Introduction

Arctic and boreal wetlands, including peatlands, lowland tundra, and other waterlogged ecosystems, are significant sources of methane (CH4), yet their contributions to the global budget remain poorly constrained, with estimates ranging from 8 to 55 Tg CH4 yr−11,2,3,4. Considerable disagreement exists between bottom-up approaches5 (field measurements and process-based modeling) and top-down approaches (atmospheric inverse modeling), particularly regarding natural CH4 sources and emissions from permafrost regions2. Bottom-up estimates of CH4 emissions at regional or global scales require both flux observations from field measurements, and spatial representation of CH4 sources and sinks, often derived from land cover datasets1,6,7,8. Each of these components contribute to the overall uncertainty of CH4 emission estimates9 in a region with relatively sparse flux measurement density4,10,11. Recent studies have emphasized the importance of thematic detail (wetland type) for differences in CH4 flux (FCH4) rates, with larger emissions from minerotrophic wetlands, such as fens, and smaller emissions from ombrotrophic wetlands, such as bogs3,11,12,13. Fens are groundwater-fed, nutrient-rich wetlands, while bogs are precipitation-fed, nutrient-poor systems. Boreal and Arctic peatlands are often mosaics of these wetland types, where low-lying areas tend to be fens and elevated microforms are more often bogs. Despite the established link between regional emissions and wetland extent and type14,15, the sensitivity of regional to global CH4 budgets to uncertainty in the distribution of CH4-emitting areas remains poorly understood1,16.

Land cover classifications used to estimate areas of CH4 emission and uptake13,17,18,19,20,21,22,23,24 may vary widely in spatial and temporal resolution, thematic detail, and methodological approaches, contributing to uncertainty in CH4 budgets9. Over recent decades, spatial resolution—the size of the pixels or grid cells that represent different land cover types—has significantly improved for wetland and lake extent maps25. However, resolutions are often still too coarse to capture the high heterogeneity of Arctic and boreal wetlands14,15. Fine-scale landscape features—such as ecotones, narrow channels, patterned ground, and isolated vegetation patches—require spatial resolutions commensurate with their characteristic dimensions (often <10 m) in order to be accurately represented, but these are generally obscured or simplified in coarser datasets15,26,27,28. For example, ecotonal transition zones at lake margins and shore vegetation often serve as CH4 emission hotspots29. Similarly, areas of CH4 uptake—underrepresented in current observations—require high-resolution mapping of wetland–upland boundaries to improve estimates of CH4 sinks in heterogeneous wetlands30. Another key consideration is that the wetland classification scheme used may not differentiate CH4 emissions, often grouping areas with distinct flux dynamics. Global vegetation maps often misclassify Arctic landscapes, with broad categories such as “grassland” failing to distinguish between wet and dry tundra, which differ in CH4 emissions31. Differentiating these wet and dry tundra landscape types can be challenging due to their similar structural and spectral characteristics32,33 but is necessary to improve high-latitude CH4 budgets34.

Resolving fine-scale landscape details, essential for FCH4, requires high spatial resolution data. While datasets at various spatial resolutions (20−1000 m) from spaceborne sensors like MODIS, Landsat, or Sentinel-2 are freely available on a global scale, very high spatial resolution data (≤2.5 m) from airborne sensors is often restricted to smaller regions. Commercially available products from providers like Planet Labs or Maxar offer high-resolution imagery but acquiring and processing these datasets for extensive areas, such as the entire circumpolar region, is time-consuming, resource-intensive, and currently cost-prohibitive. These challenges result in (1) studies focused on smaller regions that may not fully capture landscape variability14,35,36,37, or (2) large-scale analyses using coarser-resolution imagery, which fails to represent fine-scale features critical for FCH41,7,16. These limitations underlie the need to determine a minimum spatial resolution required to accurately capture landscape variability and FCH4 dynamics without compromising spatial coverage.

