Active-layer thermal monitoring on the Fildes Peninsula, King George Island, maritime Antarctica

International attention to climate change phenomena has grown in the last decade; the active layer and permafrost are of great importance in understanding processes and future trends due to their role in energy flux regulation. The objective of this paper is to present active-layer temperature data for one Circumpolar Active Layer Monitoring South hemisphere (CALM-S) site located on the Fildes Peninsula, King George Island, maritime Antarctica over an 57-month period (2008–2012). The monitoring site was installed during the summer of 2008 and consists of thermistors (accuracy of±0.2 C), arranged vertically with probes at different depths, recording data at hourly intervals in a highcapacity data logger. A series of statistical analyses was performed to describe the soil temperature time series, including a linear fit in order to identify global trends, and a series of autoregressive integrated moving average (ARIMA) models was tested in order to define the best fit for the data. The affects of weather on the thermal regime of the active layer have been identified, providing insights into the influence of climate change on permafrost. The active-layer thermal regime in the studied period was typical of periglacial environments, with extreme variation in surface during the summer resulting in frequent freeze and thaw cycles. The active-layer thickness (ALT) over the studied period shows a degree of variability related to different annual weather conditions, reaching a maximum of 117.5 cm in 2009. The ARIMA model could describe the data adequately and is an important tool for more conclusive analysis and predictions when longer data sets are available. Despite the variability when comparing temperature readings and ACT over the studied period, no trend can be identified.


Introduction
International attention to climate change phenomena has grown in the last decade, and intense modeling of climate scenarios was carried out by scientific investigations researching the sources and trends of these changes (Mora et al., 2013;IPCC, 2012;Moss et al., 2010).The cryosphere and its energy flux became the focus of many investigations, being recognized as a key component of climate for the understanding of actual (climate variability) and future trends (Ledley, 1985;Flanner et. al., 2011;van den Broeke et. al., 2008).The active layer and permafrost, part of the terrestrial cryosphere, are of great importance due to their role in energy flux regulation and sensitivity to climate change (Kane et al., 2001;Smith and Brown, 2009).Compared to other regions of the globe, our understanding of Antarctic permafrost is poor, especially in relation to its thermal state and evolution, physical properties, links to pedogenesis, hydrology, geomorphic dynamics, response to atmosphere and, thusly, to the variability and global climate change (Bockheim, 1995, Bockheim et al., 2008).An understanding of the distribution and properties of Antarctic permafrost is essential not only for the cryospheric sciences but also for the life sciences, since it will be a major effect ecosystem modification following climate-induced changes and variability (Vieira et. al., 2010).The scientific interest in King George Island has grown in the last few years due to the intensity of climate change effects such as permafrost degradation (Beyer et al., 1999).
The objective of this paper is to present active-layer temperature data for one CALM-S site located on the Fildes Peninsula, King George Island, maritime Antarctica over a 57-month period (2008)(2009)(2010)(2011)(2012).

Studied site
The archipelago of the South Shetland Islands, extending more than 400 km from southwest to northeast, lies near the northern tip of the Antarctic Peninsula.The archipelago is separated from the Antarctic Peninsula by the Bransfield Strait and from South America by Drake Passage.King George Island is the largest in the archipelago and Fildes Peninsula is at its southwestern end (Fig. 1).This peninsula is about 10 km long and 2-4 km wide.It is washed on three sides by the waters of Drake Passage, Fildes Strait and Maxwell Bay.Most of the Fildes Peninsula is free of ice; glaciers cover only the extreme northeastern part (Simonov, 1975).Fildes Peninsula has a gentle topography dominated by a wide central plain, and several other plains CE5 at different altitudes; it consists mostly of lava with small outcrops of tuffs, volcanic sandstones and agglomerates (Smellie et al., 1984).
The region experiences a subantarctic maritime climate according to the Köppen climate classification.The South Shetland Islands have an ET (tundra climate) climate, South Hemispheric Polar Oceanic (Köppen, 1936), characterized by mean annual air temperatures of −2.2 • C (data from 2000 through 2012 from the Marsh meteorological station at −62 • 11 27 S, 58 • 59 12 W and 10 m of altitude) and mean summer air temperatures above 0 • C lasting up to 4 months (Rakusa-Suszczewski et al., 1993;Wen et al., 1994).Precipitation ranges between 350 and 500 mm per year with rainfall occurring in the summer period (Øvstedal and Lewis-Smith, 2001).Ferron et al. (2004) found great climate variability when analyzing data from 1947 to 1995 and identified cycles of 5.3 years of colder conditions followed by 9.6 years of warmer conditions.
Soils in Fildes Peninsula are well developed for Antarctic standards with large areas of soils with Leptic/Lithic and Skeletic characters.Arenosols/Entisols and Cryosols/Gelisols (frequently turbated) are the most important soil classes, while Leptosols/Entisols, Gleysols/Aquents and Cambisols/Inceptisols also occur with gelic regime.Faunal activity plays a marked role in soil genesis on Fildes Peninsula and is commonly found in the north shore of the peninsula.The dominant soils are Cryosols, related to cryoturbation and active-layer processes and developed on wide areas occupied by stone fields, patterned grounds, moraines and slopes on middle and upper platforms and hills in the northern and southern areas (Michel et al., 2014).

