Tuesday, November 13, 2018

Fit-Slope Method to Analyze Sextant Sights

We have been teaching what we named the Fit-slope Method for analysing sextant sights since 1978, and it has appeared in all of our textbooks since then. Consequently we were pleased and honored that this method was referenced in the latest edition of Bowditch's American Practical Navigator, which in turn led me to realize that we do not have an explanation of this method in public that is easy to reference, although I did post a note related to it.  That note, and the Bowditch description, are short, and do not emphasize the key aspect of the method. This note is intended to remedy that.

*  *  *

For any measurement, we always get a more realistic value by making several independent measurements and then averaging the results—or maybe even applying more sophisticated statistical analysis such as least squares.  This is no different in celestial navigation when we measure the sextant height of a star from which we can compute a line of position (LOP) on the chart. We do not want to rely on just one measurement, we want many, so we can average them in some manner.

The problem we have in the sextant measurement is everything is moving. The star is moving across the sky; we are moving across the water; and the clock hands are moving across the watch.  It is not like measuring the length of a board, where we could measure it 5 times and then average the results. In the case of the sextant height of a star, we expect it to be different every time we measure it, so we have to clarify what we mean by an average.

If we were not moving at all, such as sights on land or when dead in the water, then we only have to account for the rising or setting of the star. We will get a set of sextant heights (Hs) and specific watch times (WT). Below is an example of what these might look like when plotted as a graph.
Here we have 3 sights of a star that is descending in the sky. The vertical scale might be 5' per division with say 44º 20' at the bottom and 45º 05' at the top; the horizontal scale might be 30 sec per division starting at say 16h 20m 00s on the left and 16h 24m 30s on the right. To choose the scales, we look at the range of time and range of heights and compare that to the graph paper we have, and choose some convenient scale based on that. This type of data can also be plotted digitally with a spreadsheet program such as OpenOffice Calc (Windows or Mac), and some cel nav programs like our own StarPilot do these plots automatically.

Over the duration of typical sight taking sessions (20 minutes or so), most celestial sights will follow a near linear path (meaning change in height per unit change in time is constant), providing the celestial body is not crossing our meridian at the time. So our first step analysis without further information is to just draw a line through these points using a manual form of least squares fitting. Namely the line should go through as many points as we can, leaving roughly the same number above the line as below—and maybe ignoring ones that are anomalously off the line as they are likely blunders. With little practice, navigators can actually draw this  "best fit line" as well as a computer program could using a least squares analysis.

When we choose where to draw that straight line, we have two degrees of freedom. We can move the line up and down on the page, and we can tilt the line, which is equivalent to changing its slope. The dotted line above is one choice of best fit. Once a best line is chosen, we can then choose which value will represent the "average" of the sights taken. If the line goes right through one of the points, then we can use it, and that single one will represent an average of all of them.

On the other hand, if our best line does not go through one of the actual sights, we can choose any convenient point on the line and use that for the representative average, as shown in the figure below. In these cases we might choose a point at a whole minute to simplify the later analysis. In the figure above this would be using Hs1 and WT1 to represent the average of all 3 sights, even though we did not take an actual sight at that time. With 5' per vertical interval, we then have to accept that the three sights are about ± 2' off of this average.
All of this procedure described above is what we might call standard practice, and something equivalent to this is used in most cel nav programs. This standard method, however, does not take advantage of all we know about the sights, and as such it often means we are giving up accuracy in our fix that we do not need to give up.  With that in mind,  let's look at the data again.

As we shall see, having just 3 sights is not optimum; we are better off with 4 or more, but with just three it is easier to see the point at hand.  For now we assume the slope is linear, which it is in most sight sessions—and if not, we will be able to know that as we show later. With a linear slope, we can look at these sights as the first two being consistent, and the third one being off.
Or we can think of the last two being consistent and the first one being off.

