EN324 May 14 ------ Departed Woods Hole 1530. Satellite imagery from May 10 shows complex SST structure (Figure 0514.1). A warm filament and small eddy are present in the study area, possibly under the influence of a ring / streamer / intrusion offshore. Edwin Link reports that their deep drifter south of the front is moving east; assimilation of their ADCP data is reported to capture this behavior to some extent. All these points argue the need for velocity measurements in the stratified region south of the survey track we are to occupy. A course was laid out so that we would enter the model domain at 40 40'N, 69 00'W and proceed east until reaching the 80m isobath, then proceeding ENE along the mean position of the 8m isobath up to 40 57' 67 04', a point just SE of the beginning of the survey track (Figure 0514.2). Forecasting Activities ---------------------- May 14 central forecast includes ADCP data from the Edwin Link, file el9905a2.m3d. Drifters from EN323 and EL9905 were launched in the simulation [EN323pEL9905.ind]. Subsequently an error was discovered in the parameter used to ramp up the atmospheric forcing; this effectively started the simulation with the forcing at full strength. The was corrected in EN324_FC.02, in which the atmospheric forcing was ramped up over a 1-day perid. A sensitivity experiment EN324_FC.03 was conducted with TDBC forcing turned off. We had yet to receive any of the observed drifter trajectories by the close of business on this day, so it was not possible to undertake any quantitative evaluation of the forecasts. However, it was noticed that the RMS residual velocity in these simulations was slightly higher than usual: FC.01 FC.02 FC.03 ----------------- Q4.1 20.0 20.0 17.4 Q4.2 11.7 11.7 11.2 Q4.3 10.4 10.4 10.4 In communications with the Edwin Link, Cisco and Jim indicate they had similar results. They attribute the higher residuals to noise in the ADCP data associated with frequent stops for CTD casts. Close examination of the alongtrack residuals does not reveal any systematic error (Figure 0514.3) It is interesting to note that the inclusion of TDBCS in runs 1 and 2 introduces an additional RMS velocity residual of approximately 2.5 cm/sec to the prior (comparing with run 3 with TDBCS turned off). The assimilation procedure appears to clean this up, with the final results of all three runs indistinguishable in terms of this statistic. May 15 ------ At 0246 local time the ship reached the first waypoint at the western edge of the model domain and slowed to 10knt. Examination of the ADCP data revealed a lot of noise, with only about 25% good returns. At 0530 the ship was slowed to 9knt. The return rate rose to 75%; data collection starts here. VPR trimming began in the late morning, and was completed by 1300. Rendez-vous with the Edwin Link occurred shortly thereafter. Scientific briefings took place, and supplies and data were exchanged. The term "Zodiac-net" was coined to describe the medium over which drifter trajectory data was transmitted between modeling groups on the two ships. VPR survey operations began immediately after rendez-vous. Late in the day, the extremely light winds made for a spectacular sight as we approached the tidal mixing front in the first leg of the survey. Capillary waves were clearly present in the stratified side of the front, while they were almost imperceptible in the well- mixed area. The well-mixed region took on a glassy appearance. This change was undeniably associated with water properties (as opposed to the atmosphere); there was no perceptible change in the ship's anemometer as she passed through the frontal region. The frontal boundary was razor sharp; the transition occured over a distance of meters. Perhaps the change in roughness was due to a surfactant which is present at the sea surface in the well-mixed area, but not in the stratified region. Upon closer examination, the patterns in surface roughness appeared to reveal interleaving of the two water masses. In one instance, the front was interupted by a narrow strip oriented roughly perpendicular to the front, aligned with the NE wind. Small eddies appeared to be associated with this structure, with horizontal scales of 10-20m. There was much excitement and speculation about the mechanisms which may have been responsible for this fantastic display. Forecasting activities ---------------------- The May 15 central forecast (EN324_FC.04) used ADCP data from the Edwin Link, in addition to that collected aboard Endeavor prior to bell time. Due to the fact the the NMFS model weather product did not arrive, we had to proceed with an incomplete heat flux record. The flux file was edited by hand so that zero heat flux was persisted after the end of the data record. Experience from the last leg had shown there were sometimes large differences between the AVN model wind product and the observed winds. The current period was no exception (Figure 0515.1). Sensitivity forecast EN324_FC.05 was forced with a hybrid wind record, composed of the buoy winds available at bell time and AVN model winds thereafter. Processing of the new drifter data was not yet completed by the close of business on the 15th. Quantitative evaluation of the forecasts was therefore postponed until the following day. May 16 ------ VPR survey continued until approximately 1100. A surface drifter was deployed at the front. Endeavor steamed 6km south of the front and released tracer from 1300-1435. Three drifters were deployed in the tracer as the patch was being laid out. An initial survey of the tracer patch was then carried out. Subsequently, Endeavor steamed back to the front to re-assess the distance between the patch and the front. On the way back south toward the tracer, another surface drifter was deployed in between the front and the patch; Endeavor