calibration of a transport model using drifting buoys deployed during the prestige accident
DESCRIPTION
CALIBRATION OF A TRANSPORT MODEL USING DRIFTING BUOYS DEPLOYED DURING THE PRESTIGE ACCIDENT S. CASTANEDO, A.J. ABASCAL, R. MEDINA and I.J. LOSADA. OUTLINE. 1. Introduction Data Methodology 4. Conclusions. OUTLINE. 1. Introduction Data Methodology 4. Conclusions. - PowerPoint PPT PresentationTRANSCRIPT
CALIBRATION OF A TRANSPORT MODEL USING CALIBRATION OF A TRANSPORT MODEL USING DRIFTING BUOYS DEPLOYED DURING THE DRIFTING BUOYS DEPLOYED DURING THE
PRESTIGE ACCIDENTPRESTIGE ACCIDENT
S. CASTANEDO, A.J. ABASCAL, R. MEDINA and I.J. LOSADAS. CASTANEDO, A.J. ABASCAL, R. MEDINA and I.J. LOSADA
1. Introduction1. Introduction
2.2. DataData
3.3. MethodologyMethodology
4. Conclusions4. Conclusions
OUTLINE
1. Introduction1. Introduction
2.2. DataData
3.3. MethodologyMethodology
4. Conclusions4. Conclusions
OUTLINE
1. INTRODUCTION
• Along the Spanish coast several emergency spill response systems were built during the Prestige crisis (UC, AZTI, MeteoGalicia, IMEDEA,...). In these response systems one important task was to establish operational forecasting systems for developing proper response strategies
1. INTRODUCTION
• Generally, the structure of these predictions systems was composed by collection of observations including oil slicks, numerical modelling to provide forecasts of wind, waves, currents and oil trajectories and finally, data management and dissemination.
• The emergency spill response systems were considered to be important tools in addressing the Prestige crisis.
1. INTRODUCTION
- Daily cleaning-up of the beaches- Mechanical recovery from the water surface- Protection of estuaries by means of booms
Delegación del Gobierno en CantabriaConsejería de Medio Ambiente de Cantabria
1. INTRODUCTION
• Now, we can take advantage of the experience acquired during the Prestige accident and develop a Spanish operational oceanographic system (Project ESEOO:www.eseoo.org).
• One of the main objective of the ESEOO transport model is to be used by SASEMAR in sea rescue and response to pollution of marine water.
• The success of the system will be based on the accuracy of the different numerical models involved in trajectory forecasting.
1. INTRODUCTION
The aim of this study is to calibrate a Lagrangian particle-tracking trajectory algorithm and, at the same time, investigate about the relative importance that the different forcing (wind, wave, currents) have on the oil spill fate.
CD=0.02 CD=0.03
1. Introduction1. Introduction
2.2. DataData
3.3. MethodologyMethodology
4. Conclusions4. Conclusions
OUTLINE
2. DATA
WHAT DO WE NEED?WHAT DO WE NEED?
Trajectory Analysis handbook (NOAA)
WHAT DO WE NEED?WHAT DO WE NEED?
2. DATA
FORCINGS: Wind Currents Waves
BUOYS
NUMERICAL MODEL
2. DATA 2.1. Buoys
• Among the decisions made during the management of the Prestige accident, it was proposed to release lagrangian floats to both track the biggest oil slicks position and trajectory and to provide some feedback and/or validation for the numerical models of currents and oil dispersion forecast.
• The deployment of drifting floats was organised by the National Spanish Research Council (CSIC) and AZTI Foundation using available ARGOS buoys used for oceanographic studies (García-Ladona
et al., 2005).
