Getting hi-res Landsat Images.

[Wipneus]

Intro.

As discussed in the "NØIB (Norske Øer Ice Barrier)" thread, Landsat images available in viewers are not the best for those who like to be as close to the ice as possible (without going there):

- image format is JPEG. When zooming in to the details, the artifacts become apparent;
- resolution is far worse(180m)  than the native resolution from the Landsat imager (30 meters for most channels and one 15 meter pan-chromatic (B&W) channel);
- the images are false colors based on two infrared and one visible (red) channel. Fine for land images with vegetation, but not for images with lots of water, ice and snow.

Applicable pages from the USGS Landsat FAQ's:
What are the best spectral bands to use for my study?
What band combination is used for Landsat browse images?
How does Landsat 8 differ from previous Landsat satellites?

Step 0: Get a registration on the web site of the US Geological Survey.

This is required if you want to download image files in the process that I am now describing. If you just want to explore the images available, skip this step.
Registration is free, and takes entering email+password, and responding to the confirmation email. Save the password somewhere when you consider using the bulk downloading app, you need to enter the password there as well.

Step 1: Find suitable Landsat images with the LandsatLook Viewer.

- Locate the spot of interest. You can use a name (Paris, France, or even Tobias, Greenland) or longitude,latitude coordinates.
- Zoom in sufficiently until a button "Select Scenes" appear in the selection box
- Press "Advanced Query". Make sure the OLI sensor is selected. OLI is Landsat 8, that is latest and greatest but only available in 2013 (and later, later). Select other sensors if pre-2013 data is of interest. Also set "Maximum Cloud Cover" to something high, sometimes partly clouded scenes have a clear view i just the thing that is interesting. Press Apply.
- Select "Only One" under the date slider.
-With the date slider, view the different images that are available in your area of interest.

Step 2. Order and download the Landsat data.

Convinced that the image in the viewer is worth doing some more effort on?  Has your computer enough RAM, physical and swap, installed to handle 800 MB images? Recommended is 16 GB physical RAM and perhaps double of that virtual.  Then you can continue for some serious computing.

Press the button "Metadata", a box appears with some detailed information on the selected image.
- Press "Add To Cart". You can do this actually several times before proceeding, but here I assume only one image is what we want;
- Press "View Cart"
- in the "Cart" select the image and press "Get Landsat Data";
- confirm that you go to " USGS EarthExplorer" (where you can check out);
- at some point you may be required to log in (see Step 0), unless like me you have selected the "keep logged in" option;
- a new browser window is opened with a list called "Pending Scenes". There should be only one "Scene", our image. Click on its "Entity Id";
- A new page opens with lots of details of our image, but at the bottom is a button "download". Just click on it, if you haven't yet done so (^*);
- In the last page, click on the "download" button that says "Level 1 GeoTIFF Data Product";
- Select the location where you want the file and start the download;

 ^*) If the Level 1 data seems not available, look here

Step 3: Unpack the "Level 1 GeoTIFF Data Product"

The downloaded file is a compressed archive called something like LC80030042013267LGN00.tar.gz. I recommend unpacking the archive in to an empty directory. Use your favorite software, under Linux the command is:

  tar xzf LC80030042013267LGN00.tar.gz
Once unpacked you have a directory with these files:

-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B1.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B2.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B3.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B4.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B5.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B6.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B7.TIF
-rw-r--r-- 1 672931638 sep 24 20:14 LC80030042013267LGN00_B8.TIF
-rw-r--r-- 1 168288198 sep 24 20:14 LC80030042013267LGN00_B9.TIF
-rw-r--r-- 1 168288198 sep 24 20:15 LC80030042013267LGN00_B10.TIF
-rw-r--r-- 1 168288198 sep 24 20:15 LC80030042013267LGN00_B11.TIF
-rw-r--r-- 1 168288198 sep 24 20:15 LC80030042013267LGN00_BQA.TIF
-rw-r--r-- 1      7676 sep 24 20:15 LC80030042013267LGN00_MTL.txt


All files start with the image identifier, the Bxx means "spectral band xx". Bands, and their use are listed here For this discussion B2, B3 and B4 are important as they are the visible light bands Blue, Green and Red 30 m images.  The B8 is the panchromatic B&W hi-res image, with the 15 m resolution.
Of course all other bands can be experimented with for the more advanced studies. Here we aim at a simple natural looking image.