continued through the patch and deployed another drifter seaward of the patch. Survey of the dye patch then recommenced. Forecasting activities ---------------------- The May 16 central forecast (EN324_FC.06) included Endeavor's ADCP data up through bell time. The heat flux forcing record was incomplete because the necessary data file was not received. Two sensitivity experiments were carried out; the first (EN324_FC.07) eliminated TDBC forcing, and the second (EN324_FC.08) included ADCP data from the Edwin Link (several days old at bell time). Evaluation of forecast skill: ----------------------------- The first forecast evaluation was undertaken with runs from May 14 and May 15 (simulations 01 though 05). This comparison was carried out using simulated drifters which were released at the reported locations. Due to delays in between drifter deployment and the beginning of data acquisition, the drifter observations do not begin until a finite time after deployment. In one case (drifter 234) data did not begin until a full tidal cycle after the reported release. It therefore seemed appropriate to carry out a second comparison in which simulated drifters were released at the space/time points of the first entry in each data record. Runs EN324_FC.01 through EN324_FC.05 were re-run as EN324_FC.01_dr through EN324_FC.05_dr with the new release positions. This removed all ambiguity with respect to consistency in the space/time starting points of the simulated and observed drifter trajectories. Fortunately, the results of the second comparison were similar to the first; as expected, the forecast error decreased in the second comparison because of its shorter duration (see below). This suggests that there were no problems with the release points used in the first comparison. As before, forecast error is defined as the mean distance between simulated and observed drifters at the end of the drifter record. Using this metric, forecast 08 has the highest skill; it is the one with the most up-to-date forcing and most complete ADCP data set used in the assimilation. Comparison of forecast 07 with the May 16 central (forecast 06) suggests that the TDBC forcing slightly degrades the skill. Therefore it is possible that a slight improvement in forecast 8 could result by removing the TDBC forcing from that simulation. Forecast Error (km) ------------------- Assessment # 1 2 ---------- EN324_FC.01 May 14 Central; EL ADCP 10.9 9.7 EN324_FC.02 (01) ramping up atmos 10.8 9.6 EN324_FC.03 (02) - TDBCS 10.9 9.8 EN324_FC.04 May 15 Central; EL+EN ADCP 10.8 9.8 EN324_FC.05 (04) + buoy winds 10.1 10.0 EN324_FC.06 May 16 Central; EN ADCP 10.2 EN324_FC.07 (06) - TDBCS 10.1 EN324_FC.08 (06) + EL ADCP 9.4 The data used in these evaluations consists of four drifters from the Edwin Link: two deployed in the well-mixed area at the 50m isobath (drogued at 13 and 33m), and two in the stratified region at the 63m isobath (drogued at 8m and 33m). Simulated and observed trajectories are shown for each forecast experiment in the second comparison are shown in Figures 0516.1(a-h). The results demonstrate very little sensitivity to the inputs which were varied in this set of simulations. The westward motion of the 33m drifter in the well-mixed region is fairly well represented in the model. In the best forecast (8), the distance between the simulated and observed drifter position at the end of the 2.3-day record is 6.3km. This is slightly below the mean 2-day error of 6.8km from the best forecast of the set conducted on the previous leg of the cruise (see EN323 report). The early portion of the 13m drifter record in the well-mixed area closely resembles that of the deeper drifter. About half way through the record, the 13m drifter took a turn to the southwest and accelerated. Neither the turn toward the southwest nor the speedup of the drifter motion was captured in any of the simulations. The simulated trajectory in forecast 8 does have a slightly more southerly component to its movement, but not nearly enough to make it match the observations. Moreover, the southerly component in the model is fairly monotonic; it does not contain the the abrupt change in direction present in the observations. In the stratified region, both drifters move in a mostly southerly direction. Early in the record, there is an eastward component to the motion, which is more pronounced in the shallower drifter than it is in the deeper. Although the southward component of the observed drifter motion is captured in the simulations, the east-west aspect is not. Both shallow and deep simulated drifters move to the southwest. The reason for this discrepancy is not known. However, it is interesting to note that the southwestward movement of the simulated drifters was even more pronounced in the BPE run (Figure 0516.2). Assimilation of ADCP data partially arrested this southwestward motion; the simulated drifters do not go nearly as far to the southwest after assimilation. This result is consistent with that reported by Werner and Manning on the Edwin Link. Operational Products -------------------- Operational products generated on May 16 were based on forecast EN324_FC.08. This forecast had the highest skill, not only in terms of the mean forecast error (see table above), but also the error metric for the drifter phenomenologically closest to the tracer release point (the shallow drifter in the stratified region). Surface current predictions were produced for the tracer release site via a numerical mooring (Figure 0516.3). In addition, a cloud of particles was launched around the release site in the model solutions. The results are best visualized with an animation (see web page); plots of the initial and final particle locations are provided in Figure 0516.4. A time-annotated plot of the drifter trajectory