Buoy number
TypeInitial
longitude Initial
latitudeInitial date Last date Owner
16291 PTR -5.868 45.311 15/01/2003 09/02/2003 AZTI
16651 PTR -3.518 44.278 27/12/2002 03/02/2003 AZTI
16735 PTR -6.593 45.175 29/12/2002 16/02/2003 AZTI
16751 SC40 -9.447 42.915 19/12/2002 31/01/2003 CSIC
16752 SC40 -9.356 43.155 19/12/2002 19/01/2003 CSIC
16753 SC40 -9.581 42.969 19/12/2002 30/01/2003 CSIC
16754 SC40 -9.604 42.688 19/12/2002 01/02/2003 CSIC
23249 SC40 -12.046 42.207 16/01/2003 19/02/2003 CSIC
23258 SC40 -9.58 42.662 11/01/2003 19/02/2003 CSIC
23259 SC40 -12.054 42.174 27/01/2003 19/02/2003 CSIC
23282 SC40 -3.350 45.249 02/01/2003 18/02/2003 CSIC
23289 SC40 -4.007 45.575 02/01/2003 18/02/2003 CSIC
23348 SC40 -9.416 42.861 11/01/2003 25/01/2003 CSIC
2. DATA 2.1. Buoys
December 2002 - February 2003
2. DATA 2.1. Buoys
WIND: HIRLAM model (INM) (www.inm.es)
2.2. Wind and wave conditions
2. DATA
Wind at 10 meters above the MSL
x 0.2º x 0.2º (z 22 km)
t 6 hours
Data from re-analysis corresponding to the period November 2002-November 2003
x 0.25 x 0.25º (z 28km)
t 3 hours
WAVE: WAM model (PE) (www.puertos.es)
2.2. Wind and wave conditions
2. DATA
2.3. Currents
2. DATA
CURRENTS 1: NRLPOM model (USA) (http://www.aos.princeton.edu)
CURRENTS 2: MERCATOR model (FR) (http://www.mercator-ocean.fr/)
x z 7 km, t 3 hours
x z 7 km, t 24 hours
FORCING Período FUENTE
WINDRe-analysis data
Nov. 2002 -Nov. 2003HIRLAM
(INM)
WAVEDic. 2002 – Dic. 2003 WAM
(PE)
CURRENTS 1 Dic. 2002 - Dic. 2003NRLPOM
(USA)
CURRENTS 2 Nov. 2002 -Mar. 2003 MERCATOR (FR)
2. DATA
BUOYS Dic. 02 – Feb. 03
2.4. Summary
1. Introduction1. Introduction
2.2. DataData
3.3. MethodologyMethodology
4. Conclusions4. Conclusions
OUTLINE
We want to simulate the buoy trajectory by means of a numerical model: Lagrangian transport model
XXii(t+(t+t) = t) = XXii(t) + (t) + uu(t) (t) t + diffusiont + diffusion
u(t) = ucurrents+ CD* uwind + CW * uwave
CD: wind drag coefficient
CW : wave coefficient
Difussion: (García-Martínez y Flores-Tovar,
1999; Lonin, 1999)
k: diffusion coefficient
3. METHODOLOGY
6k
t
3. METHODOLOGY
US,V CD * UVUwind: Wind-induced current
CD: 3% (Sobey, 1992)
2.5%-4.4% (ASCE, 1996)
Uwave: Wave-induced Stokes drift , 8S H
gHU
C
(Sobey y Barker, 1997)
CW: 0.01- 0.1 (FLTQ, 2003)
3. METHODOLOGY
We need to determine the coefficients CD and Cw in order to obtain the best fit between the numerical result and the observed buoy trajectory
Owing to the great quantity of variables involved in the Owing to the great quantity of variables involved in the problem, aa problem, aa optimization algorithmoptimization algorithm is used in this study is used in this study as a as a preliminary preliminary tooltool
3. METHODOLOGY
1. Coefficients that minimize the error between numerical and actual buoy trajectory: optimization algorithm
PROCEDURE:
2. Introduction of these coefficients in the Lagrangian transport model
3. Analysis of the results/conclusions
3. METHODOLOGY
Global optimization algorithm: SCE-UA (shuffled complex SCE-UA (shuffled complex evolution method – University of Arizona) evolution method – University of Arizona) (Duan et al, 1994)(Duan et al, 1994)
3.1. Automatic calibration
N: number of buoys
UB: actual buoy velocity
