Charts II: NGA - 2700 Charts

[wdkester]

Originally Posted by HappySeagull

David.Where's 28157?There isn't one.Anyways,it's Early Days with this stuff...you can always skip things you don't want to do.Leave the field blank,is all,so the kap won't get made yet.Or carry on,let it bake,download it and test it in OpenCPN.

It sounds like they are telling you how to match it with OLDER charts and OLDER Datum....say to transfer a fix from one to another chart.So you don't need to correct it at all.

OOPS The chart was 28167.

I've decided to remove the offset. Will check later.

[HappySeagull]

Test some of the simpler charts,Meark56....and don't be shy about doing someThere'll be some in here that you don't already have!


Anyways,seeing as there's still feathersin my head about this...




See?if OpenCPN's grid is seen as a gps,this is correct.
so,in the dialogue,just input what you see written.(I remember cagney mentioning that the script will "do" the opposite to the chart? so "thinking"is not required,I hope!)

[HappySeagull]

Originally Posted by wdkester

OOPS The chart was 28167.

I've decided to remove the offset. Will check later.

boy,that is really a new one you found! Plainly,we are delving into the History Of Digital Mapping.Good move, reading closely!

[Tia Bu]

The KAP file generated for chart 124 seems to be off by a degree of longitude or so. When we find a KAP file that needs to be corrected, can we just redo the calibration? Will a new KAP file be generated to replace the old one?

(This is probably answered above, but I couldn't find the answer in a quick search.)

[nohal]

Yes, the KAP is generated every time you hit the save button

Pavel

[recri]

I think this would allow a sanity check on corner coordinates in the browser.

Given the lat/lon coordinates for the corners, and the pixel locations for the corners, you can compute a vertical and horizontal (nm/pixel) for the chart being entered. These should be roughly the same.

Also, given the stated scale of the chart, the scan resolution of the jpeg images in dpi, and the pixel size of the images, you can compute an upper bound for the chart dimensions in nm, which can also be used to check values in the browser.

[siff]

Originally Posted by recri

Would it be hard to break the regions into subsets? Perhaps using the first two digits of the chart number.

So instead of reloading 619 charts for region 2 every time I refresh, I'd be getting roughly 70 charts for which ever region 2 subset I was working on.

To help somewhat, there is a pdf catalog of the charts for each region which shows the subsets. I found the region 2 at this place, but they are certainly somewhere on the nga site. It could be fine to put a link or a copy of them on the general nga status page

[nohal]

Originally Posted by recri

Given the lat/lon coordinates for the corners, and the pixel locations for the corners, you can compute a vertical and horizontal (nm/pixel) for the chart being entered. These should be roughly the same.

If the things were this easy
To make it simpler than it actually is, look at Mercator projection. The latitude and longitude resolutions are equal just on the equator and as you progress in the direction to the poles, the difference increases (that's why they teach us to place the dividers to the right or left of the chart approximately at a latitude where we want to do the measurement in the first navigation lesson, never to the bottom of the chart if we want to know the distance on the chart...) Even on British Isles, which I wouldn't call any extreme latitude, the difference is already dozens of percent, so this check would be telling us very little if anything about the quality of the calibration. A bit more math we need...

Overall status report:

• I've fixed a bug in the backend causing skewed charts rotated to the right produce shifted KAPs and regenerated all the (<40) affected charts.
• The broken charts 61200, 82377 & 93340 were updated with the correct JPGs.
• The stats page was extended to show "all" the info we have about the charts. If you want to play with it and try some QA calculations, just go to the respective page, select all, and copy&paste it into Excel or OpenOffice Calc. I will be very grateful if you get back to me with the stuff you invent to check the quality of our charts.

Pavel

[recri]

Originally Posted by nohal

If the things were this easy
To make it simpler than it actually is, look at Mercator projection. The latitude and longitude resolutions are equal just on the equator and as you progress in the direction to the poles, the difference increases (that's why they teach us to place the dividers to the right or left of the chart approximately at a latitude where we want to do the measurement in the first navigation lesson, never to the bottom of the chart if we want to know the distance on the chart...) Even on British Isles, which I wouldn't call any extreme latitude, the difference is already dozens of percent, so this check would be telling us very little if anything about the quality of the calibration. A bit more math we need...

