Present address and correspondence: Christopher J. Ellis, Royal Botanic Garden Edinburgh, 20A Inverleith Row, Edinburgh EH3 5LR, UK (tel. +44 131 248 2993; e-mail firstname.lastname@example.org).
1The response of peat-rich permafrost soils to human-induced climate change may be especially important in modifying the global C-flux. We examined the Holocene developmental record of a High Arctic peat-forming wetland to investigate its sensitivity to past climate change and aid understanding of the likely effects of future climate warming on high-latitude ecosystems.
2The microhabitat of mosses was quantified in the present-day polygon-complex at Bylot Island (73° N, 80° W) and used to interpret the radiocarbon-dated macrofossil record of three cores, comprising c. 3500 years of wetland development. Recurrent wet and dry phases in the reconstructed palaeohydrological record indicated pronounced temporal variability. Wet and dry phases were compared between cores and with palaeoclimatic proxy values, measured as percentage melt and δ18O in nearby ice cores.
3Periodic wet and dry phases appear unrelated to past climate over c. 50% of the combined stratigraphic records, and are attributable instead to geomorphological mechanisms. At other times, association of wet and dry phases with significantly lower and higher values of percentage melt and δ18O indicate a possible effect of past climate change on polygon hydrology and vegetation, although inconsistencies between cores suggest that local geomorphological processes continued to modify a regional climatic effect. However, during a period incorporating the Little Ice Age (c. 305–530 cal. years bp), reconstructed moisture and vegetation change is pronounced and consistent among all three cores.
4The results provide strong evidence for the sensitivity of a High Arctic terrestrial ecosystem to past climate change during the Holocene. The estimated magnitude of changes in soil moisture between wet and dry phases is sufficient to imply recurrent shifts in wetland function, periodically impacted upon by pronounced climatic variability, although controlled principally by autogenic processes. The structure and function of such wetlands may therefore be susceptible to predicted, human-induced climate warming.
Human-induced climate change is expected to cause severe warming at high latitudes (Mitchell et al. 1990; Maxwell 1992), prompting concerns that the role of Arctic ecosystems might consequently shift from a store (or sink) to a source of greenhouse gases (CO2 and CH4) (Shaver et al. 1992; Chapin et al. 1997). The structure and function of extensive Arctic and Subarctic wetlands (Tarnocai & Zoltai 1988; Bliss 1997) are expected to be particularly important in regulating the flux of carbon between land and atmosphere. Peat-rich soils associated with such wetlands have contributed a net sink for carbon during the Holocene and are estimated to store > 97% of the tundra carbon reserve, comprising c. 180–190 × 1015 g of soil-C (Post et al. 1982; Oechel & Vourtilis 1994). Climate change and subsequent soil drying may cause a shift in wetland function from net C-input (or storage) to C-output (Oechel et al. 1993, 1995), resulting in a significant positive feedback on the global warming effects of greenhouse gases.
Recent studies in Arctic ecology have examined the response of vegetation and soils to experimental changes in factors such as temperature, water availability and nutrient status (Chapin et al. 1995; Arft et al. 1999) to demonstrate the direct and indirect effects of proxy climate change on structure and function of tundra vegetation. These short-term experiments (typically < 10 years) suggest that the climatic sensitivity of Arctic ecosystems is balanced by factors conferring both responsiveness and resilience (Wookey & Robinson 1997), but the results lack a long-term ecological context. Palaeoecological research is therefore essential to establishing whether climate sensitivity described by short-term studies may also control tundra development over the long term (i.e. during the Holocene), providing a perspective for interpreting and predicting future climatically controlled change in ecosystem structure and function, and a template for the ‘scaling-up’ of experimental data (Davis 1994).
This paper compares the long-term sensitivity of a polygon-patterned High Arctic wetland (cf. Tarnocai & Zoltai 1988; Bliss 1997) to known insolation- and volcanically forced decade- to century-scale variability in the Arctic climate (Overpeck et al. 1997; Mann et al. 1998). Existing macrofossil evidence for century-scale shifts in tundra development lack consensus, and high-resolution palaeoecological data are especially scarce for the High Arctic. Studies in the Low to Mid Arctic support the sensitivity of tundra to past climate change (Ovenden 1982; Bennike 1992; Vardy et al. 1997), consistent with evidence for the effect of human-induced global warming at similar latitudes (Jorgenson et al. 2001; Nelson et al. 2001). However, some studies in the High Arctic have invoked slope processes and local geomorphology rather than climate as controlling factors in vegetation–soil development (LaFarge-England et al. 1991; but see Garneau 1992), and a lack of data prevents an adequate assessment of the relative importance of climate and autogenic processes (Ellis & Rochefort 2004).