UM: numerical buoy velocity (wind, wave and currents)
22
1
( , ) ( , , ) ( , ) ( , , )N
Bx Mx By Myt i
J U t U t U t U t
x x x x
( , , , , )H w w w Ca a
Objective function:Objective function:
The goal of calibration is to find those values for the coefficients that minimize J
f oB
x xU
t
( )
( )
Mx H wavex w windx c currentx
My H wavey w windy c currenty
U t a U a U a U
U t a U a U a U
aH: wave coefficient (Cw)aW: wind coefficient (CD) aC: current coefficient (indication of the error in the numerical current field)
Actual buoy velocity
Numerical buoy velocity
3. METHODOLOGY 3.1. Automatic calibration
3.2. Experiment with all buoys
3. METHODOLOGY
Numero Boya
Fecha inicial
Fecha final
Número de horas
16291 15/01/2003 09/02/2003 107
16651 27/12/2002 03/02/2003 215
16735 29/12/2002 16/02/2003 240
16751 19/12/2002 31/01/2003 234
16752 19/12/2002 19/01/2003 143
16753 19/12/2002 30/01/2003 234
16754 19/12/2002 01/02/2003 235
23249 16/01/2003 19/02/2003 80
23258 11/01/2003 19/02/2003 155
23259 27/01/2003 19/02/2003 33
23282 02/01/2003 18/02/2003 301
23289 02/01/2003 18/02/2003 157
23348 11/01/2003 25/01/2003 45
• Hipothesis:1.- Linear expression of the wind coefficient aW =w + w|uwind|
2.- Swell ( ), 8S H
gHU
C
3.2. Experiment with all buoys
3. METHODOLOGY
Correlation coefficient < 50%
Next step: We need to delimitate the problem
Calibration for each buoy
3.2. Experiment with all buoys
3. METHODOLOGY
Boya aH w w ac Rx Ry
16291 0.10085 0.022426 0.000478 0.10444 0.5818 0.6521
16651 -0.13077 0.026448 0.000384 0.01123 0.6854 0.5972
16735 0.03162 0.020349 0.000845 0.06369 0.5919 0.4331
23249 0.05341 0.031110 0.000769 0.16435 0.4812 0.5012
23258 0.05797 0.024382 -0.000186 0.38309 0.5487 0.4948
23259 -0.16932 0.024124 -0.000371 -0.02391 0.7645 0.4155
23282 0.03702 0.021286 0.000226 0.04080 0.5084 0.2750
23289 0.03398 0.029249 -0.000489 -0.27240 0.4212 0.2726
16751 0.09023 0.025523 -0.000676 0.47468 0.2193 0.1955
16752 0.39433 0.015369 0.000513 0.54086 0.3856 0.2220
16753 0.21031 0.012407 0.000902 0.17193 0.4022 0.4244
16754 0.04053 0.029186 -0.000566 0.46913 0.4392 0.3780
23348 -0.04004 0.018670 -0.000627 0.19433 0.0002 0.1553
3.3. Experiment with each buoy
3. METHODOLOGY
Boya aH w w ac Rx Ry
16291 0.10085 0.022426 0.000478 0.10444 0.5818 0.6521
16651 -0.13077 0.026448 0.000384 0.01123 0.6854 0.5972
16735 0.03162 0.020349 0.000845 0.06369 0.5919 0.4331
23249 0.05341 0.031110 0.000769 0.16435 0.4812 0.5012
23258 0.05797 0.024382 -0.000186 0.38309 0.5487 0.4948
23259 -0.16932 0.024124 -0.000371 -0.02391 0.7645 0.4155
23282 0.03702 0.021286 0.000226 0.04080 0.5084 0.2750
23289 0.03398 0.029249 -0.000489 -0.27240 0.4212 0.2726
16751 0.09023 0.025523 -0.000676 0.47468 0.2193 0.1955
16752 0.39433 0.015369 0.000513 0.54086 0.3856 0.2220
16753 0.21031 0.012407 0.000902 0.17193 0.4022 0.4244
16754 0.04053 0.029186 -0.000566 0.46913 0.4392 0.3780
23348 -0.04004 0.018670 -0.000627 0.19433 0.0002 0.1553
3.3. Experiment with each buoy
3. METHODOLOGY
Boya ac Rx Ry
16291 0.10444 0.5818 0.6521
16651 0.01123 0.6854 0.5972
16735 0.06369 0.5919 0.4331
23249 0.16435 0.4812 0.5012
23258 0.38309 0.5487 0.4948
23259 -0.02391 0.7645 0.4155
Best fit buoys
Small current coefficient
Dominant forcing : wind
3.3. Experiment with each buoy
3. METHODOLOGY
