First Results from 2018 Botswana Fodar of Elephant Habitats2019-03-162019-03-19https://fairbanksfodar.com/wp-content/uploads/2014/09/fodar_logo4mn.pngFairbanks Fodarhttps://fairbanksfodar.com/wp-content/uploads/2019/03/img_9877_acr-3.jpg200px200px
We spent most of November 2018 in Botswana studying elephant habitats using fodar, expanding on our work in 2017. This blog summarizes our 2018 mapping work from acquisitions to data validation, as well as some samples of the analyses we hope to accomplish with it. In addition to making maps, we also learned a lot about the local ecology, some of which Turner shared here.
Here is a 3D visualization of the fodar data we acquired in 2018, showing vegetation alongside a river leading into the Linyanti swamp. Note that you can not only determine tree size and shape but distinguish tree species, even if you are not a tree expert. Note also that not all of the trees have leaves yet, the rainy season was just beginning during our stay.
Here is a 3D visualization of the 2018 fodar data alongside the Linyanti swamp. Here you can not only see and measure the variety of woody vegetation that grows here, but also the extent, size and species of swamp grasses and floating vegetation. Anything on the landscape becomes part of our maps, including elephant trails through the swamps or 4WD truck trails over the ground.
Here is a 3D visualization of the 2018 fodar data flying us from the Linyanti swamp into the Savuti Channel, a fascinating drainage feature with millions of years of interesting tectonic-hydrologic history to explore. Note how well the water itself is resolved due to the floating vegetation, such that we can easily measure water levels and gradients.
Our work is focused on understanding the impacts of elephant browse on woody vegetation. Woody vegetation here basically means trees and things that would like to be trees except that elephants keep eating them or knocking them over. Foraging elephants exert a primary control on the distribution of woody vegetation and grasslands throughout the savannas of Africa, and have for millions of years. The mix of woodlands and grasslands, in turn, controls the distribution of related plant communities and the creatures that live there. So elephants exert a large control on the overall and local biodiversity by knocking over trees, munching on branches, and pulling up seedlings.
Here is a video from a documentary (The Great Rift: Africa’s Wild Heart, Episode 3) I recorded using my phone that nicely describes the impact elephants can have on woody vegetation. Episode 3 is a great way to learn more about the influence of elephants on biodiversity of the savanna.
Here is some video I shot with my phone showing a small herd of elephant knocking down trees, eating grass and seedlings, and strips leaves from branches. One thing I found interesting is that after knocking off large limbs, they didn’t spend much time with newly-accessible leaves. Our campsite is just behind the tree they knocked over; we had encounters this close probably 5-10 times per day on our game drives, giving us lots of opportunities to study their behavior.
Our hope is to get a better handle on the spatial patterns and temporal trends of the impacts elephants have on trees so we can track those impacts over time. For example, field studies have shown that woody vegetation gets taller the further you get from the major swamps (the Okavango and Linyanti in this case), presumably because the elephant populations concentrate around those swamps so have a greater impact on those trees closest to the swamps. So we hope to document whether this trend exists and can be monitored over large areas using fodar. As another example, speaking as a glaciologist pretending to be an ecologist, it doesn’t seem clear if anyone really understands how many elephants can live in Botswana before the woodland-grasslands balance they maintained for millenia changes due to over- or under-population of elephants, with a corresponding ripple effect on biodiversity. I think in Botswana a big long-term question is whether populations are increasing due to the elephants figuring out that Botswana is safer from poaching than in several of the surrounding countries. Trying to count them from the air is one method of figuring this out, but perhaps mapping the changes in forests is another; that’s something we will hopefully find out soon.
Our work this year was focused on several transects emanating from or connecting the Linyanti and Okavango swamps. These swamps persist throughout the year, but swell considerably when the floodwaters from Angola arrive starting in early winter there. I had mapped two these transects last year as a demontration, as well as some areas near Maun, so this year we were able to repeat all of those and then some.