Step 4: Inspecting the image files and introducing ImageMagick.

At this stage you might fire up your favorite image editor (the Gimp, Photoshop etc.) to look what is in those files. When I was experimenting that is what I did myself. But because the processing of these "spectral band images" is highly repeatable, I use a command line tool for that.
The free tool is ImageMagick and available for Unix, Windows, Mac and even iOS. I am using these every day, for instance to produce the ice animations in the amsr2 thread.
Make sure the software is installed and open the terminal.
Imagemagick is a set of different programs, each with an unbelievable set of options. For introduction start with identify, used to describe the format and attributes of an image:

$ identify LC80030042013267LGN00_B2.TIF
LC80030042013267LGN00_B2.TIF TIFF 9171x9171 9171x9171+0+0 16-bit Grayscale DirectClass 168.3MB 0.010u 0:00.089


For now it is important to note that the file is a 16 bit TIF image, with size 9171 pixels width and height. Different Landsat scenes may have slightly different sizes, something that will slightly complicate out processing later on.

When ImageMagick reads the TIF files from LandSat it prints several warnings:

identify.im6: LC80030042013267LGN00_B2.TIF: unknown field with tag 33550 (0x830e) encountered. `TIFFReadDirectory' @ warning/tiff.c/TIFFWarnings/768.

As far as understand these warnings are harmless, concerning non image data embedded in the files, and can be suppressed by adding an option -quit on the command line.

Get the image information of the Panchromatic channel:

$ identify -quiet LC80030042013267LGN00_B8.TIF
LC80030042013267LGN00_B8.TIF TIFF 18341x18341 18341x18341+0+0 16-bit Grayscale DirectClass 672.9MB 0.000u 0:00.060

Because this channel is 15 m instead of 30 m, the width and height in pixels has doubled but not exacly! Actually one pixel is missing in both directions, a fact that we have to solve in the next section.

The last ImageMagick tool that I would like to introduce here is [ur=http://www.imagemagick.org/script/convert.phpl]convert[/url], the work horse of the set. With an almost infinite set of options it can convert between image formats as well as resize an image, blur, crop, despeckle, dither, draw on, flip, join, re-sample, and much more.

$ convert -quiet LC80030042013267LGN00_B8.TIF LC80030042013267LGN00_B8.png

Converting the TIFF file to a png file format. Png is much more efficient in disk space:
-rw-r--r-- 1 672931638 sep 24 20:14 LC80030042013267LGN00_B8.TIF
-rw-r--r-- 1 285094673 nov  6 12:53 LC80030042013267LGN00_B8.png


This is the file from which I showed selections in  the "NØIB (Norske Øer Ice Barrier)" thread.

Step 5: Process the image files with ImageMagick commands.

Ready for the magick? Here we go.

Combine the Read, Green and Blue images to one (composite) natural color image:

Code: [Select]

$ convert -quiet LC80030042013267LGN00_B4.TIF LC80030042013267LGN00_B3.TIF LC80030042013267LGN00_B2.TIF -channel RGB -combine out-nat-30.png
Producing an intermediate file out-nat-30.png, containing a perfectly usable natural color image with 30 m resolution.
But we go for the 15 m resolution, because it is there. In order to do that we must resample the image for the increased width and height:

Code: [Select]

$ convert out-nat-30.png -filter Point -resize 200% -crop 18341x18341+0+0 out-nat-15.tmp.png
The -filter option prevents that ImageMagick does some "smart" pixel interpolating when resizing and just doubles the number of pixels (in each direction).
The -crop option is nescessary because the 15 m panchromatic image had not exactly twice the size. The -crop  18341x18341+0+0, trims the result to the same size. The "+0+0" gives the offset from where to crop, this means count from the upper left corner which is what we want.

The result is a natural color image that has the correct number of pixels, but effectively the resolution is not changed because every cluster-of-four pixels has the same RGB value. The trick is to get the 15 m panchromatic information in.

Next step is to separate the RGB values into another color space called HSL, consisting of Hue,  Saturation and Lightness:

Code: [Select]

convert out-nat-15.tmp.png -colorspace HSL -separate separate_HSL_%d.png
Producing three (intermediate) files called separate_HSL_0.png, separate_HSL_1.png and separate_HSL_2.png. Each has 15 m pixels but an effective resolution of 30 m. The trick is to replace the Lightness image by our panchromatic (B8) channel. Here is the command to rebuild the image this way:
 

Code: [Select]

convert -quiet separate_HSL_0.png separate_HSL_1.png LC80030042013267LGN00_B8.TIF -set colorspace HSL -combine -colorspace RGB out-nat-15.png
Which produces the final result a 15 m hires natural color image. Only at the highest magnification where individual pixels are visible, the 30 m color resolution becomes visible.

