Summer warming explains widespread but not uniform greening in the Arctic tundra biome

The Arctic tundra biome is experiencing rapid warming, significantly impacting climate feedbacks, wildlife, and human communities. However, comprehensively assessing these impacts remains challenging due to limited long-term field measurements, particularly across the Canadian and Eurasian Arctic. Existing studies, while valuable, document varied responses—increased plant cover and shrub dominance in some areas, minimal change in others, and even declines in plant growth in certain regions. This heterogeneity underscores the need for effective use of satellite data to quantify ecological changes across this vast and rapidly warming biome. Historically, Earth-observing satellites have been employed to infer changes in tundra greenness. Pan-Arctic assessments, however, have relied on coarse-resolution data (e.g., AVHRR), resulting in inconsistencies and discrepancies in reported trends. The coarse resolution of AVHRR data (approximately 8 km) significantly exceeds the scale of ecological changes in heterogeneous tundra landscapes, hindering attribution to specific drivers (permafrost thaw, wildfires, etc.) and comparison with field observations. These limitations necessitate the use of higher-resolution satellite data for a more accurate and nuanced understanding. Landsat satellites offer a superior alternative, providing 30 m resolution data covering a comparable time period to AVHRR. This higher resolution better aligns with the scale of field measurements and ecological changes, improving the ability to reconcile satellite-derived trends with ground-based observations. While higher resolution increases data volume, recent advancements in computing and remote sensing techniques have facilitated larger-scale assessments. This study aims to leverage Landsat data to advance our understanding of tundra greenness changes across the entire Arctic biome. This research specifically addresses four key questions: (1) The extent of tundra greenness change in recent decades; (2) The correlation between inter-annual variation in tundra greenness and summer temperatures; (3) The relationship between tundra greenness trends and climate, permafrost, topography, and fire; and (4) The correlation between satellite-observed tundra greenness and field-measured plant productivity.

Numerous studies have investigated the impacts of Arctic warming on tundra ecosystems using various approaches. Field measurements, while providing the most direct evidence, are often sparse and geographically limited. Satellite remote sensing, particularly using the Normalized Difference Vegetation Index (NDVI), offers a means to assess broader spatial patterns. However, previous pan-Arctic assessments based on coarse-resolution data like AVHRR have shown significant discrepancies in results, partly due to challenges in cross-calibrating data from multiple sensors and the mismatch between data resolution and the scale of ecological processes. Studies using AVHRR have indicated both increasing (greening) and decreasing (browning) NDVI across different Arctic regions. Regional studies using higher-resolution Landsat data have shown promising results in linking tundra greening to increases in shrub cover and changes in biomass, but a comprehensive pan-Arctic assessment using Landsat data has been lacking. This study addresses this gap by using Landsat data in conjunction with a wide range of environmental and field data sets, providing a more detailed and spatially explicit view of the phenomenon than previously available.

This study uses more than three decades (1985-2016) of high-resolution Landsat satellite imagery (Landsat 5, 7, and 8) combined with various environmental and field datasets to assess changes in Arctic tundra greenness. The primary metric used is the annual maximum summer NDVI (NDVImax), derived from surface reflectance data. A total of 50,000 random sampling sites across the Arctic were selected using Google Earth Engine. Annual NDVImax time series were generated for each vegetated sampling site, employing novel approaches to cross-calibrate NDVI data among different Landsat sensors and to minimize biases related to data scarcity during the growing season. To account for uncertainties inherent in NDVImax estimation (related to sensor calibration, cross-sensor calibration, and modeling approach), Monte Carlo simulations (n = 10³) were conducted. These simulations propagated uncertainties from multiple data sources (Landsat data, climate data, field data). Changes in tundra greenness and its covariation with summer temperatures (quantified using the Summer Warmth Index (SWI)) were assessed using rank-based trend tests and correlations, all within the Monte Carlo uncertainty framework. The analysis extended to exploring potential drivers of tundra greenness change (1985-2016 and 2000-2016). This involved examining correlations between NDVImax and SWI, and using Random Forest models to investigate the relationship between NDVImax trend categories (greening, browning, no trend) and various environmental variables: changes in SWI, soil temperature, soil moisture, elevation, permafrost characteristics, land cover, and fire occurrence. Finally, the satellite-derived NDVImax observations were validated against field measurements of plant productivity. Three metrics of plant productivity were used: graminoid aboveground net primary productivity (ANPP), shrub ring-width indices (RWIs), and ecosystem gross primary productivity (GPP). The validation process also incorporated uncertainty using Monte Carlo simulations to compare NDVImax with the various productivity measures.