While spatial resolution and thematic detail are important considerations for pan-Arctic and boreal CH4 budgets, there is a trade-off between accuracy and practicality for map resolutions in land cover products used in FCH4 upscaling. Here, we synthesize very high nominal resolution (<2.5 m pixel size) land cover classification maps in conjunction with upscaled chamber-based measurements of FCH4 at seven study sites from the Arctic and boreal region. Study sites included: Tiksi, Russia; Seida, Russia; Kilpisjärvi, Finland; Pallas, Finland; Utqiaġvik, United States; and two sites in Scotty Creek, Canada (North and South; Fig. 1). We assess how regional FCH4 estimates are affected by spatial resolution by resampling land cover maps from fine to coarse scales (Fig. 1; Supplementary Fig. 1) and quantify how resolution-dependent errors relate to wetland size, type, and landscape fragmentation.

Fig. 1: Study Sites and Methodological approach.
Fig. 1: Study Sites and Methodological approach.
Full size image

(Left) site locations used in this synthesis (1 – Scotty Creek, Canada; 2 – Utqiaġvik, United States; 3 – Tiksi, Russia; 4 – Seida, Russia; 5 – Kilpisjärvi, Finland; 6 – Pallas, Finland). Coloration within the land region shows WAD2M wetland distribution for August, 2018. The blue line shows the boundary for the Arctic region as defined by the Conservation of Arctic Flora and Fauna (CAFF) working group. The yellow line shows the treeline. Dotted line shows the Arctic Circle (66° 34' N). Gray regions denote areas with no wetlands. (Right) example resolution coarsening of LC map of Tiksi, RU.

Results and discussion

Wetland distribution and fluxes with coarse resolution mapping

The study sites are distributed across the pan-Arctic and boreal, including North America, Russia, and Fennoscandia and range from boreal and sub-Arctic forest to Arctic tundra (Fig. 1). Wetlands were present across all the study regions; the median wetland fraction for all regions was ~35% (4–50%; Supplementary Table 1) at nominal resolution (Fig. 2a). Land cover was grouped to 6-7 classes at each site. Wetland classes were harmonized to represent total wetlands at all sites. At five sites (Pallas, Tiksi, Seida, and two sites at Scotty Creek), fen and bog classes were identified. Across these sites, bogs comprised approximately 18% and fens around 14% of mapped extents. For most regions (six of seven sites), median wetland extent decreased during resampling to coarser resolutions from 35% at the 2.5 m reference resolution to 32% coverage at 25 m resolution and 20% at 100 m resolution (Fig. 2a). By 250 m spatial resolution, the median wetland fraction had fallen by nearly half, to 22% coverage. At 1250 m resolution, the median wetland fraction across the landscapes reached zero (Fig. 2a).

Fig. 2: Interaction of spatial resolution, wetland extent and FCH4 estimates across regions.
Fig. 2: Interaction of spatial resolution, wetland extent and FCH4 estimates across regions.
Full size image

a Regional wetland area (% of mapped area) as a function of spatial resolution; b Linear regression of upscaled regional FCH4 (expressed as g C ha-1 h-1) and regional wetland area (%) at the nominal resolution (n = 7). Shaded area represents confidence interval for the fitted trend; c Magnitudes of total regional FCH4 (expressed as g C ha−1 h−1) by site at increasingly coarse resolutions. Error bars represent standard error of flux measurements.

A strong positive correlation (R2 = 0.6) was found between regional upscaled estimates of CH4 emissions and wetland area across the study sites (Fig. 2b). However, the unexplained variability in FCH4 suggests that other factors—related to the thematic detail of land cover classification, such as variations in climate, vegetation community, permafrost presence, and nutrient regime—also contribute significantly to the observed emissions. Among the study sites, the region with the smallest wetland area (4.7%), Kilpisjärvi in sub-Arctic Finland, exhibited the lowest FCH4 (0.16 ± 0.32 g C-CH4 ha−1 h−1), while South Scotty Creek in the boreal forest of the Taiga Plains ecozone of the Northwest Territories, Canada, with 50% wetlands, exhibited the highest FCH4 (21 ± 4.8 g C-CH4 ha−1 h−1; Fig. 2b, c). The difference in estimated wetland extent at coarser resolutions had strong effects on the net regional FCH4 emissions (Fig. 2c). In four of seven study regions, using coarser resolution classifications led to a switch from net CH4 emissions to net uptake of atmospheric CH4, particularly at or above 1 km grid size, while in other sites regional emissions decreased by 75.5–100% (Fig. 2c). The notable exception to decreasing FCH4 using coarser resolution imagery was South Scotty Creek (Fig. 2c). Emissions from this region increased by 142% at the maximum pixel size, from 21 ± 4.8 g C-CH4 ha−1 h−1 to 50 ± 13.8 g C-CH4 ha−1 h−1, indicating that though wetland extent is driving FCH4 (Fig. 2b), the non-contiguous spatial patterns of small wetlands are lost at scale.

