On the Relationship Between Oil and Snow in the Arctic National Wildlife Refuge

Posted May 9, 2019

About a month ago I mapped snow cover of about 1/3 of the 1002 Area of the Arctic National Wildlife Refuge and here I present some initial processing results along with some thoughts on its implications for oil exploration. While there are many great reasons to measure and study snow pack in this area, in this blog I’m focusing on it’s value in conducting winter seismic measurements which measure the subsurface geology, which is funded primarily to locate oil reservoirs but also supports a better understand the fascinating geologic history and modern dynamics of this actively deforming region. In short, I measured a lot of snow-free ground at the peak of winter snow pack, raising important questions about the economic and environmental viability of the proposed overland seismic methods that require a suitable snow cover to protect the fragile tundra underneath from vehicle damage.

The methods proposed to conduct this seismic work are not without risk to the fragile tundra of the 1002 Area. In the oil fields to the west, the typical — but not only — method is to drive heavy vibroseis trucks (often called thumper trucks) over a grid pattern, such that the acoustic waves they send out echo off the subsurface geologic formations and then are recorded on geophones. The time it takes these echos to be recorded are then converted into depths using the speed of sound. To limit or prevent damage to the tundra, the trucks must work in winter when the tundra is frozen solid and sufficient snow cover exists which prevents direct contact with it. While most studies of the long-term impacts of these methods document enormous impacts of this method on tundra (such as this one or this summary), they are nonetheless in regular use today and apparently do not cause immediate, obvious damage when all is going as planned, though what is considered ‘damage’ seems to be poorly defined at best. The point I have been trying to make over the past year is that the 1002 Area has some important differences to the oil fields to the west — it is not flat but very lumpy and the wind regime is different here, therefore the snow pack and wetland dynamics are both qualitatively and quantitatively different and less conducive to these methods. The permafrost itself may also be different here — the signs of surficial thaw are ubiquitous, indicating that the terrain is more susceptible to disturbance than it was 30 years ago, or even 5 years ago for that matter. Thus I have been mapping it using my system in an effort to both highlight these issues and demonstrate that these tools can contribute towards responsible decision making.

The general idea behind my measurements is this: I measure the topography in summer, I measure the topography in winter, and then I subtract the two and the resulting difference in topography is snow depth. That’s an oversimplification, but the details are irrelevant to this blog. I can do this over huge areas efficiently — despite persistently poor weather in two weeks in March/April this year, I mapped well over a 1/4 of the 6000 km2 of the 1002 Area at 5″ spatial resolution. That’s an area about half the size of Rhode Island with sufficient resolution to see if you are texting while driving. I measured the entire 1002 Area last summer, so now my goal is to subtract the summer data from the winter data to determine snow depth. To my knowledge, which is reasonably good on this subject, this is the most massive snow thickness measurement ever in Alaska, and possibly the world, with some tens of billions of measurements.

Here is an example from the Arctic NWR of comparing summer and winter fodar data to measure snow depth. Here I’m showing the imagery, but underlying these data is summer and winter topography. When I subtract the two, as shown next, I get snow depth. The red dots are snow probe measurements we made to validate the data, also shown below. You can read this paper to learn more about my technique and see more examples. Note how you can see the snow depth is zero across much of this image simply by the dark tundra visible in the winter image.

At left is the same winter image, at right is the difference in summer and winter topography: snow depth (red is thick, blue is thin). The resolution is such that you can see the shape of the polygons by the extra thickness of snow in ice wedges surrounding them. Note where you see bare ground in the image, the snow depth measurement also shows zero, as expected.

Here are measurements from the same location in a data plot. At top is the topography measured along the thin black line in the previous image: the difference between the winter topography (red line) and summer topography (black line) is snow depth. At bottom is a comparison of our probe measurements (red, formed into a line following the arrows in the previous image) and airborne data (black) — they essentially match perfectly. Indeed the airborne data is superior as our probes were not long enough to measure the deeper drifts.

The value of these measurements to the oil industry is that without suitable snow cover, seismic work cannot proceed using it’s preferred methods, and without seismic measurement oil drilling becomes a much riskier prospect, like the dry hole that has already been drilled here. While it remains unclear how thick the snow needs to be to protect the tundra from overland travel by these heavy vehicles, most permits require 6″-9″ as the minimum, though in federal lands it appears there is no actual minimum: it’s a ‘performance-based standard’ done by the operators in the field… Is the snow suitably thick here? Well, that’s the $64M question — we don’t know because there has never been a spatially-distributed, multi-annual characterization of snow pack here — my measurements are the first such measurements on this scale. However, I would not being doing it on my own unless I thought there was an issue — my 16 years experience working in this area, as well as that of others, has led me to believe that typically the snow is much thinner than this and that much of the ground is often snow-free in winter due to wind scour. As I show below, that turned out to be the case again this winter.

