Google Earth Blog

The amazing things about Google Earth


sentinel

How to quickly preview the latest Sentinel imagery

[ Update: There were a few bugs in the first version of the KML most notably it reported no imagery for tiles starting with ‘0’. If you intend to use it, please re-download. We have also slightly improved the speed with which it checks for imagery as well as changing the sorting of the displayed images to most recent first. ]

When there are major events such as the ongoing flooding in Louisiana we often check whether the events are visible in Sentinel or Landsat imagery. Given that any given location is only covered about every 10 days by Sentinel and every 16 days by Landsat 8, the first thing we want to check is when the most recent images were taken and whether or not they are any good.

Note: Landsat 7 is offset from Landsat 8 by 8 days, so between them they cover the earth every 8 days, and the Copernicus mission plans to launch a second Sentinel-2 satellite this year, bringing down the revisit frequency to 5 days.

Checking the latest Sentinel imagery is only a few clicks away on Amazon Web Services (AWS), but we thought it would be worth making it even easier.

Simply download this KML file and when you click on a tile, it will check AWS for the latest imagery for that tile and show you previews of the most recent 6 images. If you are interested in one, simply click on the date below the image to go to the appropriate AWS download page. See this post for more on how to download it, process it and view it in Google Earth.


As you can see, quite often, most of the images are not usable as they don’t cover the part of the tile you may be interested in, or there is too much cloud cover.

The main difficulty we encountered while creating the KML was getting our modern JavaScript code to work in Google Earth’s slightly ageing internal browser. The problem is that Google Earth does not show JavaScript errors, so it is a matter of trial and error. Another difficulty was simply working with the KML, because it is so large. The original file was over 100 Mb and although we stripped it down to under 20 Mb by removing unnecessary information and reducing the precision of the coordinates, it still tended to be a bit too much for our text editors.

More about processing Sentinel imagery

We recently had a look at how to process Sentinel imagery using GIMP. GEB reader ‘DJ’ asked in the comments if the geodata supplied with the imagery can be used to automatically align the imagery in Google Earth, rather than the manual method we had suggested. So we decided to investigate. We had initially thought the geodata could easily be extracted from a file called metadata.xml that is supplied on AWS with the imagery, but it turns out that although that file does contain the geodata it is not the straightforward latitude and longitude of the images. Instead, the coordinates are supplied in the Universal Transverse Mercator coordinate system. There is also a lot of other information, such as the angle of the sun relative to the ground at any given point and the angle relative to the ground with which the satellite camera was viewing it.

In addition to the metadata.xml file, it turns out the ‘.jp2’ files also contain geodata in Geography Markup Language (GML), again using UTM for the coordinates. If you open the ‘.jp2’ files in a text editor you can see the GML data. There is also a file called tileInfo.json that again contains the coordinates in UTM format.

We decided it wasn’t worth the effort of trying to convert the UTM coordinates into latitude and longitude, as the KML file for determining tile codes already has all the information we need. So we made this JavaScript tool that accepts a tile code and creates a KML Ground Overlay with the correct coordinates and you can then open its properties in Google Earth and select the Sentinel image you have created with GIMP. Note that the corners will be correct, but we found that the imagery in other parts of the image may not line up exactly with Google Earth imagery. We believe GeoSage’s Spectral Transformer for Sentinel-2 Imagery is able to stretch the image using the information in metadata.xml for a more accurate result, but we could not confirm this as our trial licence has expired. For most casual uses, however, our method should be good enough.

Also of note is that the AWS files include a file called preview.jpg that is well worth checking before anything else, as you may find there is too much cloud cover, or the image doesn’t cover the part of the tile you are interested in, etc.

It is important to note that the Sentinel imagery has a resolution of 10 m per pixel, which is better than Landsat imagery but not as good as the high resolution satellite and aerial imagery available for most locations in Google Earth. So the main use of Sentinel imagery is for observing large scale events that are not yet visible in Google Earth. It is also good enough to see developments like road construction, deforestation or mining in areas where Google Earth has no recent imagery.