We investigate whether there is evidence that the climatic sensitivity of tundra vegetation and soils described by short-term experiments might also have occurred over longer time scales (≥ 100 years) and in response to natural climatic variability. If so, then to what degree might climatically mediated shifts in the vegetation–soil complex have affected long-term wetland function?
The study site occurs at c. 10 m a.s.l. to the south of Qunguliqtut Valley in western Bylot Island, within the Simirlik National Park (73.08° N; 80.00° W), part of the Canadian Arctic Archipelago (cf. Ellis & Rochefort 2004). Mean annual precipitation measured at Pond Inlet (1951–1980), adjacent to Baffin Island, was 159 mm, with maximum (August) and minimum (February) mean temperatures of 4.9 °C and −32.7 °C, respectively (Zoltai et al. 1983). The study area comprises low-lying terraces, which developed with the accretion of aeolian sands and silts and the concurrent deposition of peat. Terrace formation was initiated sometime after a marine transgression following glacial retreat (dated using shell fragments to c. 6020 ± 80 years bp), but before active peat deposition, which commenced at or before c. 2900 ± 90 years bp (Allard 1996). The upward growth of ice-wedges in the aggrading sediments has resulted in extensive polygon-patterned ground, and the terraces include a range of associated landforms: low-centre and high-centre polygons, polygon-ponds and thaw-lakes (cf. Billings & Peterson 1980). The polygon complex is fed by meltwater from hills to the south, and the vegetation is dominated by sedges (e.g. Carex aquatilis var. stans, Eriophorum scheuchzeri), grasses (e.g. Arctagrostis latifolium, Dupontia fischeri, Pleuropogon sabinei) and fen mosses (e.g. Drepanocladus spp., Aulocomnium spp.).
process in the formation of the polygon-patterned wetland
The development of tundra polygons is described in detail by French (1996). Low-centre polygons (ridges raised by the growth of ice-wedges, surrounding lower and wetter polygon centres) persist due to the upward growth of syngenetic ice-wedges, at a rate that is limited by the accumulation of sediment and the development of permafrost. Polygon-ponds, which form during the inundation of low-centre polygons, may coalesce to form thaw-lakes. High-centre polygons occur where ice-wedges surrounding low-centre polygons are denuded, with the consequent collapse of ridges and formation of troughs around the now higher polygon centres and the cessation of upward sediment and ice-wedge growth.
Sediment cores were extracted from the centres of a low-centre polygon and high-centre polygon in June 1999 (BY-LowC and BY-HighC, respectively, 298 cm and 230 cm deep). Sediments from the active layer (to c. 30 cm depth) were sampled as 2-cm slices from an unfrozen peat monolith (c. 10 cm2). Permafrost sediment cores were collected as contiguous 5-cm-diameter, 50-cm segments, using a machine-driven corer (built by M. Allard, Université Laval). Cored sections were examined in the field to ensure that sediments were horizontally bedded and undisturbed. The frozen 50-cm sections were cut where possible into 2-cm slices and stored in sealed polythene bags in the dark at < 4 °C.
In late July and early August 2000, present-day moss assemblages were quantified based on species presence–absence in 25 5 × 5 cm subunits within 25-cm quadrats (identification and nomenclature follows Nyholm 1954–1969). Ninety-nine quadrats were analysed including representative samples from all recognized habitats: low-centre polygon ridges, flanks and centres, high-centre polygon centres and eroding flanks, polygon-pond flanks and standing water, water courses and areas of active aeolian deposition.
Sampling for the analysis of soil properties was during a single rain-free period in early August, the warmest and wettest period of the year (Zoltai et al. 1983). A turf (5 cm2 × 5 cm deep) was cut from the centre of each quadrat and divided into two equal halves. The green ‘living’ moss layer (c. 1 cm in thickness) was removed from one half and the remaining peat–sediment sample was weighed before and after drying for 24 h at c. 60 °C.