Across the Canning River, overland seismic work was done on a tight grid pattern last winter, leaving behind the checkerboard of compressed snow seen at the top of the photo, which appeared in a New York Times article last summer. Proponents of this method claim that the only impact of these method is the temporary snow tracks seen here, and many opponents consider this snow checkerboard enough of a visual impact in the protected Arctic NWR to disqualify the use of this method there. That’s a debate of personal values. Scientifically, the unanswered question is what are these tracks doing the underlying tundra and the ecosystems it supports? That’s something my data can help answer.

Here’s a 3D visualization of my data showing one of the grid lines of the checkerboard shown in the previous photo, demonstrating that the tracks left by their large seismic vehicles are visible months after snow melt and cause a topographic depression. I could trace nearly every one of the tracks I measured like this. The yellow lines are an approximation of the grid pattern used. You can download these data using a link somewhere in this blog. Do you consider these tracks ‘damage’ to the tundra? One of the qualms I have about the rules is that ‘damage’ is not defined.

Here is the 1002 Area of the Arctic NWR, with a 200 m x 200 m grid overlain on top of it. Note the grid spacing is so tight that the yellow lines blend together and create moire patterns. I show more examples below, as well as a way to download these data. This is the proposed density of overland travel to complete the seismic work here — several dozen heavy vehicles traveling in the dead of dark winter dragging a hotel for 300 people. Seismic work done here in the 1980s when this area gained protected status used similar methods, and some of these tracks are still visible today; you can read more about those impacts here.

One issue with such dense lines traveled by heavy vehicles is that the probability of crushing polar bear dens increases. *Here is a graphic* *from a website on polar bear denning* that gives the general idea of how polar bears use snow drifts for their dens. Experienced polar bear scientists walking above these dens attempting to find them have collapsed through their roof — imagine what a convoy of 5 ton thumper trucks would do. You can learn more about the threats to polar bears from this method of seismic work here.

The Data

I’ve only processed about a quarter of what I mapped out there, but it was so exciting to me that I wanted to share it now because I think the stakeholders involved might appreciate having this information long before next winter, to prevent surprises and additional expense and hassle, as it will likely take a few more months to complete all of the processing in my spare time.

In short, I found a lot of bare ground, or barely covered ground. That is, I don’t need to do my full winter-minus-summer processing tricks to determine snow depth — I know that snow depth is zero across massive areas of this landscape because I can see the bare ground! The snow depth is zero not because it doesn’t snow here, but because the pervasive strong winds coupled with the irregular topography and lack of stabilizing vegetation simply scours the snow away. This wind blown snow either redeposits to form the deep drifts that polar bears build dens in or it simply sublimates into the atmosphere and gets carried away.

I planned my acquisitions to cover as many different ecological landscape zones as possible, denoted here by color. I think I did pretty well, though I would have liked to do all of it. In this blog, I’m focusing on the data in the large block to the left of center, between Camden Bay and the Sadlerochit Mountains. Though I have no data yet to support this, from experience I believe this area to be the most prone to having zero or minimal snow depth every winter, as the topographic is lumpiest here and the winds the strongest. There is a high concentration of polar bear dens here too, so I thought this was a good area to focus my efforts.

*Here is the block I’ve processed, laid over some landsat data inside the 1002 Area outline.*

At right is my 2019 snow data and at left is summer landsat data from some random year. Soon I will have my summer 2018 topograhic data processed, such that I can subtract it from my 2019 winter topography so that I can directly measure snow depth. The striping you see in the snow image are my individual flight lines. The over 9000 photos used here were acquired over several days and the sunlight and cloud cover were constantly changing. Plus I cranked up the contrast so that humans could see the snow features better (the computer doesn’t care), and that caused some vignetting at the edges of the photos, enhancing the striped look; the stripes do not affect the topographic data derived from the photos.

Here is the topography and imagery I created. The topography is colored by elevation, ranging from 19 m (blue) to 297 m (red). Subtracting summer topography from that will give us snow depth measurements. The image is 12.5 cm resolution (GSD) and the topography is 25 cm.