The image is quite large:

-rw-r--r-- 1 716890168 nov  7 10:01 out-nat-15.png

A reflection of the amount of information in the image. Please don't convert to JPEG, or perhaps until you made the final cut and adjustments. Then convert with the highest quality settings.

Step 5a: Some optimizations.

In the previous section I have shown the procedure in a number of small steps. That is good for an explanation, showing what is done without the clobber of a number of simultanious things.
In this section are the commands that I actually use. You may skip it altogether if you like.

First there is the problem that I already mentioned that the size of the Landsat scenes are not always the same. Since we need it for the "-crop" option, we have to extract it form the files.

Under a Unix compatible shell (Linux, Mac and Windows using CygWin) it goes like this:

Code: [Select]

cropsize=`identify -quiet LC80030042013267LGN00_B8.TIF | cut -d ' ' -f 4`
This creates from the output of the identify command a variable named cropsize  with a value:
$ echo $cropsize
18341x18341+0+0

A further complication is that I wrote the commands with full file names. If these files are the only similar named files, we can use wildcards. So instead of:
LC80030042013267LGN00_B8.TIF
write:
*_B8.TIF

This makes the command independent from the actual Landsat scene.

Lastly, not all intermediate files are needed. I have not tried (yet) to squeeze everything on a single line but this is the current script (three lines):

Code: [Select]

cropsize=`identify -quiet *_B8.TIF | cut -d ' ' -f 4`
convert -quiet *_B4.TIF *_B3.TIF *_B2.TIF -channel RGB -combine  -filter Point  -resize 200% -crop $cropsize -colorspace HSL -separate separate_HSL_%d.png
convert -quiet separate_HSL_0.png separate_HSL_1.png *_B8.TIF -set colorspace HSL -combine -colorspace RGB out-nat-15.png

The script takes about 5 minutes here. I see the memory monitor for the process convertspike at 13 GB. This will give problems when you thought 8 GB was plenty or RAM.

Step 6: Final processing.

The conversion is done, but if we look at it the image is not very brilliant, lacking contrast. It is also far too big to share on the web.
The final edit are currently done in the image editor, I am using the Gimp , but I don't expect that is critical (provided it is 64 bits and can handle the sizes).

In the Gimp, I enhance the image with the automatic white balancing tool:
   Colors->Auto->White Balance
The cutting can be done with the crop tool or use a selection tool followed by choosing "Crop to Selection" from the image menu.

Further.

The procedure described above, is not finished. I am thinking:
- the HSL color separation is just one of several slightly different possibilities. HSL is the first one I tried, I have no idea if it is the best one;
- What about other spectral bands? Does infrared carry useful information that helps understanding the scenes;
- The "white balance" solution to get brilliant colors should be explored further;
- perhaps there are other ideas, I'd like to hear them.

[sidd]

link in step 0 seems broken ...

[Wipneus]

link in step 0 seems broken ...

Fixed, thanks.

[ggelsrinc]

I was hoping you'd do something like this and only one word comes to mind. I've worn out bravo and just want to say thanks.

[Neven]

Great post, Wipneus. I might copy it for a blog post next week or so.

[Wipneus]

My description in the top post is more or less complete.

No doubt there are mistakes and things that can be explained more clearly. Let me know what you find.

[Wipneus]

In the downloading process, I am hitting a case that the data is not immediately available:

the "Level 1 GeoTIFF Data Product" is not under the downloadable option, button is grayed and says "Submit for Processing".

In this case:
- I went back to the "Pending Scenes". Noticed that the L1 products are under "ON DEMAND";
- select the "order" option, clicked on "Apply";
- in the next screen press "go to item basket"
- press "proceed to checkout"

An order confirmation is sent by email, with the message "When the data is processed, anyone who ordered it will receive download notification"

A few hours later an email is received with the message that the order was processed.

A link in the email sends you back to the "Pending Scenes" where the L1 data is now available.