False Colour
It is also relatively easy to create false colour images. To do this, just download extra bands and substitute them when combining in GIMP. For example, one of the more popular false colour combinations is to use the near-infrared band B08 as Red, the Red band B04 as Green and the Green band B03 as Blue. This is a good combination for seeing fire scars. We tested it on the area around Lake Erskine, California the site of the largest, most destructive wildfire of the 2016 California wildfire season.


Copernicus Sentinel data, 2016.
1. The scar from the Erskine fire (blackened area).
2. and 3. show two other small fire scars.

Note that the other bands are lower resolution. See this page for details on the different bands.

Here is the same area done with band B8A, B11 and B12:

Processing Sentinel imagery with GIMP

Sentinel imagery can be thought of as Europe’s equivalent of Landsat imagery. It is freely available just like Landsat imagery, but higher resolution. Today we are having a look at how to process it in order to view it in Google Earth with the help of GIMP.

Before we begin, if you intend to work with Sentinel imagery a lot, then first have a look at GeoSage’s Spectral Transformer for Sentinel-2 Imagery, as that makes the process extremely easy and adds some additional features that we simply cannot accomplish with GIMP. The only downside is it is not a free product.

In addition, the European Space Agency (ESA), provides a free tool called SNAP for processing Sentinel imagery, but we have not yet managed to figure out how to use it to get imagery into Google Earth.

Obtaining the imagery
The best way we have found for getting Sentinel imagery is from Amazon Web Services (AWS). The first step is to find out what the tile code is for the area you are interested in. To do this, download this KML file, which shows the tiles and tile codes. In our case we were interested in a landslide that occurred near Glacier Bay, Alaska on 28th June, 2016. This turned out to be tile 08VLL. The next step is to go to this page on AWS and find the data for your chosen tile. So in our case, select ‘8’ which represents the ’08’ in the tile name, next select ‘V’ and finally ‘LL’. Then you choose the date you are interested in, in year – month – day order. Imagery is captured about once a week, but it can vary by location. In our case, the only image so far captured after the event of interest was captured on July 11th, 2016. Finally click on the ‘0’ as there is typically only one image for a given day. You should now see a list of files available, and for a standard colour image you only need B02.jp2, B03.jp2 and B04.jp2. Download them by clicking on the links. Each one is about 85 MB.

The imagery can also be obtained here, which provides the imagery in a format suitable for use with SNAP, but the downloads are typically 5 to 6 GB as they include a large area and all the colour bands.

Converting to jpg
The Sentinel imagery is provided in a format known as JPEG 2000 with file extension “.jp2”. Although the JPEG 2000 standard was created in 2000, it hasn’t been very popular and not many programs support it. We believe GIMP has partial support, but it was not able to open the Sentinel imagery. So, we used a free image viewing program called Irfanview to do the conversion. Simply open the files in Irfanview then save them again as “.jpg”. Other free converters exist such as OpenJPEG and ImageMagick, both of which are command line converters.

Combining the colour bands
The next step is to open all three images in GIMP – open one first, then add the others as layers by dragging them into the ‘layers’ pane. To combine them into a single image, select Colors->Components->Compose. Choose RGB as the colour model and select B04 as the red channel, B03 as the green channel and B02 as the blue channel. This will open a new GIMP window with the three layers combined into a single image. It may still look a bit colourless at this stage. Now select Colors->Levels. In the popup window click the ‘auto’ button, then click ‘OK’. The colours should now look a lot better.

At this point our image looked like this:

Glacier Bay, Alaska, Copernicus Sentinel data, 2016.

Note that the image doesn’t fill the whole square and it is actually only part of a much larger image. However, even this piece is larger than we actually want. So, we cropped the image to the area we were interested in, then exported it as a “.jpg”.

Importing into Google Earth
When you use GeoSage’s Spectral Transformer for Sentinel-2 Imagery, as mentioned earlier, the resulting image contains the geographical coordinates and it can simply be drag and dropped into Google Earth Pro. However, our method above does not include any geolocation information, so it must be manually positioned. Open Google Earth, navigate to the approximate location the image was captured then add an image overlay. In the image overlay properties select the file previously created with GIMP. Now adjust the transparency slider (found just below where you selected the image) to about half way, so you can see both the image you are adding and the Google Earth imagery behind it. The default settings allow you to rotate and adjust the size of the image overlay, but force it to remain a rectangle. However, our Sentinel image is typically not exactly rectangular, so go to the ‘location’ tab in the overlay’s properties window and click ‘Convert to LatLngQuad’. This changes the way you adjust the overlay so that you can now move each corner individually. It can be a little difficult to get it just right, but patience usually pays off in the end. Moving each corner adjusts the whole image and puts out of alignment parts that had already been aligned. You need to look for easily recognisable features as close as possible to each corner then match up the overlay with the Google Earth imagery at each corner in turn and repeat several times until they all match. Once you are done positioning it, put the transparency slider back to the right, so that the overlay is no-longer see-through.

