Ice Concentration Images and Animations

[greatdying2]

I have been developing a project to post-process the Bremen (NIC colour) concentration maps in an attempt to remove some artifacts apparently caused by clouds. Clouds seem to usual appear as transient high concentration areas, so the idea is to combine a series of daily maps into a single map showing the minimum concentration over that period.

I feel that I've been overloading the 2017 melting thread with too many uploads lately, and as I will continue to play with this and related projects, I thought a new thread was in order. If anyone else would like to post other related graphics here (looking at you TT ), on a daily basis or otherwise, I strongly encourage it.

First I will repost related messages from recent days of the 2017 melting season thread:

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123466.html#msg123466

Nice animation and quite a change! One can almost imagine a line is showing where the ice edge might end up.

On a similar note, I have noticed that occasionally someone posts a merged image of the Bremen (I think?) visual colour map, where the darkest pixel has been chosen over several days to help remove cloud effects. I like these very much, but personally I prefer the NIC colour images for their improved contrast. So I wrote a program to do a similar merge on these, where the lowest ice concentration is shown for each pixel. So far I have only generated one for the last 3 days of July, which is attached along with the original images.

Note: this is beta so if anyone notices any errors, please do let me know.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123478.html#msg123478

Nice animation and quite a change! One can almost imagine a line is showing where the ice edge might end up.

On a similar note, I have noticed that occasionally someone posts a merged image of the Bremen (I think?) visual colour map, where the darkest pixel has been chosen over several days to help remove cloud effects. I like these very much, but personally I prefer the NIC colour images for their improved contrast. So I wrote a program to do a similar merge on these, where the lowest ice concentration is shown for each pixel. So far I have only generated one for the last 3 days of July, which is attached along with the original images.

Note: this is beta so if anyone notices any errors, please do let me know.

I was playing around with it at one stage - One warning about using the darkest pixel & using NIC, is that the colours are made up of RGB, so choosing the 'darkest' may not actually be accurate - Purple will have high red and blue but low green, but the greens and yellows will have high green and red but low blue, respectively. Is [1,0,0] brighter than [0,1,0]?. I think it may work better on the white to black RGB maps, unless you have a way to detect which colour is meant to be darker than the rest.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123480.html#msg123480

One warning about using the darkest pixel & using NIC, is that the colours are made up of RGB, so choosing the 'darkest' may not actually be accurate...

You're right and I don't think there's a simple way like that to do it. What I've done is to convert each pixel to a concentration value using the colours provided in the key, take the minimum concentration, then map it back to the colour.

Here is the 3-day minimum map ending today. (Actually today is so bad that you have to look pretty hard to find areas that were worse on the previous 2 days.)

Soon I plan to make a movie of these -- I think it would be quite informative. Anyways it would be much better than flipping between browser tabs and squinting hard to mentally eliminate clouds. 

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123481.html#msg123481

One warning about using the darkest pixel & using NIC, is that the colours are made up of RGB, so choosing the 'darkest' may not actually be accurate...

You're right and I don't think there's a simple way like that to do it. What I've done is to convert each pixel to a concentration value using the colours provided in the key, take the minimum concentration, then map it back to the colour.

Here is the 3-day minimum map ending today. (Actually today is so bad that you have to look pretty hard to find areas that were worse on the previous 2 days.)

Soon I plan to make a movie of these -- I think it would be quite informative. Anyways it would be much better than flipping between browser tabs and squinting hard to mentally eliminate clouds. 

Sounds good, one other warning - you might be using the lightest pixel in your algorithm by mistake - that gash to the sw of the Pole is only present in today's image, so it shouldn't be showing up in your merged image... 

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123484.html#msg123484

What I've done is to convert each pixel to a concentration value using the colours provided in the key, take the minimum concentration, then map it back to the colour.

Sometimes clouds create virtual 'low' concentration pixels, so you may actually be amplifying cloud effects in your program. Like that patch SW of the NP in your latest image.

That said, I'm surprised that your three-day algorithm finds most of the low concentration ice in the ice margin, where we would expect it.

Nice work !

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123486.html#msg123486

you might be using the lightest pixel

Yes that's exactly what I'm doing, but it's on purpose. Each pixel shows the lowest concentration over 3 days. The idea is to eliminate the falsely high concentration regions caused by clouds. (However, it also means that when the ice moves, lower concentration regions "overwrite" higher concentration ones -- so interpret accordingly...)