Or the first and last are consistent, but the middle one is off.  The difference between these conclusions is the slope of the data, meaning how fast does the height of the star change with time as viewed from our location. This slope also depends on the motion of our boat because that changes our position, but we will come back to that detail.  For now, we just assume we are on land or dead in the water.

This brings us to the whole point of the Fit-slope Method.  This slope is not a degree of freedom we have to vary when choosing the best line. We can easily compute this slope and then force our line to have the right slope. In other words, in the three interpretations above, only one is right. We just have to take a few minutes to figure out which one it is.

To do this, we perform standard sight reductions of the body sighted at times (T1 and T2) just before and just after the sight session, using the corresponding DR positions. This tells us what the heights (calculated height, Hc) should be at these specific times. We also get the azimuth, Zn, but we do not use that. These two Hc values can be plotted on the same plot as our data using an offset scale for the heights.  The Hc heights won't line up with Hs heights because Hc and Hs differ by all the standard sextant corrections (index, dip, and altitude). Once the two Hc values are plotted, we can see what the right slope should be.

This special determination of Hc values for the slope is best done by computation rather than using sight reduction tables. This way you can get Hc from the exact DR positions you were at times T1 and T2, which is what we need for finding the most accurate slope. You can compute Hc with a trig calculator using the basic definitions of the navigational triangle, or use a program such as the free Celestial Tools by Stan Klein. If you are online you can get Hc from the USNO—use our shortcut link starpath.com/usno. I would mention using our own commercial product StarPilot, but this step is not needed in that program as it solves the Fit-slope Method automatically.

You can if needed compute the Hc with tables alone (no computers,  no calculators), but it is extra work. Most sight reduction tables in common use require employing an assumed position, which can be as much as 30 miles away from your DR positions. The resulting Hc values are perfectly good for getting an accurate LOP, but the time slope between these two Hc values at different assumed positions is not accurate enough for what we want.



Once we have the slope plotted as above we can use parallel rulers or a roller plotter to move it up into the data to find the best fit.  We are still doing a "manual least squares" fitting by eye, but now the slope is not a variable. In many cases, this helps us rule out certain sights. In this schematic example, sight C is most likely not right, and we are likely best to just average A and B and use that to represent the set of all three. When there are 4 or more sights, the confidence in choice of outliers is more satisfying—or we learn there are no obvious outliers, and we use the spread of differences from the slope line as a measure of our uncertainty in that sight.

This procedure is not a magic bullet in any sense. It is simply imposing onto the data a property that we know (the slope) and using it to help us choose which sights might be less reliable. It is not guaranteed that the conclusions will be right, especially when there are few sights and the spread in values is not large.  But this method does give us a bit more confidence in our choices.

Since we are looking for small differences, it is important when we compute the Hc values to use the right DR positions, which will change during the sight taking session. In class we teach that each sight session starts below decks with a recording in the logbook of the time, log, course, and speed.  Then after the sights are completed, we return to the logbook and record the ending time, new log, and again record course and speed. We can use that information to mark the starting and ending DR positions on the DR track that is already plotted on our plotting sheet.  We had to have that track plotted to predict the time of the sights, and we use it again to choose a DR position for the sight reductions. (See Notes on DR and Sextant Sights.)

To show the influence of this, consider we are sailing 6 kts toward the south, taking sights of a body toward the south over a period of 30 min. Bodies near due south are crossing our meridian, so their height is not changing much with time if we were not moving, but sailing toward it we would raise its sextant angle by 3' during the 30 minutes we are sailing toward it at 6 kts.

Below is an example fix using the fit-slope method. It is from the book Hawaii by Sextant, analyzed with StarPilot. Sights are from a 1982 race, Victoria, BC to Maui, HI, near the end of the race approaching land. No other navigation was available at that time but cel nav. There was a radio beacon there, but it was not helpful for RDF at the time.