Given the varying directional shifts in regional FCH4 emissions among sites with decreasing spatial resolution (Fig. 2c), we calculated the absolute values of regional FCH4 and expressed them as percent deviations from the nominal resolution FCH4 (see Methods) to quantify the magnitude of scaling error introduced by coarser resolutions. Across all sites, deviations remained small at fine resolutions, with changes of only 6 % observed up to a 10 m pixel size (Fig. 3). However, beyond this point, error propagated rapidly: at 50 m resolution, FCH4 estimates had already diverged by 25 ± 19%. Coarser resolutions similar to many pan-Arctic and global land cover datasets—250 m and 1000 m—introduced substantial deviations of 81 ± 67% and 137 ± 75%, respectively (Fig. 3), well exceeding the average uncertainty (standard error) of the flux measurements at the highest resolution (43.7%; Fig. 2c). These results emphasize the critical need for geospatial data at spatial resolutions commensurate with emission source heterogeneity to minimize upscaling errors.

Fig. 3: Percent deviation in FCH4 with decreasing spatial resolution.
Fig. 3: Percent deviation in FCH4 with decreasing spatial resolution.
Full size image

Points and error bars represent means across sites ± standard error. Percent deviation is defined as the absolute change in FCH4 expressed as a percentage of the FCH4 at the nominal reference resolution. Higher values indicate larger error.

FCH4 percent deviations were strongly linked to underlying landscape structure (Fig. 4). We characterized landscape structure using: (1) wetland patch size, quantified by mean patch size and the standard deviation of patch area (higher values indicate larger and more size-variable patches); and (2) fragmentation, represented by the landscape division index (higher values reflect more fragmented landscapes with numerous small, isolated patches). These metrics governed the sensitivity of FCH4 estimates to decreasing spatial resolution. For example, the Kilpisjärvi region exhibited a CH4 source-to-sink transition at a finer resolution (50 m) relative to the other sites (Fig. 2c), and the most rapid percent deviation from reference resolution FCH4 (Supplementary Fig. 2). Kilpisjärvi had the highest landscape division index (Fig. 4a), smallest mean wetland patch areas (Fig. 4b), and lowest standard deviation of wetland patch area (Fig. 4c), consistent with the high micro- and meso-topographic variation at the site36. As resolution coarsened, smaller high CH4-emitting patches were increasingly eliminated during pixel aggregation, leading to an apparent net CH4 sink (Fig. 2c).

Fig. 4: Geomorphological drivers of resolution sensitivity.
Fig. 4: Geomorphological drivers of resolution sensitivity.
Full size image

Linear regressions (n = 7) of resolution sensitivity rank against (a) mean landscape division index, mean wetland patch are (b), and (c) standard deviation of patch area within each region. Resolution sensitivity rank indicates the relative sensitivity of a region’s FCH4 signal to spatial resolution coarsening, ranked from 1 (most sensitive) to 7 (least). Ranks were derived as the area under the deviation-from-reference curve for each region (Supplementary Fig. 2). Areas were calculated using the trapezoidal rule over log-transformed resolution steps to normalize distance between resolutions. Shaded regions represent 95% confidence intervals.

Conversely, regions like the South Scotty Creek site, with more aggregated wetlands (Fig. 4a) and larger wetland patch size (Fig. 4b), were less sensitive to patch loss at coarser resolutions, but instead exhibited inflated CH4 estimates as wetland areas were overrepresented (Fig. 2c). Regions with smaller, more fragmented patches are generally more sensitive to FCH4 errors introduced by coarse-resolution geospatial products, which is particularly important because these wetland types often contribute disproportionately to CH4 emissions. While fragmentation can be influenced by classification method (e.g., pixel-based vs. object-based approaches and post-classification processing), it primarily reflects the underlying wetland types present. These classification effects are most relevant at the native resolution, where changes in CH4 emission magnitudes (Fig. 2c) and FCH4 percent deviation (Fig. 3) were minimal.

