eml-physical changes needed
Changes as decided upon at the Sevilleta EML meeting, April 24-25, 2002:
1) add version and citation of format definition
2) add ability to describe BIP and BIL formats for binary raster data -- see the
IPW header format for the info needed
3) rearrange for better control of required elements when using fixed vs.
variable formats. Do this by creating "fixed" and "delimited" elements with
proper content models.
4) add "objectName" element to contain the filename or other name of the
5) add field for pointer for which connection to use to get this physical object
(using "objectName"). Question as to how the semantics of that combo work --
how does one add an object name together with connection info for arbitrary
#1 Updated by Dan Higgins over 17 years ago
with regard to item 20 ability to describe BIP and BIL formats for raster data;
there is a white paper on the ESRI site that describes the header information
used for these types of files ("Extendable Image Formats for ArcView GIS 3.1 and
The header information is in the form of keywords/values. 14 keywords are
defined as follows:
nrows -The number of rows in the image. Rows are parallel to the x-axis of the
map coordinate system. There is no default.
ncols - The number of columns in the image. Columns are parallel to the y-axis
of the map coordinate system. There is no default.
nbands - The number of spectral bands in the image. The default is 1.
nbits - The number of bits per pixel per band. Acceptable values are 1, 4, 8,
16, and 32. The default value is eight bits per pixel per band. For a true color
image with three bands (R, G, B) stored using eight bits for each pixel in each
band, nbits equals eight and nbands equals three, for a total of twenty-four
bits per pixel. For an image with nbits equal to one, nbands must also equal one.
byteorder - The byte order in which image pixel values are stored. The byte
order is important for sixteen-bit images, with two bytes per pixel. Acceptable
I - Intel byte order (Silicon Graphics, DEC Alpha, PC) Also known as littleendian.
M - Motorola byte order (Sun, HP, etc.) Also known as big-endian.
The default byte order is the same as that of the host machine executing the
layout - The organization of the bands in the image file. Acceptable values are
bil - Band interleaved by line.
bip - Band interleaved by pixel.
bsq - Band sequential.
The default layout is bil.
skipbytes - The number of bytes of data in the image file to skip in order to
reach the start of the image data. This keyword allows you to bypass any
existing image header information in the file. The default value is zero bytes.
ulxmap - The x-axis map coordinate of the center of the upper-left pixel. If you
specify this parameter, set ulymap, too, otherwise a default value is used.
ulymap - The y-axis map coordinate of the center of the upper-left pixel. If
this parameter is specified, ulxmap must also be set, otherwise a default value
xdim - The x-dimension of a pixel in map units. If this parameter is specified,
ydim must also be set, otherwise a default value is used.
ydim - The y-dimension of a pixel in map units. If this parameter is specified,
xdim must also be set, otherwise a default value is used.
bandrowbytes - The number of bytes per band per row. This must be an integer.
This keyword is used only with BIL files when there are extra bits at the end of
each band within a row that must be skipped.
totalrowbytes - The total number of bytes of data per row. Use totalrowbytes
when there are extra trailing bits at the end of each row.
bandgapbytes - The number of bytes between bands in a BSQ format image. The
default is zero.
#2 Updated by Dan Higgins over 17 years ago
proposed changes to eml-physical-2.0.0beta8; partially completed (16May2002)
1) 'version' and 'citation' attributes have been added to 'format' element. It
was assumed that the 'citation' is a simple reference rather than the full
'citation' element that used elsewhere.
2) a proposed set of elements for describing binary raster data is included.
All are included as children of a new element called 'BinaryRasterInfo'
3) No changes have been made in handling 'fixed' vs 'delimited' field
Delimiters. I am not sure what to do here. The current system seems to work for
4) 'objectName' element has been added. In my mind this is usually simply a
file name that can be restored (if desired) when a object is returned
5) field for pointer to connection - Don't know how to handle
#4 Updated by Matt Jones over 17 years ago
1) I think 'citation' should be the full citation reference (type cit:LitCItation).
2) I'll review BIP/BIL stuff separately.
3) The intention of rearrangin the delimiters was to make it clear when each was
required. I think we still need to make these changes.
5) This is tightly bound to the resolution of the "distribution" discussion for
eml-resource. What this field looks like, and even whether one is needed, is
determined by whether the top level distribtion element represents a generalized
connection or a connection to a particular resource.
