Strong and thick temperature inversions are key components of the Arctic climate system and it is important to study and better understand them. The present study quantifies the temporal and spatial variability of surface-based inversions (SBIs) and elevated inversions (EIs) over Greenland, as derived from the ERA-Interim (ERA-I) dataset for the period 1979–2017. The seasonal and multiannual variability of inversion strength, thickness, and frequency are examined. Our results clearly show regional as well as seasonal patterns of both SBIs and EIs. SBIs are more frequent and stronger than EIs, and the spatial variability of inversions is larger during winter and smaller during summer. Furthermore, during summer, there has been a trend toward stronger (0.3 K decade−1), thicker (12 m decade−1), and more frequent (3% decade−1) SBIs in the southern part of Greenland, especially in the past two decades. Evidently, the strengthening of the anticyclone over Greenland causes a reduction of cloud cover, which manifests in an increase in SBI strength and thickness, particularly in the southern part of Greenland.
The Arctic planetary boundary layer provides a favorable condition for the formation of temperature inversions. These are a form of atmospheric stratification and describe a situation of increasing air temperature with elevation, leading to a stable boundary layer. Two types of inversions are prominent for the Arctic atmosphere: surface-based inversions (SBIs; i.e., inversions whose base is at Earth’s surface) and elevated inversions (EIs; i.e., inversions whose base is above Earth’s surface). (See the appendix for a list of abbreviations.) SBIs are forced by the radiative imbalance between 1) emitted longwave radiation from snow and ice surfaces and 2) incoming solar and longwave radiation, particularly during the (polar) night. EIs are formed by two mechanisms: by subsidence of air in anticyclones, or by warm air advection over underlying cold air masses (Kahl 1990; Bradley et al. 1992). The intensity and thickness of EIs are considerably smaller than those of the SBIs (Przybylak 2016).
Both SBIs and EIs can have important and complex effects on the Arctic surface energy budget as well as in the lower troposphere, and they therefore play a key role in the Arctic climate system. For instance, SBIs, which strongly decouple the surface from the overlying free troposphere, are critical for understanding low cloud and fog formation (Gilson et al. 2018b). It was also shown that low-level temperature inversions, which lead to nonlinear air temperature lapse rates, can impact the vertical glacier mass balance gradient (Mernild et al. 2008; Hulth et al. 2010; Mernild and Liston 2010). Correlations between low-level temperature inversions and episodes of increased sea ice and glacier melt have also been detected (Chutko and Lamoureux 2009; Tjernström et al. 2015). EIs are important for estimating the geostrophic drag coefficients (Overland and Davidson 1992), which are needed in sea ice motion simulation (Hibler and Bryan 1987), and for studying air pollution (Bridgman et al. 1989). Furthermore, Bintanja et al. (2011) described how inversion strength (i.e., temperature difference across the inversion layer) can affect Arctic climate change during winter, as strengthening the atmospheric stability leads to a positive temperature lapse rate feedback in the Arctic (Bintanja et al. 2012; Pithan and Mauritsen 2013). Detecting changes of inversions in the Arctic is thus crucial for understanding the impact of climate change on the cryosphere.
Several existing Arctic inversion studies have focused on winter months (Boé et al. 2009; Medeiros et al. 2011; Pithan and Mauritsen 2013), owing to the high inversion frequency during this period. Winter SBIs are primarily the result of the large deficit in surface net radiation, whereas summer SBIs are governed by the interaction of melting (in coastal areas) and warm-air advection from the south (Serreze et al. 1992; Palo et al. 2017). Clouds may strongly influence inversions through their effect on the surface net radiation budget particularly during summer (Serreze et al. 1992; Walsh and Chapman 1998). Several studies have attempted to relate changes in clouds to temperature inversions in different locations in the Arctic, such as along the northern Alaskan coast (Kahl 1990), over the central Arctic Ocean (Wetzel and Brümmer 2011; Palo et al. 2017; Nielsen-Englyst et al. 2019), in the Eurasian Arctic (Serreze et al. 1992), and in Greenland (Miller et al. 2013, 2015; Gilson et al. 2018a,b; Nielsen-Englyst et al. 2019). Studies focusing on the Arctic temperature amplification (Graversen et al. 2008; Screen and Simmonds 2010; Bintanja et al. 2012; Pithan and Mauritsen 2013) have shown vertically nonuniform warming of the Arctic troposphere, with near-surface warming occurring at a greater rate than at higher levels. Bintanja et al. (2012) addressed the role of SBIs in amplifying Arctic warming.
Strong and thick temperature inversions are persistent in large parts of Greenland, because 80% of it is covered by the Greenland Ice Sheet (GrIS), and hence radiative cooling occurs over a large region. These manifest themselves in the form of as a quasi-permanent anticyclone in the central and northern parts of Greenland, which prevents advection of warmer Atlantic air over the GrIS during winter (Steffen and Box 2001). In contrast, the southern part of Greenland is mainly influenced by the Icelandic low, and shows comparatively less intense and shallower inversions than the central and northern part (Cappelen et al. 2001; Przybylak 2016).
The interaction between synoptic-scale climate drivers and the surface is governed by boundary layer processes, whose dynamics are important in modulating energy and moisture exchange. At the Summit research station, Berkelhammer et al. (2016) scrutinized the impacts of the SBI on boundary layer dynamics, displaying that the stable atmosphere isolates the surface from free-tropospheric moisture sources and limits accumulation. Furthermore, Noël et al. (2019) showed the latitudinal contrast in summertime meltwater runoff from GrIS in response to change in Arctic atmospheric circulation. This change in atmospheric dynamics can also lead to spatial and temporal variations in inversion characteristics in Greenland, and ultimately to variations in the associated impact on the GrIS surface energy balance (SEB) and surface mass balance (SMB). This thus highlights the necessity of regionalization when studying the heterogeneity of inversion characteristics in Greenland.