Sometimes clouds create virtual 'low' concentration pixels, so you may actually be amplifying cloud effects in your program. Like that patch SW of the NP in your latest image.

That said, I'm surprised that your three-day algorithm finds most of the low concentration ice in the ice margin, where we would expect it.

Nice work !

Thanks, and I'll keep that in mind. (Hopefully most artifacts will become apparent in movie format.)

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123491.html#msg123491

On a similar note, I have noticed that occasionally someone posts a merged image of the Bremen (I think?) visual colour map, where the darkest pixel has been chosen over several days to help remove cloud effects.

Hello GD2, I think that was me. I did exactly that last year, first manually and then, in the later versions, with the color picker in Photoshop (not the darkest pixels though, but the brighter green, yellow and red ones). This year I changed to the Hamburg map, which I feel has a stronger visual impact (especially if you push the contrast à la Wipneus), I posted my last result yesterday as a gif-animation. I think it's helpful if you want tho get a view of the whole picture, between three images lay just 48 hours, so in most weather conditions the drift is not so strong, and clouds move faster anyway. Great that you are contributing something similar, a single image says more than a thousand words!

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123494.html#msg123494

Yes that's exactly what I'm doing, but it's on purpose. Each pixel shows the lowest concentration over 3 days. The idea is to eliminate the falsely high concentration regions caused by clouds.

...but at the same time introducing other false values, because every ice movement will decrease concentration in your image, regardless of the real situation.

And as Rob pointed out: Clouds are not always responsible for higher concentration pixels, but also for lower. So I am not convinced, that this 3-day-composite, that amplifies just low concentration pixels, is anywhere nearer at reality, than the original images in the first place.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123495.html#msg123495

I really like the running-minimum sea ice concentration filter, it seems to be a quite reasonable approach to suppress the impact of clouds and water vapour. The ASI 89 GHz algorithm is quite sensitive to the atmosphere and mostly reacts with an overestimation of ice concentration when there are clouds (due to the scattering which decreases polarization). And it is reasonable because the atmosphere moves much faster than the ice.

So please go on with this kind of post processing, it seems to give useful and reasonable results!

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123496.html#msg123496

Yes, TT, it was you. Please post more of them.

I added this version because I like the colour contrast better, because I wanted an excuse to do some coding using ImageJ, because I want to make movies of them (which I think may be even more revealing), and because I have some other ideas of what to maybe do with this data later (something with classifiers and predicting the minimum). Oh and for fun 

Seaice.de, thanks for the encouragement. I have spent quite a bit of time over months and years "flipping and squinting" at these maps, and my observations bear out your statement that usually atmospheric disturbances (clouds, etc.) move quickly and cause dramatically increased apparent concentration. But "flipping and squinting" is annoying -- this way is much more pleasant.

Deconstruct, I do agree (as I said before): there are limitations, especially fast-moving ice. But by merging not too many days, maybe it isn't too much of a problem, most days.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123498.html#msg123498

...but at the same time introducing other false values, because every ice movement will decrease concentration in your image, regardless of the real situation.

As I wrote several times, I'm conscious of that effect. But without pixel peeping I think these images contribute something useful.

... I am not convinced, that this 3-day-composite, that amplifies just low concentration pixels, is anywhere nearer at reality, than the original images in the first place.

They are certainly flawed – as are even highly scientifically manufactured images. But it shows a more overall picture that you otherwise would not obtain. As I could observe last year and this year, they are useful to see where the ice is going. Of course, you have to take these images with a grain of salt – as you do with every graph here, I suppose.

... because I want to make movies of them (which I think may be even more revealing) ...

With a running three-days composite? THAT IS A LOT OF WORK!!! But the result would be utterly pleasant, and it would somehow mask away the distortions caused by drift in the continuous over all movement!

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123500.html#msg123500

... because I want to make movies of them (which I think may be even more revealing) ...

With a running three-days composite? THAT IS A LOT OF WORK!!! But the result would be utterly pleasant, and it would somehow mask away the distortions caused by drift in the continuous over all movement!

