Moving beyond the physical impervious surface impact and urban habitat fragmentation of Alaska: quantitative human footprint inference from the first large scale 30 m high-resolution Landscape metrics big data quantification in R and the cloud

Study area

Alaska is located on the far northwestern end of North America, with the northernmost point of the United States (71°23′20″N), the westernmost point of the continental United States (173°11′10″E). The western southern limit of Alaska is 51°15′44″N, and the eastern southern limit is 54°39′44″N. An overview of the study area (Alaska) in the global context can be found in Fig. 1A. The climate in Alaska highly varies, mostly depending on the proximity to the sea and the latitude, allowing for climate class ranges from arctic to maritime (Shulski & Wendler, 2007). Average temperatures in the state of Alaska range from 5.5 °C in the South to about −12 °C in the North, and extreme temperatures were recorded as low as −62 °C and as high as 38 °C (WRCC, 2024). Alaska is home to some of the most extreme mountain ranges in North America, with the St. Elias Mountains, the Chugach Range, and the tallest mountain in North America (Denali–height of 6,190 m) (Wahrhaftig, 1965). Alaska is also known for its vast tundra and taiga landscapes with an uncountable number of lakes and is characterized by almost 40% of near-surface permafrost (Bonan, Chapin & Thompson, 1995; Pastick et al., 2015).

Alaska has a population of approximately 735,000 people (as of 2024), with a median age of 35.3 years, a median income of 86,000 USD, and a poverty rate of 10.5% (additional socio-economic details about the study area can be found here: https://datausa.io/profile/geo/alaska).

Alaska has a long human (Indigenous) history and is located within that North American setting; it features rich natural resources (Huryn & Hobbie, 2012; Naske & Slotnick, 2014; Truett & Johnson, 2000) and the vast majority of the U.S. National Park System and protected land–a highly prized land entity–within (Bigelow & Borchers, 2017).

Figure 1 provides an overview of the study area (the state of Alaska) and the underlying land use classes in the National Land Cover Database (NLCD) map, of which we used all “Developed, …” ones (21, 22, 23, 24). It additionally includes the landcover classes description in the legend, and more information on the utilized data can be found in the ‘Data’ section.

Data

For this project, we used the 2016 30 × 30 m resolution landcover map of Alaska, provided by the National Land Cover Database (NLCD) (see https://www.mrlc.gov/data/nlcd-2016-land-cover-alaska). This map product represents the national standard of land cover maps for the largest state of the US, given its source (U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center). The downloaded raw data comes as a .img file, which we converted in ArcGIS Pro (optionally can also be done in OpenGIS (QGIS)) into a usable .tiff file, using the ‘Resample’ tool, which carries a total file size of 8,269.77 MB (approx. 8.2 GB). It is an eight-bit unsigned file with a total of 124,236 columns and 67,844 rows, using the WGS 84 geographic projection (WGS_1984_Albers-using ‘meters’ as the linear unit), covering the entirety of Alaska, US.

We chose this dataset because it is the official national map product of Alaska’s landcover classes. It can be easily downloaded onto a personal computer as a single file from the internet free of charge (open-source) and has a high resolution of 30 × 30 m.

This pixel dataset easily misses linear features and comes with a total of 20 categories, which hardly represent the wide spectrum of landscapes present in Alaska (see legend and class descriptions online: https://www.mrlc.gov/data/legends/national-land-cover-database-class-legend-and-description). To specifically analyze the direct physical impact of humans on the landscape of Alaska, we used the urban classes: “21”, “22”, “23”, and “24”, corresponding to “Developed, Open Space”, “Developed, Low Intensity”, “Developed, Medium Intensity”, and “Developed, High Intensity” respectively. These categories were merged and considered as the urban areas of the dataset, which can be seen in Fig. 1B and highlighted in Fig. 1C.