Soil samples were sealed in polythene bags and stored on ice in the dark, before transfer to the laboratory. The partially dried samples were then re-weighed and dried at 105 °C for 18 h. Percentage dry weight per unit volume for each quadrat was calculated from the final laboratory dry weight. The same samples were incinerated at 550 °C for 5 h: loss on ignition (LOI) was used to calculate organic content and percentage mineral material (ash weight as a fraction of the pre-LOI dry weight). Within 5 days of arrival in the laboratory, the freshly stored samples were analysed for pH, using a Fisher Scientific accumet model-10 pH meter, and for conductivity (µS cm−1), using an Orion model-122 conductivity meter (Brady 1990). Conductivity was corrected for temperature and pH according to Sjörs (1950).
stratigraphic macrofossils and loi
A 1-cm3 sediment sample was removed from each contiguous horizon. Macrofossils were extracted by treatment in 20 mL of 7% KOH and quantified as percentage frequency of occurrence according to Tallis (1985). Attempts to quantify the volume of permafrost sediment are complicated by variations in ice content and mineral content, and this method quantifies moss remains, including those from permafrost sediments (Ellis & Rochefort 2004), independently of volume (Tallis 1985), whereas standard measures of bryophyte concentration are based on peat volume (Janssens 1983, 1988). Macrofossil specimens (typically bryophyte stems and leaves) were identified using a modern reference collection at the Université Laval Herbarium (QFA) and Nyholm (1954–1969).
A second sample from each horizon was oven dried at 105 °C for 18 h, weighed and the mineral content quantified after incineration at 550 °C for 5 h. Values are expressed as the three-point weighted average, calculated to minimize high-frequency noise, and emphasizing low-frequency cycles (Green 1995).
Before being submitted for radiocarbon analysis, samples were pretreated to prevent contamination by carbon-rich Tertiary deposits (Gajewski et al. 1995), which occur on Bylot Island adjacent to and within the catchment area of the sampling site (Klassen 1993). A sediment sample of c. 2 cm3 was washed in distilled water and filtered through a 0.3-mm mesh sieve. In the absence of woody remains, bryophyte stems and leaves were collected from the organic residue and rinsed in distilled water. They were dried at 60 °C and samples of ≥ 0.1 g submitted to Beta Analytic Ltd (Miami, FL, USA), for analysis by accelerator mass spectrometry (AMS).
Material was radiocarbon-dated at horizons delimiting a significant change in the stratigraphic record. Conventional radiocarbon ages (years bp) are shown in Table 1 but are generally presented as calibrated years before present (cal. years bp, according to INTCAL 98, Stuiver et al. 1998).
Table 1. Radiocarbon dates for cores BY-LowC and BY-HighC, Qunguliqtut Valley, Bylot Island (73.08° N; 80.00° W); with calibrated ages calculated using the computer program CALIB 4.3 (Stuiver et al. 1998). Results for the additional core BY-a (formerly presented by Ellis & Rochefort 2004) are also shown
Conventional radiocarbon age (yr bp ± 1 SE)
Calibrated age (cal. yr bp) Mid-point of 1δ-range
BY-LowC; 6–10 cm
BY-LowC; 29–35 cm
2420 ± 50
2700 –2370– 2355
BY-LowC; 37–39 cm
3230 ± 40
3470 –3460– 3390
BY-LowC; 81–83 cm
910 ± 50
915 –795 – 755
BY-LowC; 127–129 cm
2430 ± 50
2705 –2415– 2355
2530 ± 40
2740 –2730– 2510
2800 ± 40
2940 –2880– 2850
BY-LowC; 296–298 cm
3100 ± 50
3370 –3345– 3255
BY-HighC; 2–6 cm
280 ± 40
425 –305 – 295
BY-HighC; 50–52 cm
670 ± 40
660 –650 – 570
BY-HighC; 100–102 cm
1170 ± 50
1165 –1070– 995
BY-HighC; 118–120 cm
1920 ± 40
1895 –1870– 1830
BY-HighC; 188–190 cm
2260 ± 40
2340 –2320– 2180
BY-HighC; 228–230 cm
2590 ± 50
2760 –2745– 2730
BY-a; 2–6 cm
120 ± 40
271 –140– 0
BY-a; 18 cm
270 ± 40
420 –300– 290
BY-a; 40 cm
390 ± 40
500 –480– 330
BY-a; 52–54 cm
520 ± 40
540 –530– 520
BY-a; 62 cm
710 ± 40
680 –660– 650
BY-a; 74 cm
620 ± 40
650 –590– 550
BY-a; 114 cm
1040 ± 40
970 –950– 930
BY-a; 140–142 cm
1590 ± 40
1530 –1510– 1420
BY-a; 162 cm
1520 ± 40
1420 –1400– 1350
BY-a; 208–210 cm
1660 ± 40
1600 –1545– 1520
moss distribution and environmental gradients