Several efforts have been undertaken in order to explain the climatology of temperature inversions in Greenland on the point scale, especially in the coastal regions and at the Summit station using radiosonde (Gilson et al. 2018b), microwave radiometer (MWR) (Miller et al. 2013), and near-surface meteorological data (Adolph et al. 2018; Nielsen-Englyst et al. 2019). For example, Zhang and Seidel (2011) found increasing SBI frequency over the period 2000–09 for three coastal radiosonde stations in Greenland. Unfortunately, as the spatial coverage with in situ observations (e.g., from radiosondes) is generally low for Greenland, this makes it hard to provide a consistent spatial inversion characterization using such data alone. On the other hand, reanalysis data such as from the European Centre Medium-Range Weather Forecasts (ECMWF) reanalyses provide the possibility for spatially coherent studies, even in data-sparse regions such as Greenland.
Several studies have extensively employed the different ECMWF reanalysis datasets [e.g., ERA-40 and ERA-Interim (herein ERA-I)] for studying the Arctic temperature stratification (Graversen et al. 2008; Tjernström and Graversen 2009; Screen and Simmonds 2010; Bintanja et al. 2011, 2012; Medeiros et al. 2011; Zhang et al. 2011; Graham et al. 2019). However, most of these studies used vertical profiles based on pressure levels starting from 1000 hPa as surface level, which in reality is extrapolated below Earth’s surface in high-altitude regions. Consequently, elevated regions such as the central GrIS are generally masked out and, therefore, such studies only partly reflect the spatial variability of inversions in Greenland. Additionally, most of the studies defined the Arctic region poleward of 64° or 70°N (Boé et al. 2009; Screen and Simmonds 2010; Bintanja et al. 2011; Pavelsky et al. 2011), thus missing the southernmost parts of Greenland (59°–64°N).
Another important advantage of using reanalysis data compared to the application of Automatic Weather Stations (AWSs), or radiosonde data, is its ability to reach into the past, and thus to provide us with the opportunity to investigate climate trends. Changes in Arctic inversion characteristics have been shown for different Arctic regions, such as for the North American Arctic (Bradley et al. 1993; Walden et al. 1996), for the Arctic Ocean (Kahl et al. 1996; Pavelsky et al. 2011; Wetzel and Brümmer 2011), and for coastal stations in Greenland (Zhang and Seidel 2011), using radiosonde, satellite, or reanalysis datasets. Screen and Simmonds (2010) reported close agreement between the vertical structures of temperature trends in ERA-I and several observations in various Arctic regions (including Danmarkshavn, a northeast coastal station in Greenland). Although several studies have increased our understanding of the vertical structures of temperature trends, the relationship between changes in inversion characteristics and clouds has not yet been fully explored. This is especially true for the most recent period where major changes such as large positive Greenland blocking index (GBI) values (Hanna et al. 2016), as well as a reduction of the SMB, have been found.
The temperature gradient within the lowest 2 m of the atmosphere plays a key role in the SEB (Adolph et al. 2018). Several studies using AWS data have described the presence of stable near-surface-based inversions (NSBIs) in polar regions as being the difference between 2-m air temperature (T2m) and skin temperature (Tskin) (Hall et al. 2008; Good 2016; Adolph et al. 2018; Nielsen-Englyst et al. 2019). Radiosondes are not able to capture this near-surface process as they are launched at levels higher than 2 m above ground level. In contrast, reanalysis data offer the potential for analyzing the lowest and very decisive layers. Nevertheless, thorough evaluation by means of independent observations such as those from AWSs is still a crucial prerequisite.
In summary, previous studies on inversions in Greenland either covered the region on a too-coarse spatial and vertical resolution, or on the point scale only. Thus, what is lacking is a more adequate and informative climatology of SBI, EI, and NSBI at the Greenland scale. To close this gap, the present study, using the ERA-I dataset, aims to analyze and better understand the spatiotemporal variations and trends of temperature inversion characteristics in Greenland for the past four decades (1979–2017). These inversion patterns exert a strong influence on many climate impacts and have important implications for spatial variations in GrIS climate dynamics, and are thus in clear need of quantification. Besides assessing the climatological changes for various subregions in Greenland, this study is also dedicated to detecting the mechanisms driving the temperature inversion changes, especially in relation to changes in cloud cover and inversion strength.
The paper is structured as follows: section 2 describes the datasets employed in the study. Section 3 explains the statistical methods, the methodology for the calculation of inversion characteristics, and the applied regionalization of Greenland. Section 4 presents the characteristics of the SBIs, EIs, and NSBIs at the regional and seasonal scales, discusses potential sources of uncertainty using comparisons to observational data, and relates inversion trends to the recent changes in cloud cover. Finally, section 5 summarizes our main findings.
For the inversion analysis over all of Greenland we used the global climate reanalysis dataset, ERA-I, provided by ECMWF, which covers the period 1979–2019 (Dee et al. 2011). In particular, we used the analysis fields such as Tskin, T2m, and total cloud cover (TCC), and also upper-air parameters such as model-level air temperature and specific humidity.
Our study focuses on daily inversion statistics calculated from 0000, 0600, 1200, and 1800 UTC instantaneous values of air temperature at the lower 15 vertical levels of hybrid sigma-pressure coordinates out of 60 model levels from the surface to 0.1 hPa. We used the ERA-I dataset with a horizontal resolution of 0.25° × 0.25°, which is bilinearly interpolated from the native model grid of 0.75° × 0.75°, and the time period from January 1979 to December 2017 for all of Greenland. The heights of the respective model levels were calculated by applying the hydrostatic equation and are unevenly spaced at the following mean heights above ground level (AGL): 9, 31, 64, 111, 173, 254, 355, 477, 623, 792, 987, 1208, 1456, 1732, and 2037 m. We also include Tskin (as surface temperature) and T2m to investigate SBIs and EIs. ERA-I Tskin is the theoretical temperature of the uppermost surface (snow, ice, or soil) layer with no heat capacity derived from closing the surface energy balance and consequently, not based on observations. By contrast, ERA-I T2m is an independent assimilation product derived by optimal interpolation using dry-bulb T2m station observations and a background field coming from the lowest model level (located at a height of about 10 m) and skin temperature from the previous 6-h background forecasts (Simmons and Poli 2015; ECMWF 2016). The number of stations with T2m observations from Greenland and used for ERA-I is low, and stations are almost exclusively close to the coast [see Simmons and Poli (2015) for details]. Given the ERA-I approach, the model level atmospheric fields, from which the background forecast for the next analysis in the assimilation sequence is derived, are only very weakly forced (via the land surface scheme) by the T2m analysis.