Acquisitions
Turner and I arrived in Maun on Saturday November 4th after a several day journey. The major stress in packing was trying to ensure that if any one bag got lost I could build a system with whatever remained. I took the essentials of one complete system with me as carry-on, which was fine except the airlines wanted to limit us to only 8 kg and gave us some grief about the lithium ion batteries, so we had to get creative on check in. But in the end all went well, everything arrived, and before long we were saying hello again to our hosts at the same hotel we stayed at last year.
The next morning we were up early to begin our flight testing. It’s one thing for all the gear to arrive, it’s another to make sure it’s still all functional. The owner of Helicopter Horizons, Andrew Baker, picked us up at the hotel and successfully navigated us through the airport security with all of our gear. It is such a pleasure to work with such a great company. Maun is the busiest airport in southern Africa — there are many dozens of caravans, 206s, and airvans parked on the ramp, as well as Andrew’s 10+ R44 and JetRangers, and they seem to all be busy each day. Andrew has been on the scene here for decades and seems to always be excited about something new and useful to do with his helicopters, so despite them being solidly booked he has been at great at finding a way to fit us into their schedule.
We spent our first full day in Botswana setting up in the hangar, testing the system, and working with our new pilot Joe to learn the ropes of flying in a straight line at a constant altitude rather than hot-dogging for tourists above the wildlife. I’m being a bit glib, but there is a major difference in these flying styles and as a pilot myself I know it’s not easy to master flying straight and level within a wingspan. But it was not long before Joe had mastered it and that night I confirmed everything worked as planned, so we were ready for some real work!
How that real work was going to occur was still in something of a gray area. Plans were developing rapidly and changing daily even after our arrival, but the general plan was that we would base in the operations camp of the Great Plains conservation company near Selinda, dubbed CSU camp, as it was central to most of our transects and would greatly reduce commute time. Great Plains appears to be one of the most active and progressive conservation groups in Botswana and southern Africa, in addition to running a series of top of the line safari lodging sites. So we spent the first half of the day at the Okavango Research Institute with the lead scientist of the project Dr Richard Fynn, giving a talk and meeting folks, then spent the second half of the day dealing with logistics like buying camp food, organizing vehicles, sorting and packing, etc. The general idea was that Turner and I would fly to CSU with Joe in the R44 while Richard and his wife made the 8-12 hour treacherous drive in a University vehicle, and we should meet up there about noon.
That plan worked out reasonably well. Turner and I met up with Joe early in the morning, rigged and tested the system, smushed all of our other gear inside, flew about an hour over the scenic landscape, tossed out the gear into a waiting truck, reorganized a bit, and were mapping by about 9AM. Partly by design, partly by luck, each of our mapping blocks was just about the ideal size and distance to complete comfortably on one tank of gas. By the time returned from our first block, Richard and Theresa had arrived, and Richard joined us for the second block. So basically within 72 hours of arrival we had almost a third of work complete!
The next two days were a blur of similar activity. I was up before dawn getting the day’s missions organized and up late into the night processing data to ensure what we collected worked as well as we hoped. Turner and I just snacked for breakfast and lunch, and usually in the evenings we ate together if not too exhausted from the work, jet lag, and heat. The lack of internet and power in the late evening made work challenging, but in the end fortunately there were no major issues to deal with and all worked out fine. We even finished a day earlier than planned!
Turner takes his turn pumping fuel into the helicopter.
On our final day based out of CSU camp, we did some bootleg ground control by reconfiguring the airborne system into one of the safari vehicles. This worked out great. And soon enough we were back in Maun, scrambling to transition into our next adventures in Botswana.
And apparently giraffes don’t like landing helicopters…
My general idea in planning this trip was spending the first week or so doing the airborne mapping and the next two weeks learning ecology on the ground and doing some field work based out of safari lodges in or near our study areas, with one week open at the end of the trip. In this way, if any of our system gear got lost or stolen or broken or the logistics fell through in some way, we’d have roughly two weeks to sort that out and could try again at the end of the trip. Fortunately that wasn’t necessary, so we were able to get a lot more learning experiences in on the ground over the next three weeks, as well as becoming increasingly sophisticated with our ground control methods using a safari vehicle. The rest of this blog discusses the data itself, how we validated it, and some sample analyses.
