The spatial and temporal influence of infrastructure and road dust on seasonal snowmelt, vegetation productivity, and early season surface water cover in the Prudhoe Bay Oilfield

Arctic regions are experiencing rapid change due to cumulative effects caused by climate change and the expansion of infrastructure driven primarily by industrial development related to resource extraction (NRC 2003; AMAP 2007; Walker et al. 2022). Oil and gas exploration and development, in particular, has led to the spread of infrastructure in many areas of the circumpolar Arctic over the last five decades (Kumpula et al. 2012; Raynolds et al. 2014, 2020; Walker et al. 2022). Areas like the Prudhoe Bay Oilfield (PBO) in Alaska (Fig. 1A) are host to many types of infrastructure, ranging from paved and gravel roads, airports, stores, and other service facilities, to pipelines, processing plants, drill rigs, gravel pits and pads, and associated features of oil and gas production (Raynolds et al. 2014). Studies documenting increased infrastructure development in the area of Prudhoe Bay have shown that it impacts the surrounding natural environment in multiple ways, including changes to hydrology, soils, vegetation, permafrost, snowmelt, and ground temperatures (Benson et al. 1975, Walker et al. 1980;, 2022; Walker 1985; Walker and Everett 1987; Raynolds et al. 2014; Kanevskiy et al. 2022).

Infrastructure actively influences the distribution and redistribution of snow by creating barriers that act like snow fences, leading to modified snow depths and snowmelt patterns (Fortier et al. 2011). Previous studies have shown that snow along roads and on road embankments can disappear earlier compared to snow in the surrounding area, leading to earlier soil thaw and vegetation green-up near roads (Benson et al. 1975; Walker and Everett 1987; Walker et al. 2022; Chen et al. 2020).

In winter, roadside snow drifts develop along the margins of the elevated roads. Heavily traveled roads, such as the Spine Road and the Dalton Highway, also collect large quantities of dust during the winter and spring. These large amounts of dust decrease the albedo of snow which induces snowmelt up to two weeks earlier than similar drifts along remote roads with lower dust accumulation (Benson et al. 1975) (Fig. 1b). Thus, the snowdrifts insulate the soils, keeping the underlying tundra relatively warm during winter. Dust-covered roadside drifts also melt out earlier than areas distant from the roads, contributing to early-summer flooding, warmer summer soil temperatures, thicker active layers near roads, and enhanced thermokarst (Walker et al. 2022). These snow-free and often-flooded roadside areas attract wildlife, including many waterbirds before nesting (Walker and Everett 1987; Truett et al. 1997).

Dust and snow near roads are strongly affected by wind patterns and orientation of the roads to the dominant wind directions. Everett (1980) documented logarithmic decreases in dust volumes with distance at several sites adjacent to heavily traveled roads in northern Alaska. The volume of dust at 1000 m from the road was 2–9 times greater at the Prudhoe Bay site than at the other sites, including Franklin Bluffs, Sagwon, and Toolik, probably reflecting very high traffic volumes, the transport of dust to greater distances from the road caused by higher wind velocities, and the contributions from the relatively dense road network in the PBO. Dust was also transported to greater distances from the road during winter when the dust can be re-transported by wind with drifting snow several times over a relatively smooth snow-covered surface.

Windblown dust in the form of loess from nearby braided river channels is a natural element of local Prudhoe Bay ecosystems. The high pH of the loess-derived soils (Everett and Parkinson 1977) promotes a relatively rich local flora with abundant basophilous plant species (Walker 1985; Walker and Everett 1991). However, extreme amounts of dust near roads cause reduced species diversity and vegetative cover, particularly of small forbs, mosses, and lichens (Walker et al. 2022).

Remote sensing has been a frequently used tool to study snow cover, snow water equivalent and other snow properties in the high and mid latitudes. Landsat data were used to study the effects of dust on snow distribution in the Prudhoe Bay area in the early days of the PBO development (Benson et al. 1975). Operational products like the Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover data set allow for timely analysis of snow persistence on a global scale (Hall et al. 2002; Salomonson and Appel 2004; Hall and Riggs 2007; Frei et al. 2012). Multispectral imagery from the Landsat family of satellites has been applied to study snow and its properties successfully in numerous settings (Benson et al. 1975; Andrews Davis and Wright 1976; Dozier 1984; Macander et al. 2015; Berman et al. 2018; Hu et al. 2019). Efforts using satellite-based microwave remote sensing data have been employed in both lowland and alpine applications (Dietz et al. 2012; Tsai et al. 2019) and have produced long-term records for the Northern Hemisphere covering a period starting in 1979 (Takala et al. 2011).

