Looking at Wildfires (and more…) – An Introduction
On July 24th, 1972, the day after the launch of the Earth Resources Technology Satellite 1 (later renamed Landsat 1), Boyce Rensberger wrote in an article in the New York Times:
A new era of earth exploration began yesterday with the successful lofting of an unmanned earth-orbiting satellite that will continuously scan the surface of the earth, radioing back many kinds of information on global environment and natural resources.
He was right. And today, over four decades later, this new era is thriving more than ever before, thanks to open-data policies. With satellite data becoming easier to obtain, and technological advancements in tools making it easier to process that data, we live in a time in which satellite images are becoming more mainstream, interesting, and accessible to everybody.
This introduction is meant for all of you who want to process satellite images, with a focus on showing wildfires. I will talk about a few ways to get there and give some, hopefully, helpful hints to produce better images. We will also take a look at some facts about wildfires that might be useful to know when trying to look for them and process images.
So let us get started with some basics.
What is a wildfire?
Wildfires are uncontrolled fires, usually occurring in, but not limited to, wildlands. Some of them are started naturally, often by lightning. Others are man-made, sometimes ill-willed as arson, sometimes a prescribed or agricultural burn that didn’t turn out as planned and then becomes a wildfire.
Wildfires come in all sizes, a fire that would be described as small in one region of the Earth could be considered big in another region.
Not all wildfires are bad, many are a part of the ecological cycle in their region. Some of them become dangerous because human settlements are spreading to areas called the wildland-urban interface, or WUI, where structures are extremely close to wild areas. Wildfires that in the past were harmless now suddenly pose a danger to humans because they get close to or even reach those settlements. Climate change does its part as well, hotter and drier conditions usually favor bigger and more frequent wildfires.
A rose by any other name…
You may call it wildfire, others might prefer different terms. Sometimes those terms characterize the fire a bit more, such as forest fire or grass fire. In Australia, the term bushfire is widely used, and some people call it a wildland fire.
Causes of wildfires
Wildfires can have different ignition sources. Lightning is often the main culprit for naturally caused wildfires. Less often you may get wildfires started by volcanoes.
Wildfires started by people may have several reasons. Burning debris, cigarettes, fireworks, unattended campfires, sparks from car engines or machinery used outdoors, and, of course, sadly, arson. Another cause are prescribed fires of any kind that burn out of control.
The ignition source of a fire is important, as, given the right conditions, dry and hot weather with enough fuel available, people can be warned to change their behavior to prevent unnecessary fires. Local governments may disallow open fires, fireworks, or other activities that could pose a risk. Behavior can be changed (see for example red flag warnings), conditions are usually out of our hands.
Many people underestimate the conditions that have to be met for a wildfire to be sustained for some time and spread to a bigger area. Still, arson isn’t as rare as one would expect, one in every five reported brush/grass/forest fires in the U.S. during 2007-2011 was intentionally set. In some states arson is the leading cause for wildfires.
One of the conditions for a fire is the available fuel. As you have read before, wildfires can, for example, burn forests or grass. Actually, wildfires can burn basically anything flammable in their way. Whatever is burning is called the fuel of the fire. In some cases that might be peat, in other cases grasses, shrubs, or trees.
You can boil down the basic requirements of a fire to heat, fuel, and oxygen, the so-called fire triangle (or combustion triangle).
Technically, the third part of the triangle is an oxidizing agent, which just most times ends up to be the available oxygen from the ambient air.
The fire tetrahedron expands these three elements with the chemical reaction needed for a fire. You can read more about that here.
Different types of fires can look different
Wildfires do not always look the same. Close-up that becomes obvious when comparing a peat fire and a forest fire. But even in satellite images, you will see differences. Some fires smoke a lot, others produce less smoke. Sometimes the smoke is bright, which often happens with peat and other damp fuel, other times really dark. In terms of general shape, grass fires usually are more wind-driven and therefore elongated than forest fires, and prescribed fires may follow straight lines as they are observed and tended to by firefighters.
