Introduction

Over the last four decades, the Arctic has warmed at a rate more than 2–3 times that of global mean surface temperature1,2,3. The warming has led to a rapid decline in Arctic sea ice and Greenland Ice Sheet surface mass balance4,5. Despite amplified polar warming taking place in winter, this phenomenon is understood to be driven primarily by the sea ice albedo feedback6,7,8,9,10,11, through an oceanic capacitor effect where heat is absorbed in summer and released in winter12,13. However, the rate of summertime sea ice decline relative to carbon dioxide forcing or global mean surface temperature change, referred to as sea ice sensitivity, in observations is outside the spread of global climate models (GCMs), suggesting missing or misrepresented drivers of Arctic sea ice change in models and uncertainty in the rate of future Arctic Amplification14,15. Following the traditional feedback framework, one might examine other processes in addition to the albedo feedback, such as the lapse-rate and Planck feedback, that have been implicated as leading drivers in annual mean studies of Arctic Amplification6,8. However, these phenomena are shown to be the aggregation of nonuniform and uniform variations in the structure of atmospheric warming and determined primarily by surface warming; thus, it can be argued that to make systematic progress in constraining the future rate of Arctic Amplification we should focus on specific processes that lead to sea ice decline and surface warming, such as the water vapor feedback16. Given the higher moisture capacity in the Arctic atmosphere during summer, in addition to the prevalence of extratropical storms near the poles during this season, summer is the key period when moisture tends to migrate into the Arctic from lower latitudes, and there is sufficient water vapor to impact the radiative budget. This impact is prevalent in studies that decompose the feedback framework into contributions to summer versus winter warming, showing the water vapor feedback as the second strongest contributor to summertime Arctic warming, following the sea ice albedo feedback and highlighting this mechanism as a potentially important source of model uncertainty in projecting the future sea ice and climate response in the Arctic8,9. A better understanding of summertime moisture variations may also be key to understanding other aspects of the Arctic hydrological cycle that contribute to surface albedo change, such as precipitation12,17.

The water vapor feedback is more often associated with the tropics, as the warmer temperatures relative to the polar regions result in much larger increases in humidity following the Clausius–Clapeyron relation18. But preferential moistening at lower latitudes results in a larger equator-to-pole moist static energy (MSE) gradient that enhances latent energy transport towards higher latitudes and is an important driver of Arctic Amplification10,19,20,21. Despite this recognition, these theories focusing on the forced response of MSE-gradient-driven moisture transport often do not fully account for land-sea differences and rely on the moisture availability and transport from free-running models22,23. For example, Graverson and Butu24 found that latent energy and moisture are largely transported to the Arctic via planetary-scale waves; however, issues such as weak or missing teleconnections and eddy feedbacks that contribute to biases in simulated circulation patterns are ubiquitous among GCMs22. The importance of poleward moisture fluxes for Arctic climate change and the uncertainties associated with transport mechanisms have spurred several studies over the recent decade attempting to track the sources and pathways of Arctic water vapor.

A key result of these tracking studies is generally that the local changes in atmospheric water vapor and precipitation associated with decreased sea ice cover and increased evaporation only explain a limited fraction of the simulated changes23,25,26,27,28,29. Rather, they highlight the contributions from remote terrestrial sources as playing a leading role in increasing Arctic water vapor during boreal summer, though these conclusions are based on idealized frameworks derived from climate model simulations28,30,31. Due to limitations related with internal variability in free-running model simulations, these previous studies have not been able to diagnose the terrestrial contribution to a continually increasing long-term trend in Arctic water vapor over the historical period32. For these sources over land to be capable of fueling the observed wetting trend of the Arctic would require a reduction in the bias associated with internal variability in climate models and a mechanism by which the land surface could maintain transport of moisture into the Arctic over four decades. In this study, we explore the transport of moisture from the tropics to the Arctic and its recycling through the land surface as it is transported poleward, or the land capacitor effect, to reconcile results from water tagging studies with meridional heat and energy transport in radiative frameworks.

These attempts can be categorized into three primary approaches to study moisture variability: conventional moisture budgets33,34, trajectory models23,25,31,35, and water tagging26,27,28 in a regional or GCM. Moisture budgets only provide an Eulerian point of view of moisture changes associated with moisture processes, i.e., moisture transport, evaporation, and precipitation. Given the difficulty in attributing the local changes to the three processes, the local change term is normally set as zero over a relatively long period to estimate the budget among the other three processes. Thus, this approach is more appropriate to study climatological features of moisture changes. Trajectory models, although constrained by reanalysis, must make important assumptions about mixing and microphysical processes in individual air masses, making them more appropriate for simulating short-duration transport processes on regional scales. Water tagging enabled models address these previous limitations by directly tracking water vapor parcels from evaporation and following them through all atmospheric processes to when they are precipitated out at the land or ocean surface. However, this technique is still sensitive to the simulated climate environments in the models, which are known to differ from observations (model biases). Therefore, a comprehensive understanding of the origins of moisture sources in the Arctic atmosphere, how moisture is transported into the Arctic, and how such moisture changes contribute to summertime radiative forcing in the Arctic remains elusive. This knowledge gap hinders our ability to assess the models’ skill in projecting future moisture, precipitation, and radiation changes and the resultant sea ice variability.

To provide a comprehensive land capacitor theory that explains trends in historical summertime Arctic water vapor and bridges its three main topics of examination (transport from the tropics to high latitudes, terrestrial influences, and radiative impacts via the water vapor feedback), we have developed a novel approach which combines two traditional modeling tools by nudging observed climate conditions in a water tagging enabled GCM, to the best of our knowledge. This method alleviates many of the limitations of the prior methods by providing a quasi-Lagrangian perspective into the hydrological cycle (from evaporation to precipitation) related to the recent Arctic moistening, constrained by observations using a nudging approach36,37. In particular, the recently available water tagging capability in the isotope-enabled Community Earth System Model (iCESM)38,39, radiative kernels built using ECMWF Reanalysis v5 (ERA5)40, and the Community Atmosphere Model version 5 (CAM5)41 enables us to further trace the sources and sinks of moisture transport and their related radiative impacts40,41,42,43,44,45. With these new, publicly available tools in iCESM, it has become possible to gain an improved understanding of a number of key processes that collectively induce the recent moistening trends and enhanced radiative forcing, including local and remote evaporation, moisture transport into the Arctic, and variations of specific humidity, precipitation, and radiative forcing within the Arctic. If proven useful in providing a new lens to investigate moisture variations in the Arctic, the method used here to generate global results can also be used to study similar problems for various regions across the globe. In the following sections, we first present our analysis of the moistening trends and the corresponding pathways the contribute to Arctic moistening in this nudging simulation, then assess the role of land processes across two primary source regions in contributing to the moistening, and finally conduct a kernel analysis of the radiative impacts of this moistening that can be attributed to each source.