Tools

Given that the most-used software for the indicated objective does not directly allow for such a proposed landscape analysis on a state-wide level due to input data size limits of approximately 50 MB (McGarigal, 2015), we attempted to compile the results using a few different options. Initially, we attempted several in-built tools in ArcGIS Pro, ArcGIS, and QGIS (including GRASS GIS). However, none of these delivered meaningful results due to a variety of reasons, e.g., discontinued tools, configuration issues, and computational size issues. GRASS GIS, with its built-in landscape analysis tools, also presents limitations in terms of the configurational errors it produces, which constrain the use of those tools. ArcGIS and ArcGIS Pro discontinued previously existing tools and plug-ins to analyze landscape metrics, limiting their use to the most recent versions. The only software allowing for the successful creation of landscape metrics was R and RStudio (version 3.3), allowing us to replicate the results Fragstats would provide. We utilized the R package ‘landscapemetrics’ (https://r-spatialecology.github.io/landscapemetrics/-Hesselbarth et al., 2019). The code we used to analyze the landscape metrics can be found in Appendix 1 (https://scholarworks.alaska.edu/handle/11122/15662). This code was used for both the desktop-based RStudio software as well as the R session in the Linux cloud-computing environment. (The only difference between the code used for the desktop and cloud-computing version was the file directory used to indicate where the data was stored and where the results should be exported).

Workflow

Our workflow (Fig. 2) starts by downloading public Open Access data (see link above in the ‘Data’ section), which was then imported into ArcGIS Pro (version 3.2). In ArcGIS Pro, we conducted several steps for data preparation to allow for the proper use of the data. This included the removal of all unused values using the ‘Set Null’ tool (part of the “Image Analyst” tools), followed by the conversion of the CRS into WGS 84 and setting the units from meters into decimal degrees (EPSG: 4326), to adhere to global CRS standards and streamline the utilized data. To utilize and merge the urban classes, the individual urban classes of the NLDC map were first extracted from the TIFF file using the “Extract by Attribute” tool and were then merged using the “Mosaic To New Raster” tool, both in ArcGIS Pro. This merged urban layer is then imported into open-source QGIS, where it needs to be converted into an ASCII (.asc) file. This resulting ASCII file can then be used for the subsequent moving window analysis in R.

In this study, we analyzed the landscape metrics Coefficient of Variation–Patch Area, Coefficient of Variation–Core Area Index, Mean Core Area, Edge Density, Mean Fractal Dimensional Index, Largest Patch Index, Number Of Patches, Patch Richness, Mean Shape Index, and Total Edge (details of the metrics can be found in the Fragstats manual and online (McGarigal, 2015) in a moving window setting. The rationale for focusing on this set of landscape metrics is to include various metrics. An overview of all landscape metrics can be found in Table 2, including the description and function of each metric.

The final results were then imported into ArcGIS Pro, where we assigned a coordinate reference system (CRS-WGS 84 EPSG: 4,326) and chose a suitable symbology for visualization. Due to the known high correlation among the computed landscape metrics (McGarigal, 2015), and for ease of comprehension, here we only display one of the computed metrics (Coefficient of Variation–Patch Area; that is, the variance of different patch areas in a given pixel as a robust measure of disturbance) in the final overview and its sub-figures in the results section below. All computed landscape metrics can be found as TIFF files in Appendix 2 (https://scholarworks.alaska.edu/handle/11122/15662).

The workflow presented here also contains an optional cloud-computing add-on. This path can be chosen if a Linux cloud computing environment is available, and the results are aimed to be achieved within a shorter timeframe. The only steps that are different for this option are in the final computational section, where the merged ASCII file must be uploaded onto the cloud, where the cloud computer can access it, in addition to a slight code modification accordingly. To make the best use of the improved performance of the cloud computing environment, the code (see Appendix 1-https://scholarworks.alaska.edu/handle/11122/15662) is suggested to be slightly adjusted so the ten different jobs (one for each landscape metric) can be distributed to different cloud computing nodes. Apart from the modification of where the data is stored for the analyses and where the results will be saved, in addition to the optional code optimization, no additional data processing steps are necessary for the cloud-computing option.