Not too much work left now. Generating the 3-day (or n-day) minimum maps is already automated. I wrote a program to do it -- that's how I made these maps. All I need to do now is generate one for each day (one more for-loop should do it) and then figure out how to join them together into a movie, which is easy I guess. And yes, I agree that artifacts such as fast moving ice should be revealed in a movie -- unless they manage to run quickly away for many days. 

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123503.html#msg123503

Not too much work left now. Generating the 3-day (or n-day) minimum maps is already automated. I wrote a program to do it -- that's how I made these maps.

I envy you computer guys!

All I need to do now is generate one for each day (one more for-loop should do it) and then figure out how to join them together into a movie, which is easy I guess.

Make a gif. You can make it online, there are free platforms and it's very easy. Or import the images in your favorite editing software for longer an higher resolution results. If your are creating a loop: Please let the last image stand a while longer. That provides orientation in a loop. Otherwise there is a continuous swirl that causes nausea ;-)

[greatdying2]

As others have commented, I would also love to see more of these TT (if it's not too much work...)
http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123511.html#msg123511

Sorry, if I'm interrupting the current discussion. Just posting an animation of the last two weeks with some slowmotion. In my opinion now you can see very well, what are clouds and what is SI. If the trend we see continues, the next two weeks could definitely wreck the ice.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123533.html#msg123533

Yes that's exactly what I'm doing, but it's on purpose. Each pixel shows the lowest concentration over 3 days. The idea is to eliminate the falsely high concentration regions caused by clouds.

...but at the same time introducing other false values, because every ice movement will decrease concentration in your image, regardless of the real situation.

And as Rob pointed out: Clouds are not always responsible for higher concentration pixels, but also for lower. So I am not convinced, that this 3-day-composite, that amplifies just low concentration pixels, is anywhere nearer at reality, than the original images in the first place.

Given how fractured and mobile the ice is, it may just be better to simply animate as the blinking on and off actually captures the state of the ice.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123535.html#msg123535

Given how fractured and mobile the ice is, it may just be better to simply animate as the blinking on and off actually captures the state of the ice.

The perfect solution would be a software with pixel tracking. No double count of pixels in still pictures anymore. That could provide a nearly lossless composite, even over several more days.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123617.html#msg123617

Sorry, if I'm interrupting the current discussion. Just posting an animation of the last two weeks with some slowmotion. In my opinion now you can see very well, what are clouds and what is SI. If the trend we see continues, the next two weeks could definitely wreck the ice.

Great! Yes, the clouds are obvious here -- you can even see the cyclonic patterns.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123641.html#msg123641

Here are my first gifs of the lagging minimum post-processed Bremen NIC maps. (Might need an acronym for that  ). For these I have used the minimum over 3 days, which seems to get rid of most but not all of the atmospheric artifacts. These are for the last week (7 days ending Aug 2nd).

I produced 2 gifs:
1.) A side-by-side next to the original unmodified image for the same end day. This is for those who prefer the originals, and to help evaluate the effect of the post-processing.
2.) One just showing just the post-processed image, which has better resolution (due to restrictions on what can be posted).

They do seem quite useful. For example, the rapid acceleration of melt in the last 2 days is much more obvious with the clouds removed.

I plan to make longer ones soon that go back to earlier dates, and also to play merges over longer than 3 days.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123643.html#msg123643

Here are my first gifs of the lagging minimum post-processed Bremen NIC maps. (Might need an acronym for that  ). For these I have used the minimum over 3 days, which seems to get rid of most but not all of the atmospheric artifacts. These are for the last week (7 days ending Aug 2nd).

I produced 2 gifs:
1.) A side-by-side next to the original unmodified image for the same end day. This is for those who prefer the originals, and to help evaluate the effect of the post-processing.
2.) One just showing just the post-processed image, which has better resolution (due to restrictions on what can be posted).

They do seem quite useful. For example, the rapid acceleration of melt in the last 2 days is much more obvious with the clouds removed.

I plan to make longer ones soon that go back to earlier dates, and also to play merges over longer than 3 days.

