Geophysical Research AbstractsVol. 21, EGU2019-2177, 2019EGU General Assembly 2019© Author(s) 2018. CC Attribution 4.0 license.
Three-Dimensional Reconstruction of Ocean Current Circulation fromCoastal Marine Observations: Challenges and MethodsIvan Manso (1), Erick Fredj (2), Gabriel Jordà (3), Maristella Berta (4), Annalisa Griffa (4), Ainhoa Caballero (1),and Anna Rubio (1)(1) AZTI, Marine Research, Pasaia, Spain , (2) Department of Computer Sciences, Jerusalem College of Technology,Jerusalem, Israel, (3) Instituto Español de Oceanografía, Centre Oceanogràfic de Balears, Palma de Mallorca, Spain, (4)ISMAR, CNR, La Spezia, Italy
Monitoring and investigating the dynamics of coastal currents is crucial for the development of environmentallysustainable coastal activities, in order to preserve marine ecosystems as well as to support marine and navigationsafety. Multiplatform observing systems are arising in several areas of the coast and this work investigates thefeasibility of combining observations from independent and complementary platforms providing data at differentspatio-temporal scales.We investigate two methods to reconstruct the Three-Dimensional current velocity field: the Reduced OrderOptimal Interpolation (ROOI) and the Discrete Cosine Transform Penalized Least Square (DTC-PLS). ROOIis an Optimal Interpolation method fed, in our case, with a spatial covariance matrix extracted from a realisticnumerical oceanic model. DCT-PLS is a gap-filling method based on penalized least squares regression, relyingon a purely statistical approximation that balances fidelity to the data and smoothness of the solution.As a proof of concept we test the methods’ skills by using pseudo-observations of currents, extracted from theIBI CMEMS model. The test set-up simulates the real observatory configuration in the study area (SE Bay ofBiscay) which includes water column in-situ observations (ADCPs) as well as sea surface remote measurements(HF radars). The outputs of the methods are compared with the corresponding IBI CMEMS model fields, whichare also used as the synthetic ‘truth’.In the case of the ROOI method, different historical datasets have been tested to infer spatial covariancesindependent to the pseudo-observations obtained from IBI CMEMS. For the DCT-PLS method, the only input arethe pseudo-observations obtained from the IBI CMEMS model.Globally, the ROOI method provides the best results for the zonal component, if we consider the whole study area.Similar results using both methods are obtained for the meridional component and, in general, in the areas withhigh density of observations. The advantage of the ROOI method is that physical relationships are used to carryout a more robust blending in areas where there is low density of observations (i.e. through the spatial covariancesobtained from the model). On the other hand, the disadvantage is that a model for the covariance matrices isneeded; and therefore, the DCT-PLS method seems to be a good option if there is high density of observationsand/or absence of an appropriate model in the area. Both methods could be used to obtain new operationalproducts integrating complementary observations as well as broadening the applications of the observationaldata for coastal risk assessment, for model validation, and for the optimal planning of future coastal infrastructures.
1
Three-Dimensional Reconstruction of Ocean Current Circulation from Coastal Marine Observations: Challenges and Methods
Ivan Manso (1), Erick Fredj (2), Gabriel Jordà (3), Maristella Berta (4), Annalisa Griffa (4), Ainhoa Caballero (1), and Anna Rubio (1)
(1) AZTI, Marine Research, Pasaia, Spain(2) Department of Computer Sciences, Jerusalem College of Technology, Jerusalem, Israel(3) Instituto Español de Oceanografía, Centre Oceanogràfic de Balears, Palma de Mallorca, Spain(4) ISMAR, CNR, La Spezia, Italy
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
✓ Monitoring and investigating the dynamics of coastal currents is crucial for the development of environmentally sustainable coastal activities.
✓ In order to preserve marine ecosystems as well as to support marine and navigation safety.
3
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
✓ Monitoring and investigating the dynamics of coastal currents is crucial for the development of environmentally sustainable coastal activities.
