Spatial and temporal variations of summer surface temperatures of high-arctic tundra on Svalbard — Implications for MODIS LST based permafrost monitoring

https://doi.org/10.1016/j.rse.2010.11.018Get rights and content

Abstract

The ground surface temperature is one of the key parameters that determine the thermal regime of permafrost soils in arctic regions. Due to remoteness of most permafrost areas, monitoring of the land surface temperature (LST) through remote sensing is desirable. However, suitable satellite platforms such as MODIS provide spatial resolutions that cannot resolve the considerable small-scale heterogeneity of the surface conditions characteristic for many permafrost areas. This study investigates the spatial variability of summer surface temperatures of high-arctic tundra on Svalbard, Norway. A thermal imaging system mounted on a mast facilitates continuous monitoring of approximately 100 × 100 m of tundra with a wide variability of different surface covers and soil moisture conditions over the entire summer season from the snow melt until fall. The net radiation is found to be a control parameter for the differences in surface temperature between wet and dry areas. Under clear-sky conditions in July, the differences in surface temperature between wet and dry areas reach up to 10 K. The spatial differences reduce strongly in weekly averages of the surface temperature, which are relevant for the soil temperature evolution of deeper layers. Nevertheless, a considerable variability remains, with maximum differences between wet and dry areas of 3 to 4 K. Furthermore, the pattern of snow patches and snow-free areas during snow melt in July causes even greater differences of more than 10 K in the weekly averages. Towards the end of the summer season, the differences in surface temperature gradually diminish. Due to the pronounced spatial variability in July, the accumulated degree-day totals of the snow-free period can differ by more than 60% throughout the study area. The terrestrial observations from the thermal imaging system are compared to measurements of the land surface temperature from the MODIS sensor. During periods with frequent clear-sky conditions and thus a high density of satellite data, weekly averages calculated from the thermal imaging system and from MODIS LST agree within less than 2 K. Larger deviations occur when prolonged cloudy periods prevent satellite measurements. Furthermore, the employed MODIS L2 LST data set contains a number of strongly biased measurements, which suggest an admixing of cloud top temperatures.

We conclude that a reliable gap filling procedure to moderate the impact of prolonged cloudy periods would be of high value for a future LST-based permafrost monitoring scheme. The occurrence of sustained subpixel variability of the summer surface temperature is a complicating factor, whose impact needs to be assessed further in conjunction with other spatially variable parameters such as the snow cover and soil properties.

Research Highlights

► Monitoring of summer surface temperatures of arctic tundra using thermal imaging. ► Average differences of up to 4K between wet and dry areas. ► Pronounced spatial differences of the thaw index. ► MODIS LST can reproduce weekly averages to within 2K most of the time.

Introduction

Permafrost occurs in about a quarter of the land masses in the Northern Hemisphere (Brown et al., 1997). Particularly in arctic and subarctic regions of Russia, Canada and Alaska, permafrost-related processes play a crucial role in the energy and water cycle and thus determine the hydrology and ecology of these regions. In the past decade, a wealth of studies has gathered convincing observational and modeling evidence for a sustained warming of the Arctic (e.g. Comiso et al., 2008, Hinzman et al., 2005, Overland et al., 2008, Serreze et al., 2000, Turner et al., 2007). The ongoing warming is reflected in increasing permafrost temperatures (e.g. Osterkamp, 2005), and a sizable reduction of the permafrost area is projected until 2100 (Delisle, 2007, Lawrence et al., 2008).