at the center of the cloud (which represents the center of mass of the tracer) is shown in Figure 0516.5. Note that the original cloud release at 0.5m depth (run EN324_FC.09, based on Q4.3 of EN324_FC.08) ran into some problems with particles breaching the surface, even though these were fixed-depth drifters. The simulation was rerun as EN324_FC.10 with particles released at 2.5m depth, and the problem was avoided. May 17 ------ Dye survey operations continued until 1610. The ROV was tested later that afternoon. A larger scale dye survey was embarked upon at 1900 to define the extent of the tracer patch. Forecasting ----------- The May 17 Central forecast (EN324_FC.13) showed a slight improvement in forecast skill (9.3km mean distance) over the best forecast from the prior day (EN324_FC.08, 9.4km mean distance). Qualitatively, however, there was very little difference in the simulated drifter trajectories. One particularly striking aspect of the observations is the shallowness of the pycnocline in the stratified region. Surface stratification is very strong, and confined to a thin layer only 6m depth in some cases. The possibility that this vertical structure is important to the fate of the dye patch cannot be ruled out. It is therefore relevant to assess the extent to which this structure is present in the model. Examination of vertical sections of temperature extracted from the simulations reveal a comparatively diffuse thermocline (Figure 0517.1a). It was speculated that the diffuse thermocline could be a result of surface mixing that was too vigorous. Experiment EN324_FC.08h was conducted to see if a sharp pycnocline could be formed by increasing the surface heat flux to five times the climatological mean for this season. Clearly, this is an unrealistically large heat flux, but the intent of the experiment was to qualitatively assess the phenomenological impact on the solution. In fact, a much sharper pycnocline forms (Figure 0517.1b). However, this increase in surface heat flux is incompatible with the present boundary conditions on temperature, which are clamped to the observed values. This results in the spinup of an anomalous baroclinic jet around the bank, making the simulated circulation very unrealistic. It is clear that realistic treatment of this strongly stratified upper layer will require some effort involving several aspects, including the turbulence model, surface fluxes, and horizontal boundary conditions. As it turns out, the observed surface stratification described above may have strong lateral gradients which are not associated with the tidal mixing front. The satellite image from May 15 (which arrived just after the numerical stratification experiment was analyzed) shows a discrete patch of warm water straddling the Schlitz mooring line, oriented northeast-southwest adjacent to the 60m isobath (Figure 0517.2). Analysis of the alongtrack hydrography from the VPR surveys suggests this feature is associated with a salinity anomaly that is approximately 0.5 psu fresher than the surrounding water. There was much speculation about the origin of this feature. Operational Products: --------------------- Surface current predictions from the numerical mooring at the tracer release site were updated (Figure 0517.3) and provided to the chief scientist. May 18 ------ Survey of the tracer patch continued. A weak dye signal remained at the apparent center of mass. Several things appear to have happened: (1) dye in the very thin upper layer was driven by the easterly wind toward the front and into the well-mixed region where it was fairly uniformly distributed in the vertical; (2) dye which penetrated below the thin surface layer in the stratified area was mixed rapidly toward the bottom; (3) dye in both the well-mixed and stratified areas was dispersed in the along-isobath direction so that the patch expanded to about 10km in size; (4) the patch translated with the subtidal flow at a rate of approximately 10km per day. Forecasting Activities: ----------------------- Given that the dye had been mixed vertically both offshore and onshore of the front, the question arose as to how that might impact advection of the tracer by the subtidal flow. The chief scientist asked that we compare our forecast trajectory for the cloud of particles released at the surface with a cloud released at depth. Figure 0518.1 compares the trajectories of a surface particle at the center of the cloud (solid line) with one at 30m depth (dashed line). Westward translation of the deeper particle is not as rapid as that of the surface particle. A separation of 5-10km develops over the 1.76 days of forecast simulation time presented here. The decrease in forecast error that was noted in the May 17 central forecast was interesting. This improvement could have resulted from any of the following: (1) better atmospheric forcing (i.e. hindcast wind/heat fluxes), (2) more velocity observations, or (3) more recent velocity observations. In an attempt to distinguish between (2) and (3), a sensitivity experiment to the previous day's central forecast (EN324_FC.13) was conducted in which the older velocity data from the Edwin Link survey was not assimilated. The forecast error of run EN324_FC.14 increased from 9.3 to 10.1km, although the qualitative characteristics of the drifter trajectories remained the same (Figure 0517.2). Thus, it appears that more velocity data is better in this case. Given that, we chose to assimilate as much data as possible into the May 18 central forecast. Updated ADCP data from the Edwin Link received via email was merged and sorted with all available EN324 data. The mean forecast error was reduced to 9.1km from the 9.3km result of the May 17 central forecast. Once again, there was little change in the character of the drifter trajectories (Figure 