Validation of Oil Spill Transport and Fate Modeling in Arctic Ice

Transcription

1 Validation of Oil Spill Transport and Fate Modeling in Arctic Ice Authors: * 1 Deborah P. French-McCay, 1 Tayebeh Tajalli-Bakhsh, 1 Kathy Jayko, 2 Malcolm L. Spaulding, and 3 Zhengkai Li 1 RPS ASA 55 Village Square Drive South Kingstown, RI USA Debbie.FrenchMcCay@rpsgroup.com Tayebeh.TajalliBakhsh@rpsgroup.com Kathy.Jayko@rpsgroup.com 2 Department of Ocean Engineering, University of Rhode Island South Ferry Road, Narragansett, RI USA spaulding@oce.uri.edu 3 PureLine Treatment Systems 1241 N. Ellis Street Bensenville, IL USA zkli06@gmail.com 1

2 SUPPLEMENTARY MATERIAL Study Area Figure S1. Locations of the Beaufort Sea and Barents-Kara Seas study areas where NERSC ice model products were provided for the study (Ólason et al. 2016). Model System Figure S2. Oil spill modeling system for ice-infested waters. 2

3 Example Model Predicted and Observed Drifter Trajectories The trajectory examples shown in Figures S3 and S4 are drawn from modeling analyses of 61 drifters that moved through the TOPAZ-EVP-WIM study domain in the Beaufort Sea where data from all ice model data products were available. Model-predicted trajectories beginning on the first day of the month of March 2008 are compared to the observed tracks of two (arbitrarilyselected) example drifters ( #53536 and #5312) released in the Beaufort Sea. The model trajectories in Figures S3 and S4 show 15-day predictions initialized from the buoy positions 15 days apart, i.e., on March 1 and on March 16, The predicted trajectory patterns are similar for all modeled trajectories, reflecting the underlying current and wind field inherent in all the meteorological and ice-ocean models. Performance is reasonably good for both TOPAZ-EVP and nextsim based simulations in both cases. For these and all drifters studied, the transport paths are characterized by periods of relatively constant directional movement, interrupted by rapid changes in direction. There are typically three to five such events during the one month simulation period, although occasionally there are periods where motion is slow and buoys move in seemingly random directions. One can see divergences accumulate over time, but most of the displacements occur at the events. In the example in Figure S3, the nextsim EB model prediction is very close to the observed drifter path for the first 15 days, whereas the predictions using the TOPAZ models move too quickly but along a similar path. The buoy stopped moving (when perhaps the ice stopped moving) about March 17th, but the models continued to predict southwestward movements, with nextsim predicting slower movement than the TOPAZ models using EVP rheology. In the example in Figure S4, the nextsim EB model prediction is very close to the observed drifter path for the first 10 days as it changed direction 5 times, whereas the predictions using the TOPAZ models are close to the observed path for the first 7 days before diverging (primarily by moving too fast). These examples show the limits of the ice-ocean models abilities to forecast as being about 7-15 days, although in other examples accuracy degrades after 3-5 days of forecasting. While the accuracy of individual oil model trajectories projected weeks to months into the future would be expected to be low, in the event of a spill, forecasts could be updated frequently (on a time scale of hours to days) with satellite information, aircraft observations, drifter data, and other observations to improve reliability. Figure S5 demonstrates the model trajectories when reinitialized at the observed drifter position on 15 March In the second 15-day period, all the models correctly predict the turn to the east, but all (the nextsim EB and TOPAZ-EVP- WIM less so than TOPAZ4 and TOPAZ-EVP) move faster than the observed drifter. 3

4 Figure S3. Model trajectories using TOPAZ-EVP-WIM (black), TOPAZ4 (blue), TOPAZ- EVP (orange) and nextsim (red), initialized at yellow symbol location on 1 March 2008, and the observed track (white) of drifter # Trajectories forecast for 15 days (left panel) and 30 days (right panel) as compared to the observed drifter track over the same periods. Figure S4. Model trajectories using TOPAZ-EVP-WIM (black), TOPAZ4 (blue), TOPAZ- EVP (orange) and nextsim (red), initialized at yellow symbol location on 1 March 2008, and the observed track (white) of drifter #5312. Trajectories forecast for 15 days (left panel) and 30 days (right panel) as compared to the observed drifter track over the same periods. 4

