Monday, December 5, 2016

GRIB Formatted Regional Wind Forecasts – Review and Updates

Each time we have a regional weather routing project to work on, I have to stop and remind myself about what data are available, and what are the specs.

By specs I mean how often and when is the weather prediction model run, when are the data available after that, how far out does its forecast extend, what are the time steps in the forecast, and what is the resolution of the data, ie a wind point every 1.3 km or every 30 km, etc. And of course, what regions do they actually cover, ie how far offshore, how far north into Canada, etc.

To help myself and others with this, I have compiled here a set of notes on available forecasts and summarized the crucial specs in a Table below for quick reference.

There are two broad categories of numerical model predicted wind data: Global and Regional. Global is for mid ocean or well offshore; regional models are for near coastal and inland waters.

The best wind data for US inland and near coastal waters (HRRR model) were only available as commercial products until the last few weeks or so when Saildocs made them available to the public. Commercial products are only legal in sailboat racing up till the starting gun, so this new development from Saildocs will have a big influence not just on sailboat racing—the data can now be used throughout any race—but this Saildocs service has a potentially huge impact on all of sailing across the US. It will likely take many months and well past next summer to fully appreciate this, but stand by for real change.

We remain grateful to the commercial outlets that did pioneer the availability of the HRRR data last year, specifically Ocens and SailFlow, and their presentations of the data within their own systems are still just as convenient as they have been. Indeed, as more mariners learn the value of this data now that it is publicly available, it should only enhance the value of the commercial services they offer.

Shortly, I will supplement this note with even newer developments from the Seattle-based Mac program LuckGrib that add to and extend the revolution. This should be announced shortly. Their next update includes HRRR, NAM, ASCAT satellite winds, and much more. The scatterometer data (satellite winds) are a positive way to confirm any of the numerical forecasts. Having these data available as GRIB files is a boon to our analysis underway.

The main focus of this note are the regional wind and pressure models. For free, public global models, the dominant model in use worldwide is the GFS model from the US, along with various ensemble presentations of that data. Ensemble means the same model is run using different initial conditions and times, and the results are averaged and compiled into a single forecast that should in principle be more dependable than any single run.  Ensemble forecasts will have lower resolution and less duration than the individual runs.

Some argue that the European global model (ECMWF) is superior to the GFS on some level in some cases, maybe even on average, but that model output remains commercial data, notably from Squid Sailing and PredictWind. With access to this EC data we have the opportunity to compare that with GFS in  more cases.  From a practical point of view, considering all related uncertainties in this process, it is difficult to anticipate many instances where the difference in forecasts (GFS vs. ECMWF) will affect our routing decisions. It is not the subject at hand, but all of these global models can be wrong in substance or in timing in various conditions.  And if carefully compared to actual weather maps and text forecasts, the GFS GRIB data should meet the global forecast needs of most mariners. There are standard precautions to be taken.  The GFS specs are included in Table 1 for comparison to the regional data.


NAM (2.5, 6, 12 km), North American Mesoscale Forecast System, is the flagship regional weather forecast model run by NCEP/NOAA.  It is available for several regions of US waters in different resolutions. A good choice for longer term forecasts than available from the 18h max of HRRR, and for comparison with the digital form of official NWS forecasts in NDFD. The convenient coverage area maps are from the LuckGrib viewer. ( In principle this model should be better than the COAMPS version from the US Navy. See model discussions and links. )

NAM North America, 12 km. Would be used when one of the other higher res NAMs was not available.


NAM AK, 6 km.  Note that it covers Pac NW at twice the resolution as the conus N. American above.
We might get the impression from this official coverage map that we have a trick play in the Pacific NW and can get 6 km data from the AK run as opposed to just 12 km from the North American run, but strangely this is not the case. For some reason, the full AK region is not populated by data and we miss the corner we care about, as shown below.

Actual NAM AK data downloaded and viewed in LuckGrib. Note the missing data in the corners.


NAM Caribbean, 12 km, same as N. America, but note the super hi res of the central region below.


NAM Puerto Rico, 2.5 km. This is about state of the art in resolution, covers Virgin Islands as well.

NAM HI, 2.5 km. Likely best for planning approaches till the HRRR is available.
The HRRR is not even as high a resolution as this one, but it is updated every hour, compared to NAM every 6h,
especially important in the presence of tropical storms. 
NAM data are available from within LuckGrib and Expedition (Expedition does not include NAM AK).

NDFD Oceanic (10 km). This is unique GRIB data in that it is not pure model output, but instead is a digitized format of the official NWS forecast. In principle this should be better than any one model for the region covered, because the human forecasters that created this data have access to all models and they have chosen which they think is best at the time.  Unfortunately it only goes out to Lon 133º W, and then is available again in HI, as shown below. It can be useful in the Strait of Juan de Fuca as well. See The National Digital Forecast Database.

The NDFD data include wind and significant wave height of combined seas. There is no pressure data.  See the NAM or HRRR for pressure.


NDFD data 3 km, goes about half way up Vancouver Island. This coverage map from the NSDF site.
Pac NW limits on NDFD. About halfway up the west coast of Vancouver Island then to top of the Gulf Islands.


NDFD HI coverage, 3 km. Another view from an Expedition screen cap.

NDFD data are available from Expedition, Saildocs, and the Ocens WeatherNet (commercial data).