Representation of wetland type and other methane-emitting areas

Thematic detail in wetland categorization is important because of the often markedly different CH4 emission rates among wetland types such as bogs, fens, and permafrost wetlands3,11. Here, fens and bogs contributed equally to total CH4 emissions at the reference resolution, but bog emissions outweighed fen emissions of CH4 at >500 m spatial resolution (Fig. 5), with the contributions at the coarsest resolutions (>2.5 km) driven by the site that retained substantial wetland area (i.e., South Scotty Creek; Fig. 2c). Analysis of landscape metrics revealed clear structural differences between bogs and fens that drive this. The landscape division index was high for both wetland types (bogs: 0.9976; fens: 0.9986), indicating extensive fragmentation overall, though fens exhibited slightly higher division, suggesting a greater degree of patch isolation. Additionally, bogs exhibited larger mean patch sizes (24.13 m² for bogs; 17.87 m² for fens) and greater variability in patch size (standard deviation: 2.01 for bogs; 1.59 for fens), indicating a wider range of patch sizes within bog systems. Collectively, these metrics suggest that bogs form larger and more spatially extensive patches, whereas fens tend to consist of smaller, more uniformly sized, and highly fragmented patches embedded within a more heterogeneous landscape matrix. These morphological differences likely explain why bog CH₄ emissions were less affected by decreasing spatial resolution, as their larger, more extensive patch structures were better preserved during coarsening.

Fig. 5: Mean proportional contributions of bog and fen to total FCH4 emissions across spatial resolutions.
Fig. 5: Mean proportional contributions of bog and fen to total FCH4 emissions across spatial resolutions.
Full size image

Bars are stacked by wetland type. Error bars indicate ± standard error of each type’s proportional contribution within each resolution.

Despite their importance to FCH4, current land cover classifications often underrepresent small, morphologically complex wetland features and aquatic systems, such as polygon troughs over degrading ice wedges, graminoid meadows, aquatic macrophyte beds, and ponds. Lakes, which constitute an average of 1.7% of the reference maps, represent an additional CH4 source in the upscaled FCH4 estimates as these areas can be significant local CH4 emitters3,38,39,40,41. Due to a lack of in-situ measurements, values for lake fluxes were taken from the BAWLD-CH4 database3, corresponding to median “medium-sized peatland lakes”. Even though lakes generally lack the same level of fragmentation and intricate morphology as wetlands, small features like shallow water-land edges or areas of emergent aquatic vegetation are hotspots for CH4 emissions and therefore critical to capture29,42,43,44. Streams are often too sinuous to be accurately mapped in large-extent land cover classifications but may also release substantial amounts of CH445.

Implications for CH4 budget estimates

Wetland distribution across the Arctic and boreal landscapes may be more extensive and densely concentrated than suggested by existing coarse-resolution wetland maps (Fig. 2a, Supplementary Fig. 3a–d), with significant implications for regional CH4 emission estimates (Fig. 2c). Many current wetland maps may omit wetland areas or types in regions characterized by high spatial heterogeneity and/or where wetlands are not the predominant land cover type (Fig. 2a). However, they also likely overestimate wetlands in wetland-rich areas when coarser-resolution, continental- to global-scale wetland products are used. In this study, wetland extent dramatically increased for wetland-rich South Scotty Creek at coarser resolutions, doubling regional CH4 emissions at the coarsest resolution (Fig. 2c). Comparisons between the reference wetland maps used here and widely used land cover products revealed a 53% underestimation of wetland extent across the three Arctic sites (Supplementary Fig. 3c), while wetland extent in the four boreal and sub-Arctic sites was approximately twice that of the reference maps (Supplementary Fig. 3c). This suggests that current estimates of wetland CH4 emissions may be overestimated in boreal regions and underestimated in Arctic regions33,34, particularly at coarse resolutions (Supplementary Fig. 3d). However, these findings are based on a limited number of sites, and broader spatial patterns will require analysis across a larger sample.