The Data
All of the transect data shared the same specifications: – Photos were acquired at 10 cm GSD and orthomosaics posted at 10 cm – Point clouds had an average point density of ~25 points per meters squared – Digital Elevation Models were posted at 20 cm – Precision/repeatability is ~15 cm – Geolocational accuracy is ~1-2 pixels horizontally and ~30 cm vertically (though see validation notes below regarding vertical) without ground control – No ground control was used in processing these airborne data, only in validating it – Data delivered in WGS84 UTM34S, no geoid model applied.
In this section I try to give some flavor of the data, including a top view of the orthoimage and DEM for each block as well as a video or two flying over the point cloud. They don’t really do the data justice, but internet sharing tools for these type of data just haven’t caught up to the acquisition technology yet so best I can do in the meantime.
Kwando1 Block
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high). This transect is about 25 km long and 2 km wide.
Here is a circular fly-over of the point cloud of these data, at the north-eastern end near the river that flows into the swamp. Note that you can clearly see the size and shape of the trees, and even if they have sprouted leaves or not. We can make measurements of them all using the right software.
Kwando2 Block
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high). This transect is about 25 km long and 2 km wide.
Here is a circular fly-over of the point cloud of these data, showing some of the fossil river channels. These channels filled with sand as they dried up millions of years ago due to subtle tectonic tilts in the landscape, and this sand now supports a vegetative community that differs substantially than the more clayey soils outside the channels. The red color is not clay, but rather leaf litter from the mopane trees that grow there; when we left in late November, the mopane were just begin to leaf out.
CH1 Block
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high). This transect is about 25 km long and 2 km wide.
Here is a 3D visualization of these data demonstrating the transitions in vegetation type cause by subtle drainage differences. We can measure both the vegetation and the topography using these data to understand these relationships better than ever before.
NG18A Block (including CSU camp)
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high). This transect is about 50 km long and 2 km wide.
Here is a 3D visualization of the point cloud of these data circling around CSU camp at the northern end of the block. The broad parallel tracks to the north of camps are one of John’s experiments to increase forage and habitat for non-woodland creatures in the area.
Spillway Block (including Selinda Spillway and Savute Channel)
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high).
Here is a 3D visualization of the point cloud of these data around the tip of the Linyanti swamp. It’s just such a cool area, I could spend hours flying around on these data or weeks flying around it for real and not be bored.
Here is a 3D visualization of the 2018 fodar data flying us from the Linyanti swamp into the Savuti Channel, a fascinating drainage feature with millions of years of interesting tectonic-hydrologic history to explore. Note how well the water itself is resolved due to the floating vegetation, such that we can easily measure water levels and gradients.
NG18B Block
At left is the fodar orthoimage, at right the elevation data colored by height (blue low, red high). This transect is about 25 km long and 2 km wide.
Here is a 3D visualization of the point cloud of these data circling around a mopane forest which has been stunted due to severe elephant browse. The leaves had yet to sprout when we mapped it, making it look lead it was all dead at first glance.
Accuracy and Precision Checks
The project itself was way underfunded so there was no budget at all for ground control or data validation, but we were able to sneak some in any way. Validation here is really mostly for blunder checking, I did a thorough validation of the 2017 Botswana data and many other validation projects here. The two basic tests I used were comparing the fodar maps to those I made in 2017 and using survey-grade GPS on the ground. Here I was mostly only able to really nail horizontal accuracy, as my field book was lost/stolen (first time in 25+ years!) within a few minutes of arriving back in Maun so I don’t have exact antenna heights to check vertical. In any case, horizontal precision/accuracy seems to be better than 1-2 pixels (10-20 cm) and vertical precision is < 20 cm at 95% as we’ve always found, and vertical accuracy is likely better than 30 cm but no worse than 2 m (due to possibly blunders); a few high quality GCPs in each location can be collected at any time to resolve any blunders.