As an Acronym I propose LAMP-B (ie LAgging Minimum post-Processing-Bremen), as they're illuminating and you need a lamp at hand when in PYJAMAS

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123670.html#msg123670

I propose LAMP-B (ie LAgging Minimum post-Processing-Bremen)

I like it! With a small modification:

Here is the 3-day LAMB (LAgging Minimum Bremen) map for 4 weeks ending Aug. 2nd:

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123675.html#msg123675

Here is the 3-day LAMB (LAgging Minimum Bremen) map for 4 weeks ending Aug. 2nd:

Very nice work !
Your 3-day LAMB is definitely more 'consistent' than the unfiltered AMSR images.
Also interesting that (in this 4 week animation) the ice North of the Chukchi disappears much faster than the ice North of the ESS or the Beaufort.
Suspect is that (persistent) cyclone that drives the ice towards the CAB from the Chukchi.

[greatdying2]

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg123719.html#msg123719

I like it! With a small modification:
Here is the 3-day LAMB (LAgging Minimum Bremen) map for 4 weeks ending Aug. 2nd:

That looks indeed very good. If used in an animation like that, it is much clearer.

[greatdying2]

... And after more work than I thought, the thread is finally up to date...

[greatdying2]

Here is a 5-day LAMB of the last month, for comparison to the 3-day posted just above. (Note that the dates are slightly different.)

Judging by this, I think maybe 5 days is overkill. There doesn't seem to be much improvement (i.e., it's not much easier to see what may be happening), and of course the longer the period that is combined, the greater the artifacts that will be introduced due esp. to moving ice. Comments?

Also, I am thinking of removing the dark purple in future versions, which I find distracting due to cloud motion and which I think adds very little. Comments?

[greatdying2]

Here is the latest 3 day still image, ending Aug 3rd

[greatdying2]

Here is a 3-day LAMB for approx. the last 3 months. Actually, maybe a 5-day would be useful... Watch the Beaufort for instance.

[Neven]

It seems to be particularly useful around this time of the year, when lots of yellows and greens pop up and stick around.

Have you considered doing the same with the UH AMSR2 SIC images (higher resolutions, which means larger files)?

[greatdying2]

I hadn't thought beyond these Bremen maps... until just now when you mentioned it. 

Seems like a good idea -- I will look into it when I get a chance.

[greatdying2]

By the way, your link is missing a colon after 'ftp'. I've seen this before with links -- the forum software seems to edit them out. Tried fixing it but can't find an easy work-around. :/ Anyways thanks for the link.

[Neven]

It may be a good idea, but it will take more time and computer resources. What I don't like about the Uni Hamburg SIC maps, is that they show so much land masses. I usually crop them in Photoshop when I want to use them for something. If you want I can send you cropped images of just the Arctic.

It looks something like this:

[Neven]

By the way, your link is missing a colon after 'ftp'. I've seen this before with links -- the forum software seems to edit them out. Tried fixing it but can't find an easy work-around. :/ Anyways thanks for the link.

Indeed, the forum software seems to add 'http:' in front of the link.

Oh, well...

[oren]

If you use the icon for "Insert FTP Link" instead of "Insert Hyperlink" it will post properly. The only annoyance is that the resulting link will not be underlined.
ftp://ftp-projects.cen.uni-hamburg.de/seaice/AMSR2/3.125km/

[greatdying2]

If you use the icon for "Insert FTP Link" instead of "Insert Hyperlink" it will post properly. The only annoyance is that the resulting link will not be underlined.
ftp://ftp-projects.cen.uni-hamburg.de/seaice/AMSR2/3.125km/

I didn't even notice that button before -- thanks Oren

[greatdying2]

What I don't like about the Uni Hamburg SIC maps, is that they show so much land masses.

Agreed. I will add a cropping step. Should help reduce file size / increase resolution a lot.

[greatdying2]

I made another improvement, which I think helps a lot. Because clouds seem to mostly appear as high concentration areas, in this version I am only replacing high concentration pixels (>90%) with the minimum over the span of days.

I have also eliminated the 100% (dark purple) bucket, which I find distracting.

An animation of the last 2 weeks is attached. This time I used 5 days rather than 3, since the new algorithm introduces far fewer artifacts. For instance, this version correctly shows the ice "bounce back" after the storm, which the previous version did not, and other more subtle changes such as in the Laptev polynya.

(NB: I still need to implement a crop.)

[Peter Ellis]

With the greatest of respect, I think in many cases you are amplifying noise.  Clouds cause areas of false low concentration as well as false high concentration.  On the untreated pictures you clearly see these are arcs of low concentration pixels following the swirling shape of the weather systems.