The study revealed widespread but not uniform greening across the Arctic tundra biome during recent decades. The mean Arctic NDVImax increased significantly from 1985 to 2016 (7.3%) and from 2000 to 2016 (3.6%), with the Low Arctic and Oro Arctic experiencing the most pronounced greening. These changes were strongly correlated with increases in summer air temperatures, as indicated by a positive correlation between NDVImax and the SWI. The relationship between NDVImax and SWI was particularly strong when considering both the current and the previous year's SWI values. The 1992 minimum in mean Arctic NDVImax coincided with the cooling following the Mount Pinatubo eruption, further supporting the influence of temperature on tundra greenness. However, greening was not universal. From 1985 to 2016, 37.3% of sampling sites exhibited significant greening, 4.7% showed browning, and 58.0% showed no significant trend. Similarly, from 2000 to 2016, 21.3% showed greening, 6.0% browning, and the remaining sites no trend. Greening was most prevalent in the Oro Arctic and Low Arctic, while browning was more common in the High Arctic. Random Forest models showed that the increase in SWI was the most important predictor variable for greening. Greening occurred frequently at warm, high-elevation sites with increased summer air temperatures, soil temperatures, and soil moisture. In contrast, browning was associated with cold, low-elevation sites experiencing decreased summer temperatures, soil temperatures, and soil moisture. Interestingly, sites with soil temperatures exceeding 0 °C in the early 2000s showed a sharp decline in greening and an increase in browning. Comparisons of Landsat NDVImax with field measurements of plant productivity revealed positive correlations between NDVImax and graminoid ANPP, shrub RWIs, and ecosystem GPP. These findings strongly support the interpretation that the observed greening reflects increased plant productivity and biomass across much of the Arctic tundra.

This study's pan-Arctic assessment of tundra greenness using high-resolution Landsat data supports the hypothesis that recent summer warming has stimulated plant productivity in many parts of the Arctic. The strong correlation between NDVImax and summer temperatures, along with the validation against field measurements of plant productivity, provides robust evidence for this conclusion. The study also highlights the complexity of Arctic ecosystems, demonstrating that vegetation responses to warming are highly variable, with some regions exhibiting browning or no significant change. The findings are generally consistent with previous regional studies and long-term field observations, but offer a more comprehensive pan-Arctic perspective. The use of Landsat data reveals subtle patterns of change that would be missed by coarser-resolution data, especially in heterogeneous landscapes where greening and browning can occur in close proximity. The discrepancies observed between Landsat trends and those reported using AVHRR and MODIS highlight the importance of data resolution and the need for careful cross-comparisons between different satellite datasets. While this study focuses primarily on temperature, future research could delve deeper into the intricate interactions between temperature, soil moisture, nutrients, permafrost dynamics, and other factors driving the observed variability in vegetation responses.

This study presents a comprehensive pan-Arctic assessment of tundra greenness changes using high-resolution Landsat data. The findings confirm a widespread greening trend linked to increased summer temperatures, yet simultaneously reveal substantial spatial variability and localized browning. This underscores the complexity of Arctic ecosystem responses to climate change and emphasizes the importance of high-resolution data for accurately capturing these diverse dynamics. Future research should investigate the complex interplay of biotic and abiotic factors influencing these responses and explore the potential implications of these changes for biodiversity, ecosystem function, and human communities.

This study acknowledges several limitations. First, the availability of Landsat data was incomplete prior to 2000, particularly in eastern Eurasia, potentially affecting the analysis of long-term trends. Second, the relatively low frequency of Landsat observations per growing season can lead to uncertainty in estimating annual NDVImax, though the employed modeling techniques mitigate this issue. Third, while the study utilizes multiple metrics of plant productivity for validation, the spatial and temporal coverage of field measurements remain limited, preventing a complete representation of all Arctic ecosystems. Lastly, the study focuses mainly on the correlation between temperature and greenness trends; the precise mechanisms driving the observed patterns require further exploration.