Ironing out the electronic wrinkles

In a recent issue, I looked at the accuracy of paper charts and the surveys that underlie them (“How accurate are our charts?” Issue 120, March/April 2002). These surveys form the basis of all electronic charts, so all underlying inaccuracies get carried over into the electronic charts.

This time we’ll examine some accuracy “wrinkles” that apply specifically to electronic charts. To understand these wrinkles, you need to have some understanding of the different types of electronic charts. There are, broadly speaking, two types: raster and vector. With ever-improving software, however, the lines between them are starting to blur.

A raster scan is an electronic “photograph” of the original. To make it, the original chart is broken down into rows and columns of tiny squares (known as pixels), each of which is referenced by its row and column number (called a bitmap). Each pixel is assigned a color, which the computer stores along with its row and column location. To recreate the image electronically, the computer simply assembles all those colored blocks in the same relationship to one another as in the original.

A vector plot is quite different. It generally starts with a raster scan, but then uses a geo-referencing system, which, in the case of charts, is likely to be a latitude and longitude grid based on the World Geodetic System 1984 datum. This grid is superimposed on the raster scan. All the features on the raster image are then traced with line-following software or by hand and given a set of coordinates. For example, a depth contour will be traced, and every time there is a change of direction the computer will store the coordinates for the point at which the change takes place. It can then recreate the contour by drawing in all the points and connecting them with lines — it connects the dots.

Specific features, such as buoys, are recorded with the coordinates for the depicted location of the feature, and then tagged with the relevant data. In use, the navigation software (which is not the province of the hydrographic office) reads this data, and places the appropriate symbol together with associated information (e.g., buoy name and/or number) at the designated location. Colored areas, such as the intertidal zone or shallow water, are recorded by enclosing the area in a line (when the loop is closed, it is known as a polygon) and then tagging this enclosed area with its color. The navigation software recreates the polygon and fills it in with the relevant color.

Note that the software can be programmed to put in any color: if it comes up with hot pink for the intertidal zone, don’t blame the hydrographic office — the software is at fault. Similarly, the software could be programmed to put in a smiley face instead of a buoy symbol — another problem with the software. In practice, the same electronic vector-based chart from a given hydrographic office will look a little different on different electronic charting systems because of software differences. But if the software manufacturer does a good job it will not look that different from a paper chart.

Errors and judgement calls

This vectorizing process can introduce a layer of errors that can be as much as 0.5 mm at the scale of the chart (i.e., if the chart being vectorized is at a scale of 1:20,000, the vectorizing error may represent as much as 20,000 x 0.5 = 10,000 mm = 10 meters). These errors are generally minimized by zooming in on the raster image (i.e., enlarging it) so that the exact placement of original lines and features can be more nearly replicated. If the image is zoomed enough, when tracing lines, for example, it is possible to stay within the width of the displayed lines. But the end result still depends to a significant extent on the skill and attention of the operator.

The operator also has to make numerous judgment calls. Soundings on a paper chart, for example, do not have a precise location. In fact, the spot that the sounding represents is assumed to be at the center of gravity of the set of figures printed on the chart. To add these soundings to a vector chart, the operator judges the center of gravity of the sounding on the raster image, and clicks on this to establish the precise sounding location for the vector chart. There is plenty of room for error. At the end of the day, the quality control process is of paramount importance with respect to vectorized charts.

The advantages and disadvantages of raster and vector are not relevant to this article. What is of interest here is the impact on chart and navigational accuracy. Note that what I have given is an oversimplified picture of raster and vector charts, but it will serve my purposes. The British Admiralty, for example, in its current raster chart production starts from a combination of electronic databases, hard copy (smooth sheets, etc.) digitized on a vector basis, and raster scans. These are used to create an image from which the final raster chart image is “burned.” The raster images are already geo-referenced before they are used to produce vector charts.