Results

Historical and model simulated moisture trends

The iCESM1 simulations with atmospheric nudging yield nearly identical replays of anomalous June-July-August (JJA) atmospheric WV within the Arctic from ERA5 (Fig. 1a, black and magenta lines), with moistening trends of 0.35 kg m–2 decade–1 in iCESM1 and 0.27 kg m–2 decade–1 in ERA5. The 40-member ensemble mean from the Community Earth System Model version 2 Large Ensemble (CESM2-LE) with smoothed biomass burning exhibits a slightly enhanced trend over 1981–2022 (0.53 kg m–2 decade–1) relative to iCESM1, with the nudged simulation largely falling within the envelope from the CESM2-LE members (Fig. 1a). While there is an increasing trend over the satellite era in atmospheric WV, trends in total JJA precipitation from the CESM2-LE, iCESM1, and ERA5 are not significant, consistent with several observation-based products (Fig. 1b, See “Methods”). Despite the large spread in the magnitude of JJA Arctic precipitation amongst our model simulations and observation-derived datasets, the iCESM1 simulation generally agrees well with the observed year-to-year variability and trends of most products examined here (detrended r = 0.44–0.79, Fig. 1, and Supplementary Fig. 1). The large uncertainty of current precipitation data in reflecting long-term variability of Arctic precipitation indicates the necessity of the current modeling analysis to aid in better assessment of water vapor variability and change in high latitudes from a new perspective.

Fig. 1: Variability and trends in moisture transport from nudged iCESM1 simulations and observation-based datasets.
Fig. 1: Variability and trends in moisture transport from nudged iCESM1 simulations and observation-based datasets.
Full size image

a JJA total column precipitable water (i.e., vertically integrated specific humidity) (kg m–2) within the Arctic (70–90°N) from CESM2-LE (ensemble mean: blue, ensemble spread: grey shading), iCESM1 (black) and ERA5 (magenta) (a). b JJA total precipitation (snow+rain, large-scale+convective) within the Arctic (70–90°N) from the spread across observation-based precipitation products (green shading), iCESM1 (black), ERA5 (magenta), and the CESM2-LE (blue, grey shading) (b). c, e 1981–2022 linear trends in total precipitable water (shading, kg m–2 decade–1), 200 hPa geopotential height (contour, m decade–1), and IVT (red vectors, kg m–1 s–1 decade–1) from ERA5 (c) and iCESM1 (e). IVT vectors in panels c, e only shown where trends in total IVT are statistically significant at the 95% confidence level according to a two-sided student t-test. d, f 1979–2022 linear trends (calculated by linear regression) in zonal mean specific humidity (shading, g kg–1 decade–1) and geopotential height (contour, m decade–1) from ERA5 (d) and iCESM1 (f). Vertical dashed line in panels (d) and (f) shows the 70°N line which we use as the southern boundary of the local Arctic region. Gray stippling in panels (c–f) indicates insignificant trends (<95% confidence level according to a two-side t-test) in total precipitable water or zonal mean specific humidity.

Prior water tagging studies have highlighted the role of high latitude surfaces; therefore, we also evaluate trends in a key reservoir for water vapor transport into the Arctic, represented by soil moisture (Supplementary Fig. 2). Fearon et al. (2023) showed that soil moisture accounted for more than 30% of this moisture flux into the Arctic31. We also find reasonable agreement between the iCESM1’s simulated soil moisture trends and those from the satellite measurement-derived Soil Moisture Active Passive product (SMAP, Supplementary Fig. 2a,b). SMAP shows a decrease in soil moisture across most of the high-latitude land surface, again consistent with the nudged model simulation. This could be interpreted as a redistribution of moisture from the land to the Arctic; however, there must be some recharge mechanism capable of sustaining a long-term decrease in soil moisture coinciding with an increasing trend in moisture sourced from these regions being transported to the Arctic. This paradox motivates the subsequent analysis and the proposition of a land capacitor effect to explain sustained moisture, facilitating water vapor transport from low to high latitudes discussed in the following sections.

Remote transport pathways supplying Arctic moistening

During boreal summer there is significant moistening throughout most of the Northern Hemisphere below 400 hPa (Fig. 1d, f). The Arctic experiences moistening of the atmospheric column (0.35 kg m–2 decade–1, black curve in Fig. 1a), except for a small patch over the Beaufort Sea and parts of the Pacific sector (Fig. 1c, e). Moistening trends follow circulation, occurring where there are increases in geopotential height and poleward integrated water vapor transport (IVT). There has been an increasing geopotential height trend extending from the North Atlantic, over Greenland, and into the Arctic that has transported WV poleward to the west of Greenland (16.8 m decade–1). Poleward WV transport has also occurred between a low-pressure center over the Northeast Atlantic and high-pressure center over Europe, as well as east of a low pressure center over central Eurasia.