5 Figure S5. Model trajectories using TOPAZ-EVP-WIM (black), TOPAZ4 (blue), TOPAZ- EVP (orange) and nextsim (red), initialized at yellow symbol location on 1 March 2008, as compared to the observed track (white) of drifter #5312. The model trajectories are reinitialized at the location of the drifter on 15 March. 5

6 Accuracy of Ice-Ocean Model and Oil Spill Trajectory Forecasts Summaries of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) for March 2008 differed from those for December 2008 and May 2009, as shown in Tables S1-S4). The LenRatio, S and SS scores for each of the 5-day, 10-day and 15-day drifter intervals using each ice-ocean model were also calculated for the summer months (Tables S5- S7). (See main text for discussion.) Table S1. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 5-day simulations for non-overlapping intervals in the period 1-25 March Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean nextsim EB model, 6-hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD Table S2. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 5-day simulations for non-overlapping intervals in the period 1-30 December 2008 and 1-15 May Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean nextsim EB model, 6-hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD

7 Table S3. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 10-day simulations for nonoverlapping intervals in the period 1-30 December 2008 and 1-15 May Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean nextsim EB model, 6-hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD Table S4. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 15-day simulations for nonoverlapping intervals in the period 1-30 December 2008 and 1-15 May Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean nextsim EB model, 6-hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD Table S5. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 5-day simulations for non-overlapping intervals in the periods 1-30 June 2008 and 1-30 September Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean 6 hr mean 24 hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD

8 Table S6. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 10-day simulations for nonoverlapping intervals in the periods 1-30 June 2008 and 1-30 September Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean 6 hr mean 24 hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD Table S7. Summary of the ratios of model:observed path length (LenRatio), separation index (S) and skill scores (SS) using each ice-ocean model for transport calculations compared to the observed drifter trajectories, using 15-day simulations for nonoverlapping intervals in the periods 1-30 June 2008 and 1-30 September Ice-Ocean Model TOPAZ4 Reanalysis, 24-hr mean TOPAZ-EVP, 6-hr mean 1 hr mean 6 hr mean 24 hr mean Drifters Intervals LenRatio LenRatio S S SS SS (#) (#) Mean SD Mean SD Mean SD

9 Example Trajectories in Operational Mode Figures S6 to S9 demonstrate for a single example drifter (#66276), tracked in March and April 2008, the TOPAZ4, TOPAZ-EVP, TOPAZ-EVP-WIM and nextsim model-predicted trajectories compared to the observed trajectory, respectively. The model-predicted trajectories are reinitialized to the observed locations every 5 days. The simulations with re-initialization show improved matches to the observations. All of the models perform well in this operational mode where the modeled locations are updated regularly. Figure S6. Model trajectory using TOPAZ4 (red), reinitialized each 5 days, as compared to the observed track (blue) of drifter # from March 1 (at solid large circle) to April 30 of (Dots indicate trajectory paths plotted at 12 hour intervals. Open circle indicates March 31.) Figure S7. Model trajectory using TOPAZ-EVP (red), reinitialized each 5 days, as compared to the observed track (blue) of drifter # from March 1 (at solid large circle) to April 30 of (Dots indicate trajectory paths plotted at 12 hour intervals. Open circle indicates March 31.) 9

10 Figure S8. Model trajectory using TOPAZ-EVP-WIM (red), reinitialized each 5 days, as compared to the observed track (blue) of drifter # from March 1 (at solid large circle) to March 31 of (Dots indicate trajectory paths plotted at 12 hour intervals. Open circle indicates March 31.) Figure S9. Model trajectory using the nextsim model (red), reinitialized each 5 days, as compared to the observed track (blue) of drifter # from March 1 (at solid large circle) to April 30 of (Dots indicate trajectory paths plotted at 12 hour intervals. Open circle indicates March 31.) 10