HRRR (3 km).  The High Resolution Rapid Refresh model is updated hourly, corrected for all the latest observations,  but it only extends out 18 hours. It is available from LuckGrib, Expedition, Saildocs, and Ocens.  If you have live Internet connections while underway, these are likely the best wind forecasts available.

The data are available about 1.5 to 2.5h after each runtime. That is, a model run at 1400 should be available to you by about 1530. Thus if you download an 18-h forecast at 1535, your first data point could be at 1400 and extend to 0800 the next day.  Download at 1525, and your first point would be 1300, extending to 0700 the next day. The resolution between wind data points is fine enough to do some level of automated sailing route optimization, although it will take more experience with this to learn the limits.  Wind and pressure are available.

The HRRR model actually has wind forecasts every 15 minutes, but the 15-min steps data are not yet available in GRIB format from any source I know of.... ie it is indeed rapid in refresh and time resolution, as it was motivated by tornado and squall forecasting, which it can indeed do!

HRRR 3 km. A huge coverage for this timely and hi-res data. Extends N to Seymour Narrows in BC, and covers all of the Great Lakes.
Screen cap of coverage from LuckGrib.
Pacific NW limit of the HRRR data. 


Sample HRRR data. 18h of forecasts. The data go north to Campbell River... ie perfect for the R2AK run to Seymour Narrows.

And for those navigating by portable device, there is a super nice way to see the HRRR wind data underway, but not in a form for numerical weather routing. Use the SailFlow app in iOS or Android, and surf around in the set up to find HRRR.


Summary of the Regional Forecasts compared to the Global GFS.
See the maps above for the coverage regions.


Public Regional Models
Resolution Timing
km nmi deg coverage model run  delay  duration  time step
HRRR 3 1.6 0.03 map HRRR hourly 1.5 to 2.5 18h 1h
NAM 12 6.5 0.11 N. America 00, 06, 12, 18z 3-4h 84h 1h/36, 3h/84
NAM 12 6.5 0.11 Caribbean 00, 06, 12, 18z 3-4h 60h 1h/24, 3h/60
NAM 2.5 1.3 0.02 Puerto Rico 00, 06, 12, 18z 3-4h 60h 1h/24, 3h/60
NAM 2.5 1.3 0.02 HI 00, 06, 12, 18z 3-4h 60h 1h/24, 3h/60
NAM 6 3.2 0.05 AK 00, 06, 12, 18z 3-4h 60h 1h/24, 3h/60
NDFD 10 5.4 0.09 map NDFD 00, 06, 12, 18z 3-4h 168h 3h/72, 6h/168
RAP 13 7.0 0.12 map RAP hourly 3-4h 21h 1h
Public Global
GFS 28 15.1 0.25 Global 00, 06, 12, 18z 3-4h 384h 3h/72, 6h/180, 12h/384
Predict Wind Commercial products, G=GFS based;  E = ECMWF based
PW/G,E 1 0.5 0.01 selected 00z, 12z 6-12h 36h 1h
PW/G,E 8 4.3 0.07 coastal 00z, 12z 6-12h 168h 3h
Public Graphic only, no GRIB
UW WRF 1.3 0.7 0.01 Pac NW 00z, 12z 6-12h 60h 1h


Other Regional Models


PredictWind PWG/PWE (1 km, 8 km). The 8-km data are available in coastal waters,  nearly globally, and the 1-km data are for many popular sailing areas worldwide.  For more data on the model specs and available times, see this link at predictwind.com.

Predict Wind commercial hi res data. Large regions are 8 km, small are 1 km. A similar distribution of data options apply globally.
This is a screen cap from PredictWind Offshore app. Set the route across your route and it will select all data needed.

This product has several advantages when preparing for a race or not racing when commercial products are restricted.  First it is super easy to access in Expedition or by email, and the 1 km hi-res models extend out 36 hours,  compared to 18h with HRRR.  They also have a convenient (Mac or PC) app called Predictwind Offshore that lets you either request and look at the data online, or what is often more convenient, it will prepare a template request that you just email to them when you need an update. The larger regions of 8 km are 390 kb each, the smaller regions of 1 km data are 250 kb each.  The 8 km (also hi res) goes out a 7 days—like the lower res GFS, more or less into the realm of the unknown.  The only drawback to the data is its age, since it takes time to incorporate the ECMWF and other adjustments they make, the data can be 8 to 12h old when it arrives.

Predict Wind also has a weather routing/optimizing feature that seems to work pretty well. You can also use custom polars for the analysis.  For inland waters, however, it does not account for currents, which leaves Expedition at an advantage for that— but I should add here, if you do not have good current data, then you could be better off without using currents at all!  PW does offer and use ocean currents for the ocean routing. They remain unique with their new inclusion of ECMWF data.


UW WRF (1.3 km).  Unique to our local Pacific NW waters, we should always remember the UW WRF model, which is likely as good as any, but only available as a graphic format;  there is no grib format. The  model is run at about 00 and 12z daily and data would typically be 8+ hr old when we get it, and indeed they might not be there when you want them. This is a public service of the UW Atmospheric Sciences Dept.,  with no guarantees.

UW WRF model, run every 6h extends out 60h.

This model could be the best model for the region, but it is not updated often enough to compete with the HRRR, which is updated with all the latest actual wind observations every hour.

Nevertheless, it is worth printing out the UW data before the start, and with a good connection offshore, perfectly legal underway.  You can request the image forecast underway using saildocs.

No comments:

Post a Comment