ERA-I cloud processes are described by prognostic equations for cloud water/ice content and cloud cover based on the mass balance for total cloud condensate (Tiedtke 1993). Here we use the TCC, which represents the fraction of a grid box covered by cloud occurring at different model levels through the atmosphere (ECMWF 2016). Only those days when inversions occur are selected for further calculation.
Here, ERA-I is employed because it assimilates quality-controlled in situ and remote sensing observations and homogenized radiosonde temperature observations (Haimberger 2007); furthermore, the modeled parameters are physically consistent with the observations (Dee et al. 2011). In addition, the magnitude and vertical structure of temperature with respect to radiosonde observations over the Arctic have been improved by the use of bias correction in satellite radiances data (Dee and Uppala 2009), which further improved the simulation accuracy with respect to Arctic cloud properties and amounts (Walsh et al. 2009; Dee et al. 2011). Moreover, ERA-I shows smaller temperature errors (Graversen et al. 2008; Lindsay et al. 2014) compared to other reanalysis products such as the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996) or the Japanese Meteorological Service 25-year Reanalysis (JRA-25) (Onogi et al. 2007). In addition, ERA-I also includes a reasonable depiction of inversion properties (Tjernström and Graversen 2009) and trends (Screen and Simmonds 2010).
One major concern in using reanalysis data for trend estimation is that changes in the observing system may reflect artifacts rather than true climate signals (Bengtsson et al. 2004). In the case of Greenland, such artifacts might be related to changes in the vertical resolution of radiosondes or to the increase in the number of satellite observations. To account for these deficiencies, ERA-I is the first assimilation scheme that adjusts for biases introduced by changes in the observation network (e.g., additional satellite observations) with the aim of removing potential inhomogeneities (Dee et al. 2011). Furthermore, compared to ERA-40, ERA-I shows large improvements in quantifying the Arctic warming trend. This improvement may be related to a change in the processing of satellite radiances in 1997 (Screen and Simmonds 2010; Dee et al. 2011).
We are aware of the successor of ERA-I, the fifth-generation reanalysis, referred to as ERA5 (Hersbach and Dee 2016). ERA5 offer several improvements compared to ERA-I, most notably better spatial and temporal resolution, and more extensive observational inputs to the data assimilation system (Hennermann and Giusti 2018). Delhasse et al. (2020) compared the near-surface climate in ERA-I and ERA5 over the GrIS with the PROMICE (Programme for Monitoring of the Greenland Ice Sheet) stations and concluded that ERA5 does not significantly outperform ERA-I. Graham et al. (2019) evaluated the skill of ERA-I and ERA5 in representing independent radiosonde data in the Fram Strait and they concluded that ERA5 performs best among five reanalyses. However, they also found that, in several cases, better simulations of SBIs and EIs were achieved by ERA-I.
To raise confidence in our results, we compared the performance of ERA-I to ERA5 using quality-controlled radiosonde observations from an enhanced version of the Integrated Global Radiosonde Archive (IGRA) from Greenland (Durre and Yin 2008). Figure S1 in the online supplemental material reveals that both reanalyses products provide a close match to observations, with root-mean-square error (RMSE) values lower than 0.3 K, with the ERA5 match being slightly better than that of ERA-I. However, the general error pattern between ERA-I and ERA5 is consistent. This adds confidence to our choice of ERA-I as we expect that the signals discussed in this study will not differ significantly between ERA-I and ERA5.
b. Observational data
To evaluate the near-surface temperature of ERA-I, we used hourly T2m from the PROMICE network (Ahlstrøm et al. 2008), operated by the Geological Survey of Denmark and Greenland (GEUS), distributed in the ablation area of Greenland, and from the Zackenberg (ZAC) research station in northeast Greenland operated by the Greenland Ecosystem Monitoring program. These data are not assimilated within ERA-I (Delhasse et al. 2020) and thus provide an independent dataset for comparison. We also used hourly values of outgoing longwave radiation from all stations in order to calculate Tskin, and only high-altitude stations from the PROMICE network were selected in order to ensure perennial snow cover with known surface emissivity. Figure 1 shows the geographical distribution and corresponding abbreviations for all stations used in this study. Table 1 shows the temporal coverage and elevation difference between the stations and the corresponding ERA-I subsampled grid cell value (ERA-Is) which denotes the median of the nearest four grid cells in order to take account of the effective model resolution (Laprise 1992).
a. Inversion layer identification and classification: Surface-based and elevated inversion
For our study we implemented the objective inversion detection algorithm designed by Kahl (1990) to determine the temperature inversion characteristics, including embedded thin layers (thickness < 100 m) of negative lapse rate. The vertical temperature profile was scrutinized for inversions from 0 to 2086 m AGL for each grid point and for all daily time steps (0000, 0600, 1200, and 1800 UTC).
First, four inversion characteristics were identified as described by Kahl (1990): inversion base (zb), which is the elevation of the level above which the temperature increases; air temperature at zb (Tb); inversion top (zt), which is the elevation of the level above (below) which the temperature decreases (increases) with height; and air temperature at zt (Tt) (see Fig. 2). Second, the following inversion characteristics were then calculated: inversion thickness (Δz = zt − zb), strength (ΔT = Tt − Tb), gradient (Γ = ΔTΔz−1), and frequency (f; percentage of days with inversions) (Kahl 1990). Furthermore, we labeled inversion as “surface-based” (SBI) if zb is 0 m AGL (Fig. 2b), and SBI strength, thickness, and frequency as ΔTSBI, ΔzSBI, and fSBI, respectively. Conversely, inversions were labeled as “elevated” (EI) if zb is high above the ground (more than or equal to 2 m AGL), and EI strength, thickness, frequency, and base as ΔTEI, ΔzEI, fEI, and zEIb, respectively (Fig. 2b).