Figure 1. Raw sight data of Problem 27. C=232T; S=7.6

Notes from the book: "We are getting close to sighting land. This could be our last chance for a good fix. We do a set of morning star sights. Usually in star sight sessions we add Polaris whenever we have time to do it because it takes no pre-computation. Just set your sextant to your latitude and look north. But in this case it was woven right into the set of all sights, which were alternated in time to get the best fix."

Here is what these sights look like after they have been reduced using a calculator.

Figure 2. Analysis of the sights

This analysis does not take into account the motion of the boat. They are all reduced from the DR given in the Polaris sights. Each is valid at the WTs shown.  Below we plot all of these, adjusting each sight to the time of the last sight using Speed = 7.6 kts and Course = 232 T.

Figure 3. Plot of all sights, adjusted for boat motion. In the StarPilot, the individual LOPs are identified with a mouse cursor rollover.

The light gray vertical and horizontal lines are meridians and parallels, which end up at unusual spacing in the StarPilot plots. Here the two latitude parallels are separated  by 40.9'.


Figure 4. Showing bad Capella sight is way off the slope line (22.7' off) and can be discarded—although we did not need the fit-slope method to know this one was bad.

We see immediately that one of the Capella sights is way off and certainly a blunder, leaving us to look at the other three. Likewise, Vega has one sight notably off the others.  There is not much we can do with the Polaris data as there is no significant slope to its motion. Below are a couple of the fit slope plots.

Figure 5. Capella sights. After discarding the bad sight, the last 3 have more information. Relative to the second two, which are close to the line, the first is clearly higher. In this case, however, the difference off the line is under 1.0' so we probably won't get much difference in the fix from using the last two only or all three. 




Figure 6. The 3 Vega sights. The second is off the slope of the other two by 4.1', which we had indication of from the plot of all LOPs. So for this sight, we choose the average of the first and last as likely the best average for these 3 sights. Without the known slope line, you would have the conditions discussed in the introduction, namely not know if 2 and 3 were right, leaving 1 high, or if 1 and 2 were right, leaving 3 very high. 





Now we can redo the plot of all LOPs employing the choices from Fit-slope and again correcting for vessel motion. This is shown below.


Figure 7. Plot of all sights corrected for vessel motion. Each LOP shown represents the average of multiple sights guided by the fit-slope analysis. This is Problem 27 in Hawaii by Sextant.

My experience is that this method generally improves the fixes from typical sights at sea. In some cases, the choices of best sights to use are obvious. In other cases, we have to look at the full set of sights combined with other sights to help us make a choice.  The method is likely of most value in helping set a confidence level on the fix. If we have four sights and three line up well, but one is off, but just by, say, 1.5', then in many cases we might not feel confident to throw that one out, but with the other three nicely on the line and that one off by that much, we are more confident to remove it. This operation can then leave us with a tighter fix with less uncertainty.

Our book Hawaii by Sextant is once way to practice the method as there are many real sights and fixes included. It can certainly be tested on land, and indeed used by those new to sight taking to help evaluate their sights. You can apply it to a set of sights using a shoreline on the other side of a lake, or use it with an artificial horizon. The application of the method will be good training for actual sights at sea.
______________

To get the ultimate accuracy from a set of sights then requires us to analyze the triangle of 3 LOPs to find the best location, or most likely position, based on the shape and size of the triangle of intersections ("cocked hat"), as well as the standard deviation or variance we assign to each LOP in the set. Often these variances are the same, but when they are not, it affects the best choice. 

One of the key outcomes of the fit-slope method is a better feeling for the uncertainty in an LOP based on the analysis of the several measurements we made to get the LOP.  In the above discussion, if we did no analysis we had to assume something like ± 2' for the variance of this LOP, based on three measurements, not counting other information that might affect this value.  After the fit-slope analysis, this variance can be fairly reduced to about half of that, or better.

We have a related discussion called:  Most Likely Position from Three LOPs.








