Google Earth Blog

The amazing things about Google Earth


sentinel/landsat on AWS

California’s Super Bloom in satellite imagery

We recently came across this interesting article which features imagery from satellite imaging company Planet, showing the flower blooms in California between December 2016 and March 2017. Although the region normally goes through an annual change, this year was especially dramatic due to heavy rains after years of drought. The article shows before and afters using imagery from Planet’s Planet Explorer Beta.

We thought it would be interesting to also check Landsat and Sentinel-2 imagery of the region.
We used our Landsat animations and Sentinel-2 animations KML files to find the best images.

Here is a ‘before and after’ using Landsat thumbnail images as provided on Amazon Web Services. This is an area just north of Los Angeles and we can see how much greener it is. However, the latest image is from late February and it seems the main flower bloom was in March.

before
after

Landsat thumbnail comparison: August 14th, 2015 vs February 24th, 2017.

So, we also downloaded Sentinel-2 imagery of the region and processed it at full resolution. Here is just a small area showing a yellow flower bloom:

before
after

Sentinel-2 imagery comparison: July 31st, 2016 vs March 28th, 2017. Copernicus Sentinel data, 2017.

And here is a region with some purple:

before
after

Sentinel-2 imagery comparison: July 31st, 2016 vs March 28th, 2017. Copernicus Sentinel data, 2017.

You can also see a portion of the Sentinel-2 image from March 28th, 2017 in Google Earth using this KML file. There is no need for a comparison image as the default imagery in Google Earth is sufficient.

‘Stabilizing’ our Landsat imagery animations

We have recently been doing a series of posts about Sentinel and Landsat imagery on Amazon Web Services (AWS), including releasing a KML file that automatically retrieves thumbnails of Landsat 8 imagery from AWS and creates animations with them.

We mentioned at the time that the Landsat images are not all perfectly aligned with each other and we had adjusted each image slightly to try and create smoother animations. To do this we used a simplified model that assumed that the imagery squares were all aligned with latitude and longitude, with up being North. It turns out that our assumptions were not valid and there was still significant ‘shake’ visible in ground features in many animations, especially those of Antarctica.

After some investigation we discovered that not only do the Landsat tiles tilt to the right as per the satellite’s orbit, but the images are placed into the thumbnails at an angle, results in further rotations overall.

The imagery on AWS is provided with a file whose name ends with MTL that contains a variety of metadata for the image. This includes the coordinates of the corners of the thumbnails (this is the whole thumbnail including the black areas). If we have two thumbnails offset from each other as shown as the red and green squares in the image below left, then in our animation we need to adjust the top of one of the images by the amount shown as the ‘top offset’. Thinking of it as purely Cartesian coordinates, to work it out in latitude and longitude it involves a series of rotations and translations, which gets rather complicated. However, we realised that instead, we could stick to proper geographic calculations, for which we already have the key routines that we worked out when working on our post on drawing circles in Google Earth. It mostly relies on an open source package called GeographicLib by Charles Karney, with a few additions also by him but not included in the main library.

Above right we see the mid-point (red circle) of the top of the red square, the mid-point (green circle) of the top of the green square, and what is known as the ‘cross track intercept’ for the green point to the thin red line. The cross track intercept is the closest point on a great circle to a location not on the circle. The distance we were looking for is from the red circle to the cross track intercept. Although this all sounds complicated, it is actually only a few lines of code, because all the hard work is done by GeographicLib. We simply repeated this for all for sides and for every frame in the animation and it worked! The animations are now much more stable even over Antarctica.

Our conclusion overall, is that although geographic coordinates can be very complicated, sometimes it is actually easier to work with them than trying to simplify things, as the heavy lifting can be done by ready-made libraries of code.

Because the thumbnails are actually higher resolution than what you can see in the popup, we have also added the ability to zoom and pan the animation. Just use the mouse scroll wheel to zoom in and out and drag the image with the mouse to pan. I am afraid we don’t have a Mac to test on, so we are not sure if this works on Mac. Let us know in the comments if it doesn’t and we will try adding keyboard controls. Remember, these are only thumbnails so don’t expect great resolution when zoomed in.

You can download the updated KML file here. We have widened the size of the popup in the standard version, but if you find it too wide for your screen, then the narrower version can be found here.

Animating Sentinel-2 imagery in Google Earth

We have recently been doing a series of posts about Sentinel and Landsat imagery on Amazon Web Services (AWS). We created tools to let you quickly preview the latest Sentinel and Landsat imagery in Google Earth. We also looked at the coverage pattern for the Sentinel and Landsat imagery and created a way to animate the Landsat imagery. Today we are releasing similar animations for the Sentinel imagery.

To see the animations, simply download this KML file, open it in Google Earth and click on any tile.

There are several differences between this one and the Landsat animations. The sentinel thumbnails are much lower resolution than the Landsat thumbnails, so we don’t provide a link to a larger version. Also, the sentinel images often do not cover the complete tile, so we have provided an extra slider to allow you to filter out tiles based on how much of the tile they cover.

The KML file also shows with colour coding how much sentinel imagery there is, with a range from green to red for 1 to 120 images per tile and white for tiles that have over 120 images. The highest numbers can be found over Europe, which is understandable given that it is a European satellite. The amount of imagery also increases towards the north of Europe, we believe this is because the paths the satellite takes overlap more near the poles, allowing more imagery to be captured. There are also hotspots over deserts suggesting that the images are selected for low cloud cover.