As poleward transport trends are concentrated along narrow pathways over Eurasia and Greenland (Fig. 1c, e)46, we calculate the amount of simulated Arctic moistening resulting from atmospheric river (AR) events, defined using the Guan and Waliser47,48 tracking algorithm, which utilizes a threshold detection approach based on seasonally dependent extreme (>85th percentile) total integrated vapor transport 47,48 (See “Methods”). The JJA total precipitable water trends in the Arctic associated with ARs is consistent with poleward IVT along the aforementioned pathways, contributing 2.56 g m–2 decade–1 or 77.4% of the total trend (See “Methods”). Using JJA troposphere mean specific humidity or poleward zonally integrated IVT at 70°N yields contributions of 0.047 g kg–1 decade–1 (80.0%) and 9.22 million kg m–1 s–1 (66.8%), respectively. Contributions of ARs to total poleward IVT along 70°N over western Greenland, northern Europe, and central Eurasia is 3.16 million kg m–1 s–1 decade–1 (65.2%), 0.67 million kg m–1 s–1 decade–1 (73.0%), and 3.08 million kg m–1 s–1 decade–1 (95.5%), respectively.

The only non-significant specific humidity trends occur near the surface in the tropics and at upper tropospheric levels (Fig. 1d, f). Geopotential height trends show the strongest increases in the uppermost troposphere that extend downward closer to the pole (contours in Fig. 1d, f). Nudged iCESM1 simulations accurately simulate the long-term variability in humidity and precipitation within the Arctic when atmospheric fields are imposed. This lends more confidence to the ability of the simulated water tagging results in the iCESM1 to reflect reanalysis moisture transport trends, allowing us to reduce the assumptions (i.e., airmass mixing, realistic atmospheric circulation) associated with previous water vapor tracking approaches.

By nudging to ERA5 atmospheric fields, the model reveals that the three major vapor transport pathways are responsible for climatological poleward moisture transport and the recent Arctic moistening in summer (Fig. 1a, b). We determine the contribution to Arctic moisture changes from each tagged region by computing an Arctic mean (area weighted within 70–90°N) of total precipitable water evaporated from individual regions (See “Methods”, contribution from each tagged region is also normalized by the area of each tagged region).

To understand the contribution of each source region, we focus on the source contributions to both mean and trends in JJA Arctic averaged (70–90°N) total precipitable water. Consistent with previous water tagging studies, we find a leading contribution being sourced from a narrow band between 50 and 70°N for both the mean (7.50 g m–2, 46.6%) and in the long-term trends (1981–2022, 1.63 g m–2 decade–1, 41%) (Fig. 2)28,31. This 50–70°N band includes a large contribution from the North American land mass, contributing 7.40% to the mean and 0.329 g m–2 decade–1 or 9.73% to the 1981–2022 trends. This strong influence of North America has not been identified in previous simulations examining poleward moisture transport without nudging, which are only forced by various external radiative forcing.

Fig. 2: Sources of mean and long-term trends in Arctic water vapor.
Fig. 2: Sources of mean and long-term trends in Arctic water vapor.
Full size image

Mean (kg m–2) (a) and linear decadal trends (1979–2022, g m–2 decade–1) (b) of JJA Arctic (70–90°N) total column precipitable water when each region is removed from each of the 54 tagged source regions. Numbers indicate the source region number or tag. Shading corresponds to the magnitude of the mean (a) or trend (b) on the left side of the color bar and the relative contribution (% of total) on the right of the color bar. Contributions are standardized by the area of the source region (See Methods).

The 44-yr trends in source-region contributions to Arctic WV show some similar relationships as their mean contribution, with the strongest contributions coming from Northern Hemisphere land masses, particularly northern Europe, central Eurasia, and northeastern North America (Fig. 2b). Remote sources outside the Arctic (south of 70°N) account for 84.2% (62% of total from land) of the moistening trend over the Arctic in the last 4 decades. These trends show less emphasis on Siberia and Alaska, as well as a decreased importance of the midlatitude oceans, especially northern extensions of these regions. Locally, there is a strong normalized increase in moisture coming from the Kara Sea and Canadian Arctic Archipelago regions.

Significant impacts of regional moisture sources on the Arctic summertime WV radiative effect

We quantify the radiative impacts of moisture transport by applying the ERA5 (Fig. 3) and CAM5 (Supplementary Fig. 4) radiative kernels to the linear trends (1981–2022) in specific humidity from iCESM138,39. The total summertime Arctic WV top-of-atmosphere (TOA) radiative feedback determined from the trends in the nudged iCESM1 simulations is 2.02 W m–2 K–1. The two largest contributors, Northeastern North America and central Eurasia contributing 0.259 W m–2 K–1 (12.9%) and 0.354 W m–2 K–1 (17.5%), respectively, to the Arctic WV feedback in terms of the long-term change over the satellite era. Locally, moisture sourced through evaporation from the Arctic Ocean contributes to 6.39% (0.129 W m–2 decade–1) of the WV radiative feedback computed over the historical period, and remote sources account for the remaining 93.6%. This local contribution primarily results from a positive radiative effect originating from the Laptev Sea (region 50, 0.0575 W m–2 K–1) and the Canadian Arctic Archipelagos (region 53, 0.0436 W m–2 K–1). Though counterintuitive with global warming driving decreased sea ice cover over the Arctic, weak air-sea temperature gradients and lower tropospheric stability limit the local contribution to moistening and the WV radiative effect during summer, leaving remote transport the dominant contributor to the JJA Arctic WV radiative effect49,50.

Fig. 3: Regional contribution to the Arctic summertime TOA water vapor radiative feedback.
Fig. 3: Regional contribution to the Arctic summertime TOA water vapor radiative feedback.
Full size image

a Troposphere (300–1000 hPa) integrated WV radiative feedback (W m–2 K–1) computed by removing WV sourced from each region using the top-of-atmosphere (TOA) ERA5 radiative kernel approach (a) “Methods”. Numbers (1–54) indicate the source region number or tag. Shading corresponds to the magnitude of the JJA Arctic WV radiative response (W m–2 K–1) on the left side of the color bar and the relative contribution (% of total) on the right of the color bar. b Vertical structure of Arctic (70–90°N) WV radiative (W m–2 K–1), computed by removing WV from each tagged source region (x-axis) (b). Region numbers on x-axis in (b) correspond to numbered labels in (a).