An important use for these data is to determine the suitability of this snowpack for use by convoys of large vehicles full of seismic and support gear. Given that the current State guidelines indicate that at least 9″ (23 cm) of snow cover must be present, it would be nice if my technique was accurate enough to measure snow to this level. I already know this to be the case, based on lots of prior experience; you can learn more about related snow-measurement projects in my initial paper on snow, my measurements of the tallest peaks in the US Arctic (paper, blog), and my blog on the measuring the height and glaciers of Denali, the tallest mountain in North America. But it’s always nice to validate each data set, so below I demonstrate that my April 2019 fodar maps are of sufficient quality to measure the smallest of snow drifts. I do this here in a qualitative way — I don’t have any measurements on the ground to compare to, so I try to find features in the image that I recognize and then confirm their topographic shape is as expected. That is, it is not necessary to perform a full ground validation study (like in the examples above) on every project, if the system is functioning properly and no blunders occurred in processing, then confidence is high that prior validation results apply.

I like to start validation tests just by making sure the topography looks like topography, to check for blunders. Here we can see a couple of ravines coalescing into a single channel, with snow scour and drifting. Looks good to me (though not so good for driving heavy trucks over…). Let’s start measuring things.

Here we can see polygons in both the imagery and topography, across a ravine. The red line is an elevation profile, shown below.

*Here is the elevation across that gully, about 6 m deep. Note the topographic variation on the tundra plateaus, within about 20 cm, as expected.*

Now let’s look at snow. Here we can see some long sastrugi in the imagery and the topography. I’ve extracted an elevational profile from one of them, below.

The horiztonal tick marks here about 5 cm apart, indicating this sastrugi is about 20 cm high. But it’s a little hard to tell because it’s on a slope. It turns out to be hard to find sastrugi on a flat surface here. In any case, being able to resolve sastrugi this small is a good sign.

Here are some larger sastrugi cutting across a slope, easily seen in both the imagery and topography.

The horizontal tick marks on this elevation profile (from the previous image) are 10 cm apart, indicating the center drift is about 70 cm tall. Here we are clearly resolving the shape of this drift and the smaller ones around it.

Here I have extracted a profile across a small drift in between two patches of bare ground — this is a proxy for measuring a change from summer to winter. In the image, we can see this profile spans across where the ground is bare and in between there is snow. In the topography, we can see the ground on either side of the snow is at about the same elevation and the snow is higher. Below we see the actual elevations.

*I think this is the money shot for this validation study.* Here the horizontal grid lines are 10 cm apart, so this drift is about 35 cm tall. Where the ground is bare (on the left at the red line), the topographic variation is only 3-5 cm, which is about the noise level we have found in many other studies. We can’t tell how much of this is real and how much is noise just from this image, but even it was all noise, that’s still really good. That is, the shape of this snow patch clearly stands out against this background, and it is only 35 cm tall, meaning that it is no problem to measure features much smaller than this. So we have no reason to believe that we cannot determine where the snow pack is thinner or thicker than 9″ (23 cm) in the 1002 Area in support of seismic vehicle maneuverability.

One last example, just trying to find more examples of really small features we can identify in the photos that we can measure in topography. Here I’ve found three small snow drifts over nearly bare ground that is sloping, and extracted their elevations, below.

The horizontal grid lines are 10 cm apart here and the red line is indicating the location of one of the snow drifts. It’s height is about 15 cm, and it’s shape is clearly resolved, indicating that many of the smaller variations seen here are real as well. Everywhere I looked the situation was the same.

What I’ve shown here is a quick and dirty validation study indicating that 1) I did not screw up my March/April 2019 acquisitions and 2) my fodar maps can easily determine where the snow is thicker or thinner than 9″ (23 cm), the minimum amount needed to support seismic vehicle travel, just like we have found in many other studies. I did not actually measure snow depth because I have not yet finished processing the summer topography from 2018, this just demonstrates that it will work. In the meantime, below I show that we don’t need to wait until I quantitatively measure snow depth because we know the snow depth is near zero in many places because we can simply see it from the imagery.

Identifying Bare Ground Visually in the Peak of Winter Snow Pack

In this section I’m trying to share some sense of how much bare ground is exposed out there on a regular basis. I realize that these data are from a single year, but we have to start somewhere and that I’m doing this at all is based on over a decade of informal observations of this same situation. I could fairly be accused of cherry picking here, but that is the point really — here I’m trying to demonstrate that there are huge regions in the 1002 Area that are windswept all of the time, I’m not developing a climatology. And I don’t really have to cherry pick very much, as you will see this is the case wherever I look. Regardless, the entire set of acquisitions I made were reasonably distributed spatially and across all of the terrain types in this area, so once those are processed we can begin making more balanced generalizations.