Monday, October 22, 2018

Marine Weather Workhorses... and Secret Sources

We are in the process of updating parts of our Weather Trainer software and ran across an article we wrote in 2009 that we refer to frequently. And sure enough there are parts that needed updating (which this note addresses), and surprisingly much is probably still new to many mariners. The "workhorses" have changed a bit, but the "secret source" is still more or less secret! It first appeared about sometime in 2009, but without any announcement, as far as we could tell at the time. It is hidden right under our noses—I mean mouse cursors—in the same place online many of us check every day to see what clothes to wear to work. But let’s come back to this jewel in a moment.

Once underway on inland and near coastal waters, the NOAA Weather Radio on VHF is likely to be the main workhorse for inland waters. It gives observations every 3 hours, forecasts every 6 hours, and synopses every 12 hours. The broadcasts are continuous, 24-hr a day. A typical broadcast is about 10 minutes long, which includes some inland and mountain weather.  The NWR site has many resources about each of the stations, from which you could piece together a picture like the one below that we made for our Modern Marine Weather, 3rd ed text, which has an extended discussion of this resource.


Figure 1. NOAA Weather Radio coverage on the NW coast, showing typical ranges as well as station overlaps and VHF channels. The Canadian counterpart is called Weatheradio.

But that is not the best place to start a study of what might take place on a planned voyage, even if you happen to have a VHF radio that will receive the signals at home.  All of the information in these radio reports are available as text. It is just a matter of sorting out the best way to access these text reports.

The NWS offers forecasts by region (zone) in various formats. Some of these are itemized in a publication called Marine Weather Information Guide, but we have frankly more information and perspective on the various categories of forecasts in our textbook—this Guide, for example, does not cover the coastal zone system we discuss below. There are multiple other groupings of zones in use, but the primary practical categories are coastal zones, offshore zones, and high seas areas. Coastal zones are typically in two bands, 0 to 10 nmi offshore, and 10 to 60 nmi offshore on the West Coast, and 0 to 20 and 20 to 60 off on the East and Gulf Coasts.  Bays, sounds, lakes, and estuaries are covered in the coastal zones using custom geographic definitions to match the waterways.

Web Access to Coastal Zone Forecasts 

Coastal zones are the most localized marine areas covered by official NWS forecasts—although we can look more locally as covered later (the secret!).  The link weather.gov/marine  has a link to the page that brings up the Coastal Zone graphic index shown below.

Figure 2. Index to US Coastal Zones. Click any colored region to zoom into that coast. You can also get to this index with Google to "NWS coastal zone forecasts."


 Figure 3. Coastal waters of the West Coast. 

Each color, named by a prominent city, identifies a group of coastal zones called the coastal waters Below we the coastal zones included in the group called "Seattle Coastal Waters."



Figure 4. Coastal Zones within the Seattle Coastal Waters area.  


When viewing online, we can click any one of these zones to get that local forecast. Below is the report we get from clicking anywhere in the PZZ132 zone (dark magenta, top right). This region is the Eastern end of the Strait of Juan de Fuca.


Figure 5. Web presentation of pzz132 forecast, with several related links given.

This internet approach is graphic and straightforward, with several interesting links to related information, but notice we have lost track of the name of the actual zone we asked for, which was pzz132. The notation "pzz100" at the top refers to the synopsis of the full region covered by the "Seattle coastal waters," which covers all the zones shown in Figure 4.

Coastal Zone Forecasts by Email

Individual West Coast coastal zone forecasts are stored online at NOAA at a link like this one: http://tgftp.nws.noaa.gov/data/forecasts/marine/coastal/pz/pzz132.txt  To get a different zone, just change the pzz132 to the zone you want.  All the files that are even 100s are synopsis files for various coastal waters regions. Other coasts are in other subdirectories:

/pz = West Coast
/an = Atlantic Lat > 31 N
/am = Atlantic Lat < 30 N
/ph = Hawaii
/gm = Gulf of Mexico
/pk = Alaska

You can get an individual zone forecast file with the NWS FTPmail service, or even more easily from Saildocs: send an email to query@saildocs.com that is all blank except for this one line:  send pzz132


That will get you the report in a few seconds by email. (If you want the regional synopsis as well as that specific zone forecast then add another line to the message with send pzz100.)