TOA or surface radiative feedbacks integrate the vertical changes in moisture and temperature that modulate radiative fluxes. Additional insight can be gained by inspecting the product of the radiative kernel and the moisture changes, which reveals that the WV radiative effect within the column can be vertically uniform or unevenly distributed when considering changes over the satellite record. The strongest TOA WV radiative effect occurs in the mid- to upper troposphere, between 300 and 750 hPa, in the vertical range of the majority of remote poleward moisture transport (Fig. 3b). Moisture sourced from land sources over northeastern North America and northern Eurasia produce the strongest positive WV radiative effect throughout the atmosphere, with more remote, subtropical source regions affecting higher altitudes (~300–700  hPa) and the local sources being confined to lower altitudes (>900 hPa). Consistent with our understanding of the mean structure of the large-scale circulation pattern, moisture sources over 50 to 70°N are located at the upward branch of the Polar cell and the Ferrel cell, where extratropical storms and ARs can easily transport this moisture toward the upper part of the Arctic troposphere. The leading contributors to the summertime WV feedback, central Eurasia and northern North America, exhibit extensive warming throughout the troposphere, extending down to the surface (Fig. 3b). In contrast lower latitude sources, such as the tropical Atlantic (regions 35 and 36) are not able to influence the lower the troposphere, consistent with current understanding of how poleward transport follows isotherms. NE North America (region 47), despite a small negative radiative effect near the surface, has the second most extensive warming throughout the lower levels of the column, creating a strong contribution to the positive Arctic WV radiative effect (Fig. 3b).

Because central and eastern Canada (region 47) appears to be particularly effective in inducing the summertime WV radiative effect (16.0%, Fig. 3a), we run an additional simulation with 9 smaller regions within the area of interest (Supplementary Fig. 5d). The vertical structure of trends originating from region 47 suggest this radiative efficiency is likely due to increased transport throughout middle to lower levels (Supplementary Fig. 5c). Other regions (i.e., 41, 43) have much weaker radiative impacts because over recent decades their transport has been pushed to higher elevations (Supplementary Fig. 5a, b), reducing their capability to travel farther distances into the Arctic and influence the energy budget.

Diverse land capacitor effects mediating transport between continental source regions

The results from the tagging suggest that the strongest source regions contributing to historical Arctic moistening originate from central and eastern Canada (region 47) and central and western Eurasia (regions 43+44). We extend the analysis from the previous sections to determine the mechanisms driving the increases in moisture supply from these two key source regions. Beginning with region 47, water vapor is transported from the southern and southwestern portions of the region, just south of Hudson Bay (Supplementary Fig. 5d). Precipitation coming into NE North America coincides with the strongest positive soil moisture trends (blue bars in Fig. 4b, Supplementary Fig. 5e). Moisture sourced from the leading regions contributing to Arctic moistening generally leaves behind drying trends both globally and over midlatitude land masses (Supplementary Fig. 5e, 7a). The one exception is over central and eastern Canada (region 47), where there is an increasing signal in soil moisture (red box in Supplementary Fig. 5e). This suggests that even though the region is a leading contributor to moisture redistribution from the midlatitudes to the Arctic, there is a capacitor effect, where recycling of moisture from the tropics or the buildup of moisture in the land surface supplies transport into the Arctic over several decades. Land and soil processes in region 47 enhance moisture storage through time, which can then be transported to the Arctic and contribute to summertime Arctic moistening and radiative feedbacks.

Fig. 4: A land capacitor effect driven by tropical-midlatitude transport in spring and transpiration in summer.
Fig. 4: A land capacitor effect driven by tropical-midlatitude transport in spring and transpiration in summer.
Full size image

a Linear trends (1981–2022, 10-3 Sv decade-1) in precipitation accumulated between January-May that falls in northeastern North America (region 47). Each grid cell shows the region from where each portion of the precipitation trend originated (a). b Linear trends (1981–2022, 10-3 Sv decade-1) in total precipitation (PNA), total evaporation (ETOT), soil evaporation (ESOIL), vegetation transpiration (EVEGT), and vegetation evaporation (EVEGE) accumulated over January–May (winter-spring, blue bars) and June-August (summer, red bars) integrated over northeastern North America (50–70°N, 240–300°E, region 47, 10-3 Sv decade-1), as well as precipitation originating from northeastern North America and falling over the Arctic (PArctic, 70–90°N) or all non-Arctic regions (PNonArctic, 90°S–70°N). These trends show the moisture convergence (PNA – ETOT) over region 47 and the proportion of moisture evaporated from region 47 that is transported into the Arctic (PArctic vs. PNonArctic) (b). c, d Same as a, b but for the western and central European region (region 43+44). Shading in (a) and (c) only shows statistically significant trends at the 95% confidence level using a two-side t-test. Positive trend values in (b) and (d) indicate increasing precipitation and increasing evaporative fluxes.

The land surface creates a land capacitor effect by storing incoming precipitation from remote, equatorward regions, which is initiated in the early part of the year, then releasing this moisture during summer. In spring, much of the moisture originates from the Caribbean and tropical Atlantic (regions 35 and 36), accounting for 54% of the precipitation trends over northeastern North America (Fig. 4a). During summer, the source of this moisture shifts closer to the Caribbean and continental United States (regions 35 and 41, 35.6%). The Caribbean and tropical Atlantic are the key source regions supplying this increase in soil moisture over central and eastern Canada via precipitation in spring (Fig. 4a). A Bermuda high type anticyclonic circulation trend over the subtropical Atlantic, paired with an elongated ridge extending over Greenland, drives this poleward moisture transport over northeastern North America (Supplementary Fig. 5e). Much of the transport originating from these tropical sources continues into the Arctic, though precipitation along the way feeds land-sourced moisture towards the end of the summer season. Changes in local recycling in both spring and summer can only explain as much as 18% of the precipitation trends driving increases in soil moisture, with most of the locally evaporated WV coming from plant transpiration before being transported northward and precipitated in the Arctic (Fig. 4a). Permafrost melt can only account for as much as 2.86% of the northeastern North American evaporation trends (See “Methods”). Despite previous studies also showing a contribution from land-sourced moisture to summertime Arctic WV increases, this moisture originally comes from the Caribbean Sea and tropical Atlantic before being deposited in the land surface in northeastern North America and ultimately terminating in the Arctic. With increasing evapotranspiration from this region, the large-scale circulation trend also directs more of this moisture into the Arctic (Fig. 4b). This highlights the importance of accurately capturing land surface processes, such as land capacitor effects, because they facilitate the interactions between the Arctic and lower latitudes, through a delayed release of moisture through vegetation following moisture convergence into the region earlier in winter and spring. These results also highlight the need for caution when interpreting tagging and backtracking approaches to understand the sources of Arctic moisture change.