11 May 2009 Field Experiment Winds were generally 5-10 m/s during the experimental period. Winds increased to 6 m/s the night of May 16, peaking at m/s during May (Faksness et al. 2010; Figures S10- S11). Figure S11 also shows ERA Interim winds at the experimental and proxy sites ( Pack Ice ), which were very similar to the measured winds (data from Faksness et al., 2010, obtained by personal communication, 10 July 2017). The measured currents at 5 m depth averaged 16.7 cm/s and varied from 0.2 cm/s to 50.7 cm/s (Faksness et al. 2010). Figure S12 shows that the modeled currents by TOPAZ-EVP-WIM were in the same range as the measured currents for the first 2.5 days (through May 17) but modeled currents were slower than observed later during the experiment. Figure S13 shows the modeled ice speeds by TOPAZ-EVP-WIM at the experimental and proxy site in the pack ice as compared to measured ice movements. (Measured current and ice speeds from Faksness et al. 2010, obtained by personal communication, 10 July 2017). In the TOPAZ-EVP-WIM model, there was <30% ice at the experimental site May Figure S10. Stickplot of wind speed and direction (pointing down-wind) during the experimental period (Faksness et al. 2010). Figure S11. Measured winds during the field experiment (data from Faksness et al., 2010) and ERA Interim winds at the experimental ( No Ice ), proxy ( Pack Ice ), and MIZ sites. 11

12 Figure S12. Measured currents during the field experiment (data from Faksness et al., 2010) and TOPAZ-EVP-WIM modeled currents at the experimental ( No Ice ), proxy ( Pack Ice ), and MIZ sites. Figure S13. Measured ice speeds during the field experiment (data from Faksness et al., 2010) and modeled ice speeds using TOPAZ-EVP-WIM at the experimental ( No Ice ), proxy ( Pack Ice ), and MIZ sites. As shown in Figure S14, the oil was observed to move with the ice to the northeast, then to the south, and then again northeast (to 10 km north and then to 40 km south of the spill site, 50 km total range of trajectory North/South). Figures S15 and S16 show the modeled trajectories from the experimental and proxy sites, respectively. In Figure S17 the centroid of the trajectory at the proxy site is shown. Figure S18 shows the model trajectory in the MIZ, which shows more eastward movement than the simulations at the FEX2009 and proxy sites. 12

13 Figure S14. Progressive vector diagram of observed oil and ice movement during March 2009 field experiment (data from Faksness et al., 2010). 13

14 Figure S15. SIMAP model cumulative trajectory from the experimental site using TOPAZ- EVP-WIM as forcing. Note that the dots are not weighted by mass and many represent surfacing sheen (e.g., those to the west of the main trajectory). Figure S16. SIMAP model cumulative trajectory from the proxy pack ice site using TOPAZ-EVP-WIM as forcing. Note that the dots are not weighted by mass and many represent surfacing sheen (e.g., those to the west of the main trajectory). 14

15 Figure S17. SIMAP model trajectory (line indicates centroid of the oil) from the proxy pack ice site on May 15 until May 29, using TOPAZ-EVP-WIM as forcing. For this simulation, the spill volume was 100 times the actual experimental release volume in order to illustrate measurable concentrations. (Still is snapshot of concentrations, winds, ice cover and currents at the end of the experiment on May 20 at 11AM; see corresponding video file AS R1supplb.PDF.) Figure S18. SIMAP model cumulative trajectory from the MIZ site using TOPAZ-EVP- WIM as forcing. Note that the dots are not weighted by mass and many represent surfacing sheen (e.g., those to the west of the main trajectory). 15

16 Time Averaging of Ice-Ocean Model Results for Use in Oil and Particle Transport Models From the sample of drifters examined in this study, it appears that the updated TOPAZ-EVP and TOPAZ-EVP-WIM ice-ocean models perform better than the publically-available TOPAZ4 reanalysis version, but this may be due in part to the daily averaging used to deliver (on the web) the TOPAZ4 product as compared to 6-hourly or hourly data delivered in the TOPAZ-EVP and TOPAZ-EVP-WIM products, respectively. While time averaging does not appear to make a significant difference for short (5-day) intervals, for longer forecasts the averaging erases sudden changes in direction seen in buoy trajectories and introduces error. Provision of TOPAZ-EVP model data at smaller time steps (than daily, as on the web server, or even than 6-hourly) and without time averaging would provide for improved performance in oil spill modeling. When the oil spill model utilized the same high-resolution ice vector data as used by NERSC in Phase 1 for simulating drifter trajectories (Figure S19), the results agreed, whereas the use of time-averaged data degraded the model performance and the drift tracks diverged somewhat from the actual buoy tracks recorded. Figure S19. Comparison of nextsim model trajectory for an example IABP buoy run by NERSC using the original model time step and grid (blue), as compared to the model trajectory run by RPS ASA with 6-hour averaged nextsim data interpolated onto the TOPAZ4 grid as provided in the netcdf file (green), and to the IABP buoy track (white). 16