 

[sidd]

ImageMagick is a wonderful thing.Look at GRASS to stick all these into a GIS database (you might have to install postgres, but thats not too bad)

[Wipneus]

For the Landsat 7 TM and ETM+ instruments, the natural color bands (RGB) are 3,2 and 1. Not 4,3 and 2 as is used the instructions.

I will update those soon.

[Patrick]

Fantastic post, Wipneus, good work!

Let me link two additional blog entries from MapBox. Especially the first one should be interesting as it's explaining and showcasing the different bands available from Landsat 8. The second post basically does the same as Wipneus', but I link it anyway...

Putting Landsat 8’s Bands to Work
Processing Landsat 8 Using Open-Source Tools

[Wipneus]

Fantastic post, Wipneus, good work!

Let me link two additional blog entries from MapBox. Especially the first one should be interesting as it's explaining and showcasing the different bands available from Landsat 8. The second post basically does the same as Wipneus', but I link it anyway...

Putting Landsat 8’s Bands to Work
Processing Landsat 8 Using Open-Source Tools

Thanks Patrick, those links look very helpful.

[Espen]

Wipneus,

Maybe it is just easier going up there?

[ChrisReynolds]

Thanks, yet again, Wipneus.

[ggelsrinc]

I know this article is about forests and a tree hugger like me has a natural fondness for it, but something in the contents started me thinking.

In this era of big data, anyone can now see how and where the world's forests are changing thanks to a new mapping project made possible, in part, by the computing resources of the tech giant Google.

The map compiles 100-foot-resolution satellite images of Earth's land area taken each season, every year between 2000 and 2012, to paint a picture of where trees were lost or gained. Globally, the map shows that 888,000 square miles of forest were lost between 2000 and 2012. In the same period, 309,000 square miles were gained.


"We've been working on global-scale land cover monitoring for a while but just with big, blurry pixels. This is the first time we've done it with a resolution and granularity where change is quite discrete and we can quantify it clearly," Matthew Hansen, a remote sensing scientist at the University of Maryland who led the team that created of the map, told NBC News.

For the project, Hansen and colleagues accessed high-resolution satellite imagery from the U.S. Geological Survey's Landsat 7 satellite, the first in the decades-long remote sensing data collection program to capture imagery from Earth's entire land area. This data was made freely available in 2008, which removed a cost barrier for this type of mapping work that researchers had previously found insurmountable.

Source: http://www.nbcnews.com/science/forests-disappearing-2000-google-cloud-maps-global-changes-2D11591792

Having been around the internet during the pre-PC days makes me wonder, why it wouldn't be easy to get many PCs involved in making hi-res Landsat images.

I don't consider the idea new, but I'm not aware of any service allowing PCs to be linked together to work on a personal project involving hi-res Landsat images or other high computational requirements.

Obviously, I'm not google, but I wouldn't mind lending my computer to assist efforts for a good cause. It isn't like I always use it.   

[Wipneus]

I figured out today why the Landsat 7 images turn out so discolored. It is the 8 bit depth of the images, where Landsat 8 uses 16 bits (12 of them significant). It is important that the calculations use enough precision and if forced to 16 bits the colors are much more realistic.

Adding "-define png:bit-depth=16" as a first option to the first "convert" command does the trick.

[A-Team]

Right, on 12-16 bit imagery, the key thing is taking a processing step or two within software that actually utilizes that bit depth information, for example contrast changes, rotations, bump maps, or re-projections in ImageJ. After that step, the file can be saved out of that software into software like gimp that does not, without any loss of information.

That is, look at the initial histogram. If it is tightly peaked like a dark Landsat, the image would get very posterized if gimp were used at the first step -- it simply does not have the information needed to normalize that histogram smoothly over the full [0.255] range. However once a change is made within ImageJ that accurately spreads out the histogram, there is no further advantage to higher bit depth.

Dropping from 16-bit Tiff to 8-bit png or lossless jpeg will dramatically decrease file size, making it faster to experiment with image enhancements.