Permafrost areas have received increased attention through scenarios, which suggest a positive climatic feedback due to massive emissions of methane from thawing organic-rich permafrost soils (e.g. Walter et al., 2006, Zimov et al., 2006). Furthermore, anticipated changes in the hydrological cycle and the landcover and vegetation may have strong impacts at least on the regional scale (e.g. Chapin et al., 2005). Considering such drastic projections, it seems imperative to continuously map and monitor the thermal state of the permafrost world-wide in order to assess its vulnerability to climate change. While a monitoring program of permafrost temperatures in boreholes has been initiated (Burgess et al., 2000), it is evident that the vast and remote permafrost areas cannot be sufficiently covered by such a labor-intensive scheme alone. It is thus desirable to better exploit the potential of remotely sensed data sets to complement terrestrially based monitoring. However, as permafrost is a ground temperature phenomenon, none of the potential measures used to characterize permafrost, such as the thaw depth, the unfrozen water content or the temperature at a certain depth, is directly accessible through remote sensing techniques. Some studies have evaluated the performance of proxies for the state of the permafrost, such as the occurrence of certain vegetation types (Stow et al., 2004, Walker et al., 2003) or the increase or shrinkage of lakes (Smith et al., 2005, Yoshikawa and Hinzman, 2003), which can be detected in high-resolution satellite images in the visible (VIS) or near-infrared (NIR) part of the electromagnetic spectrum. However, such indicators are usually only meaningful for limited areas, so that it seems questionable that they can be used for a larger-scale monitoring.

Remotely sensed land surface temperatures (LST) are a more universal proxy. In long-term records of satellite-derived LST, a warming trend over the arctic land masses has been observed (Comiso, 2006, Comiso and Parkinson, 2004), but the implications for the thermal state of the permafrost remain debatable, since the relationship between LST and subsurface temperatures is complex. In permafrost regions with a dense vegetation cover, for instance, satellites can only access the top temperature of the canopy which may be different from the actual ground surface temperature relevant for the permafrost. However, modeling schemes can be employed in order to assess the subsurface conditions based on the surface or even top-of-canopy information provided by satellite-based LST measurements. Various levels of complexity are conceivable, from a simple frost and thaw index approach (Nelson & Outcalt, 1987) to transient conductive heat transfer models, which deliver the full temperature distribution in the ground (e.g. Romanovsky & Osterkamp, 1997). Marchenko et al. (2009) propose a permafrost monitoring scheme based on the existing time series of remotely sensed LST and an equilibrium permafrost model (Sazonova & Romanovsky, 2003), which could deliver active layer thickness and mean annual ground temperatures on the pan-arctic scale. Despite of the large number of required additional input parameters, such as information on the snow cover and the composition of the soil, such concepts may possess the flexibility to incorporate the full range of different climatic, geological and ecological conditions found in permafrost areas. The application of such an LST-based scheme for Northern Quebec and Labrador, Canada (Hachem et al., 2009) highlights the prospects for large-scale permafrost monitoring.

The land surface temperature is currently accessible by a number of remote sensing platforms, some of which provide a high overpass frequency at the expense of spatial resolution (e.g. Terra MODIS, Aqua MODIS, ERS-AATSR, and NOAA-AVHRR), while others follow the opposite concept (e.g. Landsat TM and Terra ASTER). For the purpose of permafrost monitoring, daily to monthly LST averages are required, as high-frequency fluctuations of the surface temperature, such as the daily cycle, are damped close to the surface and do not affect the temperature evolution of deeper soil layers. However, accurate temporal averages of LST can only be obtained, if a sufficient overpass frequency is provided. In this point, satellite platforms such as Terra/Aqua MODIS, which under ideal conditions are even capable to resolve the daily cycle, are superior to high-resolution sensors such as Landsat TM, which can only provide a few measurements per averaging period. On the other hand, as strong variations of the surface cover and the soil moisture conditions occur on scales of meters in permafrost regions, pronounced subpixel differences of the surface temperature are conceivable for a medium-resolution sensor such as MODIS, which has a footprint area on the order of one square kilometer. Over hilly, heterogeneous terrain in China, Liu et al. (2006) detected a much larger range of LST values in 90 m-ASTER thermal images compared to 1 km MODIS scenes acquired at the same time, which reflects the area-averaging of the thermal signal performed by the medium-resolution sensor MODIS. This subpixel variability is a challenge for monitoring algorithms of environmental parameters, such as evapotranspiration, based on remotely sensed LST (Kustas et al., 2003, McCabe and Wood, 2006). For a permafrost monitoring scheme, a pronounced subpixel variability of LST would add additional complexity, as the employed heat transfer models are highly non-linear due to the phase change of water.