In contrast, moisture transport from western and central Eurasia (regions 43–44) exhibit decreasing precipitation trends in boreal summer (–0.00619 Sv decade–1) but increases in winter-spring precipitation (0.00255 Sv decade–1) (Fig. 4c, d). Between February and May, an upper level high pressure pattern over low latitude Eurasia contributes to this rainfall trend over northern central Eurasia, diverting moisture from the Mediterranean poleward into northern central Eurasia (Supplementary Fig. 6d). Sources of rainfall over the region primarily originate from lower latitudes, such as from the Mediterranean Sea (region 37, 16.8%) and local recycling (region 43+44, 24.4%) (Fig. 4c). Decreasing precipitation during summer coincides with increasing total evaporation (0.00111 Sv decade–1), transpiration (0.00225 Sv decade–1, Fig. 4d) and soil moisture drying (–0.036 mm mm–3 decade–1, Supplementary Fig. 6). These drying trends are associated with a strengthening anticyclonic circulation trend that is part of a high wavenumber feature along the jet over Europe and eastern/central Eurasia, which prevents transport to and precipitation in the region as well as induces subsidence-driven warming that enhances evapotranspiration, depleting the soil moisture (Supplementary Fig. 6). The combination of precipitation in winter-spring and the large-scale circulation trends in summer result in increasing transpiration and directs more of the resulting transport into the Arctic (70–90°N) (Fig. 4d). Statistically significant enhancement of snow melt also coincides with regions of soil moisture decrease over Eastern Europe and may have contributed to soil moisture recharge that could supply increasing evapotranspiration (Supplementary Fig. 6d)51. This decrease in snow cover, particularly in spring, could impart some memory into the trends in moisture transport via summertime soil moisture changes over Eurasia (Supplementary Fig. 7). However, there are still large uncertainties in measurements and model-simulated snow cover, its seasonality and its interactions with summertime soil moisture and atmosphere circulations. Similar to North America, these changes in Eurasian soil moisture and evapotranspiration cannot be explained by permafrost melt (–2.52%, See “Methods”). In total, the Eurasian transport pathway, originating from the tropical Atlantic Ocean and Mediterranean Sea shows an opposite influence of large-scale circulation on the sources of moisture contributing to increasing WV in the Arctic, relative to that over North America.

Discussion

Confirming previous Arctic-focused water tagging studies, remote sources of moisture—primarily terrestrial pathways—are the primary contributor to Arctic warming and moistening over the recent historical period32,52,53. Remotely sourced moisture contributes to 84.2% (61% normalized for area) of total column WV trends and 93.6% of the WV radiative effect during summer. This is understandable because weak air-sea temperature gradients in the Arctic limit the surface’s ability to influence the atmosphere through evaporation or latent heat fluxes during summer, in contrast to potential interactions between clouds and the surface during fall when there is weak static stability and strong air-sea gradients49,52,53. The dominant contribution from sources outside the Arctic highlights the need to understand the importance of linking large-scale circulation to processes driving poleward moisture transport and their impacts on polar climate. In our simulations, we also find changes in moisture sinks, represented by a decrease in precipitation efficiency (defined as the change in precipitation divided by change in total water path) through reduction of cloud ice in Arctic mixed-phase clouds, decreases cloud ice content, increasing the lifetime of liquid water and clouds in the atmosphere and their radiative impacts (Supplementary Fig. 8). Thus, warming temperatures combined with enhanced poleward heat transport and the resulting increases in water vapor residence time (+0.61 days decade–1) in the Arctic may induce stronger Arctic Amplification through alterations in the phase partitioning of local mixed-phase clouds and contribute to the moistening trends shown in this study (Supplementary Fig. 8)35,54.

The summertime water vapor feedback derived using linear trends over the satellite period, exhibits a large influence on the boreal summer TOA and surface radiative budget (2 W m2 K–1), that is comparable to other leading feedbacks, such as the alternative fixed relative humidity Planck feedback (–2 W m–2 K–1) and greater than the lapse rate feedback (–0.84 W m–2 K–1 Supplementary Fig. 9). Given that the WV feedback is a key component of summertime polar warming which can drive sea ice decline and heat absorption into the ocean, having delayed impacts on the Arctic during fall and winter when amplification is seen, more efforts should be spent to understand its significance. However, moisture transport over the historical period has shown a strong response to atmospheric circulation variability, highlighting the need to understand changes in circulation over the historical record and into the future12. In addition, models tend to be biased in their representations of large-scale sea surface temperature (SST) trends55. Therefore, it is also important to consider large biases in simulated SSTs and their influence on moisture availability and atmospheric circulation when comparing observed trends from the satellite record.

Results from the iCESM1 simulations suggest that land processes are especially important in mediating poleward transport, especially in boreal summer, despite this moisture originating from the ocean or the tropics. Soil moisture over the satellite record shows drying trends across most of the Earth, except over central and eastern Canada (Fig. 4b). However, northern North America (region 47) shows increasing WV directly above the region that likely helps to retain soil moisture, while region 41 shows drying throughout the local atmosphere and soil (Fig. 4). A similar land capacitor effect also operates in other regions (west and central Eurasia), with soil moisture being key to supplying increased poleward transport, as the only potential sources feeding the trends over Eurasia are associated with wintertime precipitation originating from the Atlantic. This indicates that the Arctic moisture sources include direct and indirect pathways that rely on high-latitude land-ocean-atmosphere interactions. This is understandable since the northward moisture transport is primarily facilitated by atmospheric eddy activities, which typically take much longer time to traverse the same distance compared to atmospheric activities moving in the zonal direction. In addition, the average residence time of WV in the atmosphere is only a few days (~2 to 10 days), making a land capacitor effect, such as that identified in central and eastern Canada, necessary to relay the stronger moisture influx originating from the lower latitudes to the Arctic via continent-based pathway35. Specifically, there is a key role of vegetation in facilitating the delay in the land capacitor effect driven by increasing summertime vapor pressure deficits. This analysis also suggests that a shift in the phase of atmospheric circulation trends could alter the sources and pathways of moisture transport to the Arctic and, therefore the importance of WV in determining the rate of Arctic Amplification.