Our analysis indicates that mapping resolutions should be near 25 m to limit uncertainty in CH4 emissions to ~13 ± 9% from mapping-related error in wetland distribution (Fig. 3). Using coarser spatial resolution introduces error related predominately to wetland size but also to wetland type - which can also substantially alter emission estimates (Figs. 3 and 5)15. While remote sensing of wetlands has advanced substantially in recent years, global-scale wetland maps are often still produced at relatively coarse spatial resolutions22,46,47. In contrast, local and regional maps already reach much finer resolutions with suitable thematic detail but limited spatial extent21. Newer global land cover products at 30 m resolution have improved wetland mapping but still lack the thematic detail needed for CH4 budget estimation, particularly for high-latitude wetlands23,48. Arctic region datasets at the 10 m resolution can either have similarly limited thematic detail24 or do not include the Boreal region49, though the latter represents a significant advancement in Arctic wetland mapping for FCH4 upscaling and provides a strong foundation for extending similar approaches to the broader Arctic-boreal region. Although the flux measurements used here are restricted to the growing season and do not resolve potential variability from seasonality, the growing season typically represents the period of the highest biogeochemical activity and FCH4 in permafrost ecosystems. Consequently, while our recommendations are most directly applicable to the growing season, and we acknowledge the potential for substantial non-growing season emissions50,51, the contribution of permafrost wetlands to annual CH4 budgets is likely significant and may extend well beyond the growing season.

To improve bottom-up estimates of Arctic and Boreal region CH4 budgets, critical advances are needed: (1) more detailed Arctic-boreal geospatial data (≤25 m to limit resolution related error to 13 ± 9 %) with FCH4-relevant themes (moisture and nutrient regime) and (2) improvements to process-based modeling implementation. Land cover classifications that base wetland classes on hydrological inputs and corresponding nutrient conditions, using vegetation type as a proxy, may more effectively represent distinct FCH4 dynamics across regions with limited spectral variability. Mapping and classifying wetland types with remote sensing requires additional information on soil drainage and peat presence that does not always correlate with current vegetation52. Regions with extensive field validation and very high-resolution imagery, as synthesized here, may aid in developing and validating new, higher resolution large-extent wetland products. In regions where wetland extent information is available only at coarse scales, there are limitations in accuracy for global and pan-Arctic and boreal soil and wetland products. This error, associated with coarse wetland representation, can be propagated in wetland data-fusion products and hinders the detailed representation of wetland extent and corresponding CH4 emissions (Fig. 2c), which ideally should be captured at a resolution near 25 m to limit error in CH4 budgets (Fig. 3). Therefore, process-based CH4 models operating on coarser scales (>1 km) should incorporate sub-grid tiling or statistics to account for the extensive distribution of smaller wetland areas across the Arctic and boreal region.

While the highest spatial resolution is desirable, practical constraints necessitate balancing resolution with available resources. Earth observation data from ESA’s Sentinel satellite constellations provide higher spatial and temporal resolution compared to single-platform missions such as Landsat. Furthermore, their fully open-access model offers broad accessibility, in contrast to commercial high-resolution constellations (e.g., Maxar, Planet Labs), making Sentinel data particularly valuable for global scientific monitoring of phenological dynamics. Advances in multispectral and radar satellite capabilities have also improved the classification of peatland types (e.g., rich vs. poor fens), moisture gradients, and topographic context. These enhanced sensing capacities allow for more accurate delineation of wetland heterogeneity, particularly in ecotonal zones that are often misrepresented or omitted in lower-resolution products. However, it is important to recognize that wetlands and peatlands remain challenging to classify, as their land cover types and vegetation communities are often highly patterned and fragmented26, and their spectral properties vary seasonally with changes in moisture conditions and vegetation phenology53. Nevertheless, processing capabilities are advancing rapidly, enabling the use of adequate resolution remote sensing datasets for global-scale land cover mapping54,55, including applications for monitoring global surface water dynamics and mapping wetland extents56. New directions include machine learning-based upscaling techniques for land cover detection or change trend analysis57,58 and AI-based Foundation Models59 that will increasingly enable rapid analysis of high-resolution, near-real time observation data from highly dynamic environments such as wetlands. Our study underscores the importance of utilizing spatially and thematically high-resolution wetland maps to capture Arctic and boreal FCH4 at sufficient detail, and with lower uncertainty, to inform carbon cycle models. The important discussion of to what extent the Arctic and boreal region C sink and CH4 source contribute to the global C cycle and how they change in a rapidly warming Arctic4,10 builds on our ability to successfully map and classify wetland distribution and type. In addition to mapping these remote areas, additional aircraft campaigns focused on atmospheric CH4 measurements can help to validate the spatial distribution of CH4 emissions60,61,62.

