In this study we provide a quantification of the main patterns of change of a subarctic peatland caused by permafrost decay monitored between 1957 and 2003. Up-thrusting of the peatland surface due to permafrost aggradation during the Little Ice Age resulted in the formation of an extensive peat plateau that gradually fragmented into residual palsas from the 19th century to the present. Only about 18% of the original surface occupied by permafrost was thawed in 1957, whereas only 13% was still surviving in 2003. Rapid permafrost melting over the last 50 years caused the concurrent formation of thermokarst ponds and fen-bog vegetation with rapid peat accumulation through natural successional processes of terrestrialization. The main climatic driver for accelerated permafrost thawing was snow precipitation which increased from 1957 to present while annual and seasonal temperatures remained relatively stable until about the mid-1990s when annual temperature rose well above the mean. Contrary to current expectations, the melting of permafrost caused by recent climate change does not transform the peatland to a carbon-source ecosystem as rapid terrestrialization exacerbates carbon-sink conditions and tends to balance the local carbon budget.
 We have followed the fate of permafrost from 1957 (enlarged aerial photographs taken in 1957, starting date) to 2003 based on the detailed monitoring of a subarctic peatland (56°11′N/75°55′W: eastern coast of Hudson Bay, northern Québec, Canada) using four field surveys at 10-year intervals in 1973, 1983, 1993 and 2003, respectively. The peatland investigated is in a pristine area of the discontinuous permafrost zone of northern Canada [Payette and Rochefort, 2001]. The site is representative of the extensive permafrost peatlands developed over clay-silt marine sediments along the eastern Hudson Bay coast between 53°N and 58°N [Allard and Seguin, 1987; Lagarec, 1982; Payette and Rochefort, 2001]. Peatland monitoring was based on the distribution of all permafrost landforms (palsas, collapse scars, thermokarst ponds) and permafrost-free sedge-shrub lawns within the geographical limits of the permafrost peatland. A dense grid of geographic coordinates was laid out for landform mapping in 1973 [Samson, 1974], whereas an optical prism was used in 1983 [Laprise and Payette, 1988] and a total station in 1993 [Laberge and Payette, 1995] and 2003. Field measurements were made between early and mid-summer. Permafrost inception and development cause the up-thrusting of the peat surface due to the growth and piling of lenses of segregation ice in peat and more particularly in mineral deposits [Allard and Rousseau, 1999]. Tree colonization occurs sometime after up-thrusting of the peat surface because of well-drained soil conditions. Thus the age distribution of the oldest black spruce (Picea mariana (Mill.)B.S.P.) trees (n = 15) growing on palsa mounds was used to determine the minimum age of permafrost development.
3. Results and Discussion
 Permafrost inception likely occurred during the Little Ice Age, and corresponds to maximum age of black spruce trees (350 years: from AD 1680 to present) colonizing the frozen peat plateau protruding above the peatland surface. Several dead trees distributed in thermokarst ponds indicate recent decay of permafrost bodies. The overall distribution and life span of pond trees and tilted (reaction wood) living trees [Laprise and Payette, 1988] on palsa mounds suggest continuous permafrost thawing during the 20th century. About 18% of the initial frozen peatland surface was melted in 1957 (Figure 1). Thereafter accelerated thawing occurred with only 38%, 28% and 13% of the original frozen surface still remaining in 1983, 1993 and 2003, respectively (Figure 2a). However, the annual loss of permafrost from one survey to the other (Figure 2b) indicates increasing decay rates from 1957–1983 (2.5% yr−1), to 1983–1993 (2.8% yr−1) and 1993–2003 (5.3% yr−1). Assuming a loss rate equivalent to that of the last decade, permafrost will disappear completely in less than 20 years from now.
 From 1957 to 2003, permafrost loss was compensated by a concurrent gain in thermokarst ponds and fen/bog vegetation (Figures 1 and 2a). The decadal trends in the area occupied by ponds and fen/bog vegetation indicate active hydroseral processes towards terrestrialization of the peatland with fast sedge (Carex aquatilis Wahlenb.) and Sphagnum riparium Angstr. invasion and establishment. Shallow thermokarst ponds and slow subsidence of palsa sides likely provided propitious conditions for effective terrestrialization. Over the last 10 years, permafrost subsidence (as measured from 10 palsa mounds in 1993 and 2003) was of the order of 40% of the original height (about 1–1.5 m height decrease) which corresponds to substantial inflow of water from melting ice within palsas. Transformation of the peatland surface shows the fen/bog vegetation to assume dominance progressively, but reversion to pond conditions occurred during the survey period (Figure 3) which suggests important fluctuations of the water table through time. Old-aged thermokarst ponds (i.e., >20 years old) also are probably the deepest of all ponds, as they are at the centre of large sunken palsas where the largest volume of segregation ice was located.
 The overall spatial pattern of permafrost degradation from 1957 to 2003 shows the progressive fragmentation of the peat plateau, with most residual palsas and thermokarst ponds having their longest axis north-south oriented (Figure 1). This repetitive pattern of elongated palsas and ponds sets the ideal topographic conditions for snow trapping from dominant westerly winds. Increased snow precipitation from the late 1960s to present (Figure 2c) facilitated permafrost thaw along palsa margins exposed from the north-east to the south-east as also indicated by the main orientation of tilted trees [Laprise and Payette, 1988]. Over the last 30 years, the fragmentation process associated with continuous permafrost thawing caused a three-fold increase of the number of residual palsas (Figure 2a) along with a sharp decrease of the size of individual palsas. The progressive fragmentation and isolation of residual palsas and development of thermokarst ponds resulted in the loss of physical inertia of the initial peat plateau, and rapid decay began after 1957 when increased snow precipitation reduced frost penetration [Seppälä, 1990, 1994] although air temperature remained rather stable until the mid-1990s. Since 1996, accelerated thawing was facilitated by the dramatic increase of precipitation and temperature (Figure 2c).
 Long-term ecosystem monitoring provides baseline information on the patterns and processes of ecosystem changes associated with recent ecological changes in pristine northern environments. Two major findings here are the rapid permafrost thawing based on well documented melting rates over the last 50 years and the resulting formation of compensatory successional trajectories relative to the global carbon budget [Christensen et al., 2004], i.e., the concurrent development of thermokarst ponds (C-source) and peat accumulation through terrestrialization (C-sink). Also of interest are the successive patterns of permafrost/thermokarst distribution and change after 1957 which suggest that the degradation of permafrost was a fast, one-way, self-organized process. Long-term ecosystem monitoring is important for realistic predictions of future behaviour of subarctic peatlands as warmer and wetter conditions are anticipated in this century for high-latitude areas [Houghton et al., 2001]. The dramatic decay of peatland permafrost since the mid-1950s is on line with several other accelerated ecosystem changes during the past 50 years across the northern and southern biomes [Crutzen and Steffen, 2003].
 We are most grateful to NSERC (Canada) and FQRNT (Québec) for financial support. We thank M. Seppälä and two anonymous reviewers for thoughtful comments on an earlier draft.