As rapid warming of the Arctic occurs, it is imperative that climate
indicators such as temperature be monitored over large areas to understand
and predict the effects of climate changes. Temperatures are traditionally
tracked using in situ 2 m air temperatures and can also be assessed using
remote sensing techniques. Remote sensing is especially valuable over the
Greenland Ice Sheet, where few ground-based air temperature measurements
exist. Because of the presence of surface-based temperature inversions in
ice-covered areas, differences between 2 m air temperature and the
temperature of the actual snow surface (referred to as “skin” temperature)
can be significant and are particularly relevant when considering validation
and application of remote sensing temperature data. We present results from
a field campaign extending from 8 June to 18 July 2015, near Summit
Station in Greenland, to study surface temperature using the following
measurements: skin temperature measured by an infrared (IR) sensor, 2 m air
temperature measured by a National Oceanic and Atmospheric Administration
(NOAA) meteorological station, and a Moderate Resolution Imaging Spectroradiometer (MODIS) surface temperature product. Our data indicate
that 2 m air temperature is often significantly higher than snow skin
temperature measured in situ, and this finding may account for apparent
biases in previous studies of MODIS products that used 2 m air temperature
for validation. This inversion is present during our study period when
incoming solar radiation and wind speed are both low. As compared to our in
situ IR skin temperature measurements, after additional cloud masking, the
MOD/MYD11 Collection 6 surface temperature standard product has an RMSE of
1.0
The Arctic is experiencing warming at a more rapid rate than the rest of the world (Stocker, 2014), but the impacts of this increased temperature extend beyond the polar region. Declining sea ice extent and retreat of glaciers contribute to a powerful ice–albedo feedback that results in further warming on a large scale. This increased warming leads to declining mass balance of the Greenland Ice Sheet, contributing to global sea level rise. Quantifying current and future ice sheet mass balance remains an active area of research (e.g., Rignot et al., 2011; Rae et al., 2012; Vernon et al., 2013) and is critical to improving projections of sea level rise. Declining Greenland Ice Sheet mass balance is driven in part by changes in surface energy balance, which drives surface temperature and surface melt (Box, 2013; van den Broeke et al., 2016). Tracking surface temperatures then allows us to monitor surface melt for mass balance considerations and also informs our understanding of key ice sheet surface processes. Surface temperature changes result from fluctuations in the surface energy balance, which controls the exchange between the snow surface and the atmospheric surface layer. The surface energy balance (dependant on net radiation, sensible and latent heat fluxes, and conduction from underlying snow and ice) affects the stability of the near-surface atmosphere and the extent to which turbulent heat exchange occurs between the snow surface and the lower atmosphere, impacting both local and regional circulation and climate. Surface temperature processes also play an important role in paleoclimate records that are stored within ice sheets (Waddington and Morse, 1994; Van Lipzig et al., 2002).
Surface temperature is a critical component for monitoring ice sheet mass balance, tracking changes in surface energy balance and atmospheric exchange, and understanding processes that affect paleoclimate records; however, making accurate measurements of surface temperature across the vast expanse of the Greenland Ice Sheet over a long period of time is challenging (Reeves Eyre and Zeng, 2017). The installation of automatic weather stations (AWS) across the ice sheet has begun to provide point meteorological data at many locations through programs such as Greenland Climate Network (GC-Net) (e.g., Steffen et al., 1996; Steffen and Box, 2001; Shuman et al., 2001) and the Programme for Monitoring of the Greenland Ice Sheet (PROMICE), which monitors both skin and air temperatures (e.g., Ahlstrøm et al., 2008; van As et al., 2011; Fausto et al., 2012). In addition, thermal infrared (IR) satellite remote sensing provides the opportunity to collect surface temperature with large spatial coverage and sub-daily to weekly temporal resolution, depending on cloud conditions. In this study, we will focus on the Moderate Resolution Imaging Spectroradiometer (MODIS) thermal IR land surface temperature (LST) product.