Our new findings regarding the land capacitor’s role in mediating poleward moisture transport still needs to be further reconciled with conceptual frameworks built on the importance of the tropics for determining polar moistening, as the pole-to-equator MSE gradient framework may oversimplify issues of sources and transport pathways. For example, the upward branch of the Ferrel cell (around 60–70°N) may play a key role in determining the moistening trend in the Arctic, as the large-scale zonal mean circulation cells are likely very important for determining poleward and upward moisture transport. In particular, transport over this zone is dominated by poleward and upward transport of moisture by eddies (v'q', ω’q’, in which v’, q’, ω’ indicate eddy components of meridional winds, specific humidity, and vertical motion, respectively), as indicated by the impact of storms and ARs from 50 to 70°N. However, ARs are sensitive to circulation changes that are also driven by internal variability of the climate system. This suggests a scenario by which internal variability interacts with anthropogenic forcing. This is a potential mechanism that needs more attention in future model evaluations.

It has been theorized that changes in the equator-to-pole temperature and moisture gradients could modify large-scale circulation patterns and storm tracks. It should be noted that during boreal summer these gradients are relatively small, and the anthropogenically forced response from climate models induces very weak regional changes, especially in atmospheric circulation. Therefore, with the impacts of wind-driven changes for moisture transport highlighted in this study, we emphasize the importance of understanding atmospheric variability and the pattern effect—which can simultaneously drive evaporation and transport—while also modulating meridional gradients and global temperature imbalances that are the foundation for simplified feedback frameworks. These results also highlight the importance of the fully coupled atmosphere-ocean-cryosphere-land system in determining regional climate change impacts. To our knowledge, this is the first study going to this extent in making the connection between the components of the Earth system, by attributing contributions of the Arctic water vapor feedback to regions involved in the land capacitor effect and explicitly quantifying the land surface’s role in multiple phenomena in the traditional and fixed relative humidity feedback frameworks. In the future, better constraints on the interaction of the forced response and internal variability can help us better project changes in Arctic hydroclimate, namely precipitation-related processes, under increasing greenhouse gas concentrations. Additionally, this is the first study directly bridging observations, reanalysis, and water tagging simulations through the use of nudging. However, much of the analysis here is focused on the nudged simulations and studies that build on this framework by further integrating model simulations and observations is necessary to better account for model biases (i.e., cloud physics, land surface, plant physiological changes and ocean components). This understanding will also help us better simulate the forced response in GCMs and reduce the uncertainty in earth system model climate sensitivity resulting from changes in moisture availability and transport. In particular, the combined approach, including water tagging, circulation nudging, and radiation calculation developed in this study may serve as a useful tool to enable us to better understand the global hydrological cycle and related climate feedbacks in reality.

Methods

Reanalysis

Gridded winds (u,v), temperature, precipitation rate, and specific/relative humidity from the ERA5 atmospheric reanalysis are used to derive year-to-year variability and trends in the Arctic hydrological cycle over the satellite era (1979–2022)56,57. ERA5 has been shown to capture precipitation relatively well in the Arctic region compared with other major reanalysis products58,59,60. We compute total column-integrated water vapor (TMQ) as vertically integrated total specific humidity at each grid cell and integrated vapor transport (IVT) to evaluate changes in moisture and transport, respectively:

$${TMQ}=-\frac{1}{g}\mathop{\int }\nolimits^{300\,{hPa}}_{1000\,{hPa}}{qdp}$$
$${uIVT}=-\frac{1}{g}\mathop{\int }\nolimits^{300\,{hPa}}_{1000\,{hPa}}{qudp}$$
$${vIVT}=-\frac{1}{g}\mathop{\int }\nolimits^{300\,{hPa}}_{1000\,{hPa}}{qvdp}$$

In these equations, q is specific humidity, u represents the zonal horizontal wind, v is the meridional horizontal wind, and g is the gravitational constant. The quantities are computed on standard pressure levels. In the reanalysis, 6-hourly horizontal winds and specific humidity are used.

Precipitation

To evaluate the ability of the model simulations to replicate observed precipitation, we examine high latitude total precipitation trends using a wide range of precipitation products to sample the large uncertainties in a data-sparse region61. Observation-derived precipitation products used include Global Precipitation Climatology Project version 2.3 (GPCP)62, Global Precipitation Climatology Centre (GPCC)63, Climate Prediction Center Merged Analysis of Precipitation (CMAP)64, the University of Delaware Terrestrial Precipitation product (UDel-TS)65, and the Integrated Multi-satellitE Retrievals for NASA-JAXA Global Precipitation Measurement (GPM-IMERG). We use the GPCP monthly satellite-gauge product at 1° × 1° resolution from 1979 to 2022. CMAP merges measurements from 5 satellite products and rain gauge data averaged to monthly means. UDel-TS interpolates monthly total precipitation from station rain gauge data over land from 1979 to 2014. GPM-IMERG is a NASA product estimating monthly mean global surface precipitation rates at a horizontal resolution of 0.1° from 2000 to the present. Detrended total precipitation over the Arctic in the nudged model simulations strongly correlates with detrended observational precipitation products from 1981 to 2022: CMAP (r = 0.79), GPCP (r = 0.59), GPCC (r = 0.58), and UDel-TS (r = 0.44, 1981–2014). CMAP, GPCC, and UDel-TS do not show a strong increase in Arctic precipitation and have insignificant trends consistent with ERA5 and model simulations shown in this study (Fig. 1b and Supplementary Fig. 1). In contrast, GPCP and GPM-IMERG show increasing precipitation trends beginning in the 2000s, with GPM-IMERG rapidly decreasing in the early 2010s (Supplementary Fig. 1). There is substantial uncertainty between the moisture products shown here and their definitions or representations of precipitation, with these products likely not able to measure total precipitation as simulated in the model.