Picture elements

Let’s assume an electronic chart is being developed from an existing paper chart. This is, in practice, how almost all have been made so far, although increasingly data will be stored electronically and electronic charts will be produced directly from this database. Nevertheless, it still helps to visualize the process by thinking in terms of digitizing a paper chart. With both raster and vector charts, the first step is generally to make a raster scan.

The paper chart will be at a particular scale. To scan it, every square inch or centimeter is divided into pixels, and the color of each is copied electronically so it can be reproduced on a computer screen. In other words, all features, lines, symbols, etc. are broken down into a mass of tiny colored squares. If we look at one of these squares under a microscope we find it really consists of two different colored areas, divided down the middle. The digitizing process will average these color differences and produce a homogeneous copy. If, however, we cut the sides of the squares being copied in half (which means this square is now broken up into four pixels), we can capture the different colors of the two halves of the square, but now we have four times as many pixels (each of which is one quarter the size of the original pixel). In other words, the smaller the size of the squares being copied, the greater the detail and definition of the electronic copy, but the larger the resulting electronic file.

The number of squares (pixels) into which the chart being copied is divided is expressed in pixels per inch (ppi) or dots per inch (dpi). In the hydrographic world, the British Admiralty scans at 1,016 dpi and NOAA, the U.S. hydrographic agency, at 762 dpi. But then both organizations output the files to end users at a much lower resolution: 127 dpi in the case of the BA, and 254 dpi for NOAA. Why these funny numbers? Partly it is a historical accident: the first drum scanner used by NOAA scanned at 762 dpi! And partly it is because we are mixing metric and imperial measurements. The BA’s 127 dpi is derived from the fact that its standards for paper chart drafting require a level of precision in the placement of lines, features, etc., of +/- 0.2 mm. When you convert this to inches, you find it is 1/127 of an inch. In other words, for electronic reproduction purposes the BA is dividing its paper charts into pixels whose sides are 0.2 mm long (the pixels have an area of 0.2 x 0.2 = 0.04 mm2), whereas NOAA is using pixels with a side length of 0.1 mm (i.e., with an area of 0.1 x 0.1 = 0.01 mm2).

Why is NOAA outputting electronic files at double the resolution? The greater the dpi, the more the detail that can be captured. Many of the features on a paper chart are displayed using small numbers and letters and fine lines (the finest lines used by NOAA on its paper charts are 0.10 mm wide). It is the opinion of NOAA that 254 dpi “is the lowest unenhanced resolution at which charted features are still fully legible.” Clearly, the British Admiralty does not agree! The key word here is “unenhanced.” The British Admiralty uses something known as anti-aliasing technology to create images that “actually look much sharper than charts using 254 dpi,” according to a letter from a British Admiralty staff member.

Above and beyond the clarity issue, underlying the debate as to whether to use 127 dpi or 254 dpi there are important considerations concerning the relationship between scanning pixel size and display pixel size, and the potential this creates for a chart to be “over zoomed.”

Displaying the image

Irrespective of whether the chart has been reduced to 127 dpi or 254 dpi, what happens when we display this electronic file on our chartplotter, laptop or desktop computer? The display screen is also broken down into individual pixels, but in terms of current display technology these have a typical side length of between 0.20 mm and 0.30 mm, with some on low-end equipment being as large as 0.40 mm (three screens I recently checked varied from 0.19 mm for a relatively expensive CRT monitor, through 0.23 mm for a new, flat-screen LCD monitor, to 0.30 mm for a laptop). Let’s assume 0.25 mm — i.e., an area of 0.25 x 0.25 = 0.0625 mm2 per pixel. If one pixel on the screen is used to display one pixel in the electronic file, and the electronic file is based on 127 dpi (0.2 mm pixels) the resulting image will be somewhat larger than the original paper chart (each 0.04 mm2 pixel has been blown up to 0.0625 mm2. The image will be a little over half as big again). But if the electronic file is based on 254 dpi (0.1 mm pixels), the resulting image will now be over six times as large as the original paper chart (the pixels have been blown up from an area of 0.01 mm2 to 0.0625 mm2). In other words, given current display screen technology, any electronic chart file output above 127 dpi will result in an image that is considerably enlarged compared to the equivalent paper chart.