Here is a July 2014 open access article from NSIDC that reviews all the main types and sources of freely available remotely sensed data with cryospheric applications, as well as free software needed to view and analyze it. This includes "aerial and satellite photography, satellite-borne visible, near-infrared and thermal infrared sensors, synthetic aperture radar, passive microwave imagers and active microwave scatterometers.... The paper concludes with a discussion of open data access within polar and cryospheric sciences, considering trends in data discoverability, access, sharing and use."

http://www.mdpi.com/2072-4292/6/7/6183/pdf

[A-Team]

Ah ha ... this procedure of sharpening color images using the higher resolution panchromatic channel even has a name and its own wikipedia article. https://en.wikipedia.org/wiki/Pansharpened_image In fact, Landsat sensors are intentionally designed around this idea:

"Pansharpening is a process of merging high-resolution panchromatic and lower resolution multispectral imagery to create a single high-resolution color image... Pansharpening produces a high-resolution color image from three, four or more low-resolution multispectral satellite bands plus a corresponding high-resolution panchromatic bands.

One of the principal reasons for configuring satellite sensors this way is to keep satellite weight, cost, bandwidth and complexity down. Pan sharpening uses spatial information in the high-resolution grayscale band and color information in the multispectral bands to create a high-resolution color image, essentially increasing the resolution of the color information in the data set to match that of the panchromatic band.

Common color-space transformation used for pan sharpening are HSI (hue-saturation-intensity), and YCbCr. The same steps can also be performed using wavelet decomposition or PCA and replacing the first component with the pan band."

Wavelet decomposition has a very fast and effective filter in Gimp --> Generic.

Principal component analysis (like sinc/Lanczos resampling) is another of these 'optimally rational' procedures from the statistical perspective that doesn't always offer much to the human perceptual point of view. I haven't seen PCA (https://en.wikipedia.org/wiki/Principal_component_analysis in Gimp but ImageJ has the plugin, http://lmb.informatik.uni-freiburg.de/resources/opensource/imagej_plugins.en.html.

We could be using PCA to reduce the band count in Landsat8 while retaining almost all the information, say B1-B6 --> 3 channels --> 2 channels + panchromatic. Rather, EarthExplorer should be pre-computing this and offering as a download. Note that same site offers an ImageJ plugin that can open netCDF and HDF5, so we aren't the first to want to exit those.

ImageJ is basically a shell for hosting plugins that do advanced digital processing, plus the odd menu command for digital editing; Gimp is focused instead on digital editing (a very different endeavor), though it too supports processing plugins. However files can be passed back and forth fairly seamlessly.

The main difference being the 80,000 lb gorilla behind ImageJ vs a ragged band of determined volunteers at Gimp.

[Ned W]

PCA was historically used pretty widely in the optical remote sensing community to reduce the dimensionality of large data sets, but it's not really necessary for most purposes nowadays and the results can be a bit unpredictable due to their dependence on the distribution of the data.

If you want to pan-sharpen the Landsat-8 imagery, there are other methods better than PCA for doing that.  Also, keep in mind the spectral range of the panchromatic band on Landsat-8 -- it overlaps with the visible bands (2, 3, 4) but it's probably not appropriate to use it with the near-IR or SWIR bands.

If you want to reduce the dimensionality of Landsat-8 data, rather than using PCA I'd recommend averaging the three visible bands (2,3,4) since they're highly intercorrelated, but keep the near-IR and SWIR bands separate (if you need to use them at all... I'm not sure what your objectives here are).

You can get almost all the meaningful information from a Landsat-image by using one visible band (2, 3, 4, or their average), the near-IR band (5) and one SWIR band (6 or 7, or their average).  It's more physically interpretable than a 3-band PCA image.

[A-Team]

Helpful comment. Our imaging objectives in Greenland are driven by its huge expanse of flat ice and snow (eg google earth overview) which are initially featureless (~ uniformly reflective) at short wavelengths and-- other than melt lakes and calving fronts -- don't provide much of a time series. Landsat, as the name suggests, images Greenland mostly out of orbital boredom.

We have longer wavelength channels such as Jaxa cloud-penetrating radar used for daily melt but its resolution is very low but with much higher resolution Sentinel 11-day return on its way imminently.  If PCA of Sentinel with Landsat is not immediately interpretable, so much the better. (We've mostly been combining disparate imagery to date as color space channels.)

Additionally, there exist published products such as surface and bedrock DEMs, radar isochrons, surface slope, aspect and velocity of variable resolution, coverage and accuracy.

Thus the overall destination is not a big folder of individual views but rather a raster GIS framework that allows coordinated computation across layers of origin. That provides the context and defines the utility of derived products from Landsat bands. As the number of layers grows, so does then need for reduction of redundancy.