“Surface” temperatures in climatological studies often refer to 2 m air temperature (Hudson and Brandt, 2005) as it is a standard measurement at meteorological stations around the globe; however, remotely sensed surface temperatures from satellite-borne sensors in the cryosphere measure the radiometric surface temperature, which is the actual “skin” temperature of the surface at the snow–air interface (Warren and Brandt, 2008). Thermal stratification near the snow surface causes differences between the 2 m air temperature and the skin temperature. Incoming solar irradiance and wind speed are two major controls on thermal stratification. Temperature inversions occur when the incoming solar irradiance is small (i.e., during night) and the snow surface emits longwave radiation; the net radiation at the surface is negative, causing heat transport from the air to the snow surface and lower temperatures at the snow surface than in the air directly above it. The opposite phenomenon of temperature lapse can occur when there is significant incoming solar irradiance resulting in net positive radiation at the surface, with higher temperatures closer to the ground surface and upward heat transport from the snow surface to the air. Winds can serve to neutralize these temperature gradients by mixing air masses. In the polar regions, the high albedo of snow in the visible part of the spectrum means relatively little solar radiation is absorbed even during periods of sunlight. Combined with high emissivity of snow at longer wavelengths as compared to the emissivity of the atmosphere, conditions in polar regions often result in the presence of inversions.
The presence of surface-based inversions in the hundreds of meters of the
lower atmosphere in the polar regions has long been established (Sverdrup,
1926), and the phenomenon can be
detected through measurements of temperature at two or more heights to
determine the magnitude and sign of the temperature difference over the
relevant height difference. Lower atmospheric inversions have been
characterized in Greenland and the wider Arctic (Reeh, 1989; Kahl, 1990;
Overland and Guest, 1991) as well as in Antarctica (Philpot and Zillman,
1970). “Surface-based” inversions have typically been studied with 2 m air
temperature as the base of the inversion and the height of the inversion
extending hundreds of meters or more into the atmosphere. However, work by
Hudson and Brandt (2005) demonstrated the presence of a surface-based
temperature inversion below 2 m in the winter of 2001 at South Pole in
Antarctica, showing that the largest temperature gradient was in the 20 cm
nearest to the snow surface. Hall et al. (2008) analyzed 2 m air temperature
data and skin temperature data from across Greenland and discussed conditions
that lead to near-surface thermal stratification over snow-covered areas.
Good (2016) presents measurements of skin temperature and 2 m air
temperature and finds that at polar sites, during snow-covered seasons in
fall, winter, and spring, these two temperatures generally agree well, with
the caveat that there is a reduced amplitude of diurnal cycle temperatures at
2 m, which would imply a temperature inversion during the night and a
temperature lapse during the day. In work using satellite data to study
warming trends in the Arctic, Comiso (2003) presents a dataset from an Arctic
sea ice study showing correlation between 2 m air temperature and skin
temperature that had been averaged monthly. Over sea ice, there was an
average offset of 0.34
In recent years, studies have been conducted on surface energy balance and near-surface processes in Greenland (e.g., Miller et al., 2013, 2015, 2017; Berkelhammer et al., 2016) and Antarctica (e.g., van As et al., 2005; van den Broeke et al., 2006; Kuipers Munneke et al., 2012). At our study site at Summit, Greenland, Miller et al. (2013) studied the inversions over 2 years but considered the 2 m air temperature to be the base of these inversions, and they did not investigate the surface processes beneath 2 m height. They find that inversions are prevalent in winter months and are less intense during summer months and that the presence of clouds results in weaker inversions. In Miller et al. (2015) the impact of clouds on the surface energy budget at Summit is further investigated, and the warming effect of clouds on 2 m air temperatures is shown in all seasons. Details of the Summit, Greenland, surface energy balance are extensively documented in Miller et al. (2017). Berkelhammer et al. (2016) discuss the impacts of the surface-based temperature inversions on boundary-layer dynamics, showing that the stability of the atmosphere prevents mixing and ultimately limits accumulation at Summit. These recent studies have investigated near-surface processes at Summit because of the importance of surface energy balance and turbulent snow–atmosphere exchange in climate monitoring and ultimately prediction of larger-scale circulation and future change in ice mass balance. Though some surface temperature measurements at Summit have been made (Berkelhammer et al., 2016), controls on surface temperature gradients in the lowest 2 m of the atmosphere, which are most relevant for the remote sensing community and also have important implications for changing ice sheet dynamics, have not been explicitly studied at Summit, Greenland.