If you want to use a multi-day averaging procedure to reduce the effect of transient noise such as clouds, why on Earth don't you take the median, which is robust to outliers in both directions?

[greatdying2]

With the greatest of respect, I think in many cases you are amplifying noise.  Clouds cause areas of false low concentration as well as false high concentration.  On the untreated pictures you clearly see these are arcs of low concentration pixels following the swirling shape of the weather systems.

If you want to use a multi-day averaging procedure to reduce the effect of transient noise such as clouds, why on Earth don't you take the median, which is robust to outliers in both directions?

Could be (perhaps you could show an example?), but if so, with this version such artifacts would probably last only 1 frame (unless they occur in an area that is over 90% concentration). Using the median won't correct areas that are covered in cloud more often than not, which I suspect from personal observation is a lot, but I appreciate the feedback and will give it a try.

[Peter Ellis]

Could be (perhaps you could show an example?),

There are some very clear "spiral" artifacts in the unprocessed images in the ESAS at around 180 degrees West, 80 degrees North, from 25-28 July and again from 30th July to 3rd August. They appear as curved "streaks" that are 20-30% lower than the surrounding ice. They vanish on the 4th.

The 28th is particularly striking, with a low concentration "streak" roughly following the line of 180 degrees West. When you look at the Earthview visible image, you can clearly see the cloud streak that is causing it.
https://worldview.earthdata.nasa.gov/?p=arctic&l=VIIRS_SNPP_CorrectedReflectance_TrueColor(hidden),MODIS_Aqua_CorrectedReflectance_TrueColor(hidden),MODIS_Terra_CorrectedReflectance_TrueColor,Reference_Labels(hidden),Reference_Features(hidden),Coastlines&t=2017-07-28&z=3&v=-4245951.750346582,-517242.99381873314,1517120.249653418,2448261.006181267

but if so, with this version such artifacts would probably last only 1 frame (unless they occur in an area that is over 90% concentration).

Quite the opposite - if the low value is a false outlier caused by cloud, your averaging procedure will retain this for the next three days, amplifying the noise.

[Peter Ellis]

In fact, if you do a careful day-by-day matchup of Earthview to the Bremen images, you will see that the clouds much more often coincide with low apparent ice concentration than with high apparent ice concentration.

The reason for this is simple.  In a visible light image, the clouds scatter/reflect light more than the surface, and so they appear whiter.  The Bremen images are not visible light images, they are microwave images.  These can distinguish between ice and water - however any water will do.  Thus, in areas where the sky is clear, they can detect the difference between ice and open water.  They are however fooled by melt ponds on the surface and by airborne moisture in clouds, in both cases giving falsely low readings.

In short: the entire premise of your averaging is wrong.

[Andreas T]

Peter is right, the effect can lead to the Bremen concentration chart showing open water where worldview shows ice which lasts for weeks longer. The strongest example I remember is of landfast ice in the Laptev sea which often is water covered when the snow melts, but there also was ice near Wrangel island last year which kept "coming back from the dead" repeatedly.
example of melt on landfast ice see here:

A reasonably good view of the fast ice between the New Siberian Islands and the Lena delta today. That Uni Bremen SIC map really fooled me  :

[greatdying2]

There are some very clear "spiral" artifacts in the unprocessed images in the ESAS at around 180 degrees West, 80 degrees North, from 25-28 July and again from 30th July to 3rd August. They appear as curved "streaks" that are 20-30% lower than the surrounding ice. They vanish on the 4th.

The 28th is particularly striking, with a low concentration "streak" roughly following the line of 180 degrees West. When you look at the Earthview visible image, you can clearly see the cloud streak that is causing it.

Thanks for the example. This particular case may or may not be an artifact on the original Bremen concentration map. After looking at it I am not convinced either way. If you look a few days later at the next clear day in this region on WV (Aug. 1), you find that the concentrations displayed for July 28 are confirmed. You can also see this on the animations I have been posting (see also below). On the other hand, if you look at earlier days (which are mostly cloudy for some time), the area appears to have some but only a small fraction of open water.