However the major point of this field work for me was as a methods test for measuring ground control in Botswana itself, rather than validation for the airborne data which I’ve validated countless times now. Along these lines I was able to collect some survey-grade GPS data based from CSU camp and Selinda Explorer’s camp several weeks later by adapting my airborne system for walking and vehicle use, progressively getting more sophisticated. First, early in the morning before we left CSU camp to return to Maun, we mounted the aircraft antenna to a monopod and walked around camp surveying ground features that might be visible in our maps, while Turner followed behind taking photos of the features I was surveying. Next we mounted the antenna onto the hood of a safari vehicle and made continuous measurements; I had some issues with that in our drive from CSU camp (partly due to overhanging canopies), but had them sorted for the others. Then I began acquiring oblique fodar data, by mounting the camera to the truck and taking bursts of photos obliquely at the scenery such that we could make points clouds from the ground for comparison with those made in the air. Finally on our self-drive trip we simply drove through the southern end of NG18A and took geolocated photos without the survey system as a means to validate our species identification from the fodar orthomosaics, using techniques that could be done by anyone without expensive gear.
After leaving CSU camp, we spent a week or so in the Central Kalahari Game Reserve to gain a better understanding of that ecosystem and its mapping needs. When we returned from there, we spent a few days at Selinda Explorer’s Camp, run by Great Plains Conservation within our Spillway Block. It is an excellent camp with a first-rate staff, and though we were there essentially as tourist they were happy to accommodate our strange photographic requests.
Next I compare our ground tracks measured with precise GPS while driving in our safari vehicles. Here at various times throughout the day I used the airborne system as we went on game drives. The antenna was located above the left-front wheel well. I couldn’t ask for a better correspondence between the ground and the airborne data based on these tests.
Here several sets of tracks overlap on different days (color) in one or both directions. The antenna was not placed on the vehicle in the exact same location on both days, but in any case the tracks are often within a few centimeters. In driving in one direction on both days, you can see by the wide-spacing in lines we veered to avoid some obstacle, but only going in one direction.
Here is a closeup of the ground track over the topography. The vehicle tracks are incised into the ground by 10-30 cm or so. They are clearly resolved in topography, but skewed a bit upwards as they are only 20-30 cm wide, so there is a bit of spatial bias caused by the higher edges of the tracks, as the elevation of the edge and the rut get averaged together a bit. But as best I could determine it, the mean vertical misfit was only a few centimeters.
In total we made over 23,000 measurements using this driving method. I filtered these down to about 8,000 by quality statistics from the GPS processing software so that I was mostly only using ones unaffected by poor sky view in the trees. I then used the middle 95% of those to calculate the range of vertical misfit of +/-13 cm for 11,851 points, about a mean of several centimeters (discrepancies in antenna heights prevent an exact value). Considering the suspension travel on the vehicle is over +/- 30 cm, the spatial biasing caused by the narrow, incised tracks, and the fact that some percentage of these points were still under a tree canopy (as in the last image at left), this is an amazing result and totally in line with what we’ve seen in dozens of other projects. Considering the horizontal accuracy shown above is nearly perfect, I have no reason to suspect the vertical accuracy is any worse, and my best guesses at antenna heights show it to be within centimeters.
We got to know many of the pride as individuals during our time there, and spent much of our time about this close to them.
We also have overlapping and independently-made maps at several locations, which gives us another opportunity to assess accuracy. Here I show several of those analyses. The first comparisons are of the Maun airport and the Boro River not far from it. Note that these maps were made not for production but for system shakedowns, so they all have some wonkiness. For example, on our first flight the camera nearly fell off because I had forgotten to tighten the clamp sufficiently (it was backed up though, so it would not really have fallen off) and between Joe learning to hold a very tight heading while ATC was pushing us around, line spacing suffered from optimal. Despite these challenges, the comparisons were outstanding, so I expect the real data to be even better.
Here is an example of the horizontal precision between 2017 and 2018 fodar maps of the Maun aiport. The shadows can make things confusing, stick to looking at the runway markings.
I’ve heard Maun described several times as a ‘drinking town with a safari problem’. The Tandurei restaurant, shown here with the courtyard with the red plane in it, is one of the main places local pilots go to escape their problems. Note that several trees have been removed near the road; I’ll revisit these later as an example of detecting elephant browse. Note also that the differences in shadows can play tricks with your mind, so when assessing horizontal precision here focus on road stripes at first.