The email version you get back will look like:



Getting your forecasts this way requires a computer, tablet, or smartphone. Throughout the Pacific NW waters and the Salish Sea we have good cell phone connections, so this generally works well.

Coastal Zone Forecasts (and Observations) by Voice Phone

Without wireless devices and network connections, you can still get these forecasts with a flip phone or princess phone along with latest observations using a NOAA service called Dial-a-Buoy.   Call

888-701-8992 

and follow instructions. If you do not know the ID of a buoy or lighthouse near the region you care about, then choose option 2 and enter a Lat-Lon. Then select a buoy to get latest observations and following that you can request the forecast for that region, which will be the coastal zone forecast of that buoy location.

This is becoming a bit old school, but it could pay to have that number in your list of contacts as a back up if you are on a vessel when other options fail.

Coastal Zones versus Coastal Waters Forecasts

Coastal waters are groups of coastal zones. You can tell from the index shown in Figure 3 how these are grouped on the West Coast. There are similar groupings along all coasts. Below shows an example from Florida

Figure 6. Coastal waters of Florida.


Figure 7. Coastal zones included in the Miami Coastal Waters region. 

A typical file here would be found at http://tgftp.nws.noaa.gov/data/forecasts/marine/coastal/am/amz630.txt for Biscayne Bay, and the file /amz600.txt would be the synopsis for the full Miami coastal waters region shown above, which will include, by the way, the location of west wall of the Gulf Stream in this region.

Coastal Waters Forecasts by Email

For local daily sailing, we would likely care about just the coastal zone that covers our waters, but when transiting a region it could be more convenient to download the coastal waters forecasts (CWF) that includes several adjacent zones.  The file names for the CWF files are more complex, but saildocs has a short cut for requesting them.


Figure 8. Coastal waters file names from Modern Marine Weather, 3rd ed. These files (CWF) are groups of coastal zone forecasts covering about 200 nmi along the coast. They are available from Saildocs, i.e., send fzus52.ktbw for coastal waters around Tampa, FL. Use just the green part of the full file name.



Observations from NDBC

When we want near live observations rather than forecasts, then the primary source is the National Data Buoy Center (NDBC)—easy to find on Google with NDBC.  This site is very easy to use. Zoom into the buoy or station of choice and you see the results.

One subtlety we confront are the units options.  Choose either "Metric" or "English." With metric we get mb for pressure but have to live with m/s for wind speeds. Choose English and you get knots for wind speed but then stuck with inches of mercury for pressure, which is intended for aviation and TV weather.

The small graph icons on the pages are links to plot the data. Below is a unique option that shows both wind and pressure, showing the units for Metric and English.


Figure 9. Data plots from NDBC. The inside uses English units; choosing Metric units puts the outside scales on the data.


Once you know the ID of a favorite station you can find it easily on Google with that ID alone. You can also use it in the Dial-a-buoy program mentioned above. Likewise you can get the latest data including sea state by email from saildocs with this query (change the file name to the one you want):

send http://tgftp.nws.noaa.gov/data/observations/marine/latest_obs/41002.txt


NWS Mobile Weather App—Sort of

The NWS does not have an actual app, but they have a mobile website set up that you can save in your phone to access much of their data. The link to the page is


Using an iPhone, you can open this in Safari, then press the share button,  and choose save to home screen. This then is effectively an app.  A few clicks gets you to a display of the coastal waters forecasts, and another few clicks gets you to a list of buoys for latest observations. You can even view the latest weather maps, called "radiofax charts" on the menu.

