One day in the summer of 2016, I was wheeling a heavy stack of chairs into the back of a truck after a long day of work when I decided I needed to go on an adventure. On this particular day, instead of rushing home to relax after a long day working in the blazing sun, I felt an irresistible urge to escape to the cold, clean air of the high peaks that enclose Salida, my hometown, and the valley surrounding it. I knew that Mikey, a friend of mine, wasn’t working that day, so I sent a text asking him if he would want to drive a nearby unexplored 4x4 road I had read about called Hayden Pass. He replied immediately, and after a pit stop for some gas, snacks, and cold drinks, we were on our way. The road begins about 10 miles east of Salida, in a mountain hamlet called Coaldale, ascending up and over the Sangre de Cristo range and exiting the mountains in the town of Villa Grove.
Upon reaching Coaldale, we turned right off of the highway onto a washboarded dirt road which led upwards into the blue-green forest surrounding the peaks of the Sangre de Cristo, passing several cabins and an RV resort complete with mini-golf and a ropes course along the way. Journeying upward through the scrub oak into the subalpine Engelmann spruce forest, we casually remarked on the amount of dead and fallen trees surrounding us. In many places, the forest looked more dead than alive, a seemingly endless matrix of grey, bare conifers covered in lichen, standing upright, leaning against each other, and littering the forest floor with fallen logs. Later, after getting out of the car to stretch our legs and do a bit of exploring, I came to a ridgeline where I could look out over the forest, and was shocked to see the dead trees extending far into the distance. I snapped a few quick photos and returned to the car, remarking to Mikey that it seemed like a wildfire was bound to happen here. I had no idea how right I was.
Figure 1: Severely beetle killed trees looking back toward Coaldale from Hayden Pass, CO.
Two days later, on July 10th, I was returning from another mountain adventure with a different friend, this time 30 miles north of Salida in the Sawatch Range. It had been a beautiful July day spent in the alpine tundra with the wildflowers at peak bloom, but on the drive home, our good mood was suddenly replaced by anxiety. To the south, we could see the silhouette of the familiar Sangre de Cristo range on the edge of the horizon, however, something was wrong. What looked like a nuclear mushroom cloud, a tall, messy, grey-black plume of smoke, was billowing from the mountains, just outside of the town we had grown up in. I ignored the 65 mph speed limit signs as I accelerated, anxious to get home and find out how bad the fire was.
Figure 2: View of the Hayden Pass fire smoke plume driving south toward Salida, CO, on the day of initial ignition, July 8 2016.
Living in Colorado, I grew up knowing that wildfire was a constant threat during the summer, especially in drought years. However, although I had seen past burn scars and the haze from distant blazes in the past, I had never seen such a dramatic wildfire up close. The fact that it was just outside of my hometown added to the surreal and terrifying experience; even with a buffer of about 10 miles from the fire to town, the constant helicopters flying overhead to bring water to the fire and the wildland fire command center temporarily stationed at my old high school made this blaze feel like a more tangible threat. The blaze threatened the neighboring town of Coaldale, destroying a mountain cabin as well as temporarily shutting down the RV resort, Cutty’s, that Mikey and I had seen on our drive of Hayden Pass just a few days prior, when we noticed the incredible amount of downed and dead timber in the area. Luckily, no lives were lost, but the 16,000 acre blaze was a reminder of the sheer power of wildfire when the conditions are right, and it continued to burn for several months, finally burning out in October 2016.
Beetles, Conifers, and Fire
Many of the large wildfires which occur in Colorado are in part the product of an unexpected, miniscule arsonist: various bark beetles native to the intermountain West. The species linked to extensive tree mortality in Colorado are the mountain pine beetle (Dendroctonus ponderosae), which infests pines including Limber, Lodgepole, and Ponderosa species, and the spruce beetle (Dendroctonus rufipennis), which primarily target Englemann spruce found in Colorado’s subalpine forests below treeline. While the mountain pine beetle has caused the most widespread forest damage historically, severely impacting nearly 3.4 million acres of forest between 1997 and 2017, the spruce beetle has overtaken the mountain pine beetle as the most damaging species for Colorado conifers, infecting an estimated 1.8 million acres since 2000, with 178,000 of those acres being infected in 2018 alone. When these beetles infect trees, the direct effect is usually tree mortality, though spruce beetles do not turn the needles of infected trees into the characteristic red color associated with mountain pine beetle infestation.
