Wednesday, February 17, 2016

Sailboat Racing on the Salish Sea—Part 1. Northern Waters

The Salish Sea refers to the combined waters of the Pacific Northwest. It is highly unlikely that any Salish native ever used the term in this sense, but there is enough interdependence of these waters and varied reasons to refer to them as a unit that the term has been adopted by the Geographic Boards of both the US and Canada. For our purposes at hand, we break this term up into Northern Waters, being the Strait of Juan de Fuca, the San Juan and Golf Islands, and the Southern Straits of Georgia, leaving the Southern Part to be made up of Admiralty Inlet and Puget Sound. Note that though we do agree with Canada on the name of this grand combined waterway, we do not agree on the name of the Strait of Juan de Fuca, which they insist on calling the Juan de Fuca Strait!

And I should also add, that the reason we break up the waters this way is because the University of Washington Department of Atmospheric Sciences produces their highest resolution numerical weather predictions on these two separate map regions.  Most local races are in one or the other of these Parts, but there are a couple notable races that sail between them, and those will give the navigator a bit more work to do.

The task at hand is to describe and compare the best possible wind forecasts for these regions. To my knowledge, the best forecast is from the UW WRF model, meaning that they start with the best regional model data and then add to that the best practical topographic and surface data.  This model is run twice a day, initialized at 00z and 12z, and produces graphic map forecasts, one per hour out to 60h. Thus when the 00z run is completed, the first valid time of the sequence will be 00z. Likewise with the 12z. The 6th map of the 12z sequence will be valid at 18z, and so on. But these computations do not run right at 00z or 12z; it takes some time for the assimilated data to get into the system, and then it takes some time to actually run the computations. On top of that, UW is running quite a few models on their computer, and the high resolution run we care about (1.33 km between data points) is the last of the set to run.

Consequently we do not get the 12z hi-res data online until about 00z the next day. Likewise the 00z data becomes available about 12z that day. In local times, we get the newest data each day about 4 am and 4 pm PST, and these data will then be 12h old when we get it. So what are in principle the best data are severely limited by being very old when we get it.

The Figures below show what the ocean maps look like over this region compared to the UW data for the same time. The global models used for the ocean data do not show any of the crucial structure to the isobars and subsequent wind patterns that are created by the shape and texture of the terrain.


Figure 1. Surface analysis from OPC.

Figure 2. UW WRF model prediction initialized at same time as the surface analysis above. This combines both UW maps they call "Northern Waters" and "Puget Sound"


To show how dynamic the patterns are over the inland waters, look at the animation below that shows how these maps evolve. We have highlighted a couple key isobars, because it can be difficult to identify the isobars in their maps. (Meteorology theorists do not have a proper respect for surface pressure. They should go sailing more often!)


Figure 3. Video showing how the UW-WRF model winds and isobars evolve hourly, along with notes on where to access the data.

* * *

As of a few months ago,  however, we now have a very attractive alternative or supplement to the UW data, thanks to the Seattle based ocens.com.  They now offer the High Resolution Rapid Refresh (HRRR) model data in grib format—NWS scientists, by the way, pronounce this abbreviation "her," and refer to this as "her-data."

You can see low-res presentations of the data online from the primary source (use domain = sea), but it is not as detailed as we would hope for in the format available there.


Figure 4. HRRR wind data for domain = sea, viewed online. It is not a hi-res presentation, but we can make rough checks.  The wind at Pt Townsend was 9 kt at 090 at the time of this map, which looks consistent. Buoy Juliette was 4 kts at 150, also consistent.
Figure 5. HRRR "surface pressure"  data for domain = sea, viewed online. BIG CAUTION HERE: What they call 'surface pressure' is actually station pressure, totally dominated by the elevation of the land, and no value at all to weather routing.  They do not seem to have mean sea level pressure online.  Again, we need more meteorologists going sailing.
For tactical use underway, we need these data in grib format so we can read digital values of wind and pressure. Then we can load the data into a program like Expedition that might give us useful route guidance, even on these inland waters. Thanks to Ocens we now can get just that.  The data they offer is a commercial product (there is a charge to set up the account, then a small fee for each download), so when racing we can only use it before the start.

Elsewhere I have posted video notes on how to download these data from Ocens using their WeatherNet application for both PC computers and for iPad. Their Mac products do not support this new data so far.  Refer to those links for the details. Even with an Ocens account there are a couple crucial nuances to actually finding and downloading it.

The Ocens HRRR data is packaged into separate wind and pressure files, with 3h of forecast in each grib file. So to get the maximum forecast of 15h of wind and pressure you will need to download a total of 10 files. Since the files are high resolution, you will want to set the Lat-Lon selection region as small as possible, not to mention that there is not much logic in larger regions for these local forecasts. They are also updated hourly with full assimilation of all observations at each new run.

The motivation for this model was to forecast tornadoes and thunderstorms, which accounts for the rapid refresh and high resolution. In fact, we have every reason to think that we could even forecast squall behavior on some level with this data.  Recall, though, this is only for US waters.

Static samples of what this HRRR data in grib format looks like is shown below, viewed in Expedition. Each download of the data provides 15 successive hours of this type of forecast, and that forecast model is updated every hour with latest information from all buoys, lighthouses, air ports, and radar, as well as from commercial aircraft flying over the area with observations of winds aloft that go into the model as well, with important influences on the surface winds and pressures.