Studies based on multispectral data as well as those based on microwave remote sensing systems lack either the spatial or temporal resolution to adequately assess the impact of separate infrastructure units on the surrounding environment on an annual basis. Data products like the MODIS snow coverage maps provide daily global measurements in 500 m spatial resolution. While the daily temporal resolution provides many advantages for timely analysis of snow cover, the impact of local infrastructure on the direct environment cannot be fully captured with this coarse spatial data. Other products such as the GlobSnow (Takala et al. 2011) are available in 1 km to 250 m resolution versions. Snow mapping efforts based on Landsat imagery have the advantage of higher spatial resolution (30 m). Macander et al. (2015) calculated the typical snow-free date for Northwest Alaska based on Landsat imagery from 1985 to 2011 (Fig. 1b). Snowmelt patterns in Arctic landscapes have also been studied on a local level using time-lapse cameras (e.g., Kępski et al. 2017) and in situ sensor observations. Sentinel-2 and the Normalized Difference Snow Index (NDSI) have been used successfully to map snow cover in alpine areas (Gascoin et al. 2020); however, studies using Sentinel-2 to map snow in tundra environments are currently lacking, to the best of our knowledge. Sentinel-2 allows for mapping snow cover at high spatial resolution (20 m) and offers the possibility of acquisition with a 5 day interval over the study area (Drusch et al. 2012; Gascon et al. 2017).

The ability to accurately map and monitor the expansion of infrastructure in the Arctic is important for assessing and mitigating risks to and from infrastructure in a climate that is warming three times faster than other places on earth (Walker and Peirce 2015). Approaches for mapping infrastructure using remote sensing have been widely studied on a global and local scale (Bartsch et al. 2020). Infrastructure and its impacts on Arctic environments can be assessed by remotely sensed methodologies. Several data sets based on remote sensing data with regional to global coverage exist (e.g., Wang et al. 2017; Esch et al. 2018), but often lack either comprehensive documentation of different types of infrastructure or suitable coverage of Arctic settlements and infrastructure hubs (Bartsch et al. 2020). Recent advancements have allowed for high spatial resolution mapping (10 m) of both linear infrastructure (e.g., roads) and built-up areas on a circumpolar scale based on Synthetic Aperture Radar and multispectral data acquired by the Sentinel-1 and Sentinel-2 constellations (Sentinel-1/2-derived arctic coastal human impact data set (SACHI); Bartsch et al. 2020).

In this study, we describe and quantify spatial and temporal patterns of snowmelt timing in relation to infrastructure in the Prudhoe Bay area. We demonstrate the potential of resources like the Sentinel-2 constellation to quantify and monitor snowmelt patterns in Arctic industrial regions as well as the resulting cumulative effects of infrastructure development on hydrology and vegetation. This study addresses multiple themes, focussing on connectivity of different landscape elements, connected to the Terrestrial Multidisciplinary distributed Observatories for the Study of Arctic Connections initiative (Vincent et al. 2019), making it highly relevant for the future study of Arctic natural and social systems.

Spatial patterns in snow cover and snow-free dates are visible in the results based on Sentinel-2 derived NDSI. Depending on data availability and cloud conditions, the Sentinel-2 constellation provides acquisitions over the Prudhoe Bay area 1–2 times per week. This is a relatively high temporal resolution compared to other multispectral satellite data sets with comparable spatial resolution (e.g., Landsat satellites). After the launch of Sentinel-2B in March 2017, the second satellite in the constellation, the temporal resolution improved significantly. Earlier years did not have enough temporal or spatial coverage of the study area during the relevant time frames to adequately derive snowmelt timing.

