The RTOFS data is very easy to obtain underway in GRIB format from several sources; the NCOM GRIB data can be obtained underway just as easily using LuckGrib on a Mac, but acquisition using a PC requires several custom steps. Racing sailors (always) and fishermen (often), and to various extents every mariner cares to have the best possible current forecasts, so the question arises as to which of these two models is best where there is overlapping coverage?
Our goal is to look into ways to make this evaluation as best we can, as we continue to look for publications that present this data. We also remain aware of the practical solution that avoids this work in some cases. Namely, we download both models underway, then measure the actual current we experience at the moment (the vector difference between COG-SOG and HDG-BSP), and see who is best... or if either is close enough to be useful. But this does not solve the main task of computing optimum routes based on wind, waves, and currents. For this we need some guidelines to what might be best current data.
Two ways to do this comes to mind. One is look for buoys that record currents and compare them over some extended period—ocean currents vary far more than our pilot charts or older textbooks might imply. But these point comparisons do not reflect the crucial flow patterns, especially the mesoscale eddies that can put much stronger current in our path than we might have guessed.
The other method is to compare coastal current measurements from HF radar stations with the two forecasts, and this note is a first step at that.
The validity of this approach is yet to be seen. Our best hope is a qualitative comparison. We would consider ± 30º and ± 0.3 kt as full agreement. Even ± 45º and ± 0.5 kts would be encouraging, meaning to know the ocean currents to that latter level of accuracy would be beneficial to ocean routing—as opposed to having them be wrong by more than that, which would be detrimental. The bar is low: do no harm.
Coastal currents include varying levels of tidal influence, and it is not clear (to me) how the models account for this. We can zoom in onto the weaker currents near shore and watch them vary on a tidal period, so some influence is definitely there. We can also see nice demos of a rotating current in some cases. HFR stations offer various resolutions from 0.5 km to 6 km along with 25-h averages over these. We have used the 6-km data as a compromise between the RTOFS (9 km) and NCOM (3 km).
For a quick overview we use a graphic analysis with readily available tools. We get the HRF graphic data from Coastal Observing Research and Development Center and we download and display the model forecasts using the Mac app LuckGrib. Then we capture the current arrows from the HFR data and overlay them on the forecasts using Photoshop.
Below is a sample of the HFR data from the link above.
The insert shows the HI data we used below. Current speeds are scaled to the colorbar on the left.
To compare speeds, we have made custom contour fill colors in the model display that match the colorbar used in the HFR display as shown below.
The color bar from the HFR (black border) is shown here on top of the gradient design page from LuckGrib, which has a color picker that allows for precise duplication of the colorbar, and an option to mute the colors with an Opacity control. A sample of the use of this scheme is below.
The colored arrows are from HFR, which are here overlaid on the model prediction for the corresponding times. The black arrows are from the model, and the speed from the model is the background color, taking into account that these colors had to be muted some (opacity control) so all arrows show. Actual colorbars used are shown in each of the pictures below.
Thus the yellow arrow at A means HFR and model speed agreed very closely. Arrow C is slightly darker (stronger current) than arrow D, but both are in agreement with the model (keeping in mind its muted color, and our lax tolerance on agreement). The orange arrow at B is about in full agreement. On the other hand the model forecast at arrow E is light blue on a yellow background, which means the model currents are too strong by maybe 0.5 kts or so. Likewise the the observed currents at F are notably stronger than forecasted.
That is the system we apply to the overlays shown below.
Below is the same location 4h later.
Again, similar with some bias to the NCOM, which does, for example, a good job on the eddy in the SW corner. Both miss details of the flow to the far west of the island. NCOM a bit better on the speeds. Again, the HFR data are 6 km resolution. At the end of this note we show all stations compared to 25-h HFR averages.
(Note we chose this comparison at 4h later by simply watching the model predictions to see a change... without further thinking! In retrospect, this would have been better done in increments of 3h because the GRIB viewer had interpolated the 4h data between the 3h and 6h forecasts. The HFR data were indeed at 16z. Again, we learn from this first pass. There is a way to shut off the LuckGrib viewer interpolation in both space and time, I just forgot to do it.)
Now we look at currents in or bordering the Gulf Stream, where the colorbar scale changes from max 2 to max 3.
Here we see notable discrepancies between forecasts and measurements, and I am not sure what to say about it. We have sent inquiries to experts. The models are similar, with the NCOM having slightly closer speeds. In both cases we see currents >3kts, so the colorbar could have been optimized. (We have learned several ways to improve this first pass study.) Note that both models call for a "two-lane" highway of current with a weaker band between them—which would be more interesting if the data actually reflected that. If the measurements are correct, the models have a problem in this region.
Going farther north...and noting that except for the 16z HI data, all else are 12z, Aug 9, with both forecasts being 12 h out from a 00z computation.
Again, it is not clear how to interpret this. RTOFS seems to have the inside wall and the speeds better matched. Also it is not clear if the upper part of the measurement data might not be from a neighboring station and what implications that might have.
In short, we present this data for the hopes of getting feedback from experts.
Below shows the 25h averages HFR data for each of the above stations at 12z.
The RTOFS does notably better in this case, with the general flow and speed about right over most of the area. The higher-res NCOM, on the other hand, still maintains eddies and flows that were not there when averaged over the past day, and is wrong more often than right.
Again, this remains difficult to understand without guidance from local mariners or scientists. There appears to be a cross Gulf Stream component that is persistent over a daily average, whereas the models just run straight up the coast.
Here we see more of what we might expect on a daily average. Namely, the average directions of the forecasts are about right, but the speeds are lower because they have averaged out flow in different directions.
Several implications have arisen. If the measurements are correct, then the 12z and 16z single-time data above shows the exceptions we might expect... in these waters. These measurements, however, are all coastal waters, so the behavior offshore could be different, presumably simpler.
We do not see reasons to change our advice over the past years. If you see a prominent eddy forecasted in your waters, then chances are there is an eddy somewhere around there. The models do after all use measured water temperatures and sea level elevations that can mark these anomalies, and they watch them over time. But they might not be precisely where they are plotted, nor of the same strength.
Such forecasted eddies or rivers of current call for careful measurements of the current plus sea water temperature on board. If you drive into one, watch how the current is changing as you proceed, and from that you can judge how to navigate to stay in it if favorable or how to likely best get out of it. If you see one that you are not in, but it could be favorable, then the layout of the forecast can dictate the value of trying to get in it. The time scale of these eddies and meandering bands of stronger currents are long enough to have initial planning done several days before getting to them. On the other hand, in near coastal waters we can see that this is a near real time exercise to find the best current.
Another point that comes clear is that we here at Starpath need to learn more. We are slowly gathering data on known model evaluation as well as known HFR accuracy. It seems we must be cautious in drawing any conclusions from this rough study.
We might learn more by comparing to the OSCAR data, which are averages over the past 10 days, without actual forecasting of the future. OSCAR currents are readily available underway in GRIB format from saildocs.
Based on notes in the References below, a first hand guess at this point is RTOFS is in general overestimating the surface current speeds and OSCAR is (understandably) underestimating it. I will follow up here as we learn more.
—A special thanks to Toby Burch for assistance with the graphics manipulations.
— GOFS 3.1 Validation data (2017, the underlying model used by RTOFS)
— OSCAR validation data. About OSCAR
— Description of RTOFS (2009) with some total flow comparisons in Florida Strait.