An overview of forest remote sensing technologies

Today I am going to try to do the impossible. I will attempt to concisely describe every type of remote sensing technology that can be used to measure the forest.??Dun-dun-DUUUUNN.

Primarily, we have five technologies: Imaging, Photogrammetry, Radar, Hyperspectral, and LiDAR. But before we dive into what those are, there are a few caveats to consider. First, these technologies behave completely differently at different scales. Imagine analyzing a drone image of a tree, where there may be hundreds of pixels representing that tree, versus a satellite image where each pixel is 30 meters wide and there are half a dozen trees inside.?Each of those images is very different and gives us vastly different information.???These technologies also vary in terms of cost, availability, and cumbersomness - the best data is expensive and rare, and the densest information may a terabyte of data or more per hectare.

All that said, let’s dive in:

Imaging –Basically, this describes taking pictures of the Earth. This can be done at any resolution, and when I am working with imagery, I categorize it into low resolution imagery with pixels larger than 50 meters, medium resolution for pixels ranging from 5-50 meters, and high resolution for anything with a pixel size less than 1 meter. These categories are important because at low and medium resolution imagery we can infer things about the trees just by looking at pixel values, whereas high resolution images need to be clumped together into multi-pixel objects.?The resulting analyses are totally different, and often more challenging for high resolution imagery.

One thing to note about imaging is that current technologies can measure many more wavelengths of light than the human eye can detect. When people look at an object, all they need to see is three colors of light (red, green, and blue), but there are an infinite number of subtle colors in the electromagnetic spectrum that can be imaged; some of the most useful wavelengths of light for measuring forests are in the shortwave infrared spectrum.

Sentinel-2 and Landsat are satellite constellations that are the most common and abundant source of medium resolution imagery out there these days. Their images are collected by governments worldwide, are freely available to the public, and are fairly easy to work with. Most people access this data on a platform called Google Earth Engine (which is not the same as Google Earth). Both of these satellite groups measure shortwave infrared and are excellent at describing forest canopy cover and history. Landsat has been operational since 1984, so it can tell us a lot about what’s happened to forests in the past.

But Imaging can’t work alone – it doesn’t provide enough information to measure forest carbon or create forest inventories. The reason for this is that these sensors saturate as forests get larger – for example, they aren’t able to tell the difference between forests with a lot of wood, and forests with a medium amount of wood. ?At a certain point in a forest’s lifecycle, their canopy color just becomes fixed, and a camera can’t tell the difference. ?

Finally, low resolution imaging might sound lame, but because we can image the entire planet every day it can be very useful for tracking wildfires, monitoring seasons, and observing climate trends. The most common low resolution sensor used these days is called MODIS, and it is also available in Google Earth Engine.

Photogrammetry – This is the reconstruction of 2D images into a 3D rendering using complicated trigonometry. We can take high resolution images from drones or planes and calculate the height of objects based on their offset in each picture. It really is magic!

Since it can measure tree height and structure, Photogrammetry can be used to estimate forest carbon - but it comes with an important caveat. If you can’t see the ground, you don’t know how high your trees are. This is a major problem, because if we’re off in estimating tree height by only a couple of meters, then our carbon numbers are going to be very wrong. So photogrammetry only really works when you have an open canopy forest, or when you pair it with another ground-sensing technology. Honestly, this is what has prevented it from being adopted for large scale use.

The other issue with photogrammetry is simply that drones run out of batteries! Nobody is going out there and surveying a 50,000 hectare REDD+ project with drones. You can get this kind of data from planes, but if you’re going to be flying a plane, wouldn’t you rather be capturing LiDAR? (You would).

The last thing to note about photogrammetry is that eventually we’ll be able to do this from space, which will bebe game changing because of the scale-up issue.?There are some commercial spaceborne imagery sources (Maxar, Airbus, Planet), but the technology is still in its infancy, and in my opinion spaceborne photogrammetry isn’t sufficient to reliably estimate forest carbon. Maxar is offering a 50 cm height map which might work… but it’s crazy expensive. I think they should offer discounts for forest carbon projects!

Radar – There’s a lot that can be said about radar. Firstly, not all radar is equal. The use-cases for radar depend a lot on the frequency of radio waves being shot out. Shorter wavelengths (C and X-bands) essentially bounce off the forest canopy; this is still okay for deforestation monitoring, but isn’t useful for forest carbon.

Longer wavelength radar (L and P-bands) can penetrate into the forest and bounce back to tell us about forest structure. This type of radar can be used to estimate forest carbon, although L-band radar doesn’t quite do the trick in large forests. Sadly, there are no P-band radar satellites in orbit right now. But in 2023 the European Space Agency is launching one that should be useful for making forest inventory maps.

The Sentinel-1 satellite constellation uses C-band radar to image the Earth about once a week, and can be accessed through Google Earth Engine to map deforestation.

Finally, I want to mention a very new technology, which is high resolution radar. There are a handful of private companies selling X-band radar from space with resolutions up to 25 cm. Like I said before, X-band radar pretty much just bounces off the forest canopy. But what makes this special is that the resolution is so high, that we can actually make out the crowns of individual trees. I hope that eventually someone does some deep learning work with this to model forest carbon.