Methods
The active-layer monitoring site (−62 • 12 12 S, 58 • 57 37 W at 60 m of altitude) was installed in the summer of 2008 and consists of thermistors (accuracy ±0.2 • C) (model 107 Temperature Probe, Campbell Scientific Inc, Utah, USA) arranged in a vertical array at different depths down to the permafrost table (10.5 cm (F1), 32.5 cm (F2), 67.5 cm (F3), 83.5 cm (F4)).The site was selected during a soil survey and represents most of Turbic Cryosols occurring on the peninsula.All probes were connected to a data logger (model CR 1000, Campbell Scientific Inc, Utah, USA) recording data at hourly intervals from 1 March 2008 until 30 November 2012.
The characteristics of the monitored site and the exact depth of the probes are presented in Tables 1 and 2. The depth of the probes was established in respect to pedological differentiation of horizons.Sites affected by cryoturbation frequently form AB horizons; the intense migration of the organic material in depth associated with the limited carbon input (mosses and lichens) makes the formation of A horizons difficult.Air temperature was obtained from the  Marsh automatic meteorological station located at the Teniente Rodolfo Marsh Martin Airport.We calculated the thawing days (days in which all hourly soil temperature measurements are positive and at least one reading is warmer than +0.5 • C), freezing days (days in which all hourly soil temperature measurements are negative and at least one reading is colder than −0.5 • C), isothermal days (days in which all the hourly measurements range only between ±0.5 • C), freeze-thaw days (days in which there are both negative and positive temperatures with at least one value greater than ±0.5 • C), thawing degree days (TDD, obtained by the cumulative sum of the mean daily temperatures above 0 • C) and freezing degree days (FDD, obtained by the cumulative sum of the mean daily temperatures below 0 • C) according to Guglielmin et al. (2008).The active-layer thickness (ALT) was calculated as the 0 • C depth by extrapolating the thermal gradient from the two deepest temperature measurements (Guglielmin, 2006).
The apparent thermal diffusivity (ATD) was estimated for different seasons from the equation of McGaw et al. (1978): (1) where α is the apparent thermal diffusivity (m 2 s −1 ), t is time increments (s), Z is space increments (m), T is the temperature, j is the temporal position and i is the depth position.Nelson et al. (1985), Outcalt and Hinkel (1989) and Hinkel et al. (1990Hinkel et al. ( , 2001) ) have used this estimative to assess the resistance to energy flux in the profile.Hourly estimations were made for intermediate depths of both profiles, and mean values were calculated and plotted for each day.
A series of statistical analyses were performed to describe the soil temperature time series.The Box-Pierce test and augmented Dickey-Fuller tests where performed to confirm the stationarity and independent distribution of the time series (data not shown).The histogram (frequency distribution of temperature readings) and first difference (the difference between consecutive hourly temperature readings) were plotted, and the time series was decomposed into its seasonal and trend components by locally weighted smoothing (Loess) using a window of 25.A linear fit was applied to the time series in order to identify a global trend.Finally, a series of autoregressive integrated moving average (ARIMA) models were tested until satisfactory results were found.In order to define the best fit for the data, we first examined the ACF (Auto correlation function) and PACF (Partial Auto correlation function) plots to determine the appropriate model and then tested a series of combinations; standardized residuals, autocorrelation plot, Ljung-Box statistics and the Akaike information criterion (AIC) (Burnham and Anderson, 2002) were the major parameters used to judge the suitability of the model.Considering the seasonal nature of the data, a seasonal component was added after the best models were selected.
For a time series of data x t , where t is time and x is the real number (soil temperature in the present case), an ARMA (q, p ) model is used: where p is autoregressive terms, q is moving-average terms, L is the lag operator, ϕ is the autoregressive parameter, θ is www.solid-earth.net/5/1361/2014/Solid Earth, 5, 1361-1374, 2014 the moving average parameter, ε is the error and i is the ith term.The errors are assumed to be independent, identically distributed variables sampled from a normal distribution with zero mean and one standard deviation [0,1].
When the term 1 − p i=1 i L i has a unitary root of multiplicity d, it can be rewritten as An ARIMA (p, d, q) model expresses this polynomial factorization property with p = p − d and is expressed by and thus can be thought of as a particular case of an ARMA (p + d, q) process with an autoregressive polynomial with d unit roots.