In remote sensing validation studies or use of remotely sensed temperatures, this distinction between 2 m air temperature and skin temperature is important and has been demonstrated in polar regions (Comiso, 2003). Indeed, best practices for thermal remote sensing validation indicate that ground-based radiance measurements that yield a skin temperature provide the best validation of remote sensing land surface temperature products (Guillevic et al., 2017). Because these data have not always been available, previous studies have used a variety of measurement types for remote sensing surface temperature validation.
A number of validation studies present results acquired over various
timescales and in different locations to determine the accuracy of the MODIS
surface temperature products in the cryosphere (Hall et al., 2004, 2008,
2015; Koenig and Hall, 2010; Westermann et al., 2012; Hachem et al., 2012;
Shuman et al., 2014; Østby et al., 2014; Shamir and Georgakakos, 2014;
Williamson et al., 2017). Table 1 provides summary statistics related to the
results of many of these validation studies and is discussed in further
detail in the discussion section. Overall, a negative bias is present in
nearly all validation studies, where the MODIS surface temperature is lower
than the measured skin or 2 m air temperatures, and this bias is
particularly prevalent at temperatures below
Summary statistics from recent literature comparing MODIS surface temperature products to in situ surface temperature measurements in snow-covered regions.
In the summer of 2015, we conducted a field campaign near Summit Station, Greenland, to measure skin and near-surface air temperature to study near-surface thermal stratification and determine its impact in validation of the MODIS land surface temperature product. We use our original dataset to determine how summertime meteorological conditions impact near-surface inversions (beneath 2 m height) on the ice sheet at Summit. Furthermore, we provide a validation of MODIS land surface temperatures and show that the use of 2 m air temperature for MODIS validation is not recommended due to the presence of near-surface inversions. Lastly, we use in situ cloud data to show that the accuracy of the MODIS surface temperature product could be improved through stricter cloud masking.
To characterize snow skin temperature, an autonomous measurement station was
installed approximately 10 km NNW of Summit, Greenland (indicated on a map
in Fig. 1), at an undisturbed site for 40 days between 8 June and 18 July
2015. A Campbell Scientific Apogee Precision IR radiometer (model SI-111) was
used to measure skin temperature of the snow. The instrument covers the
wavelength range from 8 to 14
Map indicating the location of Summit, Greenland, the study site
for remote sensing and in situ temperature comparisons. Contour lines
represent elevation change of 500 m. Latitude and longitude coordinates for
the measurement site are 72.65923
Summit Station was the location of the Greenland Ice Sheet Program 2 (GISP2)
deep core site and has operated continuously as a year-round station for
nearly a decade. The National Atmospheric and Oceanic Administration (NOAA)
has operated a meteorological station at Summit, measuring the 2 m air
temperature using a shielded Logan PT139 sensor. Additionally, wind speed
and incoming solar radiation data were also measured as part of the NOAA
station data (NOAA ESRL Global Monitoring Division, 2017). The data provided
by NOAA and used in this paper have a 1 min temporal frequency, and we
take a 30 min average of the data so that the 2 m air temperature is
comparable to the IR skin temperature measurements. Further details of the
2 m air measurements are outlined in Shuman et al. (2014). Additionally,
through the Integrated Characterization of Energy, Clouds, Atmospheric
state, and Precipitation at Summit (ICECAPS) project, a number of
instruments to monitor cloud, atmosphere, and precipitation were installed
at Summit in 2010. One of these instruments is the millimeter wavelength
cloud radar (MMCR), a custom-built Doppler 35 GHz radar that measures
reflectivity, mean Doppler velocity, Doppler spectra, and Doppler spectrum
width (data available at
Image of the IR skin temperature sensor and tripod setup.
Time series of skin temperature at Summit, Greenland, measured with SI-111 IR thermometer. Gray vertical bars indicate presence of clouds as detected by a millimeter cloud radar at Summit Station.