For a better overview of the whole study area, we summarized the cumulative human impact on the Alaskan landscape using 1 × 1 decimal degrees hexagons using the “Generate Grids and Hexagons” tool in ArcGIS Pro. For this summary, we used the ‘Join attributes by location’ tool in GIS to combine multiple impact layers and their underlying coverage. To further summarize the cumulative impact layers, we created a simple frequency distribution graph in MS Excel with the same color palette as the map overview with hexagons.

Longitudinal, latitudinal, and altitudinal assessment of urban fragmentation hotspots in Alaska

To identify the urban fragmentation hotspots in Alaska, we overlaid a 2 km lattice grid over the landscape metric and extracted the pixel values. As a result, we obtained 1-degree longitudinal, latitudinal, and altitudinal (meters above sea level) urban fragmentation indices that can be found depicted further below. We used integer degrees on a ‘group-by’ command to present averages for each degree along the latitude and longitude gradient of Alaska. With these figures, we attempt to show a profile in the X and Y direction and by altitude to highlight where urban fragmentations can predominately be found in Alaska thus far. This analysis was carried out in GIS and MS Excel and visualized using frequency histograms.

Direct and indirect human impact components affecting the Alaskan landscape

For the most inclusive overview of the impact that humans have on the modern Alaskan landscape, we additionally overlaid a series of GIS layers indicating direct and indirect impacts on the landscape. Figure 5 aims to visualize the sheer omnipresence and cumulative–true human impact on the landscape of Alaska. As elaborated later, Fig. 5 can still be considered a minimum estimate, as many other direct or indirect factors are either not available online as a digital dataset (e.g., traplines, ATV, powerlines, and snowmobile trails, etc.–see Table 3 in the Discussion section) or are highly correlated with other factors (e.g., houses and roofs). The hexagon summary of the cumulative human impacts is presented in Fig. 6, and the simple frequency distribution in Fig. 7.

Figure 5 displays the sum of human impacts on the Alaskan landscape. Figure 6 illustrates the number of impact layers within each hexagon, with red colors indicating a high number of impact layers and dark blue colors indicating a lower number of impact layers. It additionally contains Alaska’s National Parks outlined in light blue. Figure 7 illustrates the frequency distribution of the number of human impact layers within one decimal degree hexagons in the state of Alaska, and the color coding is the same as that for Fig. 6. All hexagons with a color other than blue represent areas with more than 11 cumulative impact layers within a 1 × 1 decimal degree area. This means that within each of the non-blue hexagons, more than 11 different layers impact the Alaskan landscape.

Longitudinal, latitudinal, and altitudinal assessment of urban fragmentation hotspots in Alaska

To outline fragmentation hotspots in Alaska, we also summarized the computed urban area metrics by longitude, latitude, and altitude. The results of that hotspot analysis can be found in Fig. 8 below. These figures aim to display where urban hotspots can be found in Alaska in the landscape and geographic location. It shows the known hotspots like Juneau, Homer, Anchorage, Fairbanks, the road system, and Utqiagvik by latitude, longitude, and altitude peaks. With climate change and rising sea levels, these urban hotpots will likely extend and shift (unlikely on a voluntary basis, especially urban hotspots on the coasts; see Marino, 2015).

Urban hotspots can be observed in Fig. 8A at longitude −150 decimal degrees for Anchorage, −148 for Fairbanks, −134 for Juneau, and −177 for Adak), in Fig. 8B at latitude 61 for Anchorage, 64 for Fairbanks, 59 for Juneau, 55 for Ketchikan, and 71 for Utqiagvik, and in Fig. 8C at altitude in meters, close to 0 for many coastal villages and Anchorage, approx. 130 for Fairbanks, approx. 400 for Healy, and 6,190 for Denali.

Qualitative evaluation of the results

Our results show the units and magnitude of human footprint explicit in the coordinate space and elevation, as per the input GIS layers and methods described. As one may expect from the Anthropocene (Dalby, 2016), from a qualitative perspective, the sheer entirety of the state of Alaska appears to be impacted by human activity in one way or another. Especially when considering atmospheric impacts like man-made CO₂ levels rises and methane (GHGs), no area is truly untouched and pristine (see also Bao, Jia & Xu, 2023 for ecological processes), IPCC reports for national coverage (Byerly, 2010), and the lack of summer sea ice on the north shore by 2050 (Wang & Overland, 2015). Still, the major urban impact hotspots can currently be found in central and southern Alaska (see Figs. 5 and 6) around major urban areas like the cities of Anchorage, Fairbanks, Juneau, Homer, and Prudhoe Bay, and along major waterways like the Yukon and the Tanana River, as well as major roadways like Highway 1, 2, 3, 4, 6, and 11. Figure 6 additionally displays urban impact hotspots on the Seward Peninsula (central-west).

The National Parks (outlined in light blue) appear to have overall fewer impact hotspots (compared to other areas). However, some also show high human impact hotspots, e.g., Denali National Park (Central Alaska), Bering Land Bridge National Preserve (on the Seward Peninsula), and National Parks in Southeast Alaska. Those include Sitka National Historic Park, Glacier Bay National Park, and Klondike Gold Rush National Historic Park (see details: National Park Service, 2025). The most northern National Parks appear to have the lowest human impact hotspots. However, it is noteworthy that some of the major human impact coldspots appear to be outside of the National Parks (e.g., see north of the most northern National Parks in Fig. 6, near the Arctic National Wildlife Refuge (on the eastern side) and the North slope on the western side of the Haul Road (Dalton Highway)). Notably, even the blue pixels in Fig. 6 (which can be considered coldspots) can include up to ten human impact layers.

Accordingly, the computed landscape metrics appear as expected, with high edge densities and low patch sizes in highly fragmented urban areas and a low number of total edges in areas with low expected landscape fragmentation. The other included metrics followed this trend.

Quantitative impact on the Alaskan landscape

The total 30 × 30 m pixel landcover classification map provided by NLCD includes 2,147,483,647 pixels, all urban classes combined add up to a total of 1,433,589 pixels, which only represents 0.067%. To provide a more realistic minimum-estimated human impact on the Alaskan landscape, we summarized all available direct and indirect human impact components presented in the section “Direct and indirect human impact components affecting the Alaskan landscape” using >475 hexagons. In reality, none of the hexagons show 0 (zero) human impact layers and are at least to a small extent impacted by humans as some impact layers cover the whole state (e.g., CO_2, Methane, and permafrost melting). Other major land management and governmental activities and acts affect predator control and trapping, as well as the homesteading act in Alaska as the “last frontier” (Anderson, 2011; Boertje, Keech & Paragi, 2010; Hull & Leask, 2000; Van Ballenberghe, 2006). Therefore, the true human impact on the Alaskan landscape is most likely close to 100%. A vital consideration supporting this argument is the non-infrastructural impact surrounding human structures. For example, hunting activities are likely to occur within a 1–2 km buffer off all roads and trails, and the same is true for bodies of water as landing pathways for floatplanes. So, even though some remote areas appear to have low human impact, the true human impact on nature and the landscape is prevalent. This finding sets a baseline and has many implications for Alaska, its management, and the U.S. National Park system overall, as well as how Alaska is understood and subsequently managed.

Data and method limitations

The NLDC map is already pixelated and thus widely inaccurate for its categories, while fragmentation layers dominate in the lack thereof. The limitations of the NLDC map also lie in the missing updates for the state of Alaska, where the contiguous US has received more recent NLDC updates, Alaska has not been updated since 2016. Additionally, with technical advancements since the last update, a usable accuracy, and a higher resolution could also provide additional insights, similar to the latest datasets provided by ESRI (https://livingatlas.arcgis.com/landcover/). Such higher-resolution products still come with the drawback of being composed of multiple tiles rather than a continuous map, which would be a valuable addition to the scientific community and the National Land Cover Database (NLCD).