Here is are a few example tests of spatial alignment between 2017 and 2018 fodar maps made along the Boro River, not far from Maun. In 2017, we walked across that fresh burn scar. The grasses here have adapted to quickly regrow after fire, as their roots remain largely unaffected by burning or grazing. My suspicion is also that the spatial extent of the burn says something about the change in soils and soil moisture at the boundary and could be a useful tool in mapping such soils.
The spatial alignment is spectacular, especially considering the hokeyness of how we collected these shakedown data. Not also that image clarity improved in 2018 due some subtle changes in rigging.
Though precision is different than accuracy, in the tests above they are essentially the same. In these studies and all others, when I map the same locations multiple times, they plot in the same locations to within a pixel or two. There is no way the data can be so precise without also being that accurate. Unlike the canonical textbook model of arrows being shot at a bullseye, in the realworld we have no bullseyes so we have to use repeatibility as the test for truth, just as we do for GPS measurements on the ground. And given the close correspondence between our ground-based GPS measurements (shown previously), we have no reason to believe that horizontal accuracy here is anything other than 1-2 pixels.
The airport is also probably our best location for measuring vertical precision and accuracy, largely because we can eliminate the issues of vegetation and vegetative growth. Having determined that horizontal accuracy is only a pixel or two, we can have faith that vertical comparisons here are meaningful.
The color image shows the difference in elevation measured by fodar in 2017 and 2018, with colors stretched over +/- 50 cm. So the yellow-greens that cover most of the airport are very small differences, 10-20 cm. Presumably nothing has changed here, so the differences would be zero if my data were perfect. The large changes in color at the edges are increased noise, but this is expected as this is edge data with insufficient photo overlap– in a real project I ensure that the edge data are not inside the area of interest by acquiring extra lines.
NG18B is the only location we mapped twice in 2018. The second time we flew part of the transect at a lower elevation to test how this would improve our elephant studies (described later). The transect were processed independently so make a great comparison test in many respects, though the results are conservative given the low lines were done ad hoc and not to normal standards.
Here I discuss two methods of ground control involving ground photos using the same example. As I progressed with ground-vehicle technique development, I experimented with creating fodar point clouds from the ground to compare to the airborne one, for utility in geolocation accuracy determination, canopy structure measurement, and canopy shape technique comparisons. These same photos, as well as the ones not acquired in this way, are also useful in determining species type from the airborne data, as the ground photos resolve tree and leaf shape better.
At left I have overlain the ground-based point cloud over the airborne-based point cloud. The ground one has a lot of blue in the leaves, as I was looking up into the sky through the leaves (see photos 340 and 350 above); I can filter that blue out, but I left some of it in as I thought it improved clarity in distinguish the two point clouds. This view captures the same area as shown in those two photos. One thing to note is that these ground photos were acquired 10 days after the airborne data, and many trees turned from leafless to leafy during that time.
Here is a top down view of both point clouds, roughly matching the perspective in the previous orthoimage. Notice how many more blue (ground-based) trees appear. This is in large part due to the emergence of leaves, but also the increase in resolution, in this case 3-5 cm compared to 20 cm.
Here is a perspective view of the same data in the previous comparison. Here you can clearly see the differences leaves and resolution make, as not all of the trees have leaves yet. The point is that time of year matters and that resolution matters — the differences here have nothing to do with the quality or capabilities of the technique. Indeed, the fact that the same equipment is revealing these differences indicates that the equipment and technique are not responsible for them, it’s really just knowing how to use them to optimize results and cost. In this case, the thin diameters of stunted mopane trunks and branches are below what can be resolved at 10 cm image GSD from the air, but flying lower or driving next to them is sufficient to resolve them. Doubling the resolution (5 cm instead of 10 cm) roughly doubles the flight time and costs, but for these transects that’s only $1500 so it’s always important to keep actual costs in mind rather just percentage increases. The question then becomes does the extra $1500 in cost increase the value of the science by including leafless mopane in the elevation results? I address this further later.