✓ In order to preserve marine ecosystems as well as to support marine and navigation safety.
✓ Multiplatform observing systems are arising in several areas ofthe coast.
https://doi.org/10.1007/978-3-319-66493-4_1
4
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
✓ Monitoring and investigating the dynamics of coastal currents is crucial for the development of environmentally sustainable coastal activities.
✓ In order to preserve marine ecosystems as well as to support marine and navigation safety.
✓ Multiplatform observing systems are arising in several areas ofthe coast.
https://doi.org/10.1007/978-3-319-66493-4_1
✓ This work investigates the feasibility of combining observations fromindependent and complementary platforms by means of datareconstruction methods.
5
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
?
✓ Monitoring and investigating the dynamics of coastal currents is crucial for the development of environmentally sustainable coastal activities.
✓ In order to preserve marine ecosystems as well as to support marine and navigation safety.
✓ Multiplatform observing systems are arising in several areas ofthe coast.
✓ MAIN AIM: compare the performance of two data reconstruction methods in 3D, in terms of current velocities.
https://doi.org/10.1007/978-3-319-66493-4_1
✓ This work investigates the feasibility of combining observations fromindependent and complementary platforms by means of datareconstruction methods.
6
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Discrete Cosine Transform Penalized Least Square (DCT-PLS)(García, 2010; Fredj et al., 2016)
✓ Gap-filling method based on penalized least squares regression
✓ Fitting model obtained from the trade-off between the bias of the fitting and the smoothness of the solution
7
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Discrete Cosine Transform Penalized Least Square (DCT-PLS)(García, 2010; Fredj et al., 2016)
✓ Gap-filling method based on penalized least squares regression
✓ Fitting model obtained from the trade-off between the bias of the fitting and the smoothness of the solution
Reduced Order Optimal Interpolation (ROOI)(Kaplan et al., 2000; Jordà et al., 2016)
✓ Fed with historical data (e.g. model outputs) by means of a spatial covariance matrix
✓ The reconstructed field is obtained by means of the spatial EOFs of the covariance matrix
VS
8
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
9
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
10
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
IBI CMEMS (IBI_REANALYSIS_PHYS_005_002)
11
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
IBI CMEMS (IBI_REANALYSIS_PHYS_005_002)
IBI surface grid, simulating the HF radar field
IBI grid points where the ADCPs (moorings) are simulated
12
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
Surface and water column dynamics seasonally modulated
Summer 2011: Jun-Jul-Aug-Sep
Two periods are analyzed:
Winter 2010-2011: Nov-Dec-Jan-Feb
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
IBI surface grid, simulating the HF radar field
IBI grid points where the ADCPs are simulated
13
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
RECONSTRUCTION DCT-PLS
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
14
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
RECONSTRUCTION DCT-PLS RECONSTRUCTION ROOI
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
15
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
RECONSTRUCTION DCT-PLS RECONSTRUCTION ROOI
MODEL 2 (IBI/GLORYS HR/GLORYS LR)U, V for the whole 3D grid (from 1993 to 2009)
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
IBI →IBI_REANALYSIS_PHYS_005_002
GLORYS HR→GLOBAL_REANALYSIS_PHY_001_025
GLORYS LR→GLOBAL_REANALYSIS_PHY_001_030
16
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
RECONSTRUCTION DCT-PLS
RECONSTRUCTED FIELDSU, V for the whole 3D grid
RECONSTRUCTION ROOI
MODEL 2 (IBI/GLORYS HR/GLORYS LR)U, V for the whole 3D grid (from 1993 to 2009)
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
17
MODEL 1 (IBI)U, V for the whole 3D grid(summer 2011 / winter 2010-2011)
PSEUDO–OBSERVATIONSU, V at observation points
RECONSTRUCTION DCT-PLS
RECONSTRUCTED FIELDSU, V for the whole 3D grid
REFERENCE FIELDU, V for the whole 3D grid
VS
RECONSTRUCTION ROOI
MODEL 2 (IBI/GLORYS HR/GLORYS LR)U, V for the whole 3D grid (from 1993 to 2009)
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
18
U V
-12 m -100 m -12 m -100 m
ROOI(with
model covarian
ce)
DCT-PLS