The Sentinel-2A satellite that is gathering the imagery was launched in June 2015. In comparison, Landsat 8 has been around since 2013. However, the Sentinel-2A satellite covers the globe roughly every 10 days, whereas Landsat 8 takes 16 days. In addition, the Landsat 8 archive on AWS only includes selected images from 2013 and 2014 (with significantly more of the US than other parts of the world) and only has the complete set of images from 2015 onwards.

We also find the clouds look whiter and obstruct the picture more in the Sentinel imagery than they do in the Landsat imagery. This may relate to how the imagery was processed for the thumbnail or it could reflect differences in the exact wavelengths the respective satellites use to capture the colour bands.

We came across a few errors in the data, such as mislabelled tiles or missing thumbnails, but they were not significant enough to seriously affect the operation of the animations.

As with the Landsat imagery, it is important to note that this is very low resolution imagery, so expect to only see very large scale phenomena. Also, with only a year’s worth of data there is not a lot of change to see. However, it is a continuously updated service and with the expected launch of Sentinel-2B sometime next year doubling the frequency of imagery, we can expect some spectacular animations in years to come.

Animating Landsat imagery in Google Earth

[ Update: We have updated the KML file with more stable animations and the ability to zoom in and out and pan. See this post for more details. ]

Last week we created a KML file to easily preview the latest Landsat imagery. The data comes from Amazon Web Services (AWS), which hosts a large amount of Landsat 8 imagery and also includes some handy thumbnail images. So, we have now created a KML file that automatically creates animations from the thumbnail images.

Yesterday we had a look at the coverage of Landsat 8 imagery, but we focused on how recent the imagery is. Today’s KML file instead colour codes tiles based on how many images are available. As we noted yesterday, the AWS archive is Landsat 8 imagery only and does not include all the imagery. It turns out that most of the world includes a fairly comprehensive set of imagery from 2015 onward, whereas the US (excluding Alaska) has imagery going back to 2013. The result is that most locations outside the US have about 44 scenes whereas the US typically has nearly twice as many, at about 77 scenes. Note that these figures increase with time as the data is live and a new scene is added to each tile approximately every 16 days.

To see the animations, download this KML file. Click on any coloured tile for an animation of that location. Depending on your internet speed, it may take a short while to load all the images. If you have slow internet or just want to get a quick preview of the animation, then download this KML file which uses smaller, much lower resolution thumbnails.

There are three sliders. The first slider shows the progress of the animation and allows you to manually switch between images. Doing so stops the animation, which you can restart with a button. The second slider lets you adjust the speed of the animation. The third slider allows you to filter the images by removing any with over a specified level of cloud cover.

AWS provides two thumbnails for each scene, one quite small one that we used for our ‘recent images’ KML and one much larger. We have used the larger one even though it is too large to comfortably show at full resolution in Google Earth. To see the animation running at full size, click the link at the top right of the animation window.

Remember that Landsat imagery is very low resolution (about 30m per pixel) and these are just thumbnails, which are even lower resolution than that. So expect to only see very large scale changes. Look for changes in seasons, ice cover changes near the poles, lakes shrinking or growing, and even sand dunes moving.

A major problem is the large amount of cloud cover in the images. If you select only scenes with minimal cloud cover you end up with only a few images to work with. This highlights just how difficult it is for commercial providers to get good imagery.

One problem we had was that the images are not captured from exactly the same angle, so there was a significant amount of shaking in the animations. We have tried to fix this by reading the latitude and longitude metadata included with the scenes and moving the individual images appropriately. It is still not perfect as a browser cannot position an image at sub-pixel resolution.

Let us know in the comments if you find anything particularly interesting. Ideally tell us the path and row.

Here is an interesting tile in Chad that shows some fire scars.

Landsat imagery coverage

Last week we looked at Sentinel imagery coverage and today we are doing the same for Landsat imagery.

First, note that we are discussing Landsat 8 imagery only, as Landsat 7 imagery is not currently available on Amazon Web Services (AWS), which is the source of our information. In addition, we have excluded Landsat imagery captured at night as we do not have the appropriate Row/Path data for night time imagery – although the imagery itself is available on AWS. Also, this page says that although all scenes from 2015 onwards are available, only a selection of cloud-free scenes from 2013 and 2014 are included.

You can see the colour coded coverage map in Google Earth with this KML file. Click on any square to see the date the last image was captured and how old that image is in days. Note that this KML is not dynamic and is based on the data as it was on AWS on August 20th, 2016.

For imagery captured in the last 20 days, we have used this colour range: with red for the most recent imagery and green as the oldest. Anything older than 20 days we have coloured white.

Having a look at the order in which strips are captured, we get a pattern like this:
1   10   3   12   5   14   7   16   9   2   11   4   13   6   15   8
which then repeats after the 16th day. See this post to see an animation of the Landsat orbit.

The imagery of Antarctica is mostly several months old, because it is too dark to capture good imagery over the winter months. There is also a horizontal stripe in the Atlantic and Pacific from February/March 2015 and a vertical stripe from May 2015. We do not know why they were captured, or why the oceans in general have never been captured.

Also of note is that some scenes are recorded with a slightly different scene identifier. Most of the scene identifiers start with the letters LC8, whereas these images start with either LO8 or LT8. There are only a few scenes with these different identifiers. According to this page this means that they include only data from the Operational Land Imager (OLI) or only from the Thermal Infrared Sensor (TIRS), whereas most Landsat imagery includes data from both sensors. We have included separate folders in the KML to show the coverage for the different cases. This page explains what each letter in a Landsat scene identifier means.

Our next project with this data is to create animations with the thumbnails.