Prescribed fires (also called controlled burns) are planned fires and may serve different purposes. They can be used to maintain a landscape, revitalize or create agricultural areas, clear out fuel too close to settlements, or to create firebreaks for possible wildfires in an area, making sure fires won’t spread too much and threaten human lives or structures.
Backfires are fires that are started by firefighters to battle wildfires. Usually, backfires are started ahead of an advancing wildfire to try to consume all the fuel and stop the wildfire from advancing in that direction, basically by creating a last-minute, ad-hoc firebreak.
Agricultural fires are a type of prescribed fire (controlled burn). There are different types of agricultural fires, some are started on a static agricultural field to revitalize it and get rid of weeds, and others are part of slash-and-burn agriculture. In this type of agriculture, a plot of the forest is cut down, burned, and then used as an agricultural area until the yields drop because of nutrient depletion. When this happens another plot gets cut down and burned to serve as agricultural area and the old plot is left to recover to again be used as agricultural field or as pasture for cattle or other livestock.
This type of agriculture can be the reason for massive deforestation in parts of the world, not only because more and more forest gets slashed, but also because every agricultural burn bears the risk of turning into an uncontrolled wildfire, burning even more forest than actually needed for the agricultural plot. Techniques such as reduced impact logging can help to mitigate some, but not all, of these detrimental effects.
Now let us look at wildfiresImportantThis part of the post will explain how I find and process wildfire (and other satellite) images, something I have been asked again and again to explain. Keep in mind that it will only explain a part of the process, as processing images, especially post-processing them, is different each time and a lot of the things I do in Photoshop would need their own separate post to explain them. However, it gives you a good starting point from which you can explore more on your own. Sometimes I use tools other than those mentioned here, and so should you. There are many ways to do it.
So you want to look at wildfires? Well, then your first challenge is to find them, you may have a general idea (or maybe even know an exact location from a news report), but you still have to pinpoint it.
The first thing you have to do to image fires is to find them. Fortunately, this has become easier with advances in technology and the availability of data on the Internet. Several satellites can be used to look for fires on Earth. To quickly get results I would recommend you look at the MODIS and VIIRS fire products. You can easily get a daily map of detected fires here on NASA Worldview. Once you have picked the fires that interest you, you can start looking for actual imagery.
Aqua/Terra MODIS and Suomi NPP VIIRS
Using NASA Worldview you can get natural color and false color images (which better highlight burn scars) taken by the Aqua and Terra satellites as well as by the Suomi NPP satellite. The spatial resolution isn’t that great but it will give you a first impression. Those images are also well-suited for animations showing the spread of a fire and to show the extent of smoke plumes as they image the whole Earth daily.
I recommend that you download the images as GeoTIFF files. With a GeoTIFF, you can reproject the images to the projection of your choice using GDAL or basically any GIS application. This is not always necessary but can sometimes help your audience to better be able to get a feeling for the location and size of a fire.
Landsat 7/8Note You will need to set up an account and login in order to acquire data, which is free in both cases.
With a spatial resolution of 15 or 30 meters (depending on the bands used), it can be used to create nice wildfire images and even delineate the burn scar with some accuracy. In the EO Browser, you will immediately see whether your data is cloud-free in your area of interest. If you are using EarthExplorer use the preview so you don’t spend time downloading unnecessary cloudy data.
You can download Sentinel-2 data through the EO Browser or the USGS EarthExplorer website as well. I would recommend the EO Browser if you don’t intend to process the images completely manually; It will allow you to get predefined band combinations or run custom scripts. The predefined combinations are really easy to use and will allow even beginners to immediately get decent wildfire images. With a resolution of 10, 20, or 60 meters (depending on the bands used), Sentinel-2 data can be used to get a somewhat accurate delineation of the burn scar.
I recommend you get Sentinel-3 data through the EO Browser because it offers data from the Ocean and Land Colour Instrument (OLCI) and Sea and Land Surface Temperature Radiometer (SLSTR) instruments. With OLCI you can create natural color images of big areas, showing bigger fires or smoke plumes. False color images, better showing burn scars, are equally possible. You will also be able to get daily images here. With SLSTR data you can get another false color view and brightness temperatures in different bands, which can be used to detect hot spots.