For polygonal tundra in the Lena River Delta, Siberia, Langer et al. (2010) show that small-scale variations of the surface temperature between wet and dry areas are strongly reduced in temporal averages of at least a few days. Therefore, temporal LST averages derived from Terra/Aqua MODIS can indeed be assigned to both wet and dry areas within a reasonable margin of error at the investigated site. While this is encouraging for the use of MODIS LST in permafrost monitoring, the results cannot necessarily be generalized to other permafrost areas. Thus, it is desirable to evaluate the controlling factors for the formation of subpixel variability in a series of intercomparable studies at different sites, that span the immense range of land cover types and climatic conditions found in permafrost areas.

In this study, we present spatially resolved measurements of the summer surface temperature at a high-arctic continuous permafrost area on Svalbard, Norway, using a thermal imaging system similar to the one employed by (Langer et al., 2010). The observation period includes the entire summer season from beginning of July until mid of September. We illuminate various sources of spatial variability of the surface temperature at this permafrost site and discuss the implications for the interpretation of remotely sensed LST. In a second step, we evaluate the performance of the MODIS L2 product (Wan, 2008) for the study period and give an assessment of the suitability for permafrost monitoring.

Section snippets

Study site

The study is conducted on the Brøgger peninsula in NW Svalbard, about 1.5 km SW of the village of Ny-Ålesund (Fig. 1), where a large number of measurement campaigns and long-term monitoring programs to have created a remarkable data basis on climatological and atmospheric variables (e.g. Beine et al., 2001, Winther et al., 2002). The maritime influence of the nearby Kongsfjorden and North Atlantic Ocean leads to cool summers (average air temperature July + 5 °C) and relatively mild winters

Setup

The measurements of the surface temperature are performed with a high-resolution thermal imaging system (VARIOCAM HR™, Infratec GmbH, Dresden, Germany), which is based on an uncooled microbolometer array with a resolution of 384 × 288 pixel. The thermal imaging system is sensitive in the atmospheric window from 7.5 to 14 μm, where molecular absorption and emission of electromagnetic radiation are weak, and features a 12 mm wide angle lens. A more detailed description of the thermal imaging system

Results

With the combined information of the 2008 and 2009 study periods, the surface temperature can be investigated for almost the entire snow-free period. To present the results in the correct order of the annual cycle, we commence each section with the 2009 study period (3 July to 19 August) and conclude with 2008 study period (24 July to 15 September).

Spatial variability of surface temperature

While pronounced spatial differences in the surface temperature between dry and wet areas exist at some points in time, the results of this study show that they are greatly reduced if averages on timescales of one week are considered. On the one hand, this can be explained by the occurrence of cloudy periods with reduced spatial variability in most of the considered weekly intervals. Furthermore, while the daily temperature amplitude is much larger at the dry compared to the wet sites, the

Summary and conclusion

The summer surface temperature of an area of approximately 100 × 100 m of high-arctic tundra on Svalbard is monitored at high temporal resolution using a thermal imaging system. The study covers the entire season from the melting of the perennial snow pack until fall, when refreezing occurs. The monitored area features a pronounced spatial variability of the snow cover, the soil moisture conditions and the surface cover, so that an assessment on the long-term impact on the surface temperature can

Acknowledgements

We are grateful to M. Maturilli for providing us with the data of the BSRN station. We thank M. Schumacher, E. Larmanou, M. Sieber and A. le Tressoler from the AWIPEV base in Ny-Ålesund for the ongoing support of our permafrost research, which contributed greatly to the success of this work. We would like to thank four anonymous reviewers whose comments were very helpful in improving the manuscript. We gratefully acknowledge financial support by the Helmholtz Association through a grant (VH-NG

References (62)

  • Z. Wan

    New refinements and validation of the MODIS land-surface temperature/emissivity products

    Remote Sensing of Environment

    (2008)
  • J. Winther et al.