Pavel

Maybe not too much math, when I compute:

(lat_height/pixel_height) / (lng_width*cos(mid_lat)/pixel_width)

using:

lat_height = lat_NE-lat_SW
lng_width = lng_NE-lng_SW # but beware the international date line
mid_lat = lat_SW+lat_height/2
pixel_height = (dist(NW,SW)+dist(NE,SE))/2
pixel_width = (dist(NW,NE)+dist(SW,SE))/2
dist(a,b) = sqrt((xa-xb)^2 + (ya-yb)^2)

for a series of charts from Baja California to the Straits of Magellan, I get:

21008: true 1.00448 +lat 1.00967 -lat 0.99411 +lng 0.99949 -lng 1.00953
21017: true 1.00524 +lat 1.01035 -lat 0.99502 +lng 1.00024 -lng 1.01029
21021: true 1.00496 +lat 1.00999 -lat 0.99490 +lng 0.99996 -lng 1.01001
21033: true 1.00586 +lat 1.01092 -lat 0.99574 +lng 1.00085 -lng 1.01091
21036: true 1.00363 +lat 1.00867 -lat 0.99355 +lng 0.99864 -lng 1.00867
21120: true 1.00504 +lat 1.01011 -lat 0.99490 +lng 1.00004 -lng 1.01009
21121: true 1.00642 +lat 1.01146 -lat 0.99634 +lng 1.00141 -lng 1.01148
21122: true 1.00559 +lat 1.01063 -lat 0.99553 +lng 1.00059 -lng 1.01065
21140: true 1.00393 +lat 1.00900 -lat 0.99379 +lng 0.99894 -lng 1.00898
21141: true 1.00608 +lat 1.01113 -lat 0.99598 +lng 1.00107 -lng 1.01113
21160: true 1.00436 +lat 1.00943 -lat 0.99423 +lng 0.99936 -lng 1.00941
22004: true 1.00669 +lat 1.01171 -lat 0.99666 +lng 1.00168 -lng 1.01175
22008: true 1.00605 +lat 1.01101 -lat 0.99612 +lng 1.00104 -lng 1.01110
22032: true 1.00075 +lat 1.00546 -lat 0.99133 +lng 0.99577 -lng 1.00578
22036: true 1.00114 +lat 1.00585 -lat 0.99172 +lng 0.99616 -lng 1.00617
22205: true 1.00712 +lat 1.01207 -lat 0.99722 +lng 1.00211 -lng 1.01218

The value labeled true uses the chart values. The values labeled +/- lat/lng increase/decrease the charted latitude/longitude height/width by a factor of 0.005, a half percent error, by moving the northeast coordinate.

If you take (true >= 1.0 && true <= 1.00712) as the rule, then only 13 / 64 of the errors pass the rule. A 1/2 percent span error on a 4 degree quad is a 1.2 minute error in corner coordinate. If you increase the error in span to 1%, then none of the errors pass the rule.

Give me a url that fetches the chart summaries and I'll compute the statistics.

[recri]

Oh, I can compute statistics from the summary tables. Nice job!

[nohal]

recri,
if you don't want to parse the html tables, you can of course get it in other format... Just let me know.

Pavel

[recri]

Pavel --

The html is fine, thanks,

-- rec --

[Tia Bu]

On the subject of QA, chart number 145 is giving me heartburn. The printed latitude and longitude intersections all over the (KAP) chart are consistently off by several miles when checked in OpenCPN. (They are shifted to the Southeast.)

I have checked everything I can think of on the calibration. It all seems to be entered correctly. None of the other International charts from the Miscellaneous Region I've downloaded and checked with OpenCPN have this problem. One chart (124) had one longitude coordinate mistyped, but that cleared up with a simple correction. (Surely we'll all do that on occasion.)

Do you think this is just an old paper chart that is "off"?

[nohal]

Tia Bu...
145 is one of the ~40 charts hit by the bug in skew calculation I fixed yesterday. If you download the current KAP, it looks to be quite exact for a 1:6300000 chart.

Pavel

[recri]

I processed the 489 charts in the Rectangular, Doable category about an hour ago.

I discarded all the charts with less than 5 digits in their chart numbers, all the charts which weren't MERCATOR, all the charts with missing corner coordinates, and computed the nautical miles/pixel ratio as described above.

There were 433 charts which were "square" to better than 9 parts per thousand, the ratio was greater than 0.991 and less than 1.009. The mean for those charts was 1.00436 with standard deviation of 0.00208.

I inspected all the charts outside the cutoff.

These 7 charts, all high latitude large scale, appear to be correct:

29002 29025 35000 37000 42046 43015 44000

These charts all have errors in corner coordinates:

24400 25400 44069 51002 51400 51460 52054 52262 53244 53281 71005 74002 81151

There are several places where 0, 6, and 9 got confused, there's one with the same longitude for east and west, and a couple where east ended up west of west or south north of north.