Boya ac Rx Ry
23282 0.04080 0.5084 0.2750
23289 -0.27240 0.4212 0.2726
16751 0.47468 0.2193 0.1955
16752 0.54086 0.3856 0.2220
16753 0.17193 0.4022 0.4244
16754 0.46913 0.4392 0.3780
23348 0.19433 0.0002 0.1553
Worse fit buoys
Dominant forcing : wind and current
3.3. Experiment with each buoy
3. METHODOLOGY
We obtain the best fit when wind is the dominant forcing
When currents are important (continental slope and near
the coast) the agreement between observed and numerical
trajectories is worse
The numerical current field must be improved
3.3. Experiment with each buoy
3. METHODOLOGY
3. METHODOLOGY
PROCEDURE:
1. We select the buoys located outside of the continental slope (mainly affected by wind)
• Hipothesis:In these buoys the effect of the currents is negligible
2. We obtain CD and CW with these outer buoys
3. With all buoys and with CD and CW obtained in 2., the current coefficient is carried out
Numero
BoyaFecha inicial
Fecha finalNúmero de
horas
16291 15/01/2003 20/01/2003 37
16651 14/01/2002 21/01/2003 59
16735 29/12/2002 18/01/2003 100
23282 14/01/2003 18/02/2003 37
23289 01/02/2003 16/01/2003 51
23259 28/01/2003 03/02/2003 31
23289 01/02/2003 16/01/2003 51
3.4. Outer buoys
3. METHODOLOGY
1. We select the buoys located outside of the continental slope (mainly affected by wind)
• Hipothesis:In these buoys the effect of the currents is negligible
2
2
N Bx H wavex w w wind windx
t i iBy H wavey w w wind windy
U a U U UJ
U a U U U
2. We obtain CD and CW with these outer buoys
3.4. Outer buoys
3. METHODOLOGY
( ) M H wave w windU t a U a U
aW =w + w|uwind|
3.4. Outer buoys
3. METHODOLOGY
Current fields
POM MERCATOR
0.0526 (0.0178 0.00079 ) M wind C currentawave windU U + U U U
3.5. Current coefficient
3. METHODOLOGY
3. With all buoys and with CD and CW obtained in 2., the current coefficient is carried out
2 2
0.0526 0.0178 0.00079 0.0526 0.0178 0.00079
N
Bx oleajex viento vientox c corrx By oleajey viento vientoy c corryt i i
J U U U U a U U U U U a U
POM
3.5. Current coefficient
3. METHODOLOGY
MERCATOR
3.5. Current coefficient
3. METHODOLOGY
2 2
0.0526 0.0178 0.00079 0.0526 0.0178 0.00079
N
Bx oleajex viento vientox c corrx By oleajey viento vientoy c corryt i i
J U U U U a U U U U U a U
XXii(t+(t+t) = t) = XXii(t) + (t) + uu(t) (t) t + diffusiont + diffusion
Introduction of the calculated coefficients (CD, Cw, ac) in the Lagrangian transport model
3.6. Lagrangian model
3. METHODOLOGY
u(t) =0.10312* ucurrent +(0.0178+0.000798*| uwind |)* uwind+ 0.0526* uwave
x=7.27 km, y=6.77 km, t=60 s
2
0
1 T
m b mi
RMSE x xT
RMSE: CD, Cw and ac coefficients calculated by the SCE_UA method Numerical simulation with all buoys
3.6. Lagrangian model
3. METHODOLOGY
2
0
1 T
m b mi
RMSE x xT
3.6. Lagrangian model
3. METHODOLOGY
RMSE: CD and Cw coefficients calculated by the SCA_UA method and ac=1 Numerical simulation with all buoys
Period: 15-01-2003 al 23-01-2003
3.6. Lagrangian model
3. METHODOLOGY
Numerical simulation with 5 buoys (3 outside of continental slope)
21
3
4
5
8 days
1
2
3
4
5
48 hour steps
3.6. Lagrangian model
3. METHODOLOGY
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200317/01/2003
30.29 15.49 4.57 4.67 8.20
17/01/200319/01/2003
25.62 16.02 5.47 5.94 2.78
19/01/200321/01/2003