Conclusions

High-resolution wetland extent and category data are critical for accurately estimating CH4 emissions in northern high latitudes. A spatial resolution near or finer than 25 m for global or pan-Arctic and boreal products should be used, although some regions with high landscape heterogeneity may need even finer spatial resolutions. While techniques involving aggregation to coarser resolutions while reporting subgrid land cover fractions offer methods for reducing input parameter size while retaining representativity, they rely on accurate underlying data that must be at the proper spatial resolution needed to describe land cover morphology. Similarly, spectral unmixing techniques on coarser products that provide fractional coverage of land cover type within a pixel must be able to provide fractional estimates that match that of higher resolution products while also differentiating similar land cover types that are biogeochemically relevant for CH4 (i.e., fens, bogs and other wetland types).

There are numerous thematic peatland and wetland classification systems, each employing a variety of methods for interpreting remotely sensed data. It is essential to clearly describe the classification criteria used, as there is a clear need for greater standardization in this field. Several current Earth observation datasets support the recommended resolution (e.g. Landsat, Sentinel) but must be supplemented with additional geospatial information to identify wetland types (e.g., bogs, fens) crucial for FCH4 processes. Estimates of FCH4 derived solely from surface inundation or generic wetland classes may overlook key landscape features, leading to significant errors and uncertainty. Despite the growing number of regional and global bottom-up FCH4 estimates, caution must be exercised regarding persistent biases arising from poor spatial representation, artificial redistribution of wetlands in coarse-resolution datasets, and inadequate classification of wetland types. Continued coordinated research efforts have strong potential to leverage advances in computing and Earth observation technologies to improve understanding of CH4 budgets in high-latitude ecosystems.

Methods

Data selection

To conduct this synthesis study, we identified existing land cover classification datasets encompassing Arctic and boreal region wetlands (Fig. 1). These datasets comprised published studies and/or available datasets with very high spatial resolution land cover classification maps14,35,36,37,63,64,65,66 (<2.5 m) and where chamber-based FCH4 measurements were conducted14,30,35,36,37,67,68,69 that represented the array of land cover classes in the land cover maps (Supplementary Table 1). Datasets were chosen based on spatial and temporal coverage, data quality, and alignment with our research objectives. Specifically, selected land cover classification maps needed to possess spatial resolutions of less than 2.5 m, cover areas of at least 2.5 km2, and have been generated within five years of relevant FCH4 measurements. The land cover maps are described in more detail in Supplementary Table 1 as well as in the primary sources. Due to the large spatial extent of the mapped area of the Scotty Creek watershed (~ 12 × 19 km), this region was split into northern and southern locations, North Scotty Creek and South Scotty Creek, respectively.

Land cover class harmonization

The thematic detail of the reference maps varied significantly, ranging from low resolutions where wetland types were simply labeled as ‘wetland’ to high-resolution products detailing various wetland classes based on factors like moisture gradient, nutrient availability, and successional stages in wetland development (Supplementary Table 1). This variability in thematic detail directly impacts pixel patch size, as larger objects are more likely to be represented at coarser scales. To address potential scaling issues, land cover types were aggregated and harmonized across maps, reducing the number of classes to 5–7 (Supplementary Table 1). The general categories included upland/grassy, barren, forest, wetlands, and lakes. We tested the sensitivity to distinguishing different wetland classes by combining fen and bog classes at higher-resolution land cover themes and repeating the scaling analysis. For the Utqiaġvik area, the land cover map was aggregated into ‘permafrost bog’ and ‘permafrost wetland’ categories, aligning with the BAWLD classification theme characteristic of this continuous permafrost wetland region3,13. Lakes and rivers were not included in this analysis due to limited representation in FCH4 measurements within our datasets.