Even though the temporal resolution was theoretically adequate to detect snow-free patterns over the study area after 2017 areas might become snow free between cloud free satellite acquisitions, especially toward the end of snowmelt, when melting progresses quickly. Therefore, some fine scale spatial patterns may be missed (Fig. 4B). For our analysis, due to cloud cover and gaps between acquisitions, sufficient imagery during the snowmelt period was only available during 2019 and 2020. The spatial patterns in this study showed earlier snowmelt for locations adjacent to roads and revealed linear features associated with later snow-free dates. These delayed snowmelt features represent large snowdrifts associated with elevated pipelines and roads, with the Trans-Alaska Pipeline specifically standing out (Fig. 4). This can be confirmed through field observations shown in Fig. 16. Both photographs in Fig. 16 show clear delayed snowmelt along pipeline features with Fig. 16B depicting delayed snowmelt along the Trans-Alaska Pipeline.

Previous studies that mapped snow-free dates for the region showed similar patterns of earlier snowmelt in the Prudhoe Bay area compared to surrounding areas less impacted by infrastructure and oil development (Macander et al. 2015). In addition, the study by Macander et al. (2015) showed similar patterns of later snowmelt surrounding the Trans-Alaska Pipeline, supporting our findings (see Figs. 2 and 4).

Analysing snow-free dates revealed general patterns of similar median snow-free dates, independent of distance to the nearest roads and pipelines (see Figs. 5 and 6). However, pixels within 5 km of the nearest road frequently showed outliers of earlier snow-free dates and in general a higher variability of snow-free timing compared to locations with greater distance to the nearest road segment. This suggests that while the proximity to roads does not necessarily lead to an earlier snow-free date, there is variability in snow-free timing introduced to the overall snowmelt progression by the presence of infrastructure, in particular roads (e.g., greater snow in deep drift areas and less snow in heavily dusted areas). Distinguishing between highly trafficked snow-free roads and smaller roads which typically experience lower usage revealed similar patterns to the analysis including all road types (compare Figs. 5, 6, and 7).

The classification of infrastructure according to winter snow cover (February–March) Sentinel-2 NDSI values revealed distinct differences between the main roads (e.g., Dalton Highway and Spine Road) and roads that experience less traffic, as well as pipelines and other types of infrastructure (see Figs. 2 and 8). While features like the Spine Road showed up in the classification as snow-free prior to general snowmelt onset, most other features were snow covered. Several main roads, including the Spine Road, are either transformed into ice roads during the winter by maintenance crews or naturally develop an ice layer due to compacted snow through high traffic. These roads are likely to melt out quickly due to sublimation and gravel may quickly come to the surface in spring before general snowmelt onset on the rest of the landscape. Features like the airport runway, which was classified as snow free, are purposefully kept snow free over the course of the winter. Other types of infrastructure, like gravel pits, pump stations, and gravel pads, were partially snow free. These features are often actively used during winter and are kept partially snow free or become snow free due to frequent machine activity. On pads, snow is removed from the portion of the pad that vehicles need to access. Throughout the winter, crews clear snow off the ice layer on pads and put it on the tundra to minimize mud on pads during the break up season. Snow will pile up around infrastructure in areas that are not drivable. According to information given during the interview with Hilcorp’s field operations’ supervisor for oilfield roads and pads, these results agree with their general observations (P. McCollor, personal communication, 2021). In their first-hand experience, areas with the most traffic melt earliest and roads that are elevated above the surrounding terrain are naturally cleared of snow by wind more quickly than those close to tundra level. Main roads (primary and flowline roads) are constructed at a minimum of 4 feet (approximately 1.2 m) above tundra, which helps to insulate the permafrost (by maintaining a frost bulb high up into the road) and keeps the road surface harder longer. Smaller roads are built closer to the tundra level and do not get a lot of maintenance; both factors would suggest later snow off dates.

Localized melting occurs in areas where gravel protrudes through the ice. Road maintenance crews see steam coming off roads in winter, especially where gravel is emergent. On pads, snow is removed only from the portion of the pad that vehicles need to access. Around April 15, maintenance crews start stripping ice and snow off roads and drill site pads. Primary and flowline roads are intentionally iced in October for dust abatement, and maintenance crews water the roads throughout the winter to keep the ice built up. Ice paving is not done on oilfield access roads or on paved sections of the Dalton Highway, which include the last 52 miles coming into Deadhorse, where paving was completed in 2020–2021. Alaska DOT&PF uses ice paving on all non-asphalt sections of the Dalton Highway south to Milepost 209. Few sections of road within the oilfield are paved. Around 15 April, maintenance crews start stripping ice and snow off roads and drill site pads (W.J. Russel, personal communication, 2021).