Quadrat data describing the distribution of mosses in the present-day polygon wetland were summarized using detrended correspondence analysis (DCA, with rare species down-weighted), computed using the program MVSP 3.1 (Kovach 1986–1999). The first four orthogonal axes of the DCA were compared, using a general linear model, with values of soil moisture, pH, corrected conductivity of a bulk soil–water sample and soil organic/mineral matter. Stepwise selection was used to derive the most effective multiple regression, calculated using the program Genstat 7.1 (Genstat 7.1 2003). The power of the individual environmental variables in explaining the position of moss taxa along each of the four axes was examined by comparing the adjusted R2 value for a multiple regression including the variable with the model with the variable selectively removed.
The records of cores BY-LowC and BY-HighC were supplemented by re-analysis of a third core (BY-a), from a high-centre polygon in the same complex, which has previously been shown to indicate climatic sensitivity (Ellis & Rochefort 2004).
Macrofossil records were initially tested for evidence of allochthonous deposition, which would confound palaeoenvironmental reconstruction based on the assumption that sediments are autochthonous (see Appendix S1, available as supplementary material online). The first orthogonal axis of the DCA for present-day moss assemblages was assumed to summarize the autochthonous record of macrofossil mosses in each core. Axis one values for individual species were used to calculate a weighted average (WA) score (Curtis & McIntosh 1950, 1951) for each stratigraphic horizon. The stratigraphic profiles of WA scores were smoothed using a three-point weighted average to minimize high-frequency noise (Green 1995) and linear regression was used to derive lines of best fit for continuous periods (between hiatuses) in BY-LowC, BY-HighC and BY-a. The data were then replotted for each core as the residuals. Discrete periods with residual WA values above or below the long-term mean (i.e. the regression line) were interpreted as contiguous wet and dry phases. Two different age models were used to infer sedimentation rates (Appendix S2): linear interpretation for BY-LowC and BY-HighC and linear regression for core BY-a.
A semi-quantitative estimate of past soil moisture conditions was derived by classifying present-day values as the average for each unit interval along the first axis of the DCA, and using analysis of variance and least-squared difference at the 5% level to identify discrete moisture classes (Genstat 7.1 2003). These were then used to interpret the stratigraphic record of WA ordination scores.
The radiocarbon-dated records of residual WA values were compared with proxy-climate records available for the same High Arctic region as our study site. Five-year average values of percentage melt in the stratigraphy of the Agassiz-84 ice core, Ellesmere Island (Koerner & Fisher 1990), from ad 1961 to 9 bc and 5-year average values of δ18O for the combined stratigraphies of two adjacent cores from the Devon Island ice cap (Paterson et al. 1977) from ad 1973 to 727 bc were provided by the US National Oceanic and Atmospheric Administration through the World Data Center for Paleoclimatology at Boulder, Colorado, USA. Despite noted complexities in interpreting palaeoclimatic trends (Koerner 1977; Fisher et al. 1983; Bradley 1990), it is reasonable to assume that these records provide an approximation of past climate change during the mid to late Holocene (Bradley 1985, 1990; Fisher & Koerner 1994). Percentage melt measures the changing concentration of textural melt layers with depth and is a proxy for summer temperature (Koerner 1977; Koerner & Fisher 1990), reflecting long-term variations in net radiation (Rn), which will directly control spring snow-melt [i.e. precipitation (P)] (Young & Lewkowicz 1990; Young et al. 1997), and evapotranspiration (ET) (Vourtilis & Oechel 1997). The ratio of oxygen isotopes 18O/16O, δ18O, provides a proxy-record of mean annual palaeotemperature (Paterson et al. 1977; Bradley 1985). The palaeoecological record was compared since 1800 cal. year bp with 5-year values of both percentage melt and δ18O, and with values of δ18O only for the previous 700 years.