Keep in mind that red is 85-90%, orange 80-85%, and yellow is 70-80%, It's not unreasonable for the colour to change from purple to red and downwards when just a bit of water starts to show. In fact I think that's the point of this choice of colour scheme and concentration buckets -- to give an early indication of area reduction, and it's one of the reasons that I personally like the scheme and prefer it for some purposes to their "visual" colour scheme.

but if so, with this version such artifacts would probably last only 1 frame (unless they occur in an area that is over 90% concentration).

Quite the opposite - if the low value is a false outlier caused by cloud, your averaging procedure will retain this for the next three days, amplifying the noise.

No. As I had described in an earlier post, the latest version at that post was using the most recent concentration below 90% and what I said was correct. (See my post following for a summary of the algorithms used in different versions, including new versions.)

In fact, if you do a careful day-by-day matchup of Earthview to the Bremen images, you will see that the clouds much more often coincide with low apparent ice concentration than with high apparent ice concentration.

The reason for this is simple.  In a visible light image, the clouds scatter/reflect light more than the surface, and so they appear whiter.  The Bremen images are not visible light images, they are microwave images.  These can distinguish between ice and water - however any water will do.  Thus, in areas where the sky is clear, they can detect the difference between ice and open water.  They are however fooled by melt ponds on the surface and by airborne moisture in clouds, in both cases giving falsely low readings.

In short: the entire premise of your averaging is wrong.

Basically, what you're saying is that the Bremen maps are useless. Or perhaps that the correct approach in your opinion would be to take the maximum in order to amplify the very obvious high concentration artifacts? Or just don't attempt to filter them at all, and just use the "brain filter" to attempt to see through the noise?

I disagree. I have looked at multiple examples, and compared different kinds of maps, including visual animations that have been posted recently on the 2017 melting thread. To me, it seems that these Bremen maps typically display clouds as high concentration artifacts. But certainly not always and I agree that there are also some low concentration artifacts. Like any product, the Bremen maps are not perfect and if you want to interpret them properly, you need to understand their limitations. Which is why I have been posting side-by-side images of the original maps.

All I am doing here is cleaning up the images a bit and creating animations. Like any evidence, it's up to you to decide how much weight to it in your own evaluation of what's really happening. If you don't think these filtered animations are useful... ok, don't use them. Personally I find a useful supplement to the unfiltered originals with reasonable results.

If you're not sure, have a look at my following post of several different versions, including the originals.

[greatdying2]

As mentioned above, I have implemented some new versions of the filter, including a median version, and added a crop (and label), which is a big improvement -- thanks for the suggestion Neven.

As a reminder, the filters work by evaluating the concentration value for each day over a fixed period of N days. To see how each algorithm performs, I have applied each one using N = 5 days to each day from July 1st to Aug 4th. The results are presented as animations. (Note: I have had to resize the original gifs in order to get them to display properly on the forum -- 70% size reduction using other settings at default at ezgif.com.)

Here is a summary of the algorithms used in each filter:

v1.0) unfiltered originals (this post -- see next post for the others),
v1.1) minimum,
v1.2) most recent under 90%,
v1.3) median (new),
v1.4) most recent under 90%, but only if at least 2 days are under 90% (new).

I have some thoughts about the advantages and disadvantages of each, but I will save that for a later post. I would love to hear your comments on these, especially criticisms.


Click to animate.

[greatdying2]

v1.1 -- minimum

[greatdying2]

v1.2 -- latest under 90%

Note: This was the last version presented in previous posts. For each pixel, it uses the most recent concentration value that is under 90% (i.e., not purple), if such exists.

[greatdying2]

v1.3 -- median

[greatdying2]

v1.4 -- latest value under 90%, but only if at least 2 days are under 90% (otherwise use original)

[greatdying2]

Ok, there they are. Comments encouraged.

[Sterks]

I like the median. It is like watching the normal evolution but much less noisier. For the whole season, it could make an awesome animation!
The others also nice, but they introduce somehow their own noise.
Great exploratory work, greatdying, in all cases.

[greatdying2]

Thanks Sterks. I have similar thoughts. For long-term animations the median is probably is best, nice and smooth. It may even filter out some of the low concentration artifacts in the original images. On the other hand, it is relatively insensitive to day-to-day changes. For instance, if you watch the pacific side, you can see it "breathing" in some of the other animations, which I think does reflect real physical changes to the ice (although maybe not exactly just concentration). So for short-term animations, I am leaning towards one of the noisier versions (to better watch the horse race  ).