Results and discussion
Interannual variability of the active-layer temperature shows parallel behavior despite contrasts between different years; daily temperatures records are presented in Fig.   ALT was estimated for every season and the results are summarized in Table 1 (2008 and 2012 being incomplete).Maximum ALT ranged between 89 and 106 cm with a mean of 101 cm; the totality of the active layer froze during winter every year over the studied period.During 2010 a curious phenomenon occurred: temperatures at F3 remained negative the whole year, reaching values above zero at the bottommost layer (F4); that year ATL was estimated slightly below the deepest probe, probably due to the accumulation of water over the permafrost table.
The grouping of days into freezing, thawing, isothermal and freeze-thaw offers a quick parameter for comparing different periods (Fig. 3).Most of the thawing days occurred between January and March, more frequently for the uppermost layers; only 8 days were recorded for F4 in March 2009.Freeze days were concentrated between April and November and are more evenly distributed in depth although F1 and F2 were frozen for longer periods in 2008, 2011 and 2012.Isothermal and freeze-thaw days occurred between December and May.Isothermal periods were long for F3 and F4, which shows a strong zero curtain effect (buffered temperature change due to freezing and thawing of soil moisture).Freeze-thaw days were more common in F1; the site experienced frequent freeze-thaw cycles in the surface, especially during summer.Great temperature changes in depth, especially on the −0.5 to 0.5 • C zone, were rare; only 1 day was recorded in 2011 for F4.
The cumulative sum of the daily averages reached a maximum in 2009 and a minimum in 2011 for all layers; values varied greatly over the years, with December and January being the hottest months and June always the coldest (Fig. 4).Over the 57 months, the TDD were 902 ATD was calculated for F3 and F4 by considering hourly readings and then averaging for the climatic seasons; results are shown in Table 3. Thermal diffusivity can experience considerable seasonal variations when thawing and freezing processes occur (Hinkel, 1997).The ability of the profile to transmit energy varies during the year due mostly to water content: moisture, on the one hand, enhances energy flux through percolation but, on the other hand, absorbs and emits energy on freezing and thawing processes.Average ATD for the 57 months was 4.2 × 10 −5 (F2) and 1.1 × 10 −5 m 2 s −1 (F3); these values are consistent with other findings from wet soil profiles of maritime Antarctica. De Pablo et. al. (2013)   values suggest that nonconductive effects oppose and overwhelm the conductive trend (Outcalt and Hinkel, 1989).Histograms for the studied layers show a predominance of temperatures around 0. F3 expresses the largest frequency in this region while the greatest amplitude is found for F1 (15 • C) (Fig. 5).The plot of first differences of the hourly measurements gives us an idea of periods with great temperature oscillation.Most of the strong variations are associated with summer for F1 and F2 while being evenly distributed for F3 and very limited for F4 (Fig. 5).The decomposition of the time series reveals a great seasonal component associated with summer for F1 and F2, with most of the noise also concentrated in the warmer months.F3 behaves more erratically, and the seasonal effect at F4 is shifted ahead, while noise is reduced (Fig. 6).Linear regression was performed on hourly readings in order to have an estimation of overall trend: the slope of the line is slightly negative for all layers and intercept is always negative (Fig. 7).
Although the studied period was limited to 57 months and any forecast is not suitable, an ARIMA model was tested at each layer in order to evaluate which model better fit the data.Autocorrelation and partial autocorrelation plots were estimated as a guide for the selection of the ARIMA parameters.The plots (Fig. 8) show strong correlation with data from the previous month.This correlation is smaller for the subsequent period and, after the   4).The best model for F2 includes only seasonal parameter, one seasonal autoregressive parameter and one seasonal differentiating parameter.The best fit for F4 includes one autoregressive parameter, two moving average parameters and two seasonal autoregressive parameters.The diagnostics of the models (Fig. 9) are satisfactory, the standardized residuals do not show clusters of volatility, no significant auto correlation is found between the residuals and the ρ values for the Ljung-Box statistics are all large, indicating that the residuals are pattern-less.
Monitoring of the CALM-S site on Fildes Peninsula has provided data on the thermal dynamics and frost conditions on a densely occupied maritime Antarctic site.The effects of weather on the thermal regime of the active layer have been identified, providing insights into the influence of climate change on permafrost and leading to the following conclusions: -The active-layer thermal regime in the studied period was typical of periglacial environments, with extreme variation in surface temperature during summer resulting in frequent freeze and thaw cycles.
-The ALT over the studied period shows a degree of variability related to different annual weather conditions, reaching a maximum of 117.5 cm in 2009.
-The calculated ATD suggests strong influence of water content on the soil thermal regime.
-The ARIMA model can describe the data adequately and is an important tool for more conclusive analysis and predictions when longer data sets are available.
-Despite the variability when comparing temperature readings and ACT over the studied period, no trend can be identified.