In most of the existing studies inversions are referred to as SBIs if zb is either at 2 m AGL (Zhang et al. 2011), or at 9 m AGL (Wetzel and Brümmer 2011; Palarz et al. 2018), or at any inversion layer within a boundary layer depth of 25 m (Tjernström and Graversen 2009). However, several other studies have demonstrated the presence of strong near-surface temperature inversion below 2 m (Hudson and Brandt 2005; Adolph et al. 2018; Nielsen-Englyst et al. 2019) and scanned the temperature profile starting from the “surface” (i.e., Tskin of the surface at the snow–air interface). Consequently, we based our SBI nomenclature starting from the surface, while evaluating the potential of ERA-I to capture these surface inversions by comparing with independent station observations (see section 3b).
We apply the following criteria in defining inversions (Fig. 2). If T2m is higher than Tskin, this profile is classified as SBI as well as NSBI (see detailed explanation in section 3b) and hence Tb equals Tskin. If a noninversion layer (where temperature decreases with height) of less than 100-m thickness is embedded between two inversion layers, then these two inversion profiles were combined and classified as SBI. However, after such combinations, we encountered some profiles where the upper inversion layer Tt was less than the lowest inversion layer Tb, resulting in a negative lapse rate. Whenever this situation arose, the highest level was skipped, and Tt was defined by the subsequent lower inversion layer. Additionally, this skipped inversion layer was counted as EI and the same aforementioned method was applied to determine the top of the given EI layer. If the thickness of the embedded noninversion layer was more than 100 m, then the lower inversion layer was classed as SBI and the upper as EI (Fig. 2b).
As the inversion characteristics are typically not normally distributed, we calculated median values to represent their central tendency (Kahl 1990; Kahl et al. 1992; Tjernström et al. 2012; Palo et al. 2017). First, we calculated the daily median of the inversion characteristics for each grid cell whenever more than one profile (0000, 0600, 1200, 1800 UTC) from a single day contained a temperature inversion. Whenever there was just one inversion profile a day, we used this profile to represent that day (Palo et al. 2017). Second, monthly median values were calculated by considering only those days with inversions. Furthermore, monthly f was calculated by dividing the number of days with at least one inversion by the number of days in a month. Finally, seasonal statistics of the inversion characteristics were calculated based on the meteorological seasons: winter [December–February (DJF); for December 1979 data are not included], spring [March–May (MAM)], summer [June–August (JJA)], and autumn [September–November (SON)]. To aid comparability across the subregions, we also show the spatial distribution of the inversion characteristics.
b. Deriving near-surface-based inversion from automatic weather stations
To investigate the presence of near-surface-based inversions, we analyzed the NSBI strength (ΔTNSBI) within the lowest 2-m vertical profile (i.e., the difference between T2m and Tskin). As previously mentioned, ERA-I Tskin is the numerical model result, which makes it necessary to assess the ability of the reanalysis system to represent the near-surface layer. Even though ERA-I T2m is constrained by T2m observations in Greenland, T2m observations from the PROMICE network are not assimilated in ERA-I (Delhasse et al. 2020), thus providing an independent dataset. Keeping this in mind, we present the direct comparison between ERA-IsTskin, T2m, and the daily median ΔTNSBI, with temporally, and spatially collocated AWS-based ΔTNSBI.
The Tskin value for all stations was estimated from the outgoing longwave radiation applying the Stefan–Boltzmann law using a surface emissivity of 0.97 for snow and ice (Van As 2011). To maintain procedural consistency, daily median ΔTNSBI for all the stations was calculated in the same way as daily median ΔTNSBI from ERA-Is. To constrain the data to the presence of snow and ice, only temperature differences (T2m minus Tskin) with a surface albedo greater than 0.3 were selected (Nielsen-Englyst et al. 2019). Furthermore, the median of the positive temperature differences at 0000, 0600, 1200, and 1800 UTC was calculated to represent the ΔTNSBI for that day for every station.
The corresponding daily ΔTNSBI for ERA-I was calculated first by extracting the hourly T2m and Tskin for each station’s location by linear interpolation, and second, by computing the median of the positive differences between the interpolated T2m and Tskin. Only the days with at least one inversion in both datasets were included for comparison.
One source of systematic bias in ERA-I might be due to elevation differences between ERA-Is and stations (Delhasse et al. 2020). The ERA-I elevation (m) for Greenland is computed by dividing the surface geopotential (m2 s−2) by Earth’s gravity (i.e., 9.8 m s−2; Fig. S2) and the corresponding ERA-Is elevation is extracted by linearly interpolating to the respective station’s location. Figure S3 shows the linear fit between the seasonal mean bias of Tskin, T2m, and ΔTNSBI as a function of the elevation bias between the ERA-Is and the respective AWS. Furthermore, the mean error (ME; Fig. S3) and the median absolute deviation (MAD; see Table 1) are used as a measure of accuracy.
c. Regionalization of Greenland
Greenland is characterized by extreme orographic and climatic heterogeneity (Steffen and Box 2001; Abermann et al. 2017). Furthermore, due to the presence of wide and dense polar pack ice drifting along the east coast with the East Greenland Current, the high Arctic zone extends much farther south on the east coast than on the west coast. Thus, the east coast is characterized by a more continental climate, with very cold winters and generally drier conditions (Stendel et al. 2008), compared to its western counterpart. These differences create strong horizontal and vertical temperature gradients, and, hence, gradient of temperature inversion conditions.
To account for this strong regional variability, we divided Greenland into seven climatic regions as described in Cappelen et al. (2001), namely Central (C) [called “Ice cap” in Cappelen et al. (2001)], North (N), Northeast (NE), Southeast (SE), South (S), Southwest (SW), and Northwest (NW) (see Fig. 1). Regional inversion characteristics presented here are spatially and temporally aggregated grid cell values within the given regions. The C region mostly covers the GrIS, while all other marginal regions along the coasts are influenced by the ice sheet, peripheral glaciers, and the ocean, and therefore show a significant variety of climate.