RRMSD-U/ref.field vs ROOI/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-52m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30' 20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
RRMSD-V/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
Relative RMSD winter
summer
19
U V
-12 m -100 m -12 m -100 m
ROOI(with
model covarian
ce)
DCT-PLS
RRMSD-U/ref.field vs ROOI/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-52m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30' 20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
✓ Lowest differences near the
areas of high density of
observations
RRMSD-V/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Relative RMSD winter
summer
20
U V
-12 m -100 m -12 m -100 m
ROOI(with
model covarian
ce)
DCT-PLS
RRMSD-U/ref.field vs ROOI/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-52m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30' 20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
✓ Lowest differences near the
areas of high density of
observations
✓ Lower differences in winter
than in summerRRMSD-V/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Relative RMSD winter
summer
21
U V
-12 m -100 m -12 m -100 m
ROOI(with
model covarian
ce)
DCT-PLS
RRMSD-U/ref.field vs ROOI/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-52m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30' 20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
✓ Lowest differences near the
areas of high density of
observations
✓ Lower differences in winter
than in summer
✓ Lower error near the areas of
high density of observations
for the DCT-PLS
RRMSD-V/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Relative RMSD winter
summer
22
U V
-12 m -100 m -12 m -100 m
ROOI(with
model covarian
ce)
DCT-PLS
RRMSD-U/ref.field vs ROOI/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-52m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/summer/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs ROOI/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-U/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30' 20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-U/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2RRMSD-V/ref.field vs DCT-PLS/winter/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
RRMSD-V/ref.field vs DCT-PLS/winter/-100m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
✓ Lowest differences near the
areas of high density of
observations
✓ Lower differences in winter
than in summer
✓ Lower error near the areas of
high density of observations
for the DCT-PLS
✓ Better results for ROOI at the
rest of the areas
RRMSD-V/ref.field vs DCT-PLS/summer/-12m
4oW
30' 3
oW
30' 2
oW
30'
20'
40'
44oN
20'
40'
0
0.5
1
1.5
2
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Relative RMSD winter
summer
23
Whole grid Reduced grid
U
V
U
V
1000500
200
2000
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Mean Relative RMSD
24
Whole grid Reduced grid
U
V
U
V
1000500
200
2000
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Whole grid case:✓ Better performance of the ROOI✓ Lower relative RMSD values in
winter than in summer
Mean Relative RMSD
25
Whole grid Reduced grid
U
V
U
V
1000500
200
2000
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Whole grid case:✓ Better performance of the ROOI✓ Lower relative RMSD values in
winter than in summerReduced grid case:✓ Better performance of the DCT-PLS
Mean Relative RMSD
26
Whole grid Reduced grid
U
V
U
V
1000500
200
2000
Whole grid case:✓ Better performance of the ROOI✓ Lower relative RMSD values in
winter than in summerReduced grid case:✓ Better performance of the DCT-PLSIn general:✓ The DCT-PLS is more sensitive to
the distance from the areas of highdensity of observations
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
Mean Relative RMSD
27
✓ Satisfactory 3D data reconstruction
✓ The DCT-PLS method provides better results in the areas of high density of observations
✓ The ROOI method provides better results out of those areas
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
28
✓ Satisfactory 3D data reconstruction
✓ The DCT-PLS method provides better results in the areas of high density of observations
✓ The ROOI method provides better results out of those areas
✓ Pros and cons of the DCT-PLS:
Pros: No need for extra information of the area
Cons: The absence of extra information also makes the blending less robust out of the observations areas.
✓ Pros and cons of the ROOI:
Pros: Robust data blending with physical information of the area. Provides good results even for areas far from the observations.
Cons: Need for good models in the study area.