Values of percentage melt and δ18O were compared between contiguous wet and dry phases using analysis of variance and least-squares differences at the 5% level to determine significance (Genstat 7.1 2003). A combined record of palaeohydrology, based on the contemporaneous occurrence in different cores of wet or dry phases that corresponded to significantly different values of percentage melt and δ18O, was constructed. This was then compared with palaeoclimatic proxy values using a two-tailed unpaired t-test (Genstat 7.1 2003) on values of percentage melt and δ18O between wet or dry phases that occurred (i) only in one core or contemporaneously in either (ii) two or (iii) three cores.
mosses – present-day habitats
The DCA identifies soil moisture as the single most important variable in the present-day distribution and abundance of mosses (Table 2) and analysis of variance demonstrates that first axis sample scores can be divided into two classes corresponding to significantly wetter and drier soil moisture around a DCA score of 3 (Fig. 1).
Table 2. Comparison of environmental variables with sample scores (DCA) for the distribution of mosses in the present-day polygon wetland. Eigenvalues (λ) and the percentage variation explained are shown for the first four orthogonal axes. Values for each environmental variable show the effect on adjusted R2 of removing it from a general linear model, generated as a stepwise multiple regression to explain sample scores (DCA) using the four measured variables
% variation explained (cumulative)
Adjusted R2 and F-value (improvement if variables dropped)
Macrofossil mosses in cores BY-LowC and BY-HighC exhibit striking variability during polygon development (Figs 2 & 3, respectively). However, there are sections in the macrofossil records of each core where moss remains are few or absent, with lenses of mineral sediment suspended in segregation ice at 127–81 cm and 116–102 cm depth in BY-LowC and BY-HighC, respectively (Figs 2 & 3), and a second, lesser hiatus at 27–14 cm in BY-LowC accompanying a peak in mineral content (Fig. 2). The stratigraphic records of each core nevertheless provide clear evidence for vegetation change during polygon development.
Radiocarbon dates for two horizons in the stratigraphy of BY-LowC were thought to be erroneous and cautiously rejected from the timeframe (Appendix S4).
Based on linear interpolation between radiocarbon-dated horizons in BY-LowC and BY-HighC, the sedimentation rate of peat-rich deposits varies between c. 0.22 cm yr−1 and 0.057 cm yr−1, with a combined average rate of c. 0.11 cm yr−1 (Table 1, Fig. 4). This is consistent with data from comparable High Arctic habitats (LaFarge-England et al. 1991; Bennike 1992; Vardy et al. 1997). However, sedimentation appears to have been slower (c. 0.022 cm yr−1) at 128–82 cm depth in BY-LowC and 119–101 cm depth in BY-HighC (Table 1, Fig. 4), matching the shared hiatus in the macrofossil records (Figs 2 & 3). There was no segregation ice in the stratigraphy of BY-a (Ellis & Rochefort 2004) and a linear regression between the scatter of radiocarbon-dated horizons suggests a sedimentation rate of 0.13 cm yr−1 (Fig. 4), similar to the average estimated for BY-LowC and BY-HighC, and within the 95% confidence interval for the range of growth rates in these two cores (i.e. 0.14–0.08 cm yr−1). Linear interpolation, rather than regression, estimates sedimentation rates of up to 2 cm yr−1 for BY-a, inconsistent with data from this and other studies (LaFarge-England et al. 1991; Bennike 1992; Vardy et al. 1997). The balance of evidence suggests that two different but consistent age models – linear interpolation in BY-LowC and BY-HighC and linear regression in BY-a – provides the most realistic scenario for the comparison of cores across a common timeframe.