You should feel free to check this into CVS when you are ready, even if it is
not complete. The only reason Owen and Dan are using Bugzilla attachments is
because they don't have write access to the eml module, which is a side-effect
of moving to the ecoinfo cvs server.
#5 Updated by Matt Jones over 17 years ago
About raster metadata -- looks good. A few comments:
1) nrows & ncols should be required. The rest of the fields should be optional,
with the default values explicitly encoded in the schema.
2) you are missing all of the documentation tags. Please add them as I have
described in other bugs.
3) use camel caps for element names as described in other bugs. Elements should
be initially lowercase. Types should be initially uppercase. So
"BinaryRasterInfo" should be "binaryRasterInfo"
4) I think we need to reorganize the placement of the binaryRasterInfo element.
Right now it is possible to provide a field delimiter and raster info, which is
inappropriate. Maybe we should cluster these into a top-level choice. What
happens if we want to add other physical descriptors later? Right now we
support various text character encodings for tabular data, and binary raster
data. What about text-encoded raster data? I think we need to figure out how
physical can be extensible like entity is. Not sure how this should happen.
5) Could you also review the Image Processing Workbench (IPW) to make sure that
we accomodate everything it can handle as well. IPW allows raster images to be
viewed in standard programs like xv, and is well-used in the remote sensing
community. IPW information can be found at: http://www.icess.ucsb.edu/~ipw2/
Look in particular at the "mkbih" command. Thanks.
#6 Updated by Peter McCartney over 17 years ago
The raster parsing info looks good. I have some notes that we put together
based in the Erdas import tools that i will check this against to see if theres
anything else. offhand i dont see where we indicate the origin or whether to
read rowfirst or columnfirst.
I hope we are still going to see some struture in this module so that we can
start setting elements to required when they should be. I think there should be
a primary division at the top between the description of the physical object,
description of its format, and the reference to the connection it is found at.
the physical object description would include its name, size, owner?, etc.
the connection is merely a pointer by name, or idref or whatever we decide.
the format section needs to be further subdivided into at least three choices
thus far: a named format (with optional version and citation), an ASCII format
description (im willing to try working with your mixed model for
delimited/fixedlength), and the binary raster. Others are likely to be defined
as time goes on.
Ill leave you with an ever further leap to say that i would like to see the
object description repeat within a single physical module and attach an
optional extent or coverage description to each. this allows you to deal easily
with multiple files produced by cutting a single data entity into tiles or
series PURELY for the purposes of storage/transport considerations and it is
expected that it would be reassembled prior to making use of it. the coverage
or extent module would provide the guidelines as to how to reassemble the
pieces. In our ASU use of this, it is assumed in this case that the coordinates
provided are in the projected units as they have to be quite precise in order
to properly put the images together.
I will attach a copy of our earlier draft just so you can see what i mean.
#9 Updated by Chris Jones over 17 years ago
The distribution element underneath the individual entities such as dataTable is
redundant since Matt included distribution in the ResourceGroup after the notes
from the Sevilleta Meeting. I suggest we remove this and keep it in Resource to
minimize confusion as to where it goes.
#12 Updated by Peter McCartney over 17 years ago
I just checked out the files and here are my lingering comments.
a. I still suggest changing dataFormat. FormatName is only needed if you
anre NOT providing the parsing information inline. This structure is confusing
because someone could enter ascii fixed info but also enter dbase under format.
b. Distribution element repeats and contains a repeating choice. This
doesnt make sense unless theres something about the inline element that might
occur more than once per entity.
c. Ascii fixed wont work as it is. Start column needs to repeat with field
length and you need to add physical record information.
d. drop genericBinary unless someone has a definition for it
#13 Updated by Owen Eddins over 17 years ago
I'm passing the following comments from Tim Bergsma the data manager at Kellog
Biological Station in Michagen. He made them in a eml-dev email. I posting to
bugzilla just to make sure they don't fall through the cracks.
11. It looks from my printout as though <distribution> is defined
somewhat differently under <resourceGroup> vs. <physical>, i.e. no
#14 Updated by Matt Jones over 17 years ago
Distribution has now been changed in resource and physical. They are
essentially the same now (both include an inline element), but the resource
DistributionType allows for an online/connectionDefinition to stand by itself,
whereas the PhysicalDistributionType only allows connectionDefinition inside of
connection (as in "online/connection/connectionDefinition"). All issues in this
bug are resolved with these changes. FIXED.