Example 1: Zooming out from tussock-scale to 1002-scale

Below is sequence of screenshots starting zoomed in enough to see individual tussocks then gradually zooming out to put the entire 1002 Area in perspective. The red lines are spaced 200 m apart — this grid represents what is currently being proposed for overland travel by a convoy of dozens of heavy vehicles dragging a hotel for 300 people: some 40,000 miles of trail in total. As you can see here, most of these grid lines cross bare ground at the peak of winter snow pack. At the end of this blog, I show how you can download these data for yourself and explore them in Google Earth as I have done in this example.

The dark colors are tundra tussocks poking through the snow. The deeper snow (where you don’t see tussocks) are in the ice wedges surrounding high-centered polygons. The red lines are 200 m apart for scale, which is also the grid spacing of the proposed overland seismic survey.

This image is zoomed out from the previous one, the pushpin is in the same location for all of the images in this sequence. At right I see a single cell that is almost fully covered with snow (but not quite) — really one has to cherry pick the non-bare-ground in this area, as across all of the lines here I can find bare ground.

As the topography gets lumpier, the percentage of bare ground increases. The number of snow drifts also increase, but in general they do not span a seismic grid cell.

*Within this area, I could not find more than a handful of seismic cells that did not have obvious bare ground in them.*

The image shown here is what I have made available below for download, such that you can fly around over it yourself in Google Earth, just as I have done here. You will find very few seismic grid cells that are fully covered in snow. *This is about 5% of the block I have processed.*

*Even at this zoom level, the amount of seismic travel proposed is so dense that the lines blur together creating a moire pattern.*

In the download provided below, I’ve also included the seismic grid pattern. It’s not the actual one, just one I created at 200 m x 200 m spacing as indicated in the proposal. It totals to about 40,000 miles of travel. That’s a lot of travel by dozens of heavy vehicles dragging a hotel for 300 people 24/7 in a hurry in the dead of the dark winter at -40F. I find it hard to believe they would be able to use ground measurements to keep from driving over the pervasive thin-to-nonexistent snow pack seen here.

Example Pairs

It’s difficult to show both detail and spatial extent in 2D images, which is why ultimately I hope to share the entire summer/winter data set using an online tool similar to Google Earth such that you can zoom in and out as you please. In the meantime, I’ve somewhat randomly pulled pairs screenshots below, first showing detail then showing surrounding extent, to give an impression for the nature of wind scour here. Just to repeat to emphasize the point, this blog is just showing the data qualitatively — once I fully process the data we will have actual snow thickness measurements (like 12 cm, 18 cm, etc) for each pixel you see here. The examples shown in this section are outside the data snippet I provided for downloaded, since you can look at those data on your own.

*Here we can clearly see ice wedges filled with snow scoured to the level of bare high-centered polygons. Assessing snow depth here is easy!*

*This is zoomed out from the previous image. Using that previous image should help in interpreting the same features over a broader area.*

*Some cool snow dunes over what appears to be otherwise nearly bare tundra.*

*A zoom out on the previous image. Most of these seismic grid cells cross visibly bare ground.*

The dune patterns are really cool and vary considerably across the terrain, controlled by wind speed and direction, terrain shape, and snow availability. Here they are formed directly over bare ground.

A zoom out on the previous image. Most of these seismic grid cells cross visibly bare ground, even where the terrain is fairly even. The change in brightness towards the bottom is due to acquisition spanning between two days.

The thinnest and thickest snow packs are often adjacent to each other, with scour occurring on the top of bluffs and deep drifts along the bluff faces. This looks like perfect polar bear habitat. The implications for seismic are that even where the snow is thick in regions like this it is still a bad idea to drive over it due to likelihood of polar bear dens.

*A zoom out of the previous image. Nearly all of the seismic cells cross visibly bare ground.*

The snow here beautifully captures the shape of the permafrost polygons. The fact that we can see their shape suggests to me that the snow too thin to drive on, as it means the high-centered polygons are not completely covered.