Figure 6. HRRR data from Ocens viewed in Expedition. Top is Eastern Straits of Juan de Fuca, Bottom is central Puget Sound.
Though the focus here is Northern Waters, I pause for a quick look at Figure 6 in motion, because it shows what detailed forecasts we have now for all of our inland waters.

Figure 8. Video example of HRRR data in Expedition, also illustrating their meteogram function.

* * *

Now with that long introduction, we get to the main point at hand: How do we gather together the forecasts from these two sources (UW-WRF graphics and HRRR gribs) to make our own best guess of what will take place?

The most elemental approach would be to download the UW images of the most recent forecast that covers the race you care about. Then print out each page in color and put them in a notebook, carefully labeled with the valid times. Likewise, without further tools, you could print out the low res HRRR data from the images online. Not much detail there, but almost enough to check for consistency with the UW data.

But with due respect to all my good friends who use them, that would be a sort of flip-phone approach. We have more powerful tools now, and indeed Expedition is one of them, primarily because it has such a very nice image import function, which I have discussed in several videos: Part 1 describes the procedure, and discovers a good example of why we do this, and Part 2 optimizes the image display and introduces the NDFD grib view of the graphic forecast, and finally Part 3 uses the Expedition ensemble display to show how the GFS and NDFD begin to diverge as a front approaches.

With that technique in hand, we can now directly overlay the HRRR grib data on top of georeferenced images of the UW data. A static example is shown in Figure 9.

Figure 9. Static capture from the video of Figure 10, showing HRRR grib data from Ocens overlaid on the graphic forecast from UW. Black isobars and winds are the HRRR grib data. The blue line on the UW graph is a highway.  The agreement is pretty good in most areas. Although the isobars do not overlap very often, the actual pressure difference at any point is typically less than 1 mb and the general shapes are about right. See also notes to Figure 10 video.


The video of Figure 10 illustrates the process of getting the UW images into Expedition. This seems to me the best way we can study and compare these two models, with the not so hidden agenda of wanting to show that the HRRR data will be just as good in practice, because we can get that much easier and it is digital.

 Figure 10. How to import and georeference UW graphic forecasts and then overlay the HRRR data in grib format.  Use the color scale on the graphics for the UW wind speeds. Use the Tool Tip in expedition to read wind and pressure from HRRR.

To reproduce the georeferencing of these Northern Water maps shown in the Figure 10 video, you can cut and paste these values mentioned in the video:

     SW coast line = 47 44.660N 124 27.829W

     NW coast line = 48 42.278N 124 57.555W

     Foul Weather Bluff, NE corner = 47 56.423N 122 36.265W

In striving to find the best forecast, we have to keep in mind that a simple overlay of the latest data corresponding to the same times is not the end-all determination.  In fact, we cannot even make the comparison any time we choose.  Although we get new HRRR data every hour, the only comparison we can make with UW-WRF is at 00z or 12z, the initiation times at UW, at which times we can compare what each predicts for the next 15h.

Just a few hours later, however, we can get a brand new HRRR forecast for 15 hours, fully assimilated, but we will not get any new info from UW for 12h. In other words, even if the UW model happened to be better at one particular time, a few hours later the HRRR will be better by default, because it has all new inputs and there is no UW option until the next 12h cycle. Thus in rapidly changing conditions, the HRRR predictions have to win out—not to mention that the HRRR data are available for all US waters across the country, very few of which are lucky enough to have the excellent forecasts we get from UW.

The forecasts can be easily checked with the convenient index we have online at starpath.com/local.  Just look at the Observations pages for direct links to key points around the waterway.

When preparing for an actual race we would also want to fold in what the local NWS office is telling us in text forecasts. Then looking at the model forecasts we will better understand why they say what they say, and we will be able to better interpret those brief text statements, and greatly expand their  implications.  We have the text forecasts online as well in a very convenient format at our Zone Forecasts page.

When folding in the text forecasts for the Northern Waters you may notice that our Canadian friends are very conservative with wind warnings. This is a well known policy, so having these model forecasts at hand can often help in spotting the differences between Canadian and US wind forecasts for the same waters.

By "conservative" I mean that if their in-house predictions call for 15 to 20 with some probability of 25, they are more likely to call this 20 to 25, than the US counterpart might do.  That is not at all to say this is a bad policy, but it is just a known tendency that is valuable to keep in mind when making tactical interpretations of the txt forecasts.

When used for sailboat racing, I have to stress again, that the HRRR grib data is only available at present as a commercial product. It must be purchased and as such cannot be used during the race, but only up to the starting time.

Needless to say, however, this commercial data can be a great boon to all commercial maritime applications that need best possible forecasts, such as ferry companies and tow boat companies, as it could be for all recreational mariners who do not have rules on outside assistance.

To me, the availability of the HRRR data in grib format is the most important marine weather development for US inland waters that we have seen in many years, and we remain grateful to Ocens for making this pioneering step into this field.  For more widespread use in racing, we have to wait till more sailors learn about it, and eventually there may be a public source.

(Later on we will figure out the georeferencing of the UW maps for the Southern Waters of Admiralty Inlet and Puget Sound and post them as well. To proceed on your own with that, just make your best guess of 3 points, then load it into Expedition, and then use that display to see how to fine tune the coordinates to line up with the base map geographic outlines. This is a bit harder than it should be because the UW maps do not use a Lat-Lon grid, but some other coordinate system.)










No comments: