Permafrost in the northern high latitudes is undergoing rapid changes driven by enhanced atmospheric warming at roughly four times the mean global rate (Rantanen et al., 2022). Complex environmental changes that accompany degrading permafrost include widespread earlier spring thawing and lengthening of the thaw season, shifts in seasonal snow cover properties, contrasting wetting and drying patterns, vegetation greening and browning, and increasing disturbances (Heijmans et al., 2022; Foster et al., 2022). Permafrost degradation, which involves active layer thickening, may lead to ground surface deformation (Du et al., 2019). In addition, thawing permafrost could potentially mobilize a vast reservoir of soil organic carbon previously stabilized in perennially frozen ground, resulting in accelerated soil decomposition and greenhouse gas emissions that may further amplify global warming (Turetsky et al., 2020; Schuur et al., 2022; Miner et al., 2022).

Active layer thickness (ALT), defined as “the thickness of the layer of the ground that is subject to annual thawing and freezing in areas underlain by permafrost” (Lewkowicz et al., 2025), is an essential climate variable for monitoring permafrost degradation (Smith and Brown, 2009). Accurate mapping of ALT spatial and temporal variability is critical for understanding impacts of climate change and disturbance on the permafrost energy balance, terrestrial water budget, organic matter decomposition, and land-atmosphere carbon exchange (Zhang et al., 2005; Schaefer et al., 2015; Garibaldi et al., 2022; Liu et al., 2024). ALT spatial patterns are closely linked to ambient air temperature at regional scales while demonstrating large spatial heterogeneity at local scales due to complex interactions with precipitation, vegetation, slope, aspect, soil and snow properties, topography, and disturbance (Leibman et al., 2012; Widhalm et al., 2017; Loranty et al., 2018; Chen et al., 2019; Garibaldi et al., 2022).

Despite the importance of permafrost landscapes in monitoring and projecting Arctic-boreal system changes, there is only limited understanding of high-resolution (e.g., sub-meter to 10 m) ALT patterns and underlying driving factors over the highly heterogeneous permafrost regions due to a paucity of high-resolution ALT measurements. It is also unclear how local ALT patterns and its associated governing mechanisms scale up to the broader landscape when quantified at coarser spatial resolutions (e.g., 10 to 1000 m) using global satellite observations and regional process models (Yi et al., 2020). However, a better understanding of the high-resolution patterns, mechanisms, and their scaling effects is needed to clarify permafrost vulnerability and associated feedbacks to climate change, and to improve representation of the hydrological, ecological, and biogeochemical processes of Arctic ecosystems in global Earth system models as well as remote sensing inversion algorithms (Koven et al., 2013; Chen et al., 2019; Hantson et al., 2025).

Remote sensing of ALT at high resolutions ranges from relatively direct electromagnetic retrievals from low frequency ground penetrating radar (GPR) to more indirect measures of surface topographic features from Light Detection and Ranging (LiDAR) (Gangodagamage et al., 2014) and soil freeze-thaw (FT) driven land surface deformations from Interferometric synthetic aperture radar (InSAR) measures (Schaefer et al., 2015). The distribution of ALT over ice-wedge polygon landscapes was empirically inferred using relevant topographic features quantified from LiDAR surface elevation measurements and data fusion approaches (Gangodagamage et al., 2014). In addition, the active layer heaves when frozen and subsides when thawed due to denser liquid water than ice. The seasonal vertical ground movement measured using InSAR is thus related to the ALT and ice/water content (Schaefer et al., 2015). Low-frequency (e.g., L- and P-band) microwave measurements are capable of penetrating through vegetation and soil layers and show strong potential for mapping active layer properties, including soil moisture and FT dynamics, organic matter content, and ALT (Tabatabaeenejad et al., 2014; Bakian-Dogaheh et al., 2025). This is due to the strong microwave sensitivity to the changes of soil dielectric properties, which are affected by soil moisture, texture, and freeze/thaw state (Kneisel et al., 2008; Du et al., 2019). In particular, the contrast in dielectric permittivity at the interface between thawed and frozen soil layers may lead to measurable microwave backscattering or reflection signals for estimating ALT (Kneisel et al., 2008). For example, retrieval algorithms were developed by exploiting airborne P-band radar data for ALT mapping in Alaska (Chen et al., 2019), which showed favorable retrieval accuracy for ALT up to ∼60 cm. Another study on post-fire recovery in tundra regions further confirmed the unique sensitivity of low-frequency airborne radar in detecting the relatively fine-resolution heterogeneity in active layer conditions and post fire recovery that may be missing from coarser-resolution and higher-frequency satellite observations and model simulations (Yi et al., 2021). More recently, advanced physics-based computational radar retrieval algorithms have been developed to map the active layer soil organic matter and soil moisture profile of the Arctic-foothills tundra using P-band radar data (Bakian-Dogaheh et al., 2022). In sum, the remote sensing techniques provide information necessary to infer or model ALT.

Considering its dependence on surface conditions (Kelley et al., 2004), ALT can also be indirectly inferred from optical vegetation observations and relatively high-frequency radar backscatter signals considering the ALT dependence on surface conditions using regression analysis (Kelley et al., 2004; Gangodagamage et al., 2014; Widhalm et al., 2017). Recent machine-learning based studies using airborne hyperspectral imaging data achieved meter-level ALT predictions over Interior Alaska with favorable accuracy (Zhang et al., 2021). Thus, complementary information available from multi-sensor and multi-resolution remote sensing provides the basis for developing more effective models and ALT retrievals capable of resolving local patterns and environmental linkages spanning Arctic-boreal permafrost landscapes (Brodylo et al., 2024; Hantson et al., 2025).

Although remote sensing is a viable option for inferring ALT, it remains unclear how local ALT patterns and their governing mechanisms (e.g., sub-meter to 10 m) scale up to broader landscape footprints (e.g., 10 to 1000 m) represented by global satellite observations and Earth system models. Recent studies have begun to resolve multi-resolution ALT patterns using multi-sensor remote sensing (Brodylo et al., 2024; Hantson et al., 2025), but these efforts have been limited to Interior Alaska (Brodylo et al., 2024) and the Seward Peninsula (Hantson et al., 2025). For the Arctic-foothills tundra environment, dense in-situ sampling is extremely sparse, but is needed as a ground-truth benchmark for evaluating remote sensing retrievals and to quantify how uncertainty changes with spatial resolution. In addition, there is also a lack of multi-scale observations integrating field samples, high-resolution drone/airborne images, and satellite observations. Yet such measurements are essential to resolving how ALT patterns scale across resolutions. This study addresses this gap through a field experiment that integrated in-situ measurements with multi-scale remote sensing observations for the understudied Arctic tundra.

The objectives of this study are to: (1) assess ALT spatial heterogeneity and scaling properties from 0.1 to 1000 m resolution within the tundra foothills region of the North Slope of Alaska; and to (2) assess the underlying environmental controls on ALT patterns manifesting at different resolutions and the potential influence of spatial resolution on regional permafrost remote sensing and modeling uncertainty.

Our study focused on the Imnavait Creek (68.6167° N, 149.3167° W) area within the Alaskan North Slope tundra foothills region (Fig. 1). The study region experiences a cold climate with a mean annual temperature of -7.4 °C, ranging from an average of -17 °C in January to 9.4 °C in July (Schramm et al., 2007). Annual precipitation averages 340 mm, two-thirds of which falls as light rainfall during the summer (Schramm et al., 2007). The dominant vegetation consists of water-tolerant plants like tussock sedges and mosses, grasses and low shrubs (Schramm et al., 2007). Vegetation and soil patterns vary with slope, aspect, and drainage conditions, exhibiting a patchy distribution across the study domain (Hinkel and Nelson, 2003). The region is also characterized by diverse glacial landforms, including deposits, stream networks, and bedrock outcrops (Hinkel and Nelson, 2003). The region has gently rolling terrain, with elevations ranging from ∼750 to 980 m. Vegetation and soil patterns vary with slope, aspect, and drainage conditions, exhibiting a patchy distribution across the study domain (Hinkel and Nelson, 2003). The region is also characterized by diverse glacial landforms, including deposits, stream networks, and bedrock outcrops (Hinkel and Nelson, 2003). Increasing ALT was observed from both in-situ measurements at the Circumpolar Active Layer Monitoring (CALM) sites (https://www2.gwu.edu/~calm/data/north.htm, last access: 1 July 2025) and regional ALT records at 1 km resolution, which were generated using machine learning (ML) by combining in situ ALT observations with a suite of observational biophysical variables (Liu et al., 2024).

The local study area consisted of four intensively sampled field plots (Plot 3, Plot 4, Plot 5, and Plot 6 in Fig. 1; 90 m × 90 m each) distributed across an elevation gradient along west facing hill slopes, and surrounded by a larger (5 km × 5 km) landscape domain used for analyzing ALT across different resolutions (Fig. 1; Table 1). The field plots are dominated by tussock-forming sedges and moss-lichen mats, along with scattered dwarf birch shrubs, and riparian grasses (Figs. 1–2, S1 in the Supplement). The common species identified near the ALT sampling points include dwarf birch (Betula nana), alpine blueberry (Vaccinium uliginosum), black bearberry (Arctous spp.), crowberry (Empetrum nigrum), and Arctostaphylos uva-ursi, among other less-recognizable species. Our soil coring samples indicated high organic matter content of the topsoil (0 to 10 cm depth) for Plots 3, 4 and 5, with the respective values of 77.3 %, 75.1 %, and 71.8 %, contrasting with a relatively low value of 45.5 % for Plot 6. In-situ measurements showed that ALT varies with terrain, vegetation cover types, and surface water conditions, with overall greater active layer thickness found at the higher-elevation Plot 6 (58 cm) than at the other plots (39 to 49 cm).

Plot 3 and Plot 4 are located on west-facing downhill slopes characterized by gradual elevation changes (10 to 15 m across the plot), small water tracks, and scattered glacial erratics. Plot 3 is crossed by multiple drainage features that are transverse to the Kuparuk River (Fig. 1). These features are evident in the topography as well as the vegetation, with alternating bands of dwarf willow, grasses, and tussocks (Fig. S1a). Standing water is present in many places, particularly along the drainage channels. Plot 4 is more homogeneous in vegetation character and height, consisting mostly of grass tussocks and moss, with some larger glacial erratics present (Fig. S1b).

Plots 5 and 6 are on the east side of the access road and are more level than Plots 3 and 4. Plot 5 is located at a valley bottom and is partitioned by Imnavait Creek, which winds directly through the middle of the plot. Shallow drainage channels transverse to Imnavait Creek are found on the west side of the plot, while the east side is upland and drier. Vegetation cover in Plot 5 follows the typical mix of grasses, mosses, and sedges, with some moss forming humps about 20 cm tall, while shrub density and height increase close to the creek. Along Imnavait Creek there are taller (0.5 to 1.0 m) birch shrubs and more grasses (Fig. S1c). Plot 6 is characterized by widespread subsurface rocks (Fig. 1) and overall short-statured grass-sedge tussocks and mosses, lichens, small bunchgrasses, Arctous spp., and alpine blueberry, interrupted by areas of bare rock and gravel (Fig. S1d). Bare areas are common in the center of the plot.

As part of the NASA Arctic Boreal Vulnerability (ABoVE) field campaign, our study used the ABoVE standard projection, which is in the Canada Albers Equal Area projection, for data visualization and analysis (Loboda et al., 2017).

Recent studies have focused on resolving multi-resolution ALT patterns and their environmental linkages using multi-sensor remote sensing (Brodylo et al., 2024; Hantson et al., 2025). Compared with the existing studies, our study differs in three respects, which include (a) advancing from transect-based (1-dimensional) sampling to intensive 2-dimensional grid-cell sampling, which provided a more comprehensive representation of spatial variability to support multi-resolution analysis; (b) characterizing ALT patterns in the unique Arctic-foothills tundra environment, thereby complementing existing studies for Interior Alaska (Brodylo et al., 2024) and the Seward Peninsula (Hantson et al., 2025); and (c) quantifying resolution-dependent uncertainties in ALT retrievals, offering additional support for improved interpretation of multi-resolution ALT products derived from remote sensing and process-based simulations for the changing Arctic.

This study represents a multi-resolution assessment of ALT spatial heterogeneity within the Arctic foothills of northern Alaska, a region characterized by continuous permafrost cover and relatively complex terrain within the Arctic domain. We employed a ML framework for ALT mapping ranging from local plot (0.1 m) to landscape (10 m) and regional (30 to 1000 m) levels using multi-resolution observations, including intensive field sampling, and drone, airborne and satellite remote sensing.

Our study showed that vegetation, soil, surface water, and topography are all key drivers of meter-resolution ALT patterns, while terrain topography becomes increasingly dominant in controlling ALT distribution at coarser resolutions. Resolution effect analysis further revealed a quadratic functional relationship describing the growth of resolution-dependent uncertainties in ALT mapping, indicating a rapid increase in uncertainty at sub-meter resolutions, a moderate error increase at intermediate (1 to 30 m) resolutions, and a more conservative increase at coarser (30 to 1000 m) resolutions representative of many global satellite and model based assessments. For areas where soil, vegetation, and terrain conditions are markedly different from the Arctic foothills tundra, their ALT scaling effects can be independently quantified through similar approaches from this study.