*A zoom out of the previous image. Close inspection shows that most of these seismic grid cells are crossing visibly bare ground.*

Towards the southern end of this block, the snow was noticeably thicker, often obscuring the shape of the polygons and not showing easily-identifiable bare ground. Here we will have to use the topographic measurements to determine snow depth quantitatively, rather than just looking for bare ground within the image.

*A zoom out of the previous image. Here is it difficult to determine snow depth visibly, because the visual method only works when snow depth is zero or a thin dusting.*

Here is just a little warning about visual interpretation. Through the center of this image is a change in lighting conditions and possibly some processing differences as the photos were acquired on different sorties — the change in contrast seen here does not indicate a change in snow cover. This visual difference really only matters to humans, the computer can measure much smaller changes in contrast and the topographic method I use to determine snow thickness is essentially unaffected by this.

*A zoom out on the previous image. Here we will have to use the quantitative methods of subtracting summer topography from winter topography to determine snow depth.*

Want to Download the Data and See for Yourself?

I’ve provided two ways you can view some of the data on your own if you have Google Earth. The full block of data I’ve processed is 8 GB when compressed fairly heavily, about 150 GB uncompressed. So I cut out a snippet that I thought was of manageable size to share via Google Earth. Note too that the imagery is heavily compressed to facilitate download, such that my version is probably twice as sharp as seen here.

The first method is to download a KMZ file here, then you can fly around on it locally but you will have to download a 375 MB file. It also includes the seismic checkerboard for the 1002 Area. If you have the bandwidth and are nerdy about these things, this is the better option.

The second method is to download this network link KMZ that is only a few kB and same as above, but the imagery may be a bit sluggish in loading to your computer as it is not stored on your computer.

If you just want the 200 m x 200 m seismic checkerboard grid, you can download that here, it’s small; if you have read this far, I think you would find it jaw dropping regardless of your interest in snow or fodar.

I have created a KMZ file for Google Earth which you can download and use on your own. It includes a snippet of the imagery data (left) and the 200 m seismic grid I created (right).

The Implications of these Data for Overland Seismic Work

What I’ve shown in this blog is really just the quick-and-dirty tip of the iceberg in terms of what I plan to do with the data I have in hand and what is possible in the future, but it nonetheless clearly demonstrates that vast areas of the Arctic NWR were snow free this winter, confirming quantitatively what we have anecdotally observed in many previous winters.

I see the implications of these measurements for oil exploration falling into two main categories.

First, by law (and simply because it’s a good idea) we need to protect the ecosystems here from potentially massive impacts, and ensuring a minimum snow cover provides at least some hope in that direction. Perhaps as importantly, there is already a lot of momentum to reverse the laws allowing development access in this otherwise highly protected and pristine wildlife refuge, but if the first thing that occurs here is that 40,000 miles of tire ruts get left behind, the resulting public outrage will virtually guarantee that no drilling will occur out here for decades. So to prevent everyone from losing (no drilling and a devastated pristine landscape), we need to measure snow like this several times per year for many years to develop an appropriate characterization of both snow cover and snow dynamics, as well as include some field studies to measure things like density and hardness. This approach is nothing new to science. And we need to have an independent monitoring program during and after after seismic work to validate that the permitted protocols were followed, as well as a long-term monitoring program during summers to determine the long-term impacts of this overland travel along with better definitions for what is considered ‘damage’ and who is responsible for mitigating that damage.

Second, the cost of this seismic work is in the hundreds of millions of dollars — can you imagine being an investor in this work and finding out the project had to be cancelled mid-stream because no one thought to check to see if the snow was ever thick enough to meet the legal, minimum standards? It’s one thing to say in a permit application that “We will measure snow depth ahead of our trucks to ensure the legal minimums are met” and it’s quite another to say to an investor “Sorry, we just spent $50M of your money on mobilization for something that’s apparently never going to happen, so can you write us another check so we can use more appropriate methods?” So, yeah, I think there is real economic value to getting an accurate understanding of snow pack out here before committing huge resources to a plan that is likely to fail and start considering alternatives now, like eliminating the overland travel by heavy vehicles and using using point sources for acoustic energy (which has been previously demonstrated to work fine here) deployed by helicopter, allowing work to proceed in any season to eliminate many wildlife issues as well. Even simply identifying which areas are likely to be too thin for overland seismic could pay for itself 100x given the economics in play here — that’s a pretty healthy ROI in any industry!

So whether you support oil development or oppose it, I hope you agree that the implications of these data are clear — we need to make high-resolution photogrammetric mapping of snow in the 1002 Area a priority.