0517.3). A note on the atmospheric forcing used in the May 18 central forecast: there was a period of several days in which the model forecast heat fluxes were not received from NMFS due to an email problem. That problem was corrected and transmissions were resumed in time to be incorporated in this day's central forecast. However, there was a gap in the heat flux record between the end of the May 14 transmission (day 136.50) and the beginning of the May 18 transmission (day 137.12). This gap was filled manually by copying the heat fluxes from day 137 into the those needed to complete the record for day 136. May 19 ------ The final survey of the tracer patch was concluded in the wee hours of the morning. A VPR section out to the 200m isobath and back was then occupied. We appeared to have reached a warm core ring at the most offshore extent of the track. Upon return to the study site, the six drifters which had been deployed prior to, during, and just after the tracer release were recovered. ROV operations were then undertaken, lasting until approximately 2300. A VPR survey of the study site was then begun in preparation for the second tracer release experiment, to take place in the pycnocline. Forecasting Activities ---------------------- A disk failure on the shore-based mail server prevented any email from reaching the ship. Therefore, we did not have the atmospheric weather or far-field ocean model forecasts needed to produce our operational products. We could have proceeded without the far-field ocean model, and forced the forecast system with wind predictions based on NOAA weather radio reports. However, it was not necessary to do so because shipboard operations planned for the following day (continuation of the VPR survey of the study area) did not require any input from the modeling team. The arrival of new drifter data from both Endeavor and Edwin Link sources provided additional opportunity for forecast evaluation. Observed drifter trajectories were compared with model results from the most up-to-date forecast, run EN324_FC.16. This simulation was re-run as EN324_FC.20 with the new drifter release points, shown in Figure 0519.1. The discussion below is split between surface drifters and those with subsurface drogues because there is a dramatic difference in our ability to forecast the motion of these two groups. Deep Drifters: In the well-mixed region, both the 16m and 26m drifters move southwest (Figure 0519.2). The shallower drogue moves slightly slower (approximately 10 km/day) than the deeper (approximately 5 km/day), indicating the presence of vertical shear even though water mass properties appear well-mixed. Simulated trajectories are quite similar to the observations, with separations at the end of the two-day record of only 1.6 and 3.7km for the shallow and deep drifters, respectively. The flow in the stratified area is more sluggish. Both the 12m and 36m drifters move west at speeds of 3 km/day or less. The simulated drifter trajectory at 36m agrees quite well with the observed trajectory, with only 3.9km separation developing by the end of the record. A larger separation (6.8 km) develops in the shallower case because the model drifter moves in a more southerly direction than is observed. Mean forecast error for the deep drifters is 4.0 km. Surface Drifters: Whereas the trend in the deeper drifters was movement toward the west and southwest, the surface drifters have a pronounced northwest component to their motion (Figure 0519.3). This aspect of the observed drifter motion is not captured by the simulations, resulting in a mean forecast error (11.9km) that is much higher than that for the deeper drogues. It is possible that this discrepancy may be associated with windage on the surface structure of the instruments; the directionality of the additional displacement is not inconsistent with what would be expected under the easterly wind conditions present during the deployment. May 20 ------ VPR survey continued; expected completion at 0400 May 21. Rain and fog during the day, and 20-30 knt NW wind blows up during the night. Email system operational again. Forecasting Activities: ----------------------- UNC forecast products were not received, so the central forecast EN324_FC.28 had to proceed without TDBC forcing, using winds extracted from NMFS source. The May 20 central forecast, its two predecessors and associated sensitivity experiments were evaluated against the available drifter data set described in the May 19 narrative. The results for deep and shallow drifters are presented in tabular form below: Mean Forecast Error (km) Surface Deep EN324_FC.23 May 17 central [FC.13 rerun] 13.5 3.4 EN324_FC.20 May 18 central [FC.16 rerun] 11.9 4.0 EN324_FC.24 (20) + EL9905 Mass Fields 12.3 3.0 EN324_FC.28 May 20 central 12.7 4.0 It appears that we have finally hit the point of diminishing returns with respect to "more data (and commensurately longer runs) are better." The May 17 central forecast has a lower forecast error than both the May 18 and May 20 central runs. This is a welcome result, as the longer simulations (up to 18 days from cold start to finish) were taking several hours to run. Also, it is interesting to note the improvment in forecast skill when the mass fields were updated with the EL9905 bongo data (run 24). This is consistent with the results reported by the Edwin Link team. Operational Products: --------------------- During the VPR survey, a remnant from the tracer release was found in a small patch in the pycnocline of one of the survey legs. When this came to light, the chief scientist requested a forecast of where that patch would be the following day so it could be resampled. At that time, the central forecast had yet to be run due to delays in the delivery of emailed forecast inputs. Instead, the