Things I would like to extract from Landsat: a better definition of icesheet collection channels from false-color, detection via elevated temperature at shear margins of ice streams, feature motion detection over a season from ultra-oblique sun angles, and surface classification of crevasse fields, stagnant ice, stress field etc, manifestations of bedrock topgraphy and bottom freezeups, and so on.

The fact is, field logistics are substantial so remote sensing needs to be exploited to the max, the best possible history and current attributes being needed for initialization and burn-in of models of future behavior.

[Ned W]

Yes, Landsat-8 does collect imagery on most of its passes over Greenland.  That's a big improvement over its predecessors, L-5 and L-7, which only sporadically acquired images.  My recollection is that the US Congress required L-7 to collect imagery on every orbit within US boundaries, regardless of predicted cloud cover etc., but outside the US the long-term tasking plan called for just a few images per season in remote areas.

For one thing, this makes it difficult to look at changes over time, if you want to go back to the pre-Landsat-8 period for comparison.

What you want to get out of these images will be challenging, especially if you are envisioning doing it on an operational basis over large areas.  I have had grad students work on some of those things at particular sites, for specific projects.  Automating them is more difficult.

Merging optical and SAR imagery can be useful for visual interpretation, less so for highly automated, operational processes.  It sounds like you're currently just coregistering the SAR data with the optical imagery, stacking them together, and viewing the multisensor stacks as a false-color composite.  Is that right?  In that case, sure, do some tests using the pan-sharpening methods to merge the SAR data with the optical imagery and see if the results are better for visual analysis.

Another possibility would be to look at some texture indices derived from the SAR data.  If I were developing something for ongoing operational use, I'd feel more comfortable treating the optical and SAR-derived inputs as separate things, representing spectral absorptance vs surface roughness, rather than smooshing them together.  The thing with SAR though is that backscatter is so hugely sensitive to viewing geometry and surface conditions.  Using derived products like texture might be safer than the raw amplitude values but they still can vary drastically from day to day. 

Apologies if I'm just restating the obvious and you already know all this.  I only sporadically follow the discussions here and don't know how extensive your work in this area is.

[A-Team]

NedW, your input again is much appreciated. I hope you or your grad students can become active posters here, start new forums as needed. There are lots of good projects left in Greenland for someone who can do a quick turnaround.

I am just sticking to Landsat-8 and breaking events at a local scale. Three new papers have largely drained the opportunities for island-wide mapping of bedrock, surface DEM histories, ice margin retreat, and ice thickness and ice velocity changes over the satellite era. These required a whole lot of proprietary imagery and 8-10 person collaborations.

However only first passes have been made at surface classification, layer cake horizons, post-Eemian ice trajectories, bedrock drapes and freezeup deformations. We could think about crowd-sourcing a catalog for the freezeups though I imagine I can produce one myself without busting a cpu.

Yes, mostly I am looking at whatever might be effective as a false-color composite. One thing that could be done systematically is encode variables such as z-height and horizon age as hue and saturation in HSV instead of wasting the color space with a canned palette. I will re-read your comments on SAR as the free Sentinel data comes in this month. No plans to do anything more with it than special locations.

[Patrick]

There is a new tool available to preview and download Landsat 8 imagery:

Blog post: Announcing Libra - the Landsat imagery browser you will love
and direct link: http://libra.developmentseed.org/

[Jim Hunt]

Only slightly off topic, thanks to Arcticio for pointing out the easy peasy way to get Sentinel-1A images (amongst others):

http://www.polarview.aq/arctic

[A-Team]

Landsat kicked again on 5 Feb 15 at 69º N ... Jakobshavn a bit too cloudy to see where things stand. A pair of images a year apart with very low sun angle, matching path and row -- that's the perfect formula for feature correlation software and mapping the slowest moving ice.

[Jim Hunt]

Jakobshavn a bit too cloudy to see where things stand.

I can't stop playing with PolarView. Sentinel sees through the clouds. This from February 9th:

[A-Team]

That is convenient indeed to have the main satellite resources bundled at one site with download sizes clearly marked ... now it's up to us to do some interpretation.

This 09 Feb 15 Jakobshavn Sentinel image is surprising to me for mid-winter. The outer fjord does not seem that frozen over, I suppose because Baffin Bay waters aren't that cold any more. The remaining mélange then doesn't provide as much buttressing to the glacier but perhaps enough to suppress calving, though a time series is needed to date when the last bergs calved off.