The application of DCA scores in the WA interpretation of macrofossil mosses is assumed to provide a reasonable proxy for past relative wetness, in which positive deviations in WA macrofossil horizon scores above the long-term mean indicate periods of increased soil moisture and vice versa. Comparing these scores for stratigraphic horizons with DCA sample scores for moss assemblages in the present-day wetland (Fig. 1) produces a semi-quantitative paleohydrological proxy, in which DCA or WA values above 3 are considered significant (Figs 1 & 5).
The lower stratigraphy of BY-LowC (below the hiatus with segregation ice) pre-dates the palaeoclimatic records. However, averages for 5-year values in percentage melt and/or δ18O were consistently significantly different between contiguous wet and dry phases during the period common to all cores (wetter phases have lower percentage melt in all three cores, but values of δ18O did not change significantly in BY-LowC; Fig. 6). A combined palaeohydrological record demonstrates periods during which phases associated with significantly different palaeoclimatic proxy values occur in one core alone or contemporaneously in two or three cores. Values of percentage melt were significantly different between wet and dry phases when these were registered simultaneously in two or three cores (but not when registered in one core alone), whereas values of δ18O were significantly different between wet and dry phases both when registered simultaneously in two cores and when registered in one core alone (Fig. 7).
phases in polygon hydrology: geomorphological and climatic controls
The long-term development of the three cores examined here is characterized by recurrent, contrasting phases in soil moisture and vegetation composition (Figs 2, 3 & 5), supporting a growing body of evidence for the dynamic nature of High Arctic wetland development (LaFarge-England et al. 1991; Ellis & Rochefort 2004) rather than the previously generally perceived view of Arctic vegetation as resistant. The development of three low-centre wetland polygons (cf. Tarnocai & Zoltai 1988; French 1996) was reconstructed up to the present-day (BY-LowC) or until the cessation of sediment growth with the shift to high-centre polygon conditions at c. 305 cal. years bp in BY-HighC and c. 140 cal. years bp in BY-a. Individual low-centre polygons are hydrologically discrete, comprising water-shedding polygonal ridges, raised by the growth of ice-wedges, around water-accumulating, lower and wetter polygon centres. Reconstructed changes in past soil moisture can be considered specific to individual polygons. The water balance (ΔS) of a plot within a given polygon can be approximated as:
ΔS = P + Qg + Qs − ET( eqn 1 )
influenced by climate, i.e. precipitation (P) and evapotranspiration (ET), and/or geomorphology, i.e. topography, which controls groundwater and surface-water input and output (Qg and Qs, respectively) and modifies inputs from precipitation (P) (Rovansek et al. 1996; Young et al. 1997). Climatic control may also be indirect, through the effect on topography of ice-wedge growth (Kasper & Allard 2001), which will in turn affect soil moisture (P, Qg and Qs) (Rovansek et al. 1996; Young et al. 1997). Garneau (1992) and Vardy et al. (1997) suggested that the long-term development of Arctic wetlands had been sensitive to past climatic variation, and Ellis & Rochefort (2004) observed a possible effect of the Little Ice Age (c. 300–465 cal. years bp) in the development of a High Arctic wetland polygon. However, they reported a generally low correspondence between proxy climate records and reconstructed soil moisture, and LaFarge-England et al. (1991) also suggested that local geomorphology was the principal factor controlling soil moisture during the development of High Arctic peat deposits. These palaeoecological studies (LaFarge-England et al. 1991; Ellis & Rochefort 2004) raise uncertainty about the extent to which a direct climatic effect on tundra vegetation described by short-term studies (Chapin et al. 1995; Arft et al. 1999) might also contribute to inherent, long-term vegetation change, including the direct effects of future climate warming. It is possible that soil moisture conditions and vegetation during development of tundra polygons are controlled principally by autogenic processes, i.e. the balance between sediment accumulation and upward development of syngenetic ice-wedges, which will control the topographic amplitude between polygon ridges and centres, and therefore values of Qg, Qs and P (Ellis & Rochefort 2004).