Figure 1 .
Figure 1.Location of Fildes Peninsula within Antarctica and the South Shetland Islands.
2. The temperature at 10.5 cm reaches a maximum daily average (4.1 • C ± 0.5) in early January and a minimum (−8.0 • C ± 1.4) between late July and early August.At 83.5 cm the maximum temperature (0.3 • C ± 0.2) occurs in late March and the minimum reading (−4.1 • C ± 1.0) around mid-August.Disparities can be noticed when comparing the different years: 2008 had a mild winter (21 freezing days and −1 freezing degree days at 83.5 cm in July) contrasted by a severe winter in 2011 (31 freezing days and −80 freezing degree days at 83.5 cm in July), whereas the summer of 2009 was considerably warmer (31 thawing days and 65 thawing degree days at 10.5 cm in January) compared to the summer of 2010 (17 thawing days and 21 thawing degree days at 10.5 cm in January).Daily air temperatures on Fildes Peninsula over the studied period averaged −2.3 • C (±4.1), reaching a maximum and a minimum of 5.8 and −21.2 • C in early January and late June, respectively.The difference between maximum and minimum daily averages was greater for 2009 and 2011; in these years, more extreme minimum temperatures were recorded (−17.4 • C for 2009 and −18.6 • C for 2011).The freezing season started in late May and the thawing season in mid-December with small variability between the years.Soil temperature averaged −1.2 • C (av max 2.1 • C and av min −7.3 • C) over all layers: using hourly measurements, maximum soil temperature for the upper-most layer (F1) was 8.7 • C and the minimum temperatures were −9.8, 0.5 and −5.4 • C at the bottom-most layer (F4).
• C day (F1), 449 • C day (F2), 96 • C day (F3) and 64 • C day (F4).The FDD were −3229 • C day (F1), −2623 • C day (F2), −2433 • C day (F3) and −2040 • C day (F4).Contrast between different years was significant: in 2011 FDD accumulated −819 • C day in the surface and −516 • C day at F4; in contrast, 2011 accumulated −527 • C in the surface and −382 • C at F4.There was a clear preponderance of negative soil temperatures in the studied profile; despite the percolation of liquid water and above-freezing temperatures during the summer, positive temperatures were mild in the soil profile.

Table 1 .
General characteristics of the monitored site.

Table 2 .
Soil texture of the studied profile.
and its seasonal component are always small or null, as indicated by the partial autocorrelations.The best model for F1 and F3 includes one autoregressive parameter, one seasonal autoregressive parameter and one seasonal differentiating parameter (Table model