central forecast from May 18 was rerun as EN324_FC26 with a numerical drifter inserted at the observed and place. A time annotated plot of the forecast drifter trajectory was provided to the chief scientist. That forecast simulation ended at 0800 the following morning, approximately one hour after the search for the patch was to begin. The same run was extended one day under the assumption of constant forcing (Figure 0520.1). Knowing that there had been a dramatic change in the weather, the May 20 central forecast was launched with the products that were available at the time; this included NMFS wind stress and heat flux. The strong northwest wind caused the simulated drifter to move much farther southwest, as to be expected with a strong northwest wind (Figure 0520.2). May 21 ------ VPR grid survey completed. It was decided to continue survey work for one more day before proceeding with the second tracer release. During those operations, the predicted location of the remnant patch of the previous tracer release (see May 20 "Operational Products" section) was sampled, and only background levels of dye were found. In retrospect this is not too surprising. The signal of the remnant patch at 6m was only just above background before the wind event began. During the wind event, the mixed layer deepened to approximately 15m. Given the dilution associated with both vertical and horizontal mixing, the remnant patch most likely became undetectable. It was therefore decided to render a "No Decision" in evaluation of the forecast skill in this case. Forecasting Activities ---------------------- Launch of the May 21 central forecast was delayed until after the 1930 email download in the hope that the UNC atmospheric and far-field model forecasts would arrive. Although those products were not received, we were fortunate that the NMFS extraction of the atmospheric forecast did. The central forecast was thus launched using NMFS winds and heat fluxes, but without TDBC forcing. During the day, additional experiments were carried out to follow up on other issues relating to the drifter comparisons conducted on May 20. Results from the Edwin Link suggested that forecast skill could be improved by decreasing the vertical mixing coefficient EKMIN and using shipboard winds when available. Run EN324_FC.24m was an identical twin to EN324_FC.24 except that the vertical mixing coefficient EKMIN was decreased from 2.0E-3 to 2.0E-5. Run EN324_FC.24mw was an identical twin to EN324_FC.24 except that ship observations were used to construct the wind stress and heat flux forcing files for the portion of the simulation time in which they were available. The results of these two experiments are shown below, together with the those from the control run: Mean Forecast Error (km) Surface Deep EN324_FC.24 (20) + EL9905 Mass Fields 12.3 3.0 EN324_FC.24m (24) EKMIN = 2.0E-5 13.4 4.0 EN324_FC.24mw(24) EL Atmos forcing 9.5 2.0 Contrary to the results reported by the Edwin Link team, decreasing the vertical mixing coefficient did not improve the forecast skill; in fact, the mean forecast error increased for both surface and deep drifters. However, it is important to note that the Endeavor experiment was not identical to that performed on the Link. For example, the Endeavor control run used model-derived winds, while the Link control run was based on surface fluxes derived from shipboard observations. It is quite likely there were other important differences, such as forecast timing parameters as well as the ADCP data used in the assimilation procedure. Furthermore, the drifter data sets used in the forecast evaluation are slightly different. The Link used their four drifters, while we used their four plus our six, separated out into surface and deep categories. In my opinion, the only conclusion that can be drawn from the combined experience thus far is that the smaller value of EKMIN does not necessarily improve forecast skill; the results are sensitive to which control run was used. This does not appear to be the case with respect to the atmospheric forcing sensitivity. Mean forecast error in run EN324_FC.24mw was 66 percent less than in the control run. This reduction in error is comparable to the results reported by the Link team, who experienced a 54 percent decrease. Thus it seems reasonable to conclude that shipboard wind and heat fluxes should be used whenever possible. We also experimented with the use of the weighted least squares inversion procedure rather than the fourier method that has been used up to now. Three different experiments have been performed, with the following values used for the condition number and error term weights: Rcond W0 W1 EN324_FC.21 0.0001 0.25 250.0 EN324_FC.22 0.1 0.25 250.0 EN324_FC.25 0.0001 0.1 1000.0 The weights (0.25, 250.0) were chosen for this particular case using the recipe derived in the December meeting at NCSC. The set of parameters used in EN324_FC.25 were the same as in the EN302 practice case from our December meeting. In each of these three experiments, the pertubation boundary conditions computed in the inversion procedure were so small (order 10^-12 in one case) that they did not make a perceptible impact on the BPE run. I am inclined to believe that either (1) I am making an error, or (2) there is a scaling problem in the output of the WLS inversion procedure. Operational Products -------------------- Early in the day, the chief scientist requested forecast products for the next day's tracer release in the pycnocline. As the central forecast had yet to be run, the planned position and time of the the release was incorporated into the previous day's central forecast and re-run as EN324_FC.29. Current predictions and forecast trajectory are shown in Figures 0521.1a,b. When the central forecast EN324_FC.30 became available later