Soil moisture

Simulated soil moisture trends are compared with trends from SMAP near surface soil moisture product, which combines measurements from 15 different satellite products from 1981 to 2022 (Supplementary Fig. 2). This product measures moisture within the upper 10 cm of the soil column. While we compare total column soil moisture from the model with this surface soil moisture product. JJA Arctic total precipitation (70–90°N, mm day–1) from each product is shown in Supplementary Fig. 1.

Nudged, water isotope enabled CESM1 simulations

We use the isotope-enabled Community Earth System Model version 1 (iCESM1), which uses as its base the CESM1.266. The atmosphere and land components have the nominal 1° (0.9° × 1.5° finite volume grid) and 2° (1.9° × 2.5° finite volume grid) resolution and finite-volume dynamical core (CESM1-FV2). Previously, iCESM1 has only been run with the nominal 2° resolution grid; therefore, we examine both to gauge the impacts of horizontal resolution on moisture transport and precipitation. In the future, higher resolutions with regional-refinement or global cloud resolving models would be more ideal to address issues of resolution. CAM5.3 has 30 active atmospheric levels in hybrid-sigma coordinates. The same ocean (POP2) and sea ice model (CICE4) components used in the CESM1-LE66 are also run on the nominal 1° resolution grid with displaced poles over Greenland and Antarctica. However, the iCESM experiment is run in an atmosphere-land only configuration with prescribed monthly SSTs and sea ice concentrations taken from ERA5 that are remapped to the iCESM1 ocean model grid. Tagged WV and specific humidity reflects the grid point or region from which it evaporates.

The CESM model does not explicitly diagnose permafrost. Therefore, we define simulated permafrost change as 1981–2022 trends in cumulative decline per year of vertically integrated soil ice that has remained above 273.15 K for 24 consecutive months. We use this to determine the contribution of the melting to evaporative fluxes as the ratio of permafrost melting trends (1981–2022, in Sv decade–1) relative to total evaporation and precipitation trends (Sv decade–1).

iCESM1 experiment design

The isotope-enable Community Earth System Model (iCESM1) uses numerical water tracers to follow movement and phase changes beginning from evaporation, when moisture is tagged based on its region, to when it is precipitated out of the atmosphere38,39. A single simulation is run from 1979 to 2022 (with 2 years of spinup), with the model nudged to ERA5 horizontal winds (u, v) and temperature throughout the atmosphere, and specific humidity in the lowest model level nearest the surface. In previous studies, such as Ding et al.37, the constraints imposed by the nudging (u, v, and T) result in very little member to member spreads resulting from initial states. For this reason, we only examine only this single iCESM1 realization. In addition, a number of sensitivity experiments are conducted, and the nudging simulation does not appear to be sensitive to vertical extent of the nudging or whether observed specific humidity is constrained at the lowest model level (not shown). Greenhouse gas concentrations are fixed at year-2000 levels, so as not to double-count the effects from greenhouse gas-driven warming and thereby introduce artificial heating. Previous moisture tagging experiments have employed iCESM1 simulations without nudging, but the combination of nudging and tagging in this manner allows us to replay observed atmospheric conditions (derived from ERA5) in iCESM1 so that the hydrological cycle (i.e., evaporation, transport, and precipitation) can be decomposed into its observed sources and pathways.

We construct 54 evenly dispersed regions (in terms of latitude and longitude, rather than area) around the globe without consideration for land or ocean surfaces. We also use an additional water tagging simulation, with 2 individual regions over North America and Eurasia land masses, to verify that the majority of moisture being sourced from 50 to 70°N originates from the land rather than ocean (not shown). To ensure that the iCESM1 run correctly simulates the climatological patterns of humidity and rainfall, we check the time series of total column water vapor and precipitation in the iCESM1 simulation against reanalysis, finding that the two products agree well and the results from the iCESM1 simulation are representative of the changes over the satellite era (Fig. 1a, b). We also include an additional 9-region tagging simulation, with the 9 smaller domains within central and eastern Canada (region 47 in the 54-region simulation) to show the moisture sources within this key area (Supplementary Fig. 5d).

Since the 54 tagging regions are evenly distributed based on latitude and longitude, larger source region areas closer to the equator may disproportionately contribute to Arctic WV changes solely by virtue of their area. Therefore, when reporting source-receptor region statistics we normalize specific humidity from each region by its total area, giving greater weighting to the smaller, high latitude regions. This shifts relative contributions to regions that are closer to or within our target region in the Arctic but does not change the overall qualitative results. In the text, we present both raw and normalized contributions.

CESM2 large ensemble

To evaluate the differences when using the model’s inherent circulation variability, we examine available output from the CESM2 Large Ensemble (CESM2-LE)67,68 simulations. We examine total precipitation, IVT, and 200 hPa geopotential height trends in 40-members from the CESM2-LE with smoothed biomass burning. Although CESM2-LE provides a total of 100 members for public use, only the 40 members (with smoothed biomass burning) out of the entire 100 ensemble members provides 6-h data.

Atmospheric river detection

While 6-hourly data is ideal for examining atmospheric rivers (ARs), the highest frequency the iCESM1 output was archived at was daily mean timesteps. We find that the overall pathways and trends in frequency on seasonal mean timescales are consistent between the 6-hourly ERA5 and daily mean iCESM1 data.

ARs are detected and tracked using the Guan and Waliser48 tARget v3 AR detection and tracking algorithm. It is an updated version of Guan and Waliser47, one of the earliest and most widely used AR detection tools. The tARget algorithm has been recommended by the Atmospheric River Tracking Method Intercomparison Project as a scheme suitable for both global and regional studies, particularly regions with climatologically low IVT such as at high latitudes52, due to the season- and location-dependent AR thresholds used in this algorithm.