21.73 55.38 12.75 3.17
21/01/200323/01/2003
43.94 49.92 15.27
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200323/01/2003
95.51 83.19 26.31 22.77 19.37
RMSE (48 HOUR STEPS)
RMSE (8 DAYS SIMULATION)
3.6. Lagrangian model
3. METHODOLOGY
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200317/01/2003
30.29 15.49 4.57 4.67 8.20
17/01/200319/01/2003
25.62 16.02 5.47 5.94 2.78
19/01/200321/01/2003
21.73 55.38 12.75 3.17
21/01/200323/01/2003
43.94 49.92 15.27
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200323/01/2003
95.51 83.19 26.31 22.77 19.37
3.6. Lagrangian model
3. METHODOLOGY
RMSE (48 HOUR STEPS)
RMSE (8 DAYS SIMULATION)
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200317/01/2003
30.29 15.49 4.57 4.67 8.20
17/01/200319/01/2003
25.62 16.02 5.47 5.94 2.78
19/01/200321/01/2003
21.73 55.38 12.75 3.17
21/01/200323/01/2003
43.94 49.92 15.27
FechaRMSEm (km)
Boya 1
RMSEm (km)
Boya 2
RMSEm (km)
Boya 3
RMSEm (km)
Boya 4
RMSEm
(km) Boya 5
15/01/200323/01/2003
95.51 83.19 26.31 22.77 19.37
3.6. Lagrangian model
3. METHODOLOGY
RMSE (8 DAYS SIMULATION)
RMSE (48 HOUR STEPS)
1. Introduction1. Introduction
2.2. DataData
3.3. MethodologyMethodology
4. Conclusions4. Conclusions
OUTLINE
A global optimization method (SCE-UA), developed for calibrating A global optimization method (SCE-UA), developed for calibrating
watershed models, has been used in this study. The goal of this watershed models, has been used in this study. The goal of this
method was to find the optimal forcing coefficients to be applied in method was to find the optimal forcing coefficients to be applied in
a numerical transport model. a numerical transport model.
The forcing coefficients that minimize the error between the The forcing coefficients that minimize the error between the
numerical and the observed buoy trayectories were obtained. numerical and the observed buoy trayectories were obtained.
A linear relation between wind velocity and wind drag coefficient A linear relation between wind velocity and wind drag coefficient
was found.was found.
4. CONCLUSIONS
Regarding the wave action, the separation of the sea and swell Regarding the wave action, the separation of the sea and swell
effect on the buoy trajectory provided the best result.effect on the buoy trajectory provided the best result.
We obtained the best solution when the wind was the dominant
forcing.
When it is not possible to neglect the currents (continental slope When it is not possible to neglect the currents (continental slope
and near the coast) the agreement between actual and numerical and near the coast) the agreement between actual and numerical
trajectories was worsetrajectories was worse The numerical current fields were no correct to simulate the buoy The numerical current fields were no correct to simulate the buoy
trajectories. Further research in this area is needed. trajectories. Further research in this area is needed.
4. CONCLUSIONS
CALIBRATION OF A TRANSPORT MODEL USING CALIBRATION OF A TRANSPORT MODEL USING DRIFTING BUOYS DEPLOYED DURING THE DRIFTING BUOYS DEPLOYED DURING THE
PRESTIGE ACCIDENTPRESTIGE ACCIDENT
S. CASTANEDO, A.J. ABASCAL, R. MEDINA and I.J. LOSADAS. CASTANEDO, A.J. ABASCAL, R. MEDINA and I.J. LOSADA