The traditional Fragstats software is limited by computational aspects such as a modern operating system and a major input file size constraint, hence the reason we attempted to overcome it in this study. This allows for a more realistic assessment as a future reference to work from. To overcome the computational limitation of the leading landscape ecology analysis software and its metrics, here we successfully show several different approaches to explore pathways that fulfill the computational needs required by this dataset. By exploring several software options, e.g., in-built tools in ArcGIS Pro, ArcGIS, and QGIS (including GRASS GIS), we were able to prove that it is not possible to recreate the results for our large study area that are similar to the ones Fragstats would produce when solely working with the GIS tools outlined above.

RStudio and the ‘landscape metrics’ package are the only meaningful options that allow computing landscape analyses using larger input files. However, this approach is quite time-consuming. One option to overcome this time-intensive approach is to run the code on a cloud computer. However, that comes with its own constraints and hurdles, including administration, running costs, code modifications, etc.

Each landscape metric for the merged urban landcover class took approximately 3–5 days to run on a personal computer (using the R code provided in Appendix 1-https://scholarworks.alaska.edu/handle/11122/15662; and a PC with a 13^th Gen Intel i7-13700 processor). This limits the computation of a high number of metrics and landcover classes on a personal computer within a short period of time. For a more exhaustive study and the inclusion of, e.g., other landcover classes and up to 50 landscape metrics, elaborate cloud computing pathways are recommended.

For the cumulative human impact layers, we attempted to compile an inclusive set of data, yet not all the data we initially anticipated to include were digitally available. Table 3 provides an overview of the layers we anticipated to include but were not able to obtain and thereby include in this study. For an even more complete and inclusive study of the actual human impact on the Alaskan landscape, those layers could provide additional insights. Adding archeology layers would greatly help in that effort but none are readily available and on a meaningful scale.

Closing statement

Alaska lacks a proper quantified digital assessment of the human footprint and impact. Here we present the first workflow and product to do so. From our experience, using R/RStudio is the most effective, yet time-consuming and only realistic approach to analyzing landscape metrics on larger scales with low pixel size and high resolution. The leading software for landscape metrics analysis (“Fragstats”) is most efficient for smaller study areas, whereas integrated tools in GIS still contain algorithm problems and bugs and are therefore not recommended at this stage. This project has shown that landscape metrics can be computed for an area as large as the whole state of Alaska with a high-resolution input image of 30 × 30 m, as a proof-of-concept. While the overall human footprint was found to be excessive for all of Alaska, it shows a promising outlook for more work on such high-resolution scales and large landscape areas. Future studies can be more exhaustive with the applied methods and analyze all available landcover classes and more landscape metrics. For such scenarios, cloud computing workflows are suggested, where the same R code, as presented here in Appendix 1 (https://scholarworks.alaska.edu/handle/11122/15662), can be utilized.

When looking at the computational outcome, it is clear that the physical human footprint needs to be covered better by the existing best-available official data layers for Alaska and, thus, in public discourse or policy. Looking at a more reality-based and holistic approach, Alaska is far from a pristine state, with a massive impact in just the last three decades–the acceleration–. This reality is not present yet in any wildlife, habitat, or landscape and freshwater water table management across agencies and actors, including the U.S. National Park Service. Relevant policies feature instead the ‘originalism’ and are widely based on the ‘current’ and outdated landscape management scheme. In times of man-made climate change, Alaska is not getting more pristine and wilder, nor is the U.S. or its protected areas. The current trend does not indicate a fair and earnest sophistication or apparent vision to reach a sustainable landscape and resource management. Our findings have major implications for native lands, as well as nationally protected areas like the U.S. National Park system and its well-being and meaning. It is of national and international value. With a greater understanding of human impact on the landscape, we as a society can utilize this understanding to manage the Alaskan land more sustainably and potentially set aside land excluded explicitly from direct human impacts, for increased wilderness on this planet.