 -JAN -2013 10:00 TIME : 11-JAN -2013 20:00

39.014 - 4

33.014-
3
4
31.014 - 4
I I I I 1 1 I 1 I 1 1 1 1

22E 261E 30E 341E 2215 26E 3015 3415

cosses: LONGITUDE e0RNI1R: LONGITUDE
Positive/Negative differences with US NCOM model Positive/Negative differences with US NCOM model

Z (level) : 0
11Á1E0: 1)2 JAN -2013 10:00 TIME : 12-JAN -2013 20:00

5 3

4 4
3 3

2 2

1 1

0 _ 0
1

3 3

22 °E 2I PE 3015 3415 2215 2615 3015 3415

cowrout LONGITUDE e01eWR: LONGITUDE
Positive /Negative differences with US NCOM model Positive/Negative differences with US NCOM model

Figure 5. Computed SST differences between the present DINEOF application and the US NCOM model data.

Proc. of SPIE Vol. 8795 87950L-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 Z (level) : O Z (level) : O
TIME : 09 -JAN -2013 10:00 TIME 09-JAN -2013 20:00

38.0.14-

g 30.01 O a 35.N -
5 - 1 5
2 2
3 {.O11 - 34.014 -
3

33.0.9
31 OE S20E 3.1.E 34.OÆ 35.0E
:
36.0E
-5 33.14
SLOE 32.0E 33.0Æ 34.0E 15.E 36.0E
4

coleoue LONGITUDE CON1w0. LONGITUDE

Positive/Negative differences with CYCOFOS model Positive/Negative differences with CYCOFOS model

Z (level) : O Z (level) : O
TIME 10 -JAN -2013 10:00 OJAN-2013 20:00

5
4
36.014-

pJS.ON - 0

t 2 _2
34.0N -

33.0.11 5
31.0Æ 320E 33.0E 34.0 35.0 38.0E 31.0°E 32.0°E 33.0 34.0.E 35.0°E 30.0°E
COMO/At LONGITUDE LONGITUDE
Positive/Negative differences with CYCOFOS model Positive/Negative differences with CYCOFOS model

Z (Tavel) : 0 Z pavea : G
TIME : 11 -JAN -2013 10:00 TIME : 11 -JAN -2013 20:00

5
4

j
2
3í.0N -
3
4
33.0N 5
31.01E 3201E 33.E 34.0E 35.0E 313.0E 32.0.E 33.0 34.0E 35.0E 36.0.E
Ua. LONGITUDE LONGITUDE
PoslUve /Negative differences with CYCOPOS model Positive /Negative differences with CYCOFOS model

Z bvel) : O Z (Tavel :
TI : 12-JAN-201310:00 TE: 12-) JOAN-2013 20:00

36.0MI - 36.N -
2
-
E
235.01N-
3 -
1

2
z35.0N-
Aire-
34.01N - 34.0n-
3

34'0.1N 1 I 1 I I 33.0N 1

0
1

31.04 3201E 33.0E 34.04 35.0E 30.0E 31.0 E 32.E 33.0 4 34.04 35.0E 38.0
Min .: LONGTUDE CONTOUR. LONGITUDE
Positive /Negative differences with CYCOPOS model Positive/Negative differences with CYCOPOS model

Figure 6. Computed SST differences between the present DINEOF application and the CYCOFOS model data.

Proc. of SPIE Vol. 8795 87950L-9

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms

 5. CONCLUSIONS
Data Interpolating Empirical Orthogonal Functions is a special technique, for the reconstruction of missing data from
satellite images. It is an innovative method, especially useful for filling in missing data from geophysical fields. SST is a
physical parameter commonly used in most oceanographic and meteorological applications and among others is a
principal factor for relevant arithmetic model calibration, assimilation and initialization. Past works reported successful
implementation of DINEOF in recovering missing SST data over the Western Mediterranean area. In this paper, the
DINEOF method was applied in order to recover a fully filled satellite derived sea surface temperature field, for the first
time over Eastern Mediterranean, Levantine sea and Cyprus coasts. This is a crucial step, both for the research as for the
operational implementation of the method. The results are really impressive, recovering data from very cloudy images
with remarkable efficiency. The comparison against widely used highly accurate simulations, revealed close agreement
with differences not exceeding 1 oC degree. The same quantitative agreement holds also against in-situ measurments
from Paphos tide-gauge station. The overall success suggests the existence of a strong basis available for further testing
and application. Besides the impressive results of the DINEOF algorithm, in recovering SST data the application of the
method does not stop necessarily onto satellite processing. Sirjacobs et al. (2011), effectively applied the DINEOF
method to the case of suspended matter in the North Sea. Also exists, a simplified R-project package, specially designed
for adoption into this statistical environment. Due to the simplicity of the method, and to the good quality of the results,
it is easy to predict a rapid propagation of the DINEOF usage, guiding to a better quality of results on data with missing
values.

REFERENCES

[1] Beckers, Jean-Marie, and M. Rixen. "EOF Calculations and Data Filling from Incomplete Oceanographic Datasets.",
Journal of Atmospheric and Oceanic Technology 20.12 1839-1856 (2003):.
[2] Sirjacobs D., Alvera-Azcárate A., Barth A., Lacroix G., Park Y., Nechad B., Ruddick K., Beckers J.-M., Cloud filling
of ocean colour and sea surface temperature remote sensing products over the Southern North Sea by the Data
Interpolating Empirical Orthogonal Functions methodology, Journal of Sea Research, Volume 65 (1), pp. 114-130
(2011)
[3] Peixoto, J.P., and A.H. Oort,: Physics of Climate. American Institute of Physics New York,NASA GES DISC
(Goddard Earth Sciences Data and Information Services Center) (1992)
http://disc.sci.gsfc.nasa.gov/precipitation/additional/science-focus/TDST_SCI/statis_EOF.shtml#100
[4] A. Alvera-Azcárate, A. Barth, D. Sirjacobs, J.-M. Beckers. Enhancing temporal correlations in EOF expansions for
the reconstruction of missing data using DINEOF. Ocean Science, 5, 475-485, (2009).
[5] A. Alvera-Azcárate, A. Barth, D. Sirjacobs, F.Lenartz, J.-M. Beckers. Data Interpolating Empirical Orthogonal
Functions (DINEOF): a tool for geophysical data analyses. Mediterranean Marine Science, (2010),
[6] A. Alvera-Azcárate, A. Barth, J.-M. Beckers, and R. H. Weisberg. Multivariate reconstruction of missing data in sea
surface temperature, chlorophyll and wind satellite fields. Journal of Geophysical Research, 112:C03008,
doi:10.1029/2006JC003660 (2007)
[7] J.-M. Beckers, A. Barth, and A. Alvera-Azcarate. DINEOF reconstruction of clouded images including error maps.
Application to the Sea Surface Temperature around Corsican Island. Ocean Science, 2(2):183-199, (2006)
[8] A. Alvera-Azcárate, A. Barth, M. Rixen, and J.-M. Beckers. Reconstruction of incomplete oceanographic data sets
using Empirical Orthogonal Functions. Application to the Adriatic Sea. Ocean Modelling, 9:325-346, (2005)
[9] J.-M. Beckers and M. Rixen. EOF calculations and data filling from incomplete oceanographic data sets. Journal of
Atmospheric and Oceanic Technology, 20(12):1839-1856, (2003).

*nikolaidis.andreas@ucy.ac.cy; phone +365 22893991; Oceanography Centre, University of Cyprus

Proc. of SPIE Vol. 8795 87950L-10

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 05/12/2015 Terms of Use: http://spiedl.org/terms