that day, forecast products were re-issued (Figures 0521.2a,b). That forecast simulation only ran through approximately 2000h local time the following day; it was extended an additional day in run EN324_FC.31 assuming persistence in forcing (Figures 0521.3a,b). The updated forecast suggested stronger advection toward the southwest, as to be expected with strong winds from the northwest. May 22 ------ During the night a patch of Calanus was tracked along an isopycnal surface far to the west, leaving Endeavor approximately 18 miles away from the operational area where Oceanus and Edwin Link were located. The VPR was brought on deck, and Endeavor steamed back to the operational area. VPR survey operations were begun to identify a location suitable for the tracer release. However, two aspects of the observed conditions were not favorable. First, the pycnocline was extremely sharp, 2 meters or less in vertical extent. This made it nearly impossible to inject without contaminating waters both above and below the pycnocline with dye. Second, Calanus was less abundant overall, and there was not a patch of organisms coincident with the proposed location for release of the dye. For both of these reasons, the chief scientist decided to postpone the tracer release until the following day. In the meantime, survey operations continued. Later that night, it became clear that the fluorometer on the VPR was picking up a signal from Houghton's dye released in the bottom boundary layer earlier in the day. Given the potential for interference between the two dye experiments, the chief scientist decided to relocate further west. Forecasting Activity -------------------- The May 22 Central forecast EN324_FC.32 was computed with all of the latest ADCP data and forcing products. As expected, the forecast skill based on the drifter deployment during days 135-137 continued to degrade slightly (see below). The duration of the forecast runs (14.5 days) is approaching the outer bounds of our experience thus far. A parallel forecast was initiated with a starting date oriented around the next set of drifter deployments and dye experiments, to begin on day 142. A new starting date on day 132 was chosen via the following rationale, working backwards in time: (1) we would like to have about 3 days of ADCP data to assimilate prior to the beginning of drifter/dye operations; (2) approximately 3 days of adjustment to the perturbation boundary conditions is needed prior to assimilation of the ADCP record; and (3) approximately 3 days of spinup from cold start is needed to adjust the BPE run. This suggests the following timeline: ---------------------------------------- Q4.2,3 * HS [Dye---> ------------------------------------------------------- Q4.1 * [Drifters--] [Drifters CS [ADCP-----------] Year Day 132 133 134 135 136 137 138 139 140 141 142 143 144 May 13 14 15 16 17 18 19 20 21 22 23 24 25 [Wind] Unfortunately, there is a wind event right in the middle of the ADCP record used in this run. It is clear that the wind event had a significant impact on the hydrographic structure and currents on the bank. The thin layer of stratification was mixed downward, and there appeared to be significant wind-driven flow toward the southwest. How to best accomodate these changes in our upcoming forecasting acitivities remains an open question. Would it be best to re-initialize after the wind event, or run right through it? How much data should be incorporated before and after? There is a myriad of choices to be made. Any suggestions would be most appreciated! Unfortunately, we have little new information with which to evaluate the new forecast; it will be a day or two before new drifter data start to come in. Although the run does span the earlier drifter deployment, those records begin in the model's adjustment period just after the hot start on day 135. Therefore comparison with those data is a poor measure of forecast skill. As to be expected, the skill is quite poor relative to prior runs (see below). The final run completed on May 22 was an attempt to improve on the central forecast based on the experience of the preceding days. It was initialized with the mass fields from the Edwin Link, and forced with their wind observations when available. Consistent with prior experience, forecast skill improved significantly. Mean Forecast Error (km) Surface Deep EN324_FC.32 May 22 Central 11.6 4.4 EN324_FC.33 (32) started later 14.7 9.0 EN324_FC.34 (32) EL Mass fields, Wind 11.4 2.0 Operational Products -------------------- Forecast products were issued for the tracer release based on the day's central forecast (Figures 0522.1a,b). When the improved forecast EN324_FC.34 became available, operational products were re-issued (Figures 0522.2a,b). May 23 ------ The morning of May 23 found Endeavor further to the west, engaged in survey operations to find a location suitable for the tracer release. Generally the conditions were more favorable than what had been observed further to the east the day before: the pycnocline was thicker, and patches of Calanus were more abundant. Tracer release operations began at approximately 1345 and were completed by 1430. Drifters were deployed both inside and outside the tracer patch. A survey of the initial tracer distribution ensued, followed by ROV operations. A larger scale survey to delimit the outer bounds of the patch was then begun. Forecasting Operations ---------------------- The May 24 Central forecast (EN324_FC.36) was computed using all the latest forcing products and ADCP data sets. In a continuing effort to study forecast timing issues, a parallel forecast (EN324_FC.37) was computed with a later starting date. This forecast pair is analogous to runs 32/33 conducted on the May 22. Operational Products -------------------- At 0900 the chief scientist