While these beetles are both native to the Rocky Mountain region, their prolific rise in recent decades is unprecedented and likely driven by forest management practices during the 20th century. Extensive logging and subsequent intensive fire suppression have resulted in dense, homogenous forests, made up of trees of similar size and age, a perfect scenario for a bark beetle epidemic. The bark beetle epidemics seen in Colorado and other western states are likely amplified by climate change signals. Warmer winters allow more beetles to survive from year to year, and hotter, drier summers leave host trees weakened and more susceptible to a beetle infestation.
At the same time, climate change and other factors including population increases in the wildland-urban interface (WUI) are leading to more ignitions and more large fires. The combination of increased ignitions, hotter and drier fire seasons, and an ample fuel supply due to the bark beetle outbreaks of the past several decades is the perfect storm to drive longer, more destructive, and more dangerous fire seasons in the intermountain West. Indeed, the past two years have been the costliest on record in terms of wildfire suppression costs nationwide, with most of the increase driven by a spike in western fires. That some significant portion of this increased wildfire activity is directly related to the bark beetle epidemics that are a legacy of past forestry practices speaks volumes to the impact that we as humans can have on our environment, and in turn, how our environment can reflexively impact us.
Back to Hayden Pass - Using Free Data to Learn About Fire
Even several years later, the fire on Hayden Pass sticks with me for a number of reasons. The fact that I drove the pass just two days before it burned means that Mikey and I were likely among the last people to see the area before the fire. We both noted that beetle kill had severely impacted the forest surrounding the pass and how unsurprised we would be if it were to burn at some point in the future. We were completely shocked to find out that it had happened so quickly, though.
While the official cause of the fire was designated as a lightning strike, we had seen smoke from a campfire a little ways from the road near the top of the pass two days prior, so for us, the possibility of human ignition feels very real. However, regardless of the cause, the Hayden Pass fire was a wake up call, making it very evident to us that the idyllic mountain forests surrounding Salida were also incredibly susceptible to wildfire, especially during the scorching hot, bone-dry summer months.
As a geographer, I have always wanted to take a closer look at the Hayden Pass fire to see what the techniques I had learned for analyzing satellite and other remotely sensed data could show me about an event with plenty of personal significance. Last spring, I had a chance to do just that for a remote sensing class project. I was able to quantify both burned area and the burn severity across the wildfire, allowing for some further analysis of the dynamics of the blaze. One of the most exciting aspects of this analysis is that it was done using only free, public data, and most steps could be easily performed using open-source software such as R or QGIS. Thus, the basic steps I outline below could be applied to analyze any fire of interest quickly and easily.
Burn Indices - Visualizing Wildfire
The first analysis I performed on the Hayden Pass Fire was the calculation of several “burn indices”, which are essentially simple calculations that can make burned areas stand out in a satellite image. While I used ENVI, a software designed for manipulating satellite and other remotely sensed data, the approach used could also be accomplished using QGIS or R, which are free and open-source. First, I obtained Landsat satellite imagery (available for free through USGS), which included the Hayden Pass area, selecting one image from before the burn in June, and one from after the fire was contained in October.
Landsat imagery, like many satellite data sources, comes in the form of several “bands” that contain pixel grids in which each pixel’s value corresponds to the brightness of light at a particular portion of the spectrum. Thus, a high value for a pixel in the second, “green” band signifies that a lot of green light is being reflected by a particular object. The most recent Landsat satellite takes imagery across ten bands, including three visible light bands, a near infrared (NIR) band, two shortwave infrared (SWIR) bands, and two infrared thermal bands. These band values are useful for environmental analysis, as the particular pattern of values for a pixel across the various bands can often reveal whether the pixel represents a forest, ocean, or even a burn scar. Additionally, the bands can be manipulated to create burn indices which can highlight burned areas, allowing the analyst to extract things such as burn intensity or area.
I calculated three main burn indices for the Hayden Pass fire that I knew would be useful for the further analysis I performed. Using the Landsat images acquired post-burn, I computed the Normalized Burn Ratio index, the Normalized Burn Ratio with Thermal index, and the Burn Area Index for the area surrounding the fire.
Figure 3: Burned Area Index (BAI) raster calculated for Hayden Pass Fire. The white pixels represent likely burned areas, showing the Hayden Pass burn perimeter clearly in the center of the image. However, some white pixels do not represent burned areas, such as the clouds picked up to the bottom right of the image. BAI is commonly used to measure burned area in satellite images.
Figure 4: Left: Normalized Burn Ratio (NBR) calculated for Hayden Pass fire and surrounding areas. Right: Normalized Burn Ratio with Thermal (NBRt) calculated for Hayden Pass fire and surrounding areas. NBRt differs from NBR in its inclusion of thermal bands in index calculation, leading to subtle but important differences in the resulting image. NBR and NBRt are both commonly used to measure burn intensity from satellite images.