Now we get into a minefield. In practice, some (but by no means all) electronic charting systems recognize the scale of the original paper chart and display the electronic chart at a similar scale (generally with a small degree of enlargement). This is, in fact, a legal requirement for high-end equipment. Let’s consider the implications of this for a paper chart outputted at 254 dpi and displayed on our monitor with 0.25 mm pixels: Each individual pixel is being enlarged more than six times. To keep the displayed image more or less the same size and scale as the paper chart, five out of six of the pixels in the electronic file must be omitted from the display. If the paper chart is outputted at 127 dpi, the file pixels are enlarged a little more than 1.5 times, so only one in three file pixels needs to be omitted to keep the scale the same.

The chart display software makes the decision as to what to drop. Typically, it looks at all the pixels in the electronic file that have to be fitted into a single pixel on the screen, and either averages the colors, or, if one color is predominant, uses the predominant color. Something similar happens any time the zoom out function on a raster electronic chart system is used to display a larger chart area at a smaller scale. Because a larger area of the paper chart is being looked at, there are now many more pixels in the electronic file than can be displayed on the screen. The display software must drop pixels from the electronic file (sometimes as much as 95 percent of the pixels). One result is that letters (labels) and numbers (e.g., soundings) become increasingly hard to read.

When zooming in with a raster chart, a smaller area of the original paper chart is displayed electronically. There are still the same number of pixels on the display screen as there always were, at the same physical size as before, but these are now being applied to a smaller area of the paper chart and thus to a smaller electronic file — more of the available pixels in the electronic file get used. If the file is based on 254 dpi, and the display pixel size is 0.0625 mm2, it will be possible to zoom in until the displayed image is 6.25 times larger than the original before we run out of pixels in the electronic file. However, if the file is based on 127 dpi, we will run out of pixels once we have zoomed to a little more than 1.5 times the paper chart size. Beyond this point, if we zoom in further, there will be, in effect, blank pixels on the screen, with nothing left in the electronic file to fill them. Instead, the display software fills them, and once again, it has to make a decision as to how to do this. If you keep on zooming in, the software has more and more blanks to fill, which it does by mimicking the color of adjacent pixels, causing the display image to break down into blocks of color (each block that we can now see is the data from one pixel extended to cover many adjacent pixels), ultimately becoming illegible once again.

Over zooming

With raster charts, the best display image (clearest, most detailed, with the crispest lettering and numbering) is created when there is a 1:1 relationship between the pixels in the electronic file and the pixels on the screen. If we once again assume a screen pixel of 0.0625 mm2, and the electronic file is based on 127 dpi, the displayed image will be something over 1.5 times the size of the original; if the electronic file is based on 254 dpi it will be 6.25 times larger. Let’s assume the original chart was at a scale of 1:24,000. It is now being displayed at a scale of 1:15,360 (127 dpi file), or 1:3,840 (254 dpi file). For the end user, symbols and text that normally are small and cramped are now larger and farther apart, making them easier to read.

The end user may like the clarity of this display, but in reality it can be dangerous. The survey standards and plotting accuracy applied to the original chart (1:24,000) will not be rigorous enough for the larger-scale version. For example, at 1:24,000 the chart compiler’s plotting accuracy is +/- 4.8 meters, on top of the survey errors which, even at contemporary standards, may be another five meters or more. In other words, even with up-to-date surveys, the charted position of features may be as much as 10 meters off, while with older surveys these positions may be considerably further off. For an original chart drawn at a scale of 1:3,840 the chart compiler’s plotting error is +/- 0.76 meters, while the underlying survey work will likely have been done to Special Order standards (even if the survey is still technically designated as a First Order survey), so the total displayed positioning error is likely to be less than three meters.