evidence for climatic control during polygon development
Evidence for the climatic control of soil moisture and vegetation during long-term development is circumscribed by a lack of replication and possible errors in both the radiocarbon dating scheme (Table 1, Fig. 4) and the dating of ice cores used as a paleoclimatic proxy (Fujii 1995). However, dating of ice cores over the past c. 5000 years is expected to be relatively precise (Paterson et al. 1977), and comparisons between ice cores suggest that a general interpretation of palaeotemperature is reasonable for the Holocene (Bradley 1985, 1990). Rates of sedimentation for the three cores appear consistent (Fig. 4), and the 95% confidence intervals for individual radiocarbon dates (75–100 years) are, on average, c. 45 years less than the mean duration of wet and dry phases (130 ± 15 years), suggesting that a circumspect comparison of phases may be justified. Based on the comparison between palaeoecological records and palaeoclimatic values of both percentage melt and δ18O, our evidence suggests that soil moisture of the wetland polygons investigated may have become drier during periods of climatic warming (dry phases associated with higher values of percentage melt and δ18O) and wetter during periods of climatic cooling (wet phases associated with lower values of percentage melt and δ18O). However, the relative importance of climate in controlling soil moisture is unlikely to have been consistent during polygon development and a climatic effect is therefore more strongly supported when similar phases are registered contemporaneously in the three cores (Fig. 7).
Given a combined palaeoecological record of c. 3885 years (i.e. the combined time span of BY-LowC, BY-HighC and BY-a over the period for which a comparison with palaeoclimatic records is possible), 50% of the period shows no evidence for a correspondence between wet and dry phases and palaeoclimatic proxy values in any core. There is evidence for a correspondence in only one core, or else contrasting evidence for a correspondence between phases and palaeoclimatic proxy values in several cores (i.e. significantly higher or lower values of percentage melt or δ18O related to contrasting coeval wet and dry phases) over c. 44% of the palaeoecological record (i.e. c. 1710 years). Over only c. 6% of the palaeoecological record (i.e. c. 225 years) is there evidence for a correspondence between wet and dry phases and consistent change in the records of percentage melt (two to three cores) and δ18O (two cores or one core).
The balance of evidence therefore supports an important role for autogenic geomorphological–vegetation change during the long-term development of the three wetland polygons examined (LaFarge-England et al. 1991; Ellis & Rochefort 2004). Periodic changes in vegetation and soil moisture during the development of Arctic wetland polygons, where unrelated to palaeoclimatic proxy records, might instead reflect local periglacial processes (Ellis & Rochefort 2004) resulting in amplitude changes between polygon ridges and centres. Deeper polygons will accumulate more snow in winter (resulting in higher values of P, Young et al. 1997) and will have greater rates of surface (Qs) and groundwater (Qg) input (Rovansek et al. 1996; Young et al. 1997), causing them to be wetter than polygons with a shallower topology. Wet and dry phases can then be explained by a feedback between sediment accumulation, controlled by the input of organic and mineral material, and the upward growth of ice-wedges, which is limited by the rate of sediment accumulation (Harry & Gozdzik 1988; Mackay 2000). Wet phases may correspond to periods when the amplitude between a polygon's ridges and centre is relatively large (higher values for Qg, Qs and P) although upward growth of ice-wedges, and the continued development of ridges, will gradually slow or cease, to be resumed only when sufficient sediments have accumulated in the polygon centre. This intervening period will lower the amplitude between polygon ridges and centre, and therefore lower values of Qg, Qs and P, causing the shift to drier soil moisture conditions evident in the palaeoecological record (Ellis & Rochefort 2004). A process of staggered ice-wedge formation resulting in a chevron pattern of growth (Dostovalov & Popov 1966; Mackay 1974; Lewkowicz 1994) might therefore explain recurrent shifts between wet and dry phases during the long-term development of low-centre polygons (Ellis & Rochefort 2004).
Where evidence for a climatic effect is based on a correspondence between palaeoecological and palaeoclimatic records in one core alone, or points to a contrasting response across several cores, it must be considered equivocal. However, a polygon in a dry state (with lower ridges relative to and surrounding the polygon centre and low values of Qg, Qs and P) may register a shift to a wetter and cooler climate, whereas one in a wet state (with higher ridges and values of Qg, Qs and P) may not. Thus, wet and dry phases in the combined record of palaeohydrological change matching significantly higher and lower palaeoclimatic proxy values (Fig. 7) may point to an underlying long-term climatic influence, which is modified by local geomorphological conditions so that it is only registered in individual cores.