In the following sections, we report on the differences between the inversion characteristics analyzed among these seven subregions of Greenland. We applied statistical tests for each inversion characteristic in order to distinguish robust differences. The collective statistical tests employed comprise the nonparametric Wilcoxon rank-sum test and the Kruskal–Wallis test, followed by Dunn’s post hoc test with p adjustment using the Benjamini-Hochberg method to control the false discovery rate (Benjamini and Hochberg 1995) at a 95% confidence level.
Furthermore, we explore whether any significant trends can be detected in the inversion parameters for the given periods. Here, trends in all the subregions were estimated by applying the nonparametric Mann–Kendall test at the 5% significance level in each grid cell and trends were quantified by applying Sen’s slope estimator. Moreover, Spearman correlation is computed between detrended variables in order to prevent the trend artificially affecting the correlation. Only statistically significant trends, correlation coefficients (r), and coefficients of determination (R2) are presented, and the corresponding significant areal percentages reported upon. To compute respective region area, we first calculated the area covered by each grid cell within the given region, and then summed up the area of all the grid cells (denoted by the n symbol in Fig. S2) within the given region. This was done by using a general approximation of 1° corresponding to 111 km. Since the longitudes converge toward the poles, the distance between two longitudes were weighted by the cosine of the corresponding latitude. The total area of each grid cell is then calculated by multiplying the latitude distance with weighted longitude distance.
4. Results and discussion
a. Climatology of inversions in Greenland
The results of seasonal median inversion strength (ΔTSBI and ΔTEI), thickness (ΔzSBI and ΔzEI), frequency (fSBI and fEI), and height of EI base (zEIb) for the different climatic regions of Greenland, and for the different seasons are presented in Figs. 3 and 4. They show clear seasonal and regional patterns spanning from the low to the high Arctic regions in Greenland.
1) Seasonal patterns of inversion frequency
SBIs occur more frequently and intensively than EIs, and exhibit different characteristics in all the regions and seasons (Figs. 3 and 4). SBIs are more frequent during winter and autumn (94%–100%, 25th–75th percentiles) than summer and spring (fSBI = 50%–100%), whereas EIs show an opposite pattern (for DJF and SON fEI = 5%–22%; for JJA and MAM fEI = 9%–44%). SBIs are more frequent in autumn compared to spring (when also a clear north to south contrast is visible), while EIs show the opposite pattern (Table S1).
The high fSBI during the melt season in the low-lying coastal regions occurs due to surface melt consuming energy and hence cooling the layers above the surface. In the interior of the GrIS however, where melting is rare, high fSBI is due to strong radiation losses as a result of high albedo and a negative longwave radiation balance. This keeps the surface temperature well below 0°C and develops strong and shallow SBIs (Busch et al. 1982; Palo et al. 2017). Additionally, Wetzel and Brümmer (2011) suggest that the frequency of inversions during summer is additionally sustained by a downward direction of the conductive heat flux in the cold glaciers.
Miller et al. (2013) found a fSBI of 40% at the Summit station in Greenland in July 2011 using MWR, whereas ERA-IsfSBI is 17% higher for the summer (JJA median) of the same year. This discrepancy between observed and ERA-IsfSBI could be attributed to several factors, such as different time referencing, variations in the criteria for defining inversion (Zhang et al. 2011), surface elevation difference (e.g., for Summit, the difference with ERA-Is is 82 m), or subgrid cell spatial variability. However, despite the difference in magnitude, ERA-Is shows an annual cycle of inversion frequency statistics similar to that observed at the Summit station; SBIs are more frequent in winter (fSBI = 96%) compared to summer (fSBI = 57%) in ERA-Is for the same time period as in Miller et al. (2013) (see their Fig. 6; fSBI for DJF is ~95% and for JJA is ~47%). Also, the ERA-IsfEI shows a similar pattern, with a minimum in winter (9%) and a maximum in summer (52%) compared to Miller et al. (2013): fEI for DJF is ~3% and for JJA is ~13%.
2) Regional patterns of inversion frequency
Even though all the regions show high winter median fSBI (98%–100%), we find statistically significant regional differences that are likely associated with a gradient in insolation, ranging from no polar night in the S region to more than four months of polar night in the N region (Abermann et al. 2017). During winter, the development and migration of the low pressure systems from S toward the area between the east coast of Greenland and Iceland (Turton et al. 2019) might additionally cause differences in regional frequency. With the transition from winter to spring and the onset of summer, fSBI decreases overall with maximum occurrence in the cold and dry northern regions (N, NE, NW) (79%–98%) and lower occurrence in the warmer southern regions (S, SW, SE) (50%–93%). The lowest median summer fSBI is found in SW (72%) and the highest in NE (96%).
Likewise, winter fEI shows a statistically significant regional difference with maximum occurrence in NE (10%–17%), followed by S and SW (9%–15%) (Fig. 4 and Table S1). Contrary to fSBI, fEI increases in all regions during summer, showing a maximum in the southern (S, SE, SW) and central regions (13%–44%) and a minimum in the northern regions (N, NE, NW; 8%–33%).
Based on the radiosonde data from Tasiilaq, Danmarkshavn, and Ittoqqortoormiit for the period 1980–2016, Gilson et al. (2018b) found a higher fEI than fSBI in summer, while ERA-Is shows the opposite. However, fEI derived from ERA-Is is higher at NE stations (55%) than at the SE station (25%), which is in line with the spatial differences presented by Gilson et al. (2018b) (79% at NE stations and 69% at the SE station). It has been shown that it is very challenging to directly derive SBI characteristics from radiosondes as the lowest level of the radiosondes does not correspond to that of the ground level (Mernild and Liston 2010; Zhang et al. 2011). Another factor limiting the direct comparability of radiosondes and ERA-Is is the difference in surface elevations between actual radiosonde stations and corresponding ERA-Is (difference is −291 m MSL for Tasiilaq, −25 m MSL for Ittoqqortoormiit, and −75 m MSL for Danmarkshavn).