I believe these tests paint a picture consistent with all of our previous studies — these data have a repeatabililty of better than +/- 20 cm at 95% and a geolocational accuracy of better than 30 cm, which can be reduced to the level of precision with quality ground control. My ground data was overwhelmingly sufficient to validate the horizontal accuracy and precision and vertical precision, but only mediocre in vertical accuracy due to some issues I had with my hokey ground system. An equally important point is that I got a great start in developing cheap methods for ground control, such that on my next trip I can implement something more efficient and reliable. These tests, as well as some below, also indicate that resolution (more properly Ground Sample Distance or GSD) and timing relative to leaf out are controlling factors in resolving the shape and structure of woody vegetation.
Sample Analyses
The main point of this blog is just to say that everything worked great on the technical side, but of course it is difficult to resist doing some sample tests to see how these data can contribute to our overarching questions about elephant habitats and dynamics, including exploring options we might consider for the future. So here are just a few quick examples I did while processing and validating the data, along with a few thoughts of what we could do next, in addition to what I have discussed already.
The goal of the project was to measure tree height as function of distance from water bodies and how it varies over time. I’ve already given a bunch of examples of measuring tree height, such as in the videos, but here is another just demonstrating that everywhere there is leafy vegetation (left, orthoimage) we are also measuring its topography (right, digital elevation model).
Sometime between our November 2017 and November 2018 visits, several mopane trees got knocked down in front of the Tandurei restaurant, one of our favorite hangouts, to limit interference with power lines. Thus it makes a convenient local site for testing tree dynamics in the wild, since I know what happened on the ground and could access it. So I’m using what happened there as an example of what elephants do in the field.
The Tandurei is the complex of buildings just below center with the red airplane in the parking lot. Just north (up) from there you will notice several trees have been removed, as well as kitty-kornered across the street to the left. So right away we know that the fodar imagery alone is capable of seeing when trees disappear.
Here is an oblique view of 3D fodar data — here the image data are overlaid onto the topography. Note how we can easily we can see the change in shape of trees (if you look closely you can see most appear to be thinner, due to less leaf canopy growth in 2018) or their elimination completely with these data (see upper left and just up from the Tandurei courtyard)
Here are those same data, this time without the image overlay. Blue is low elevation, red is high elevation. Note that Digital Elevation Models (DEMs) like these using any data (fodar, lidar, insar, etc) are incapable of resolving two elevations at once (like underneath a bridge or tree) so the tree canopies here come straight to the ground.
Trees are not the only vegetation we can measure change in. Here are the Boro River maps we made in 2017 and 2018. We took a mokoro trip here in 2017, and still remember some of the individual features and trees that we saw then. Here you can clearly see some differences in vegetation related to differences in the timing of flood dynamics, which are localized around the river.
Here is a stretch of the waterlogged part of the acquisition. The colors at right are +/- 50 cm of elevation difference between the 2017 and 2018 data. You can see how the color distribution is not random or aligned with sharp photo edges or corners — these are real variations, largely caused by marsh grasses I believe.
This is from a different location, but gives the general sense of what these swamp grasses look like.
One of the primary objectives of this study was to determine the height of trees as a function of distance from the various swamps. In the previous examples and in our 2017 study, we demonstrate that this no problem for trees with canopies. As fate would have it, during our 2018 visit most of the mopane were bare and the terminalia had just started sprouting. In 2017, I didn’t know anything about Botswana ecology and was here only to demonstrate cool things fodar could do, so I didn’t realize at that time that the bare mopane trees were even alive, so I was mostly thinking about mapping tree canopies. In 2018, the focus of my learning experience here was trees, but it was not until we were actually airborne that I learned that these leafless mopane were actually alive and a primary food source for elephants. Given what I saw on the ground about their size, I suspect we planned our flight lines too high, optimizing for spatial coverage over resolution. So on our last flight, I added a few lines at lower altitude so we could assess best methods for the future. In this case, I changed ground sampling distance of the imagery from 10 cm to 6 cm, and show some results below.