There are of course other satellite systems you might use to show wildfires. A lot of them are commercial and offer a higher spatial resolution. Commercial data can be quite expensive, other sources, however, can be free. I have used free data by the Indian Resourcesat-2 satellite before. If you want to give that a try, it is available on the USGS EarthExplorer website.
Sentinel-5 data can also be of interest when creating wildfire images, especially when looking at the atmosüheric effects of fires. With data from the TROPOspheric Monitoring Instrument (TROPOMI), you can visualize nitrogen dioxide and carbon monoxide data (and more), which could be interesting with wildfires. For this, you can use the EO Browser as well, which allows you to preview and visualize that data prior to taking the time to download it.
Creating wildfire satellite images
Processing satellite images can be done in many different ways. Begin to experiment and don’t shy away from trying new things. Don’t focus too much on one single tool, and be open to give others a shot. I’ve had, different levels of, success with a number of tools, but will focus mainly on the EO Browser as a starting point and post-processing as the second step for the rest of this post. However, for those of you who process completely manually, most of the things I write about post-processing are the same.
Using the EO Browser
Using the EO Browser might be the easiest way to get images. By using its default settings and, if you’re feeling adventurous, even the browser’s custom scripts it will get you some really good results in a very short time.
First, find a wildfire and try the simple preset natural color visualization. Try to find an L2A image, which I recommend because it is already atmospherically corrected. But beware, in scenes with wildfires, this correction sometimes produces results that just don’t look intuitive. If the L2A looks odd to you don’t hesitate to switch back to the L1C image. Now go to the settings and play with the Gain, Gamma, and RGB values. Tweak it until you like the result.
Next thing you should try is to switch the visualization to False Color (Urban) which is based on bands 12, 11, and 4. This way the shortwave infrared (SWIR) bands (12, 11) will pick up hot spots and also penetrate smoke to some degree. The result should be an image that shows you the hot spots as well as the burn scar of the wildfire.Important Understand, that what you are seeing are not the fires directly. While the hot spots, detected and visualized through the shortwave infrared bands, usually align well with active fires, you still only see infrared emissions, not fires.
The resulting image can often be used right from the EO Browser without any post-processing, although you, of course, can post-process it as well.
Another typical visualization for wildfires is the False Color (8, 4, 3) combination, but I believe most people will prefer the results of the (12, 11, 4) combination as it shows hot spots nicely.
Aside from the predefined combinations in the EO Browser, you can also use custom scripts. I wrote a script for wildfire visualization a while back, and it allows you to visualize the wildfire in many different ways. You can read a little bit about it here in this blog post. Not only can you superimpose burn scars and severity on natural color images, but you can also create images just showing the fire perimeter which you can then use in the GIS application of your choice. Read the blog post and don’t be cautious; I encourage you to play with the variables and see what results you can get. The script can actually be used for Sentinel-2 image visualization in general, not just wildfires. Use the script to prepare an image – or even several – which you can then, in a second step, post-process and refine with other applications.
Tricks for better results
Before you read this, you should know that for most of it you will need at least a little bit of knowledge in some kind of graphics program such as GIMP or Photoshop. Don’t let this stop you. You can get useful images without these tricks and you can always start to get some experience in a graphics program.
When dealing with wildfire images you will often have scenes with high dynamics, meaning if you correctly set the variables to show the fire and the ground, the smoke or clouds may be too bright which results in white clipping (the clouds or smoke partly being completely white). Here you lose detail in all bright parts. The same is true if you set variables so clouds and smoke look just right. Now your ground might be too dark and parts of it might by clipping the other way around, black clipping (parts showing completely black), again resulting in loss of detail.
My advice to you is to get two images. One with variables set so you get an image that has no white clipping at all (it will look quite dark, that is okay!). Then get another image in which the ground looks as close to what you would want it to look like and then go a bit brighter even (clouds and smoke in this image might clip a lot, doesn’t matter).