provided the forecasting team with an approximate time and location for the tracer release. As the May 23 central forecast had yet to be run, pycnocline current predictions and forecast trajectories for the release were based on the prior day's best forecast (EN324_FC.34), re-run as EN324_FC.35. The results were analyzed and provided to the chief scientist (Figures 0523.1a,b). When the pair of May 23 forecasts (runs 36 and 37) were completed, clouds of particles were launched at both tracer release locations (Houghton and Ledwell). See animations for best presentation of these results. Forecast trajectories for the center of mass of the Ledwell tracer are shown in Figures 0523.2a,b. Little net motion is forecast in run 36 until the strong southerly winds predicted to arrive the night of May 24 accelerate the patch toward the east. Qualitatively, forecast 37 is similar, except that the tendency for eastward motion is more pronounced earlier in the deployment. As yet, there is no discernable motion in the tracer patch, which would tend to favor the central forecast (run 36) as having more skill. However, this evaluation is highly qualitative. As more information becomes available on the movement of the tracer patch and drifters, more quantitative evaluation will be possible. Forecast trajectories for the Houghton release are shown in Figures 0532.3a,b. Movement of this deeper patch is more sluggish than that predicted for the pycnocline. Again, eastward motion is more pronounced in run 37 than run 36. In this case, there are some qualitative observations of tracer movement with which to compare the model forecasts. In the radio communications amongst Endeavor, Oceanus and Edwin Link on the morning of May 23, it was reported that the tracer patch was bounded to the south by 41 05'N, to the east by 67 22'W, and to the west by 67 30'W; the northern extent was yet to be determined. These delineations are shown as a dashed line on Figure 0532.3a. In radio communications on the morning of the May 24, it was reported that the patch was bounded to the north by 41 08'N, to the east by 67 25.5', to the south by 41 04'N, and to the west by 67 31'W. These delineations are shown as a solid line in Figure 0532.3a. Thus, there appears to be a slight westward drift in the observed tracer patch. Again, the central forecast (EN324_FC.36) appears to be qualitatively more consistent with the observations. May 24 ------ Survey operations continued. Once the large scale survey of the dye distribution was completed, an extensive set of lines were occupied to map the Calanus patches in the area. At approximately 2000h, survey of the dye patch recommenced. Throughout these survey operations, periodic lines to the front were occupied to measure the distance between the front and the dye/Calanus patches of interest. Forecasting Operations ---------------------- The May 24 Central forecast EN324_FC.38 was computed with all the latest forcing products and data records. Once again a twin experiment was conducted with a delayed starting time (EN324_FC.39). The first set of drifter observations from the latest deployment arrived (3 tracks from Manning), permitting some quantitative evaluation of forecast skill. Figure 0524.1 compares the observed drifter trajectories with simulated tracks from the central forecast. In general, forecast skill is not as good as we have become accustomed to in earlier deployments. The 8m drifter released at the 65m isobath is the only case in which there is appreciable skill; a 2.8km separation develops between the simulated and observed trajectories as they move to the southwest during the two tidal cycles of observations. Comparisons with the 13m and 33m drogues are poor, with separations in excess of 9km developing during this relatively short data record. The 33m simulated drifter moves south-southeast in contrast with the observed movement to the southwest. The west-southwest direction of the simulated 13m drogue is approximately correct, but the movement is far too slow. Furthermore, there is a large error in the tidal phasing by the end of the record. We have not seen this type of error in any of our forecast experiments thus far. Perhaps a mistake may have been made either in data processing or in incorporation of the simulated drifter into the model solution. We will look into this further. Even if there is a problem with the 13m drogue, there is no denying this forecast is of inferior skill. In radio communications the morning of May 25, the modeling team on board the Edwin Link reported similar results. The discussion led to some speculation that decrease in forecast skill may have something to do with the lack of spatial overlap between the drifter data and the current ADCP velocity measurements which are being assimilated. For reasons described in earlier transmissions, Edwin Link and Endeavor have separated by about 30km in terms of their operational area. Given the recent outage of the ADCP on the Edwin Link, it is primarily Endeavor's ADCP data which are being assimilated. Thus, in the forecast evaluation described above, we are essentially trying to predict the path of Edwin Link's drifters some 30 km east of the area in which recent velocity measurements are available. This could explain the lack of forecast skill. Shortly we will have the opportunity to compare the model results with Endeavor's drifter tracks, which are closer to the area of velocity observations; such a comparison should help shed light on these issues. Operational Products -------------------- An updated forecast trajectory for the dye patch based on the May 24 central run was provided to the chief scientist (Figure 0524.2) The results suggest a brief period of eastward movement associated with the southerly wind event, then motion back toward the west. Given