Hyperspectral – Next, we come to the most expensive technology on this list, but also the coolest. Hyperspectral imagery differs from regular imagery in that instead of imaging about 10 colors of light, it can image up to 1000 different colors!

Now, trees may all look green to us, but each tree species has a slightly different shade of light that can be distinguished with a hyperspectral camera. So hyperspectral imagery is the only technology that can tell us tree species and biodiversity.

The other really cool feature of hyperspectral imagery is that it’s able to distinguish just about every ecological characteristic about the forest that you would want to know. Each and every molecule emits a slightly different set of colors, and so ecologists and physiologists can actually use this to measure chlorophyll content, nitrogen content, phosphorus, lignin, carbon, and a dozen more trace elements. This is invaluable information for determining tree health and ecosystem functional types.

Most importantly though, hyperspectral comes with some major drawbacks. Firstly, it’s wildly expensive. The government isn’t acquiring this data over many places and forest ecology is sadly not high on many politician’s priorities, what with half of them trying to ruin democracy.

Working with this data is also extremely difficult. For each and every pixel in an image, there may be over 1000 data points/numbers associated with it. This means enormous file sizes, georectification issues, and in general it’s just a monstrous machine learning challenge. In a data science sense, hyperspectral imagery is still the wild west, since nobody to my satisfaction has really been able to completely scrape all the available information out of these images.

Lastly, I need to mention that not all hyperspectral data is equal. There’s full hyperspectral and partial hyperspectral. Partial hyperspectral imagery doesn’t include shortwave infrared bands, and so is fairly useless for everything that I just listed above. Full hyperspectral imagery is much less common and much more valuable, however the government does provide it over some experimental forests. There are a few hyperspectral satellite provides (Satellogic and DESIS/Teledyne), but they’re only providing partial hyperspectral data, so think carefully before you buy. The only full hyperspectral satellite was Hyperion, and sadly it was deorbited a few years ago. Nevertheless, Hyperion data is still out there and public, and could be very useful for forest ecologists.

LiDAR – At last, we come to LiDAR, the undisputed King of forest remote sensing. LiDAR works by shooting a laser out of an aircraft (or satellite), and measuring the time that it takes for that light to bounce off an object and come back to measure the distance. Doing that over and over again allows us to construct 3D scenes of the forest, like photogrammetry – but in this case the laser can penetrate through the forest down to the ground.

Because LiDAR gives us a full 3D rendering of forest structure, we can use it to estimate carbon, tree number, basal area, wood value, and just about anything else that you’d want to know about the forest except for species. What’s more, most developed nations have flown LiDAR over their land and a lot of this data is free and available, but likely a few years out of date.?

When you’re not relying on free LiDAR though, it tends to be fairly expensive (about the same cost as a traditional ground-based forest inventory). Nevertheless, every major timber company has their own LiDAR inventory program, so it’s pretty well accepted by the community at this point. Sadly, forest carbon registry bodies still don’t allow LiDAR to be used for verification. Partially because it’s difficult to get a grasp on carbon uncertainty, but mostly because they’re a bunch of old farts with little practical forestry experience and no remote sensing expertise. They sit in downtown Los Angeles or San Francisco skyscrapers where there are no forests and they make pronouncements abo... you know, this is going off the rails.

The last thing to know about LiDAR is that while there is spaceborne LiDAR, it is a lot different than its airborne counterpart. Instead of shooting tons of little lasers out, spaceborne LiDAR shoots out a handful of lasers that end up being about 25 meters wide when they hit the ground. They’re not able to provide 3D renderings of the forest, but they instead provide a set of samples of forest height along the transect where the laser happened to hit the ground. The most important LiDAR satellite (module) in orbit right now is called GEDI, and it can certainly be useful for estimating the forest carbon of whole countries, but hasn’t quite been demonstrated to work for small scales (like individual forest carbon projects).

Conclusion- So that’s a rapid rundown of the most important remote sensing technologies. Let’s review which technologies can solve which problems:

(1)??Forest carbon and inventories

a.?????LiDAR for sure.

b.?????Some radar (P-Band and some L-Band).

c.?????Some photogrammetry (under the right circumstances)

(2)??Forest history

a.?????Moderate resolution imagery (Landsat).

(3)??Disturbance and deforestation detection

a.?????Moderate and high resolution imagery.

b.?????Radar (all bands, but especially the cheap and abundant X and C-bands).

(4)??Tree Species and biodiversity

a.??????Hyperspectral imagery

(5)??Forest health and ecology

a.?????Hyperspectral imagery

What are some of the most exciting upcoming technologies?

·???????The ESA’s P-band biomass satellite

·???????Photogrammetry from space (cheaply)

·???????High resolution radar from space

Finally, we have to remember that not all remote sensing technologies are suitable for everything. Despite this you will find scientists using every technology for everything, mostly because they can’t afford the expensive technologies, and have to publish something. It takes a detailed scientific eye to read an academic paper and read between the lines to determine if the map is any good. You’ll find no shortage of folks modeling biomass using Landsat or Sentinel, even though those maps are only good “at scale.” Be careful before you commit time and money to working with data, and consult a down-to-earth remote sensing scientist rather than going it alone. Best of luck and reach out to me if you have any questions!