For AR detection, we use daily output from the iCESM1 simulations and 6-hourly outputs from the CESM2-LE and ERA5. The contributions and magnitudes of AR frequency and poleward IVT between ERA5 and iCESM1 are nearly identical, even despite using daily output from iCESM1 for AR detection and tracking. We calculate the AR contribution to Arctic (70–90°N) TMQ (Equation 1), 300–1000 hPa mean specific humidity, and poleward IVT at 70°N using the AR mask determined by the Guan and Waliser48 ARDT algorithm. The AR mask is expanded by 6° in all directions to account for the potential spatial influence of the events. Since water vapor and precipitation impacts associated with ARs can precede or lag an event by a day, we also extend the grid points flagged as an AR by 1 day before formation and 1 day after termination. All grid points that fall within these expanded AR shapes or masks are considered to be associated with an AR event. We then compute JJA means using total, AR only, and non-AR TMQ, specific humidity, and 70°N poleward IVT. The differences using TMQ or specific humidity are small. The AR contribution to Arctic moistening is then determined by the ratio of AR associated TMQ/specific humidity/IVT trends divided by the total trend over 1981–2022.

Radiative feedback analysis

To better understand how increasing Arctic moisture contributes to a change in radiative forcing, we follow the traditional radiative kernel approach using the ERA5 and CAM5 TOA and surface, all-sky TOA radiative kernels from Huang and Huang40 and Pendergrass et al.41, respectively. iCESM1 outputs are regridded to the same resolution (2.5° × 2.5°) and 37 vertical levels as the ERA5 and CAM5 kernels. The moisture kernel quantifies the radiative response to the moistening caused by 1 K warming, assuming constant relative humidity. The WV radiative effect is integrated through the tropospheric column (below ~300 hPa), and we use the local feedback, which considers the surface temperature response at each latitude or an area-weighted average over the Arctic region (70–90°N). The contribution of each moisture source region to the Arctic WV radiative effect is computed by subtracting the WV tagged from each region individually from the total WV at each level and grid point, then computing the radiative effect. All analyzes focus on monthly mean or JJA means to examine the summertime Arctic WV feedback and its associated moisture transport.

Following Huang and Huang40, the traditional water vapor feedback associated with total and regional humidity change is computed as:

$${{dR}}_{i}={K}_{q}{{dT}}_{i}$$

where,

$${{dT}}_{i}=\frac{{{dq}}_{i}}{{q}_{0}}\frac{{R}_{v}}{{L}_{v}}{{T}_{0}}^{2}$$

Here, the subscript i denotes each tagged region (1–54), dqi is the trend in total specific humidity minus the trend in specific humidity from each tagged region, q0 and T0 are baseline or climatological specific humidity and temperature based on the 1981–1999 reference period, Rv is the gas constant for water vapor (J), and Lv is the latent heat of vaporization for liquid water at 0 °C for water vapor (J kg–1). The local feedback is then computed by dividing by the trend in surface temperature (Ts) at each latitude (ϕ):

$${\lambda }_{{qi}}={{dR}}_{i}/\varDelta {T}_{s}\left(\phi \right)$$

Note that after computing the local feedback, the sum of the individual regions (1.70 W m–2 K–1) is not equal to the total feedback parameter (1.83 W m–2 K–1). The vertically integrated feedback parameter is computed from 1000 hPa to the tropopause, estimated as 100 hPa in the tropics then linearly decreasing to 300 hPa in the polar regions.

We also provide additional radiative feedback calculations to place our traditional kernel approach into a broader context of the established feedback framework. Following the relative humidity framework detailed in Held and Shell44, we first compute the traditional total water vapor (LW+SW), Planck, and lapse rate feedbacks, using trends in specific humidity and temperature from the iCESM1 simulations convolved with the ERA5 radiative kernels (Supplementary Fig. 9). Our results are consistent with previous feedback studies that examine seasonality, showing a large role of the water vapor feedback in summer8,9. We then compute the alternative Planck, lapse rate, and relative humidity feedbacks, where the alternative Planck and lapse rate feedbacks combine the effects of temperature and humidity changes assuming fixed relative humidity (Supplementary Fig. 9).

During JJA, the surface longwave water vapor radiative feedback (2.92 W m–2 K–1) is dominated by the response to departures from fixed relative humidity (1.79 W m–2 K–1), with the required adjustment to maintain fixed relative humidity in response to uniform (alternative Planck minus Planck, 1.09 W m–2 K–1) and nonuniform warming (alternative lapse rate minus lapse rate, 0.03 W m–2 K–1) being second order (Supplementary Fig. 9b). In contrast, the opposite conclusions can be drawn using TOA radiative kernels, being dominated by uniform temperature warming through the Planck TOA radiative response (Supplementary Fig. 9b). The additional decomposition suggests that the temperature based feedbacks do not play a large role in determining the JJA TOA WV feedback and that future studies should more thoroughly consider the impacts of relative humidity trends on Arctic climate change. Additionally, future examination of other seasons as well as cloud radiative effects and their interactions with these feedbacks would also be beneficial.

We also test the radiative impacts of moisture transport on the Arctic surface using radiative kernels derived from both ERA5 and CAM5 (Supplementary Fig. 4). The primary differences in the patterns occur within the central Arctic. Specifically, the radiative kernels computed using a single-year simulation with CAM5 show a much stronger positive magnitude from 850 to 600 hPa, north of 75°N (Supplementary Fig. 4). In contrast, the ERA5 kernels use the average of 5 years (2011–2015), that provides a relatively broad sampling of the variability in the Arctic, with 2011–2012 and 2015 being dominated by strong anticyclones and 2013–2014 being dominated by strong cyclonic patterns. Anticyclonic circulation over the Arctic, similar to the trends examined in this study, is known to induce cloud increases along the periphery of the anticyclone and in the lowest model layers but decreases in the mid-troposphere where there is strong subsidence in ERA5. The cloud component generates a cooling effect during boreal summer that dampens the net warming in the all-sky WV radiative effect. Therefore, the differences between ERA5 and CAM5 are likely the result of a combination of internal variability driven atmospheric processes (e.g., subsidence over the central Arctic) and model physics. Ultimately, we find small differences between the two products; therefore, we focus on results using the ERA5 kernels.