the issues brought to light by the most recent evaluation, the operational products for this day were provided with the caveat that forecast skill was "uncertain relative to prior results obtained thus far on the cruise." May 25 ------ VPR survey operations continued. Forecasting Activities ---------------------- Much of the day was spent investigating the potential causes of the decrease in forecast skill we have experienced in the last drifter deployment. A set of five sensitivity experiments were conducted around the May 24 central forecast EN324_FC.38. The first (run 40) represented an attempt to "restart" the forecast system based on the canonical forecast timing parameters used at the December meeting at NCSC. This differs from the timing scenario described in the May 22 narrative in that ADCP data during the drifter deployment is assimilated. Thus, such a run test our ability to hindcast the drifter trajectories rather than to forecast them. The timing diagram is as follows: ########### EN323_FC.40 ########### ---------------------------------------- Q4.2,3 * HS [Dye---> ------------------------------------------------------- Q4.1 * [Drifters CS [ADCP-----------] Year Day 134 135 136 137 138 139 140 141 142 143 144 145 146 May 15 16 17 18 19 20 21 22 23 24 25 26 27 [Wind] Analysis of the results from EN323_FC.40 revealed some large spikes in the ADCP data record. The bad values were edited out, and simulation EN323_FC.41 was run with the new data record. Run EN323_FC.42 reverted back to the mass fields created from the last broadscale survey conducted on Oceanus; TDBCS were turned off in run EN323_FC.43, and atmospheric forcing was turned off altogether in run EN323_FC.44. The table below shows the mean forecast error for each run, in addition to the separations for each drifter. As these runs were being carried out, it appeared to us that there may be an error in the launch point location of the 13m drifter provided in the .ind file. This was corrected in run EN324_FC.41cdr, and resulted in significant improvement in the forecast skill. Beyond that, the results of the sensitivity experiments were qualitatively consistent with our experience up to this point (i.e. runs 42-44). Still, the forecast error remains higher than it was preceding the series of wind events we have had recently. Moreover, there are qualitative differences between the simulated and observed drifter trajectories (Figures 0525.1a-f). The causes are still under investigation. 13m 8m 33m Mean EN324_FC.40 (38) new timing (10.2) 2.9 8.2 (7.1) EN324_FC.41 edited ADCP file (10.2) 5.7 5.5 (7.1) EN324_FC.41cdr edited ADCP file 5.4 5.5 2.1 4.4 ## EN324_FC.42 OC Mass fields 5.0 4.9 4.8 4.9 EN324_FC.43 TDBCS off 5.1 4.8 5.2 5.0 EN324_FC.44 Tau, Q off 5.0 6.2 4.3 5.1 ** () indicates statistics which are affected by the suspected problem in the release location of the 13m drifter. ## these numbers were subsequently found to be in error; see below. Operational Products -------------------- No new operational products were issued on May 25. May 26 ------ VPR and dye surveys continued. Forecasting Activities ---------------------- A second battery of sensitivity tests was conducted using an updated set of drifter data, including seven drifters deployed from Endeavor. We take EN324_FC.45 to be the central case for the comparison; it is a re-run of EN324_FC41cdr in which the numerical drifters maintain a fixed depth. Note that the set of runs described in the May 25th narrative mistakenly used fluid-following drifters. Comparison of runs 41cdr and 45 shows that it actually makes little difference in this particular case: 13m 8m 33m Mean EN324_FC.41cdr Drifter w~=0 6.2 6.1 6.1 6.1 EN324_FC.45 Drifter w==0 6.3 6.0 5.9 6.1 We now compare runs EN324_FC.45 through EN324_FC.48 for all ten drifters, separated into shallow and deep categories. Run EN324_FC.46 uses wind observations from Endeavor; TDBC forcing is turned off in run EN324_FC.47. Finally, TDBC forcing was turned back on after correcting an error in the code being used on board Endeavor to generate the .cbc file from the UNC products. Note that this problem was particular to the version of detide_adcirc being used on Endeavor, and should not affect other users. Its effect was that input data from only the first file (01-02 May) actually made it into the .cbc file. The corrected TDBC forcing was turned back on in run EN324_FC.48. Deep drifters: ID 087 234 200 037 022 010 003 006 z 13m 8m 33m 19m 19m 19m 16m 34m Mean 45 Control 6.3 6.0 5.9 5.5 6.3 2.3 10.6 2.1 5.6 46 EN Winds 6.4 6.3 7.0 4.1 4.0 2.0 7.8 4.3 5.2 47 TDBCS off 6.2 7.8 6.1 4.0 4.1 2.1 8.0 3.7 5.3 48 TDBCS2 6.5 9.9 6.4 8.0 7.7 3.1 9.2 6.0 7.1 Shallow drifters: ID 009 011 z 2.5m 2.5m Mean 45 Control 16.1 6.5 11.3 46 EN Winds 12.3 6.6 9.5 47 TDBCS off 12.4 6.9 9.7 48 TDBCS2 11.5 8.8 10.2 See Figures 0526.1a-l for comparisons of the simulated and observed trajectories. Once again, the forecast error for the surface drifters is much higher than that of the deeper drogues. The use of shipboard observations for hindcast winds improved the forecast skill for both deep and shallow groups. Turning the original TDBC forcing off caused a slight increase in forecast error. Turning this forcing back on after the Endeavor pre-processing procedure was corrected significantly degraded forecast skill. This led to a closer examination of the space-time structure of the elevations contained in the .cbc file. Comparison with files generated on the Edwin Link are currently underway; those findings will be reported at a later date. The May 26 Central forecast EN324_FC.49 was computed with all the latest forcing products and ADCP data. Just after its completion, new drifter data arrived. Its evaluation will be conducted using the new data, and will be reported in the May 27 narrative.