3) Seasonal patterns of inversion strength, thickness, and base
Over an annual cycle, SBIs are strongest and thickest in winter (6–24 K and 89–628 m) and get weaker in spring (4–18 K and 56–418 m) and even more so in summer (2.3–6.4 K and 60–179 m), then again increase during autumn (4.4–15.7 K and 79–319 m). Values of ΔTEI and ΔzEI also show a similar seasonal pattern: ΔTSBI values are approximately one order of magnitude higher than those of ΔTEI (Table S2), and ΔzEI is higher than ΔzSBI during summer (Table S3). To keep replicability and comparability to other studies, we did not apply an inversion strength criterion for defining inversion. So, the results for ΔzEI and ΔzSBI could differ depending on the use of such criteria as noted by Zhang et al. (2011).
The temperature gradient (Γ = ΔT Δz−1; K m−1) within the inversion layer is larger during winter than during summer (Fig. 5). However, the median winter Γ within the SBI layer (ΓSBI) for different regions of Greenland ranges from 0.03 to 0.04 K m−1 within a thick inversion layer, whereas for summer it lies between 0.04 and 0.05 K m−1 within a shallow inversion layer (Table 2). This pattern is in line with that observed by Bradley et al. (1992) for a coastal area in North American Arctic. Przybylak (2016) attributed such a strong Γ to the combination of subsidence-induced EI in the lower troposphere and strong SBI; we classified this kind of thermal profile as SBI. The Γ within the EI layer (ΓEI) is less steep than within the SBI layer for all the seasons (Table 2).
The EI base (zEIb) also shows a seasonality (Fig. 4); zEIb is higher during winter and the base is lower (even close to the ground) during summer; for example, zEIb is at 2 m in the SW during spring (Table S4).
4) Regional patterns of inversion strength, thickness, and base
Figures 3 and 4 clearly show that the spatial variability of ΔT and Δz is large during winter and small during summer. The north–south differences of ΔT and Δz are higher during winter (8.0 K and 351 m) compared to summer (1.5 K and 38 m) (Tables S2 and S3). During winter, ΔTSBI and ΔzSBI are highest in the N region (16.5 K, 462 m) and lowest in the SW region (7.2 K, 141 m) (Fig. 3). The reason for this spatial difference might be the dominance of anticyclonic activity in the N region, resulting in very low cloudiness and precipitation, especially at the leeward side of the ice sheet (Przybylak 2016). Furthermore, during winter, the regions near the east coast (especially NE) show stronger and thicker SBIs (9.3–18.1 K and 247–342 m) than those near the west coast, particularly SW (7.2 K and 141 m) (Fig. 3; Tables S2 and S3). This zonal difference can be attributed to the influence of sea ice along the east coast during winter, combined with clear, calm weather, and strong radiative cooling leading to cold surface temperatures. This contrasts with the southwestern region, which has much less sea ice (Cappelen et al. 2001) along the coast and is therefore relatively warmer. Additionally, different regional circulation regimes (e.g., the Icelandic and the Baffin Bay lows) govern the eastern and western regions of Greenland, respectively (Steffen and Box 2001), and may explain some of these regional differences.
The Γ values are comparatively higher in the southern regions (0.06–0.08 K m−1; see Fig. 5) in all the seasons. This is likely to be due to their proximity to mildly warmer air that is advected aloft (Bradley et al. 1992). This indicates that despite fewer and shallower inversions in the southern regions compared to northern regions, those in the southern regions can be very stable.
The zEIb values also show a clear regional pattern (Table S4). In the cold and dry northern regions of Greenland, where SBIs are dominant, the inversion base is higher (116–561 m AGL) than in the southern regions (2–285 m AGL), (Table S4). The zEIb values show largest regional variability during spring and summer.
b. Near-surface-based inversions in Greenland
We included the analysis of the ERA-I NSBIs in our study as they not only play a crucial role in near-surface atmospheric processes, but also allow for comparison with independent observations from surface stations. Such a comparison is important as NSBIs derived from ERA-I combine different reanalysis products (like T2m and Tskin). The accuracy of the inversion strength derived by ERA-I can be evaluated by directly comparing it with the station-derived inversion strength. However, stations represent point measurements, while ERA-I represents larger grid cells and both may be at different elevations; because of the strong elevation dependency of temperature, this could induce biases. To quantify these differences, we compared the Tskin, T2m, and ΔTNSBI values derived from stations with those derived from ERA-Is, and analyzed the associated elevation differences.
While biases are apparent in ERA-IsTskin, T2m (Figs. S4 and S5; Table S5), and ΔTNSBI (Fig. 6 and Table 1) in comparison with observations, the general agreement of frequency distributions is evident, in particular when the surface is covered with snow or ice in both data sources. The mean error in ΔTNSBI is a combination of biases in Tskin and T2m, which can at least partly be explained by the elevation difference. Summer mean error of ΔTNSBI shows a weak correlation with the mean error in elevation (i.e., R2 = 0.32) but is not statistically significant (Fig. S3).
Figure 6 shows the frequency distribution of ΔTNSBI for each station and its corresponding ERA-Is for the given period (see Table 1). Despite exhibiting a similar distribution, the MAD of the ERA-Is-based NSBI compared to the observations is comparatively small for stations located on the moderately flat ice sheet terrain (e.g., EGP; 2.7 K for DJF and 0.7 K for JJA) and comparatively large for stations located in complex mountainous terrain (e.g., ZAC; 5.9 K for DJF and 1.5 K for JJA) (Table 1). This relatively large value can be attributed to various factors such as the elevation difference between the actual station location and corresponding interpolated grid cell values (22 m for EGP; 530 m for ZAC), observation errors, model interpolation errors (Gao et al. 2012), and change in measurement height of the AWS during snowfall and melt (Nielsen-Englyst et al. 2019). However, for station EGP, as the winter MAD is based on only one winter (2017) of ΔTNSBI, this might explain the larger winter MAD value for this station.