Once aligned, this is what our image looked like in Google Earth:


Glacier Bay, Alaska, Copernicus Sentinel data, 2016.

Zooming in to the location of the Landslide:


Landslide near Glacier Bay, Alaska, Copernicus Sentinel data, 2016.

We can also use Google Earth’s measuring tools to find that the area affect by the landslide is about 10 km in length.

To see the above image in Google Earth download this KML file. To get an idea of the size of the event, look at the northern edge of the overlay. There are two cruise ships visible, one in the Google Earth imagery and one in the Sentinel image. They look tiny in comparison to the landslide. If the landslide had gone into the water it could have caused a catastrophic tsunami.

[ Update: Also see this post for more on processing sentinel imagery. ]

Spectral Transformer for Sentinel-2 Imagery

We have previously had a look at a tool by GeoSage for processing Landsat Data. At the time it was free for non-commercial use, but is no longer. It remains, however, the best tool we have come across for that purpose.

GeoSage has recently released Spectral Transformer for Sentinel-2 Imagery, a tool for processing Sentinel imagery. It is not free software, but well worth a look if you work with Sentinel Imagery. More of interest to us at GEB, is the wealth of information about Sentinel imagery and how to obtain it that they have provided on the product page.

The Sentinel program provides their own software called SNAP for processing Sentinel imagery, but we have not yet been able to figure out how to process imagery with it for use in Google Earth.

We downloaded the trial version of Spectral Transformer for Sentinel-2 Imagery and tried it out on some Sentinel imagery downloaded from Amazon Web Services. It was fairly straightforward to use and the resulting image can simply be drag-and-dropped into Google Earth Pro.

We tried it for the region around Katie, Oklahoma which experienced a number of tornadoes on May 9th, 2016. However, the only image available after that date was captured on May 25th and has excessive cloud cover. We tried false colour imagery using one infrared band (B08,B04,B03) but that was no better. So we thought we would try the shortwave infrared bands (SWIR), but that caused Spectral Transformer to crash on the final step, where it uses one of the 10m resolution colour bands to “Pan” sharpen the imagery (the SWIR bands are 20m resolution. So it looks like they still need to iron out a few bugs. However, the SWIR image at 20m resolution was created and could be viewed in Google Earth Pro.

[ Update: I tried running the same imagery combination the next day and it worked without crashing. I also got feedback from GeoSage suggesting it might be a RAM related issue (not enough free RAM). So if you encounter this problem try closing all other programs before running it to see if that helps (and possibly reboot first). If that does not work then contact GeoSage. ]


Try as we might, we just couldn’t see through the cloud cover. Copernicus Sentinel data, 2016.

Sentinel vs Landsat imagery

Yesterday we had a look at Snapsat, a useful website for obtaining Landsat imagery. The location we chose was Dallas, Texas and a track made by a tornado in December last year.

We thought it would be interesting to compare the Landsat imagery with the European equivalent – Sentinel imagery.

Landsat 8 captures colour images at 30 m per pixel, but also captures has a panchromatic band at 15 m per pixel, which can be combined (using a process known as pansharpening) with the colour bands to essentially achieve close to 15 m per pixel resolution. Sentinel 2A, on the other hand, captures images in colour at 10 m per pixel and does not use pansharpening. So the Sentinel imagery should be slightly better quality and we found this was indeed the case.

We are still learning the best ways to process imagery and neither image has been processed ideally.

before
after

Landsat 8 imagery vs Sentinel 2A imagery.
Images courtesy of USGS/NASA Landsat and Copernicus Sentinel data 2016.

To see the images in Google Earth download this KML file.