Resampling

To investigate the impact of spatial resolution on CH4 emission estimates, emissions derived from low spatial resolution land cover maps were compared to those derived from the reference maps (original high resolutions). Using QGIS software70, we generated coarser land cover classifications by mode-aggregating the reference maps from the base resolutions to coarser resolutions at 5, 10, 25, 50, 100, 250, 500, 1000, 1250, and 5000 meter steps. Mode aggregation assigns a pixel as the dominant (majority) land cover class within the pixel extent, replicating how many land cover products are constructed. This approach allowed us to coarsen the high-resolution reference maps in a way that is representative of many existing products. The coarsened resolutions were specifically selected to align precisely with the spatial extent of the maps. Each resolution step ensures that no partial pixels are left at the map edges. This approach preserves spatial integrity and avoids artifacts introduced by edge truncation, allowing a consistent grid structure across all scales. As such, we would compare 5, 10, 25, and 1000 m resolutions to SPOT, Sentinel, Landsat, and MODIS image resolutions respectively. CH4 emissions presented here were calculated based on the progressively coarser land cover map to analyze the conservation of CH4 emissions magnitudes at different spatial resolutions and identify the threshold at which significant deviations from the reference maps occurred. Prior to resampling, each land cover map was cropped to ensure alignment compatibility with the chosen resolution steps, eliminating any potential overhang that could affect emission estimates. It is important to acknowledge that the spatial resolutions and boundary definitions in the reference maps at nominal resolution could influence our CH4 emission estimates and the subsequent analysis.

FCH4 measurements and upscaling

Methane fluxes were measured and averaged for each land cover class at each site (Supplementary Table 2) using the manual chamber technique. The number of plots per site ranged from 12 to 279 with chamber areas between 0.03 to 0.36 m2 and were taken during the growing seasons from 2007 to 2019. Specific methodologies and measurement frequencies for fluxes and land cover maps are described in the primary sources (Supplementary Table 1). Negative FCH4 values indicate uptake while positive values indicate emission. The high-resolution land cover maps were used to upscale FCH4 across the various regions. To calculate regional FCH4, we used an upscaling approach:

$${{FCH}4}_{{region}}={\sum }_{1}^{n}({A}_{i}\times {{FCH}4}_{i})\div{A}_{{region}}$$

where i represents the individual land-cover classes within each region, n represents the total number of land cover classes (Supplementary Table 1), A represents the area and FCH4 represents the mean of the measured FCH4 for the land cover class. This methodology, akin to a “paint-by-numbers” approach, simplifies the estimation of regional emissions, offering only a snapshot of emissions at a specific point in time. This method does not capture temporal variability or provide mechanistic insights into FCH4 and CH4 budgets but instead establishes a framework for evaluating resolution requirements for accurately upscaling CH4 emissions. This approach assumes that each land cover class has a representative flux and does not capture fine-scale hydrological or microtopographic heterogeneity within classes. Fluxes for inland waters, which comprised an average of 1.7% of the land area, were not typically collected during in-situ campaigns. Instead, we used values from the BAWLD-CH4 database3 for “medium-sized peatland lakes”, where the median total flux—summing diffusive and ebullitive contributions—was 63.5 mg CH4 m−2 d−1 (diffusion: 18.4; ebullition: 45.1). To assess variability, we examined the 25th and 75th percentile estimates, which were 31.8 mg CH4 m−2 d−1 (diffusion: 11; ebullition: 20.8) and 122.5 mg CH4 m−1 d−1 (diffusion: 42; ebullition: 80.5), respectively. This variability had little impact on the deviation from the regional reference flux estimates (Supplementary Fig. 4).

Statistical analysis

Regressions and plots were generated in R software (v4.3.3)71 using the ‘ggplot2’72, ‘ggsignif ’73, ‘data.table’74 packages. Percent deviation from reference (Fig. 3; Supplementary Fig. 2) is the absolute value of the difference of regional FCH4 at the nominal resolution and regional FCH4 at a spatial resolution step divided by the regional FCH4 at the nominal resolution. Landscape division index and patch statistics (Fig. 4) were computed using FragStats software75. Resolution sensitivity scores (Fig. 4) indicate the relative sensitivity of a region’s FCH4 signal to spatial resolution coarsening, ranked from 1 (most sensitive) to 7 (least). Scores were derived as the area under the deviation-from-reference curve for each region (Supplementary Fig. 2). Areas were calculated using the trapezoidal rule over log-transformed resolution steps to normalize distance between resolutions.