ERA-I-derived NSBIs show similar seasonal and regional patterns similar to those of SBIs. In winter, the median ΔTNSBI is strongest (4.0–5.1 K), with greater spatial variability (2.6–8.8 K) in all regions compared to its summer counterpart (median, 1.6–2.5 K; range, 1.4–3.1 K). This can be attributed to the frequent passages of cold and warm air masses in wintertime compared to summertime (Steffen 1995) and the long period of the midnight sun reducing the north–south ΔTNSBI spatial variation during summer (Cappelen et al. 2001). Near Summit station for the summer 2015, we found fNSBI of 69% based on ERA-Is, which is in agreement with Adolph et al. (2018), who found a value of 68% based on near-surface meteorological observations. The ERA-Is ΔTNSBI (~1.4 K) is smaller compared with that of Adolph et al. (2018) [3.5 K (±2.4 K)]; however, the ERA-Is ΔTNSBI magnitude is not significantly different when observation uncertainty is taken into account.
c. Multidecadal trends of temperature inversion characteristics in Greenland
We limit our further analysis to summer and winter seasons considering the strong contrast in the seasonal pattern of inversion characteristics. Taking this into account, we show the multidecadal (1979–2017) change in the strength of different inversion types (ΔTNSBI, ΔTSBI, and ΔTEI) in Fig. 7.
There are distinct spatial and seasonal differences in the trends of all aforementioned variables. In winter, ΔTNSBI and ΔTSBI decrease significantly with a median trend of −0.1 and −0.8 K decade−1, in 24% and 12% of Greenland’s area (AG), respectively, over the period of 1979–2017 (Tables S6 and S7). The negative trends of ΔTNSBI and ΔTSBI are strongest in the warmer SE region (Tables S8 and S9). Additionally, ΔTNSBI decreases significantly in 27% of the C region, whereas a negative trend of ΔTSBI is significant in less than 5% of the C region. During the summer, on the other hand, there is a significant increase in ΔTNSBI and ΔTSBI with median values of 0.15 and 0.3 K decade−1 in 51% and 32% of AG, respectively, over 1979–2017, while these parameters exhibit a significant decrease in only 4% and 1% of AG. In the southern half of Greenland, both summer ΔTNSBI and ΔTSBI show a strong and significant positive trend (>0.3 K decade−1) (see Tables S8 and S9).
Interestingly, during the summer, we find for both ΔTNSBI and ΔTSBI, trends with two opposite signs in the C region. While the inversion strength is decreasing in the central part of the ice sheet, it increases toward the periphery and the south dome. Nonuniform warming of the surface and the air above leads to different inversion trends depending on the surface type: a warming of the air above a melting snow or ice surface increases the ΔT as Tskin cannot warm to more than 0°C (van den Broeke et al. 2009). In the central part of the ice sheet, Tskin is well below 0°C and increasing at a higher rate than the air above, which weakens inversions there.
While trends in ΔTEI also show spatial variation, they exhibit smaller seasonal variations; we find significant trends only in small areas of Greenland (a positive trend in the C and N regions). The trend pattern reverses as one moves from the center of the ice sheet toward the periphery (Fig. 7). However, for the most part, the decreasing trend is not statistically significant.
During winter, both ΔzSBI and ΔzEI show no significant trend for the largest part of Greenland, whereas during summer they show the strongest positive trend (8–16 m decade−1), especially in the western regions (Fig. S6, Tables S6 and S8). The trends for fSBI and fEI are clear during summer, but during winter no distinct trends are visible (Fig. S6). In summer, fSBI increased significantly, with a median trend of 3% decade−1 in 50% of AG, especially in the southern half and in the western regions. In contrast, the trend for summer fEI was significant in the opposite direction (−2% decade−1 in 29% of AG) in the areas where the fSBI trend is significant.
d. Temperature inversion strength and clouds
Changes in cloudiness, such as that caused by variation in the Greenland anticyclone (Hofer et al. 2017), may be an important driver for the surface temperature trends (Miller et al. 2015) and, consequently, for the inversions. To investigate the potential relationship between inversion characteristics and TCC, we computed their correlations. There is a close correspondence between changes in TCC and changes in ΔT for detrended data (Fig. S7) for the period 1979–2017. Notably, ΔT (especially ΔTNSBI and ΔTSBI) shows a significant negative correlation with TCC for both winter and summer seasons. We find a stronger anticorrelation between ΔTSBI and TCC during summer than during winter (for ΔTSBI, JJA r = −0.64, in 98% of AG compared to DJF r = −0.5, in 65% of AG). While ΔTNSBI showed a significant, albeit weaker anticorrelation with TCC than ΔTSBI, no seasonal differences were found (r = −0.4, in 28% of AG). This indicates that decreasing cloud cover correlates more with increasing atmospheric temperature than with near-surface temperature.
For the period 2002–15, studies show a positive anomaly in the Greenland Blocking Index (GBI) (Hanna et al. 2016), and accordingly, a decrease in cloud cover especially in southern Greenland (Hofer et al. 2017; Noël et al. 2019). Furthermore, Hofer et al. (2017) showed that the decreasing trend of summer cloud cover coincides with a reduction of the GrIS SMB, with the latter resulting from enhanced melt–albedo feedback. Emphasizing the period 2002–15, it can be seen that the increasing trends for JJA SBI characteristics coincide with decreasing TCC (Fig. 8), suggesting that TCC and SBI may be connected (note: no significant trend was found for DJF TCC and SBI; Fig. S8). On looking more closely, we find that TCC shows a strong anticorrelation with ΔTSBI for the time period 2002–15, explaining 35%–54% (in DJF for 44% of AG) and 41%–67% (in JJA for 75% of AG) of the variability in ΔTSBI (Fig. 9). The low R2 value (0.35–0.54) during DJF indicates that TCC is only one of the factors causing the ΔTSBI to change, and indicates that surface radiation deficit is likely to be the main reason (Serreze et al. 1992). In contrast, for JJA over 2002–15, the reduction of TCC explains on average 54% of the variability in ΔTSBI. Likewise, during JJA, 32%–46% of ΔTNSBI variability can be explained by changes in TCC (r = −0.62), but this is significant only in 14% of AG.