There are many different remote sensing instruments that measure radiance in
the thermal IR part of the electromagnetic spectrum to determine skin
temperature, including the Advanced Very High Resolution Radiometer (AVHRR),
the Advanced Thermal Emission and Reflection Radiometer (ASTER), the
Enhanced Thematic Mapper Plus (ETM
The MODIS instrument produces widely used LST,
which we use as the remote sensing product in this work. This instrument,
aboard the Terra and Aqua satellites, has been collecting radiance data from
24 February 2000 to present. The surface temperature products of the
Greenland Ice Sheet are used as a baseline to investigate surface
temperature trends (e.g., Hall et al., 2012), to monitor melt events on the
ice sheet (Hall et al., 2013), and as input for surface mass balance or
snowpack modeling (Fréville et al., 2014; Shamir and Georgakakos, 2014;
Navari et al., 2016). In this study, we use the MOD/MYD11 Collection 6 (C6)
product, where MOD refers to the Terra MODIS product and MYD refers to the
Aqua MODIS product. This product has a pixel size of 1 km
The MOD/MYD11 algorithm was developed to map land surface temperature (Wan
and Dozier, 1996; Wan, 2008, 2014) using radiance in MODIS bands 31 and 32,
which correspond to center
wavelengths of 11 and
12
Previous MODIS surface temperature validation studies have used Collection 5
(C5) products; C6 products started to become available in
2014. Improvements were made in the C6 MODIS algorithms, most notably to
rectify degradation of some sensors on the Terra satellite. However, the
sensor degradation was largely affecting bands in the visible part of the
spectrum and not in the thermal IR part of the spectrum used to
calculate surface temperature (Lyapustin et al., 2014; Polashenski et al.,
2015; Casey et al., 2017). MOD/MYD11 C6 benefits from improved stability of
emissivity values and improved algorithms to account for viewing angle over
its C5 counterpart (Wan, 2014). Additionally, in C6, the calibration of
bands 31 and 32 (used in surface temperature calculation) is improved.
Supplement Fig. S1 shows comparisons of C5 and C6 data at our study site
over the time period of interest. On average, C6 results in temperatures
0.2
Time series of IR skin temperature and 2 m air temperature during a clear sky period near Summit, Greenland.
The high-latitude location of Summit, Greenland, puts it within the field of
view of the MODIS instruments on Terra and Aqua multiple times each day. To
compare in situ measurements to the temporally coincident MODIS collections,
we use swath-level products whose file names contain the UTC time of
collection within
The IR skin temperature measurements operated continuously during the 40-day
campaign. The station was visited several times between 8 June and 25 June,
though no maintenance was required, and then left unmaintained for the
remainder of the measurement period. A time series of the IR skin
temperature is presented in Fig. 3. The snow skin temperature varied
between approximately
Difference between 2 m air temperature and IR skin temperature showing the presence of strong surface-based inversions at low wind speeds and low values of incoming solar radiation (indicated by the marker color).
Our IR skin temperature measurements are compared in a subset time series to
the 2 m air temperature measurements at Summit Station in Fig. 4. This
time window shows a clear sky period when diurnal cycles are clear and
conditions for inversion are most favorable. Thermal stratification in the
lowest several meters of the atmosphere is prominently seen in the
difference between 2 m air temperature and IR skin temperature (Fig. 4). The 2 m air temperature and IR skin temperature are similar during peak solar
irradiance, with the mean difference in temperature equal to
Figure 6 shows the magnitude of the temperature difference between 2 m and
snow skin temperature as a function of concurrent wind speed, with the color
of the marker indicating the concurrent incoming solar radiation. It is
clear that increasing wind speed serves to reduce any temperature gradient
in the lower meters of the atmosphere and that at peak solar radiation
there are no inversions present. These differences are much higher at lower
wind speeds; a stronger wind shear allows the system to overcome the
stability in temperature and promotes heat flux from the air to the snow
surface. Weaker winds cannot overcome the temperature stability so the
temperature differences persist. Specifically, for the data presented here,
at incoming solar radiation above 600 W m
The presence of this near-surface thermal inversion is of particular
interest in the context of previous MODIS surface temperature comparison
studies. Several studies have used 2 m air temperature to compare to MODIS
surface temperature products (Hall et al., 2004, 2008; Shuman et al., 2014).
These studies consistently report a “cold bias” in the MODIS surface
temperatures (see Table 1), where MODIS surface temperature is lower than
concurrently measured 2 m air temperature. In Shuman et al. (2014), a
comparison of MOD29 to 2 m air temperature results in a cold bias of
approximately 3
Hall et al. (2008) present a figure (their Fig. 2) similar to our Fig. 5a, in which measured IR skin temperature is plotted vs. 2 m air temperature
measured at Summit Station in Greenland from 2000 to 2001. However, they
found a consistent offset between 2 m air temperature and skin temperature
(of approximately 1
Time series as shown in Fig. 3 with only a temporal subset of data presented to clearly show the diurnal cycle of temperature during fairly clear conditions. Note that the MOD/MYD11 product shows good agreement with IR skin temperature throughout the diurnal cycle.