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
29
✓ Satisfactory 3D data reconstruction
✓ The DCT-PLS method provides better results in the areas of high density of observations
✓ The ROOI method provides better results out of those areas
✓ Pros and cons of the DCT-PLS:
Pros: No need for extra information of the area
Cons: The absence of extra information also makes the blending less robust out of the observations areas.
✓ Pros and cons of the ROOI:
Pros: Robust data blending with physical information of the area. Provides good results even for areas far from the observations.
Cons: Need for good models in the study area.
✓ Possible applications:
New operational products (very fast run)
Broaden the applications of the observational data for coastal risk assessment
Model validation
Optimal planning of future coastal observatories
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
30
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
✓ Previous work has shown that:
➢ HF radar and ADCP measurements provide a good
monitorization of the seasonal and mesoscale dynamics
of the area (Rubio et al., 2019 (accepted))
➢ Are suitable for being used in data reconstruction
methods
31
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
✓ Previous work has shown that:
➢ HF radar and ADCP measurements provide a good
monitorization of the seasonal and mesoscale dynamics
of the area (Rubio et al., 2019 (accepted))
➢ Are suitable for being used in data reconstruction
methods
✓ Next step is to use ROOI method with real HF radar and ADCP current data sets.
1- For estimating subsurface transport 2- For filling data gaps
HFR
ADCP
32
INTRODUCTION|MAIN APPROACH|RESULTS|CONCLUSIONS|FUTURE WORK
2- For the BB-Trans glider campaign (TNA JERICO-NEXT) period (17 May- 14 June
2018) by blending:
✓ Comet and Sebastian glider data
✓ HF radar data
✓ Donostia buoy ADCP data
✓ Jason-3 (track 248) and Sentinel-3A (track 257) altimeter data
http://www.euskoos.eus/es/datos/red-oceano-meteorologica-de-la-cae/
http://www.euskoos.eus/wp-content/uploads/2016/02/boya_c.jpgAZTIhttps://phys.org/news/2012-07-nasa-
contractor-jason-mission.html
33
תודהeskerrik asko
gràcies
Acknowledgements:This study has been supported by the JERICO-NEXT project, funded by the European Union's Horizon 2020 research and innovation program under grant agreement No 654410 and the COMBATproject supported by the 2nd call of the Service Evolution of CMEMS. This study has been also undertaken with the financial support of the Department of Environment, Regional Planning, Agricultureand Fisheries of the Basque Government (Marco Program). I. Manso was supported by a PhD fellowship from also the Department of Environment, Regional Planning, Agriculture and Fisheries of theBasque Government. The HFR and buoy system, whose data have been used herein, is owned to the Directorate of Emergency Attention and Meteorology of the Basque Government. Model data wereproduced and distributed by CMEMS.
34
35
✓ Gap-filling method based on penalized least squares regression
✓ The fitting model based on DCTs and a smoothing (fitting)
parameter s.
✓ Test the fitting for each s → by cross validation → GCV score
➢ GCV score estimated by the expected trade-off (F) between:
• the bias of the fitting (RSS)
• the variance of the results of the created model (P).
✓ The best fitting model is obtained from the minimization of the
GCV score.
F(𝑠) = RSS + P = 𝑦 − ො𝑦 2 + s 𝐷 ො𝑦 2
𝐸 F → GCV
min(GCV) → s
Discrete Cosine Transform Penalized Least Square (DCT-PLS)(García, 2010; Fredj et al., 2016)
✓ Fed with historical data by means of a spatial covariance matrix
✓ The reconstructed field (Z) is obtained by means of:
➢ the spatial EOFs (U) of the covariance matrix
➢ the corresponding set of amplitudes (α = UTZ)
✓ Only the leading M modes that explain a high percentage of
variance are considered (reducing the order).
✓ The amplitude (αM) is obtained from the minimization of a cost
function that accounts for:
➢ the difference with the available observations
➢ the observational error
➢ not giving too much energy to higher order modes
Z = U ∙ α
ZM = UM ∙ αM
Reduced Order Optimal Interpolation (ROOI)(Kaplan et al., 2000; Jordà et al., 2016)