    Surface albedo in Ny-Ålesund, Svalbard: Variability and trends during 1981–1997

    Global and Planetary Change

    (2002)
  • R. Wolfe et al.

    Achieving sub-pixel geolocation accuracy in support of MODIS land science

    Remote Sensing of Environment

    (2002)
  • F. Becker et al.

    Angular variation of the bidirectional reflectance of bare soils in the thermal infrared band

    Applied Optics

    (1985)
  • H. Beine et al.

    The local wind field at Ny-Ålesund and the Zeppelin mountain at Svalbard

    Meteorology and Atmospheric Physics

    (2001)
  • J. Boike

    SPARC and disappearing permafrost — A story from Bayelva in Svalbard. Svalbard Science Forum

  • J. Boike et al.

    Seasonal snow cover on frozen ground: Energy balance calculations of a permafrost site near Ny-Ålesund, Spitsbergen

    Journal of Geophysical Research, [Atmospheres]

    (2003)
  • Brown, J., Ferrians Jr, O., Heginbottom, J., Melnikov, E., 1997. Circum-Arctic map of permafrost and ground-ice...
  • M. Burgess et al.

    The Global Terrestrial Network for Permafrost (GTNet-P): Permafrost monitoring contributing to global climate observations

  • N. Bussières

    Thermal features of the Mackenzie Basin from NOAA AVHRR observations for summer 1994

    Atmosphere–Ocean

    (2002)
  • F. Chapin et al.

    Role of land-surface changes in Arctic summer warming

    Science

    (2005)
  • I. Chernykh et al.

    Determination of cloud amount and level from radiosonde soundings

    Journal of Applied Meteorology

    (1996)
  • J. Comiso

    Arctic warming signals from satellite observations

    Weather

    (2006)
  • J. Comiso et al.

    Satellite-observed changes in the Arctic

    Physics Today

    (2004)
  • J. Comiso et al.

    Accelerated decline in the Arctic sea ice cover

    Geophysical Research Letters

    (2008)
  • G. Delisle

    Near-surface permafrost degradation: How severe during the 21st century?

    Geophysical Research Letters

    (2007)
  • I. Durre et al.

    Overview of the integrated global radiosonde archive

    Journal of Climate

    (2006)
  • Y. Guo et al.

    An analysis of the effect of local heat advection on evaporation over wet and dry surface strips

    Journal of Climate

    (1994)
  • S. Hachem et al.

    Using the MODIS land surface temperature product for mapping permafrost: An application to northern Québec and Labrador, Canada

    Permafrost and Periglacial Processes

    (2009)
  • L. Hinzman et al.

    Evidence and implications of recent climate change in northern Alaska and other arctic regions

    Climatic Change

    (2005)
  • J. Labed et al.

    Angular variation of land surface spectral emissivity in the thermal infrared: Laboratory investigations on bare soils

    International Journal of Remote Sensing

    (1991)
  • Cited by (95)

    • Cloudy-sky land surface temperature from VIIRS and MODIS satellite data using a surface energy balance-based method

      2021, Remote Sensing of Environment
      Citation Excerpt :

      Multiple spatial-temporal interpolation methods have been proposed to resolve problems caused by cloud contamination. Basic spatial interpolation methods include inverse distance weighting (IDW), kriging interpolation (Neteler, 2010; Westermann et al., 2011), and representative temporal reconstruction methods using the harmonic analysis of time series (HANTS) algorithm (Xu and Shen, 2013), temporal Fourier analysis, and asymmetric Gaussian function fitting method. Spatial-temporal interpolation methods treat spatiotemporally neighboring pixels as references.

    View all citing articles on Scopus
    View full text