Figure 7 shows a time series of a subset of the measurement period with the
30 min IR skin temperature measurements overlain with the MOD/MYD11 LSTs.
MOD/MYD11 does not provide a surface temperature when the cloud mask
indicates that there are clouds present, which is why there are some gaps in
the data (i.e., at day 186/187). Most of the time series shown in Fig. 7 is
during a consistently cloudless period. Terra (MOD) passes over Summit
several times in the latter half of the day as temperatures are dropping.
Aqua (MYD) passes over Summit as temperatures are typically increasing
within the diurnal cycle. The algorithm to calculate temperature from
measured radiance is the same in the two different satellites. Figure 7
shows that there is generally good agreement between IR skin temperature and
both MOD11 and MYD11 products. This is also evident in Fig. 8a, where
MOD/MYD11 products combine to yield and RMSE of 1.6
Difference in temperature measured from MOD/MYD11 and in situ IR
skin temperature measurements as a function of
While we do not believe that 2 m air temperature is a good proxy for skin
temperature, for demonstration purposes we have compared the 2 m air
temperature measurements to the MOD/MYD11 product in Fig. 8b. In doing so,
we find an RMSE of 3.1
As compared to other MODIS validation studies, these results indicate a
closer match between in situ measurements and MODIS temperature products, as
indicated by smaller RMSE and mean bias (see Table 1). While the length of
our study is short in comparison to many of the other works referenced, the
use of a different in situ sensor is likely a key factor, and there is still
a significant range of temperatures captured within our study. In comparing
our results to other studies, it is also important to consider that we are
using a C6 product, which has seen improvements from previous
versions. The C5 cloud mask was more conservative over the
Greenland Ice Sheet than the C6 cloud mask. If we consider only
swaths that are considered cloud-free by both C5 and C6 (
However, there are still some differences between IR skin temperature and
MODIS surface temperature in our validation study. To investigate the root of
these discrepancies, we consider the sensitivity of the difference between
MOD/MYD11 surface temperature and in situ skin temperature as a function of
the following parameters: IR skin temperature, solar zenith angle, and sensor
viewing angle. These results are presented in Fig. 9. The only significant
relationship is between temperature difference and MODIS sensor view angle
(
Using the MMCR data from Summit, we identify periods when there were clouds
present above Summit Station. While our IR skin temperature measurements
were 10 km away, we believe that this is still a relatively good proxy for
cloudiness, as we resample the data to cover a 30 min window, so we feel
it is more reflective of a larger area. Figure 10 shows a comparison of IR
skin temperature to the MOD/MYD11 reduced data, when cloud-affected pixels
are removed. There is an improvement in the RMSE of the data comparison when
the cloud-affected data are removed (from 1.6 to 1.0
Comparison of MOD/MYD11 to in situ IR skin temperature after
cloud-affected data are removed. The RMSE is 1.0
Data collected during a 40-day field campaign at Summit, Greenland, in June
and July of 2015 are used to improve understanding of near-surface
temperature on an ice sheet, particularly with respect to MODIS LST
retrieval products. We find that at Summit, 2 m air temperature is often
significantly higher than skin temperature during the summer months,
particularly at periods of low incoming solar radiation and low wind speed.
This result is important because previous studies that have used 2 m air
temperature to validate MODIS surface temperature products have concluded
that there was a cold bias in the MODIS data, but our results indicate that
the MODIS data have only a very slight cold bias (
IR skin temperature data are made publicly available
through the NSF Arctic Data Center (Adolph et al., 2018). Data including 2 m
air temperature, wind speed, and irradiance were acquired from NOAA's Earth
System Research Laboratory Global Monitoring Division
(
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
We thank the National Science Foundation and Vasilii Petrenko for the opportunity to conduct this field work and Polar Field Services and staff at Summit Station for logistical support. We would like to acknowledge the editor and two anonymous reviewers for their constructive feedback. This work was funded by NSF-GRFP 2014186404 and NSF-1506155. Dorothy Hall was funded by NASA-NNX16AP80A. Edited by: Michiel van den Broeke Reviewed by: two anonymous referees