There are, however, two distinct periods (Figs 5 & 6) during which a climatic effect on polygon development is well supported (c. 305–420 and 420–530 cal. years bp). Period I, characterized by contemporaneous wet phases in all three cores, and associated with significantly lower values of percentage melt (BY-LowC, BY-HighC and BY-a) and δ18O (BY-a), is broadly coeval with a period of climatic cooling well documented from sites in the North Atlantic region (Lamb 1965; Williams & Wigley 1983; Millar & Woolfenden 1999) including the Arctic (Williams & Wigley 1983; Gajewski & Atkinson 2003): i.e. the Little Ice Age (LIA). The change from inferred drier soil moisture conditions during Period II to wetter conditions during Period I is also supported by evidence for a pronounced climatic effect at other High Arctic sites. Period I is coeval with geomorphological evidence for the effects of the LIA in the Qunguliqtut Valley (Klassen 1993; Allard 1996) and paleolimnological evidence for LIA cooling from Baffin Island (Hughen et al. 2000) and Ellesmere Island (Lamoureux & Bradley 1996). The direction of change is also consistent with evidence for colder though wetter LIA climatic conditions from Cornwall Island in the Canadian Arctic Archipelago (Lamoureux 2000; Lamoureux et al. 2001) and from central west Greenland (Bennike 1992).
the functional significance of palaeoenvironmental variation
The estimated values of past soil moisture, based on macrofossil mosses, suggest significant variation in hydrology during the development of low-centre polygons (Figs 1 & 5). Soil moisture exerts a control on tundra ecophysiology through production, decomposition and nutrient cycling (Miller et al. 1984), and a lowered water-table and increased thaw might be expected to accelerate the rate of soil decomposition (CO2 source) over photosynthesis (CO2 sink), so that the balance in tundra soils shifts from one of C-input, or storage, to C-output (Billings et al. 1982, 1983; Johnson et al. 1996). If the effect of decomposition were to increase nutrient availability, there may be an additional uptake of CO2 owing to higher rates of photosynthesis (Shaver & Chapin 1986; Shaver et al. 1998; Johnson et al. 2000), although as sink strength in vascular plants decreases, productivity may be offset by substrate-controlled or nutrient-limited CO2 loss from soil respiration by micro-organisms (Nadelhoffer et al. 1991; Hobbie 1996; Jonasson et al. 1999). Larger sinks of CO2 are accordingly associated with lower respiration rates in wetter habitats (Vourtilis et al. 2000), while short-term experiments designed to explain the net effect of climate warming on soil moisture and the C-balance of tundra plots (Johnson et al. 1996) support observational data demonstrating that a shift from net C-input to C-output accompanies the recent drying of tundra habitats (Oechel et al. 1993, 1995; Weller et al. 1995).
Inferred values for past changes in soil moisture are of a sufficient magnitude to affect CO2 flux in wetland ecophysiology, varying periodically between values ranging from c. 705 ± 126.5% d. wt. (the average and SE for DCA axis 1 scores in the range 0–3) to c. 1891 ± 175% d. wt. (range 3–4) (Figs 1 & 5). A corollary is that the observed and predicted effects on tundra soil moisture of human-induced climate warming are, in High Arctic polygon-patterned wetlands, recent additions to previous, long-term fluctuations in ecosystem function. If the inferences of this study are confirmed for Arctic wetlands in general, an inherent variation in ecosystem function controlled by climate change will have to be taken into account when interpreting the results of short-term environmental manipulations (e.g. Arft et al. 1999), and of studies attempting to scale-up results from short-term experiments to longer-term processes (McKane et al. 1997; Epstein et al. 2001).
The work was funded by NSERC, FCAR, Centre d’études nordiques and Faculté des sciences des l’agriculture et de l’alimentation, Université Laval, as a University postdoctoral scholarship awarded to C.J.E. Fieldwork was carried out from the Bylot Island Goose Camp, funded by a Canadian Wildlife Service Grant to Gilles Gauthier. Gilles Gauthier (Université Laval) provided important help and advice throughout the study. We gratefully acknowledge logistic support and additional funding provided by Polar Continental Shelf Project. The Hunter and Trapper Association of Pond Inlet (Baffin Island) enabled access to Bylot Island. We thank Lindsay Haddon, Colin Osborne and anonymous referees for suggesting improvements to the manuscript.