Wind during the winter season is known to be a major influence on snow distribution and redistribution (Pomeroy et al. 2006). The combination of a predominant wind direction and infrastructure acting as artificial barriers on the relatively flat landscape results in snow build up along roads and increased snow depth (Fortier et al. 2011). This influences the snow-free timing, as the areas with increased snow depth tend to take longer to become fully snow free. Simultaneously, dust from snow-free infrastructure or natural sources (e.g., dust emanating from the Sagavanirktok River banks) is distributed by the wind as well, and deposited along raised infrastructure segments such as elevated roadbeds. Dust on the snow changes the albedo and can lead to increased snowmelt for a specific location. Considering both the increased snow depths caused by snow redistribution through wind and the effect of dust causing earlier snowmelt, wind leads to two distinct influences on snow that have the opposite effect on snowmelt timing. The higher variability in snowmelt timing near roads shown in our results (Fig. 5) agrees with these general observations and past research (e.g., Fortier et al. 2011). Water area derived from NDWI showed strong correlation with the distance to infrastructure for roads, pipelines, and other infrastructure areas within the study area (Fig. 15). The strong connection between water area and distance to infrastructure suggests a connection of delayed snowmelt, road-related damming affecting the cross-drainage of water, and increased thermokarst ponding and snow depth close to the road (Fig. 10) increasing the amount of water from snowmelt. In addition to the increased snow depth, infrastructure also provides barriers to surface drainage of melt water, creating conditions for early season flooding (Walker et al. 2022).

Influences from infrastructure on the natural landscape are varied and touch other natural processes in addition to snowmelt, including ice-wedge degradation, both natural and triggered by construction, a process which is prevalent in our study region (Raynolds et al. 2014; Kanevskiy et al. 2022). Deepening of ice-wedge troughs caused by ice-wedge degradation are areas of greater snow accumulation, and can be seen in the snow depth measurements along the individual transects (Fig. 10). Variables such as surface water area can be influenced by infrastructure through a change in surface hydrology, particularly in low-lying drained lake basin landforms, where water-pooling is exacerbated by the linear infrastructure. Vegetation parameters such as greenness and general productivity are influenced by water and dust distribution during the growing season (Ackerman and Finlay 2019). However, those variables are highly dependent on each other with differences in snowmelt patterns influencing post-snowmelt flooding and ponding along infrastructure. NDVI values show no clear connection to distance from roads or pipelines (Fig. 12) but show differences in seasonal NDVI progression between upwind and downwind areas adjacent to roads (Fig. 13). The interdependency of these factors makes it difficult to separate direct influences from infrastructure (such as acting as a snow fence or a source of dust) and the influence these variables have on each other (such as increased snow depth leading to more water near infrastructure or later snowmelt leading to a delay in NDVI increase).

The availability of relatively high-resolution satellite imagery (Sentinel-2) and the access to relatively high-resolution infrastructure data proved to be essential for the quantification of spatial and temporal patterns of snowmelt timing in relation to infrastructure elements. This study shows the suitability of multispectral Sentinel-2 data to study spatial patterns of snowmelt timing in Arctic industrial areas. This study highlights the impact of infrastructure on a large area extending past the direct human footprint as well as the interconnectedness between vegetation, hydrology, and near-surface ground temperatures. The influence of infrastructure on different environmental factors was shown to be highly variable both spatially and temporally and interdependent on each other. The availability of relatively high-resolution satellite imagery as well as relatively high-resolution infrastructure data proved to be essential in visualizing and quantifying spatial patterns in snowmelt in relation to infrastructure elements. Snow-free dates vary strongly with their distance to infrastructure, especially roads with downwind and upwind areas experienced different snow and dust effects. Both vegetation and surface water indices (NDVI and NDWI, respectively) were shown to be correlated to distance to infrastructure. Our study shows that the presence of infrastructure introduces variability in snowmelt timing and related phenomena in Arctic areas such as increased near-surface ground temperatures. The resulting interacting effects are complex and difficult to separate. It is clear, however, that the presence of infrastructure directly and indirectly modifies patterns of landscape change in Arctic areas.

We would like to thank Pat McCollor (field operations supervisor) from ASRC Energy Services and William J. Russel from Alaska DOT&PF (Northern Region Maintenance and Operation) for their input and the fruitful discussions of early results of this study.