Normally the glacier continues to move west all winter though at a perhaps half the speed of July, so without the calving the front 'recovers' lost ground during the winter and historically forms a short floating ice shelf, all of which has to be undone before calving front can retreat to a new record upstream. If that indeed is the trend, Jakobshavn is off to a flying start this season.

I marked up a few things on the image below. The person at PolarView who laid on the 50º0'W longitude in 50 point font over this expensive imagery has been terminated with extreme cartographic prejudice.

[Jim Hunt]

NASA announce a Landsat 8 themed special issue of Remote Sensing. 18 open access papers:

http://landsat.gsfc.nasa.gov/?p=9756

This Special Issue aims to provide the user community with a good understanding of the radiometric and geometric properties of the Landsat 8 instruments and their data. This understanding will enable the community to effectively use the data in conjunction with data from other earlier Landsat sensors.

[Espen]

Is there anyone who knows how at use  ImageMagick on a windows platform, I dont even know how to install?

[Jim Hunt]

Is there anyone who knows how at use  ImageMagick on a windows platform, I dont even know how to install?

I use it on both Linux & Windows Espen. Maybe continue the conversation over here, where the conversation is already discussing such things?

https://forum.arctic-sea-ice.net/index.php/topic,1259.0.html

[oren]

Thanks Wipneus and all for the very detailed explanations. I managed to download an image, unpack it, and convert it. It ain't easy initially, but I hope to become more fluent soon. I am using ImageMagick on Windows7. I just found out the importance of keeping checked the installation option of "add application directory to your system path". This allows you to run it in the directory of the image files.

[A-Team]

I use Fiji (ImageJ2) for 16 and 32 bit and Gimp after that. Fiji is neat in that on each startup, it loads all the plugin fixes since last start. These stand-alones are largely written by academic biomedical researchers. The primary focus is improving contrast and managing xray, mri, and gel slices. Thus it is very well-suited for visible snow and ice imagery.

Documentation varies from excellent to too sketchy. Undo is rarely supported. I have not used 10% of the capabilities to date -- it can do some amazing things with time series and volumetric.

Gimp, like Photoshop, is more arts oriented -- scientific uses are a sideline. The interface is awkward. It has unlimited levels of undo which is a lifesaver and the layer feature is very powerful. Again, no end of free plugins that can be installed. G'mic, a free set of some 250 mostly scientific operations contributed by a French group, adds a whole bunch of 2nd year calculus commands and things like contouring to Gimp. G'mic can be run entirely online by uploading your image.

ImageJ2 and Gimp both made a seemingly trivial improvement a month back that however greatly improved my workflow: 'copy to [and paste from] system clipboard'. Before that, 'copy' lived in parallel universes. Now I can go back and forth seamlessly to pick up a critical command such as the biggie, adaptive contrast (CLAHE) in Fiji and continue on in Gimp.

There are still some features missing, notable map re-projections. There is a need to accurately inter-convert mercator, polar, Greenland stereographic, and Google Earth projections. Seems like it should be online for images with known lat,lon corners and pj.

This can be done to a certain extent by warping to a set of corresponding fixed points but these are far and few between for non-coastal Arctic Ocean and Greenland.

Quite a few satellite images and journal articles do not furnish either underlying geomapped x,y data or map projection. We spend a lot of time trying to fix 'presentation quality' images, ie poor resolution flattened-layer jpgs that someone made for an auditorium powerpoint and was too lazy to revise for the journal.

I would say only 10% of published cryosphere maps provide a palette key correctly indexed to map colors (even though this is explained in cartography 101). At the other extreme to this bs, researchers like I. Howat at Byrd are posting cutting edge photogrammetry-level products.

Two practical tips (that people on these forums aren't often doing):

• Fiji can 'Extract images from PDF' This provides the full resolution image the authors sent to the journal -- much better than screen capture from resizing in a pdf reader. Gimp also gives excellent control over pdfs.
  
• Reverse-search the initial image on the whole internet. Often the image, say in a press release, is dumbed way down in resolution. However you can quickly track down the original resolution if it's online. The three best tools for this are tin eye, google image search, and image raider. They don't use the same algorithm.

https://www.tineye.com/
https://support.google.com/websearch/answer/1325808?hl=en
https://www.imageraider.com/

[oren]

Wow. Thanks for all the info.