AND FINALLY... The Secret Source

The secret source of NOAA's marine weather is actually just a normal part of NOAA's land weather. It remains a bit of a secret because landsman are very unlikely to use it, and mariners are very unlikely to know to look for it there.  When this first came available (maybe 10 years ago), it did not have much written about it from the NWS... but we did have our original "secret sources" article. Now we can find more official discussion, and we know more about what it really is.

In short, it is a graphic interface to a point forecast, based on the National Digital Forecast Database (NDFD). The way to access this is to ask online for NWS weather for any town next to the water you care about. Once we get close to the water, we can go from there.  For example, let's ask Google for "NWS Seattle weather."  You must include the NWS or you will be bombarded by commercial weather pages, which vary from bad to worst.  The right NWS official page will have a URL  like https://www.weather.gov/sew/.  If you asked for "NWS Chicago weather," you would get https://www.weather.gov/lot/, and so on.  That type of format is what we want.

Alternatively, you can get to the right place by searching on or going directly to


Then click the national map near where you want the marine weather. Once near the waters you want, click more specifically to get something like shown below, where we have clicked in Puget Sound, at the top edge of Elliott Bay, just inside West Point. Then in the figure we look at the forecast at another location in the Bay.


Figure 10. Point forecast for just south of WEst Point Lighthouse

The details don't matter much at this point, other than to note these are different forecasts, inside and outside of the Bay. Though not too different now, they could be notably different in other conditions.


Figure 11. Point forecast for inside Elliott Bay

Now compare both of these "point forecasts" to the official coastal zone forecast for this region (pzz135) shown below.

Figure 12. Coastal zone forecast for Puget Sound (pzz135)

These point forecasts are in principle more accurate than the coastal zone forecast that covers a larger region, which might easily be expected for inland waters like these where the terrain and shape of the waterway is so varied over the forecast region.

We can also get more granular information on other weather factors using the meteograms at the forecast points.   These are at the bottom right of any of the above weather forecast pages. They are called Hourly Weather Graphs. Below is an example showing temp and dew point just north of West Point (Figure 10).

Figure 13. Meteogram from NWS land weather page, clicked offshore, just north of West point. Note this is a forecast, not a report.

The present time (3pm PDT) is on the left side here, which shows a forecast dew point and air temp about 2º apart, which would normally be no fog.... or patchy fog. This is one of those cases where we would want the forecaster to look out the window—or at least read the reports from the lighthouse on the point shown below.  These temperatures have been the same all morning, which we can learn from the reports at West Point Lighthouse, shown below.


Figure 14. Data from WPOW1 showing dew point equal air temp all day.

It is pea soup on Puget Sound all day today. From the Figure 13 forecast, we can expect it to clear up about  9 or 10 am tomorrow, which (looking back now) it did in fact do on time.  So despite being off on Monday with regard to fog, we can see the value of this type of presentation.  Frankly, these one or two day NDFD forecasts are right way more often than wrong.  We also see an informative presentation of the wind forecasts in these meteograms, which is especially valuable for planning if you have a front coming through during a race.


Figure 15. View of Puget Sound at 3 pm PDT as I write, viewed  from about 270 ft elevation, 1.8 nmi NE of West Point. Although fog is not uncommon, this very thick fog is rare here.

With that said about the workhorse sources of local forecasting, let me add that there is more to this topic. Please refer to our textbook for details and recommendations. None of the sources covered, for example, would be considered the best possible forecasting for sailors. The best would depend on the time frame in mind—plus there is some extra work involved in accessing them.  For the next 12 hr or so, we would likely do best with the HRRR forecasts that extend out 18 hr and are updated every hour. For longer term, say three to five days, then it depends a bit on the waters we care about, but in the not too distant future, the answer will probably be the National Blend of Models (NBM) forecasts regardless of waters.


Figure 16. Cover of Modern Marine Weather, 3rd ed.  Also available in all ebook formats.