Burn Area Index (BAI) manipulates bands 4 (red) and 5 (near-infrared). It is useful for identifying burned areas and measuring burned area, as burned areas typically show up as dramatically higher pixel values compared to un-burned areas. The formula for calculating BAI for each pixel is:
Normalized Burn Ratio (NBR) manipulates bands 5 (near-infrared) and 7 (shortwave infrared). It is useful for both identifying burned areas and estimating burn severity. High NBR values indicate healthy, well-vegetated areas, while low NBR values indicate burned areas and sometimes bare ground. The formula for calculating NBR for each pixel is:
Similar to NBR, Normalized Burn Ratio with Thermal (NBRt) manipulates bands 5 and 7, but NBRt additionally incorporates band 10, Landsat’s thermal band. This index, when calculated using a post-burn image, can identify burned areas and indicate fire severity similarly to NBR, but with better ability to distinguish between bare ground and burned areas due to the inclusion of a thermal signal in its calculation. The formula used to compute NBRt for each pixel is:
Burn Area - How Big was the Fire?
After calculating burn indices, I was able to use them to calculate the total area burned by the Hayden Pass fire as well as the burn intensity across the fire perimeter.
To do so, I first designated some areas which were clearly burned in the October BAI image with the land cover class “burned”, and some areas outside of the fire perimeter as “unburned”. I then gave this information to a maximum likelihood classification algorithm, which used these pre-labelled classes to classify burned and unburned areas across the entire image. In simplified terms, the maximum likelihood classifier essentially uses statistics based on the normal distribution to determine the probability that a pixel in an image belongs to a particular pre-designated class. While it can have issues distinguishing between very similar classes (i.e. grassland and agriculture), maximum likelihood classification can successfully distinguish classes which have distinct patterns of pixel values across the Landsat bands.
The maximum likelihood classifier outputs a classified raster grid with pixels assigned to either the “burned” or “unburned” class. Since the spatial size of each pixel is known to be 30 m x 30 m, it is possible to then calculate the total area burned for the fire by counting the total number of pixels in the “burned” class and multiplying by 900 square meters. Using this technique and converting to square miles, I calculated a total burned area of 16.42 sq. miles. This was an interesting discovery; the official fire perimeter reported by the United States Forest Service encompassed an area of 25.9 square miles. I believe that this difference stems from the different methods used to calculate the size of the fire, as the forest service typically uses the furthest extents of the fire when creating perimeter data which does not take into account for unburned patches of forest within the burned area. Furthermore, based on visual inspection of both the burn indices created and of the post-burn Landsat image, the area I calculated using the maximum likelihood method appears to potentially be more accurate than the area reported by the USFS. Thus, the use of some simple image processing techniques at home can often refine and improve on the findings of established organizations, showing the power of using free data for personal analysis.
Figure 5: Burned area classification results (red pixels indicate burned areas) for the Hayden Pass fire overlain by official USFS fire perimeter (black line). Note the differences between the two methods of quantifying burn area, with the USFS perimeter noticeably overestimating the size of the fire.
What's the Damage - Analyzing Fire Severity
The final analysis I performed for the Hayden Pass fire took a closer look at the severity of the blaze, which is important for a number of reasons, from tree mortality to post-fire rehabilitation. While it is inherently a qualitative measurement describing the amount of physical change (to soils, vegetation, etc.) that a fire imparted on a landscape, burn severity can be categorized into high, moderate, and low severity categories using the NBR or NBRt burn indexes I initially created. I chose to try to use both indices, but found that the NBRt burn severity calculation was more successful at picking out the fire itself, likely due to its use of thermal (heat) bands. The key to investigating burn severity is to create two NBRt images, one from immediately before the fire and one from after, and then overlay them and subtract the pixel values from the post-fire NBRt image from those in the pre-fire NBRt. This creates another pseudo-burn-index, called the differenced normalized burn ratio, or NBRt. High pixel values in the resulting image indicate more severe burning. The pixel values can then be color coded to visualize the places which experienced the most, and least, severe fire.
Figure 6: Top: Differenced NBR with Thermal image produced for the Hayden Pass fire. Note that snow-capped mountains in the left and bottom portions of the image are not detected as burned areas when using NBRt. Bottom: Differenced NBR image produced for the Hayden Pass fire, without using thermal bands. While the fire is detected, burn severity tends to be overestimated and many other features in the surrounding area including the peaks to the northwest and south of the fire are wrongly shown as burned areas.