Using a 1:24,000 chart, no navigator in his or her right mind would be tempted to try a passage through a channel, or reef entry, or between rocks, that was shown as 1 mm wide on the chart (24 meters in the real world), but on the enlarged version the same channel is now 6.25 mm wide, and might look passable. However, because the large-scale chart has been created by zooming in on a smaller scale chart, all the inherent errors in the smaller scale chart are still present, as opposed to the lesser errors that would normally be expected at this larger scale. The only thing that has changed is the navigator’s perception. There is a distinct chance that if this passage is attempted the boat will be run aground. In fact, the passage may not even exist in the first place. If it is an apparent passage between a couple of rocks, there may in fact be dozens of rocks, but at 1:24,000 the cartographer only had room to show a couple, so that is what is on the chart.

Because of the potential over-zooming problems when raster chart files are displayed, the British Admiralty developed the Hydrographic Chart Raster Format for its ARCS (raster) charts, and retains tight control over how its files are used by electronic chart software producers and other end users. These controls are designed to ensure that a chart will be displayed to the navigator at the scale the chart compiler intended, and that it will retain the integrity of the image. HCRF has now been adopted into the performance standards governing the legally acceptable format for raster charts used by the world’s shipping industry. Several national hydrographic offices now produce HCRF versions of their national charts. NOAA raster charts, on the other hand, do not use HCRF. Plus, the Americans are much less controlling in how their chart files get used; they frequently end up over zoomed. Private producers of raster charts commonly also exercise limited control regarding over zooming and other display issues.

With a raster chart, you do get a sense of when you are over zooming in as much as the letters, symbols and numbers are all displayed larger than normal, and eventually the image starts to break down as the electronic file runs out of pixels and the software begins to fill in the blanks. With vector charts, the image itself provides very little evidence of over zooming, because whatever the scale, the software recreates the various points and lines, fills in the colors, and adds the labels, numbers and so on at the prescribed font size (i.e., the same font size, regardless of scale). With increased zooming, curved lines start to break down into a series of points connected by straight lines, and soundings become increasingly widely spaced, but that’s about it for clues. Unless the user is specifically warned via a message box, or the software includes provisions to prevent over zooming, it is very easy to use a chart at a scale for which it was not designed.

There is another danger inherent in over zooming a vector chart, which is that at each new zoom level the software recreates symbols at the prescribed (constant) display size. Let’s return to our example of a chart at 1:24,000, with a couple of rocks that indicate a rock-strewn area. With raster charts, the displayed image of these rocks increases in size with zooming, so that although the distance between them on the display increases with the zoom level, it does so proportionately. With vector charts, irrespective of the zoom level the rocks get displayed at the same size, which disproportionately increases the distance between them, reinforcing the impression that there’s a clear channel where there is none.

Let’s consider another example — a wreck symbol. At 1:24,000, the symbol almost certainly covers a greater area than the wreck itself. But at some point when zooming in with a vector chart, the symbol (which gets recreated at the same size at each zoom level) will cover less area than that occupied by the wreck: anyone using this over-zoomed chart to navigate a track that runs close to the symbol will now hit the wreck.

Zooming in on a chart, whether raster or vector, violates one of the immutable rules of navigation: a chart should never be used at a scale larger than that at which it was compiled. This rule is routinely violated by users of electronic charts.

All high-end electronic charting systems warn the user if a chart is being used overscale (over zoomed), and, in fact, generally limit zooming to a maximum of two times the scale of the chart from which the electronic chart was made. Many low-end systems have neither the warning nor the limit. Because of this the zoom function must be used with discretion and a clear understanding of its limits and dangers, remembering that just because a system seems to suggest a particular passage can be made does not mean it is safe.

User beware

The user of any chart, paper or electronic, should not be lulled into a false sense of security about its accuracy. Electronic charts, in particular, have to be used with caution. Before you go shaving any corners, you need to have a thorough grasp of, and a healthy respect for, the limits of accuracy of the charts and tools you are using!

This is an excerpt from contributing editor Nigel Calder’s, How to Read a Nautical Chart, published by International Marine/Ragged Mountain Press.

By Ocean Navigator