e. Vertical temperature and humidity trends in the southwestern part of Greenland
The relevance of the SBIs for the surface energy balance motivated us to look more closely into the potential differences in the inversion development on the accumulation and the ablation area of the GrIS. We focus on the southwestern part of the GrIS for JJA. This shows statistically significant trends in summer TCC and ΔTSBI for 2002–15. Following Noël et al. (2019), we use an equilibrium line altitude of 1450 m MSL for the southern GrIS in order to distinguish between accumulation and ablation areas. Figure 10 shows that in the accumulation area of the southwestern part of the GrIS, the increase in ΔTSBI and ΔzSBI during summer is forced by near-surface cooling, as well as by warming of the layers above (starting from ~100 m AGL). In contrast, for the ablation area, the increase in ΔTSBI and ΔzSBI is caused by an unchanged surface temperature and by a temperature increase in the layers above. These results are in line with what could be expected from an increased GBI and reduced TCC for JJA. On average, an intensified Greenland anticyclone will increase the downward vertical motion of air, consequently increasing the atmospheric temperature by adiabatic warming. Other effects such as increased outgoing longwave radiation from reduced TCC may also play an important role in enhancing surface cooling. Interestingly, the study from Westergaard-Nielsen et al. (2018) reported mostly the presence of cooling trends in surface air temperature for the period 2001–15 for the ice-free part of Greenland, which is in line with our results (not shown). Furthermore, both in the accumulation and ablation areas, the humidity deficit [HD; i.e., saturation specific humidity (qs) minus specific humidity (q)] is increasing at a smaller rate near the surface compared to the air above, which further supports the decreasing TCC trend. Given that Berkelhammer et al. (2016) have already indicated the role of inversions in decoupling the surface from free-tropospheric moisture above the Summit station, the HD finding tends to become even more relevant. In the ablation area, this confined moisture within the lower atmosphere may act to increase the latent heat flux toward the surface (Niwano et al. 2019). Consequently, the vertical gradients in temperature and humidity between the surface and atmosphere, and hence, inversions (especially SBI) can play a significant role in modulating surface–atmosphere energy and moisture exchange in the GrIS.
Our study has aimed at analyzing and better understanding the spatiotemporal variations and trends of temperature inversion characteristics in Greenland over the past four decades (1979–2017). For this purpose, we used ERA-I, as a physically coherent dataset covering the entirety of Greenland, and compared this with station data and previous studies from different regions of Greenland. Three types of inversions (SBI, EI, and NSBI) were defined, and we analyzed the strength, thickness, frequency, and base of each type in seven climatic regions of Greenland for various seasons. In general, the ERA-I dataset proved to be reliable in capturing the spatiotemporal patterns of inversions in Greenland. This was confirmed by comparing ERA-I results with radiosonde data, independent ground stations data, and with results from other studies. Further conclusions of our study are presented below.
a. Regional and seasonal patterns of inversions
SBIs occurred more frequently and intensively than EIs in all climatic regions for all seasons. SBI frequency is higher during winter (98%–100%) compared to summer (72%–96%), whereas EI frequency shows the opposite seasonal pattern (winter: 7%–14%, summer: 15%–31%). During winter, SBIs are stronger and thicker in the north (16.6 K and 462 m) compared to the south (8.5 K and 111 m), and stronger in the east (10.8 K and 196 m) compared to the west (7.8 K and 151 m). This is possibly due to the dominance of the Greenland anticyclone in the north, and to the influence of sea ice in the east. However, in summer, north-to-south inversion variability decreases, which is probably related to the smaller latitudinal differences in incoming solar radiation. Also, EI characteristics depict similar but weaker regional variability compared to SBI. The strongest temperature gradient is noticeable closest to the surface (within 2 m AGL). Our results of spatial and temporal patterns of inversions are generally consistent with previous studies for single locations or selected parts of Greenland, thus providing a consistent overall picture.
b. Trends in SBI
We point out, that during the summer, SBIs become stronger (0.3 K decade−1), thicker (12 m decade−1), and more frequent (3% decade−1) in the southern part of Greenland, especially in the past two decades. The persistent, positive summer GBI, indicating a strengthening of the Greenland anticyclone, and hence less summer total cloud cover (along with several associated feedbacks), plays a significant role in increasing inversion strength and thickness (mostly SBI), particularly in southwest Greenland. While for the accumulation area of the Greenland Ice Sheet the increased SBI strength is forced by near-surface cooling, the ablation area is characterized by constant surface temperatures. However, for both regions, the warming of the layers above is a key driver behind increased SBI strength during summer over the period 2002–15. In this context, we also argue that SBI might play a critical role in confining moisture within the lower atmosphere by decreasing the height of the atmospheric boundary layer.
It is clear that the large-scale analysis presented here cannot address all complex and interlinked processes with respect to inversions (e.g., the radiative effect due to changes in inversions). However, the present analysis does provide a solid footing for more detailed process-oriented studies and analysis concerning the latitudinal contrasting effects of inversion characteristics on the GrIS mass budget.
We sincerely thank all providers of data. ERA-Interim data were supported by ECMWF. Observational data were provided by PROMICE (operated by the Geological Survey of Denmark and Greenland) and Asiaq (Greenland survey). The University of Graz is acknowledged for funding the article publication. We are grateful to Leopold Haimberger, Paul Berrisford, and Gianpaolo Balsamo for their help in providing detailed information on the ECMWF datasets. We would also like to express our deep gratitude to the editor, Shawn Marshall, and all three anonymous reviewers for their valuable time and for their constructive comments and suggestions.
List of Abbreviations
a. Subregions of Greenland
b. Inversion types
c. Inversion characteristics
d. Surface-based and elevated inversion characteristics
Elevated inversion frequency
Surface-based inversion frequency
Elevated inversion base
Elevated inversion strength
Near surface-based inversion strength
Surface-based inversion strength
Elevated inversion thickness
Surface-based inversion thickness
Temperature gradient within elevated inversion layer
Temperature gradient within near surface-based inversion layer
Temperature gradient within surface-based inver sion layer
ERA-Interim subsampled grid cell value
Greenland Blocking Index
Greenland Ice Sheet
Saturation specific humidity
Surface energy balance
Surface mass balance
Total cloud cover
Air temperature in 2 m above ground level
Denotes content that is immediately available upon publication as open access.