Some interesting patterns can be observed in the burn severity images produced. The majority of the most severely burned areas are located at high elevations relative to the surrounding area, appearing on ridgelines and hillcrests. Wildfire growth and movement is controlled in large part by the steepness and direction of the slope the fire is burning on, as fire generally has an easier time moving uphill than downhill. This is mostly because hot air travels upwards, carrying with it embers and other pieces of burned material which spread the fire rapidly from tree to tree uphill. This can lead to the most intense areas of burning within a fire to be located on ridges where the fire has built energy and has nowhere left to go, which is evident in the burn severity analysis for the Hayden Pass fire. Conversely, many areas of relatively low elevation in the burn severity image display low burn severity, with the bottoms of stream-cut valleys between ridges showing up as unburned in both the burn severity analysis and the burn area classification. This suggests that these local low points can act as inhibitors to wildfire spread, because fire has a difficult time moving downhill. Finally, there are some areas of high burn severity which are not located on ridges or hillcrests, but instead on the middle slopes between valleys and ridges. While it is hard to know for sure without further analysis, I suspect that these are the areas with large amounts of beetlekilled trees which provided enough fuel for the fire to reach high intensities and therefore burn more severely.
Figure 7: Burn severity raster (NBRt ) for the Hayden Pass fire overlying 3D Google Earth Imagery, allowing for topographic features to be seen. Many of the most intensely burned areas (redder colors) appear on ridges and hillcrests, while low-intensity burn and unburned areas generally follow valleys between ridges.
Using pre-established values to classify these burn severities (i.e. as high severity, moderate-high, etc.), I was able to calculate the percentages of the burn area which burned at each severity level. The Hayden Pass fire caused severe physical effects on the landscape across the majority of its extent, with 74.2% of the total burned area being classified as either high severity or moderate-to-high severity burn, and a total of 25.8% being classified as low or low-to-moderate severity. The overall severity of the fire was likely fueled in large part by the extensive beetlekilled spruce that I saw in the area before the fire, which has been linked to increased fire severity in the past as it provides abundant fuel sources to the inferno.
Revisiting Hayden Pass
Through my analysis of the Hayden Pass fire, I was able to gain a much deeper understanding of the nature of the blaze and also give myself some closure after seeing the fire ignite so soon after I had visited the area. However, I found myself drawn to the idea of returning to Hayden Pass over a year later when visiting Salida over Thanksgiving Break 2017. I wanted to see firsthand the effects of the blaze and hoped that I would leave with a positive outlook as to the recovery of the area’s forests.
I set out with my younger brother one afternoon before Thanksgiving and we quickly found ourselves at the turnoff to the road in Coaldale. Signs were in place warning of washed out conditions, landslides, and road damage as a result of a year of rain and snowmelt severely eroding the newly exposed forest floor. I decided not to turn back and instead to see how far I could make it before having to turn around due to snow or damaged roads. Unfortunately, some deep snow-banks accumulated over the fall forced us to cut the drive short, less than halfway to the top of the pass. However, the turn around point was just inside of the burned area so I was able to observe some of the post-fire forest.
I am happy to report that the portion of the forest surrounding Hayden Pass that I saw on my return appeared to be recovering well a year and a half after the fire. While many trees were scorched several feet up their trunks, and some appeared to be damaged due to the blaze, overall the forest looked much healthier than when I had seen it before the fire. Much of the dead and downed beetlekill Mikey and I saw was cleared away by the burn, leaving mostly healthy trees standing. The sunlight trickled down to the forest floor in the newly cleared forest, and the less densely packed stands of Englemann spruce will likely allow for new saplings to take root in the coming years.
Figure 8: Photo taken on Hayden Pass in November 2017, a year and a half after the Hayden Pass fire ignited. While the trunks of healthy trees were scorched, much of the dead and downed beetlekill was cleared during the blaze, leaving a more sparsely populated, sun-dappled, healthier forest and allowing room for new growth.
Seeing the seemingly improved condition of the forest was a reminder that wildfire is a natural mechanism to clear overgrown and dying forests and allow for new growth. Despite the massive wildfire that occured here, and the worry for the safety of my hometown and the people who live nearby, I left Hayden Pass that day feeling a sense of optimism and reverence for the ability of the natural world to maintain itself. The immense amount of beetlekill I saw on Hayden Pass is a legacy of improper forest management and human impacts on the climate, but in this case, nature was able to correct itself through apparent destruction, righting the wrongs of decades of human impact. I can only hope that a similar balance will be achieved in other beetlekilled forests across the West.