Another thing to keep in mind, if you are not interested in the clouds or smoke then white clipping with details lost in clouds is always better then black clipping with details lost in the land area. In other words, if you just quickly process an image and the clouds are of no interest, it doesn’t matter as much if they are completely clipped. Aesthetically, a non-clipped version might look better, but the clipped version will still show what you want to highlight.
In the graphics program of your choice, you can now combine the bright and the dark image to get an image that shows all detail in the dark parts, as well as in the bright parts. How to do this is out of the scope of this article, but you can find many very easy-to-follow instructional videos, showing different ways to combine images, for GIMP as well as Photoshop on YouTube.
You can also use your graphics program to manipulate the colors of the images. With some experience, you will be able to reduce atmospheric effects, which is especially needed when you have used L1C data. You can change the image to better match the expectations of your audience. You can get rid of the bluish tint caused by atmospheric scattering of light. Forests look way darker from above than many people would believe. Brighter and more saturated greens can help to make the image look more “natural” for many people. A little more contrast, brightness, and saturation can help to make dull images look more vivid.
How far you go with your alterations of images really is up to you. Some changes are more useful than others. I would encourage you to read my blog post “Satellite Image Guide for Journalists and Media” to get some more information on that.
Aside from this, even if you only have one image to work with and not two or more, you can use the graphics program to alter brightness, contrast, saturation, levels, color balance, and so much more. Make use of it.
Sharpening can be helpful sometimes, but don’t overdo it or images will look artificial and only gain more and more artifacts. Sharpening in multiple steps can yield better results.
Noise reduction can be useful if you know what you are doing. However, if you reduce it too much you may lose valuable details and end up with a very plastic-like look. Sometimes this may be wanted, but usually, it will just make your image look odd to the audience. Should you have a very noisy image, try to reduce the noise in small steps. Depending on what program/add-on you use, and its abilities, try to build a noise profile for the most affected area and reduce that noise in a targeted manner, instead of plainly reducing all the noise in the image.
You can also use your graphics program to create mixed visualizations. For example, if you have a natural color and a NIR/SWIR image, you could combine those and create a new visualization that might look interesting. In fact, the image at the very beginning of this article is a combination of an enhanced natural color and a NIR/SWIR/VIS image.
This is all quite basic, depending on how much experience you have with GIMP or Photoshop you can do a lot more. You can combine bands in different ways, for example subtracting one from another, use more than three bands for an image by using a combination of two (or more) for one (or more) of the color channels, and use layers to treat different parts of the images differently. This can sometimes be helpful to process the water part of an image different from the land part. In this case, you can enhance the contrast on the water part way more and maybe get some structure and show differences without the land part looking all wrong because of too much contrast. The same can be true for the smoke in a wildfire image, sometimes it can be useful to create an extra layer for the smoke. This way you can prevent your smoke from becoming too bright in your image while still be able to get the ground bright enough to look normal and show sufficient details.
If you are using Photoshop you should give the “Camera Raw Filter” a try. It can help you to quickly change the look of your image to something much more clear and pleasing. It takes some getting used to, but it can be really helpful to get the image’s look to where you would like it.
Manually processing images
Manually processing images is another way to get nice visualizations. It has the advantage that from the beginning you are working in your graphics program which will help you prevent things like clipping (and while 8 bits/channel processing is usually okay to get nice results, consider doing all your processing in 16 bits/channel for even better results, especially in scenes with very dark and bright areas).
Otherwise, the process is straight-forward. You open the raw images of the bands you would like to use and combine them into a single image. In the case of a natural color image with Sentinel-2 that would be a 4-3-2 combination. You pretty much do manually a lot of the stuff the EO Browser could do for you, which will take you much longer but will also give you the freedom to do some things you can’t do in there or things you just don’t know how to transfer to working EO Browser code. You are then free to do all the stuff I talked about in post-processing.