Here are views of point clouds of different resolution from NG18B. Though the difference is only 4 cm in GSD, because the leafless branches of these mopane are so thin, the lower resolution makes a big difference in capturing them digitally. If we wished, we could also digitally filter out all of the trees so that we could studying hydrology, such as the tree filtering I demonstrate in this blog.
Here is another way to understand the improvement that higher resolution makes. These are the two point clouds made from different flying heights, with red colored as high elevations and mostly indicating the location and height of trees. As you flicker between them, you will notice that there are more red dots (trees) visible in the higher resolution data, and also that larger red dots at 10 cm turn into several smaller red dots at 6 cm as clumps of trees get resolved as individual trees. You will also see a color change in many dots, as higher resolution data will resolve the height of a small spiky tree better than lower resolution, meaning that measured tree height will be a function of resolution for leafless trees. Note that this is not revealing a deficiency in fodar, as such spatial biasing issues are common to every technique, but rather simply why and how resolution matters so that the best optimizations between time, money, spatial extent, and analysis abilities can be made.
Here is an example of determining when a tree has been knocked down (near center), using its shadow (left) and its elevation (right).
Though fodar cannot measure water surface elevations accurately, it can measure the elevations of things floating in the water accurately and their elevations can be used as a proxy for water surface elevation. In the example below, I show how lily pads are a great proxy along these lines.
The Future
My analysis of the 2018 fodar data shown here is that it technically performed exactly as planned and that this opens the door to a wide variety of future work. We discovered here that the measurement of tree heights is possible from the elevation data when the proper resolution is selected and complicated by temporal variations in leafiness, but with proper care in planning these constraints can be overcome as the change is logistical not technical. In this and our 2017 work, we learned that we also measure the subtle variations in topography that control hydrology here at the process level, such that we also measure topographic change at the scale that allows us to predict changes at the process level, whether that’s making flood inundation maps using the dry topography, measuring dry pans as bathymetry to determine their volume later using only a photograph, or measuring water levels in the swamps in realtime using the floating vegetative mat as a proxy for water height then determining flood water volumes by comparing it to the dry topography as bathymetric hypsometry. Previously we also determined that we could not only identify and count large mammals in our data, but measure body morphometrics using the topography.
In 2017 we spent some time mapping animals to see how well it would work. So this video is not an actual video or something from a drone, this is a 3D visualization of our fodar data.
Here is a similar example using elephants. By mapping both the game and the landscape at this resolution monthly over huge areas, we can begin to answer much more sophisticated questions about wildlife ecology, as described below.
All of these findings to me lead up to an intriguing and potentially game-changing possibility — we could map the entire Okavango Delta in 2 days using the entire fleet of Helicopter Horizon’s ships to limit temporal change during the acquisition, then do this on a monthly basis to capture the full range of seasonal variation in ecological dynamics between landscape, hydrology, botany, and wildlife. The price to equip 10 helicopters with my fodar systems and operate them like this is far less than a single lidar unit costs to purchase. That is, flying 10 helicopters in echelon formation reduces our mission duration by a factor of 10 without increasing our flight time, allowing us to capture swaths 10x wider, effectively eliminating double-counting of moving game while simultaneously eliminating under-counting since the counting is occurring in the office rather than visually while in the air, with the bonus that we can not only count elephants and nearly all game visible at the selected resolution but also get information on their size and health, and does so operationally for less cost than simply buying one unit of the state-of-the-art alternative with no sacrifice in quality. Doing this monthly for a year allows us to asses the changes in distribution in game as they may relate to changes in flood stage, vegetative state, etc. For example, we can not only learn the seasonal impacts of elephants on woody vegetation, but we can determine how long their presence in a particular area relates to their impacts as a function of vegetation type or even individual trees. As another example, if we compare such game counts to traditional game counting methods, we have a means to assess that method and apply those finding to past or future counts using that method. But of course, these are ideas on African ecology coming from an Alaskan glaciologist, so how valuable such analyses are is beyond my expertise to say, I’m just opening the door to discussions on that by saying that these analysis are technically feasible should they be deemed valuable.
Contact Us
We're not around right now. But you can send us an email and we'll get back to you, asap.