Using different tools
Sometimes it can be helpful to create 3d images to visualize the data. Among other tools, QGIS and the Qgis2threejs plugin can be used to create visualizations such as the ones that can be seen here. They can also be used to create simple animations.
Elevation data can be downloaded from multiple sources, including the already mentioned USGS EarthExplorer website which offers the ASTER GDEM dataset, which can then be used in QGIS.
Animations as the one shown below are especially interesting in mountainous areas as they help to get a better feeling for the actual location of fires (or here lava-flows) than a two-dimensional image. They can, however, be used in any kind of image to help the audience to get a better idea of an area.
Annotations – The little extra
Annotations can be helpful to explain the image to your audience. They are not always needed but sometimes can be very useful. Even if you have never used a graphics program like GIMP or Photoshop, you should not shy away from trying it, if you believe your image would profit from annotations. Just adding basic annotations is easy and you will be able to do so in minutes after starting to play around with GIMP or Photoshop.
If you aren’t processing an image with a usage case already in mind, you might also want to think about producing two versions of your image – one with and one without annotations. That gives you, or whoever might be using your image, the flexibility to decide how to use them.
Sometimes annotations can be placed directly in your image, maybe with some arrows to look extra cool and direct the viewers to the part of the image you want to highlight, and other times annotations are just a simple line of text at the top or bottom of the image saying what can be seen and when it was taken.
Don’t just look at the fire
Instead of just looking at the wildfire make sure to also look at what else can be seen in the vicinity. Are there settlements or even cities? Maybe mark them in the image.
And even close to the fire you might see things you would want to annotate or explain in an image. This could be a visible firebreak (a stripe of vegetation cleared by firefighters to try to stop a fire from spreading in that direction), or maybe lines of fire retardant. Fire retardant dropped from planes or helicopters is usually dyed red and can, depending on your band combination and spatial resolution, be visible in satellite images. You may even be able to predict how likely different types of vegetation near an active fire are likely to burn by looking at how water-deprived they are using the Normalized Difference Water Index (NDWI). Read more on this here.
For some cool real-life images and videos of fire retardant drops you should follow Global SuperTanker on Twitter; the account shares videos and images of the Spirit of John Muir, a Boeing 747-400, the World’s biggest Very Large Air Tanker (VLAT).
Feedback / Questions
In case you have general questions about the EO Browser you should read this FAQ and register at the Sentinel Hub Forum here to ask questions or maybe contribute your own scripts for the EO Browser. You can find all scripts in this Custom Script Repository, and a very good tutorial for custom scripting here.
Sentinel Hub Custom Script Contest
I’d also like to guide your attention to the second round of the Sentinel Hub Custom Script Contest, which will be starting on October 15th and once more will give you the opportunity to win attractive prizes with your script for the EO Browser.
Acknowledgments and some links
You should follow CopernicusEMS the Copernicus Emergency Management Service to get the latest wildfire information and CopernicusECMWF the European Centre for Medium-Range Weather Forecasts for information about the atmospheric effects of wildfires.
And while I possibly could recommend at least dozens of accounts to follow on Twitter regarding fires, I will just give you a few that always make me learn some new stuff and are interesting to read. Without big explanations, you should check out Jessica McCarty, Mark Parrington, Prof. Guillermo Rein, and Dr. Thomas Smith.
I’d like to especially thank the wonderful Krista West, who was so kind as to read a draft of this to make sure my over-caffeinated brain didn’t go bonkers and provided welcome feedback, ideas, and corrections. Krista has been part of the remote sensing world for over a decade now, and is currently embarking on a new adventure – she is returning to school to pursue her Ph.D., focusing on remote sensing of wildland fire. A while ago she wrote about imaging wildfires as well, read her blog post here.
I cannot stress enough how useful Twitter has become to quickly get answers from experts in different fields and to contact journalists. If you are interested in science communication and outreach, you really should be using Twitter.
Thanks to Sentinel Hub for their continuous support and excellent service.
Thank you also to all my followers for retweeting my images and many journalists for using them in their reporting.