Precipitation observations from 1996 to 2016 interpolated for Poplar Fen averaged 380 ± 17(SE) mm total precipitation with 284 ± 15 mm falling as rain and 96 ± 6 mm as snow (Table 2). For the 4 hydrologic years of high burned area in the spring, total winter snowfall was below average for all years except for 1997–1998. The lowest total snowfall was measured in hydrologic years 1998–1999, 1999–2000, and 2008–2009 – all years with low burned area in the spring. Peak SWE from the Gordon Lake snow pillow from 2009 to 2016 (Fig. 3) averaged 120 ± 10 mm. Peak SWE prior to snowmelt in hydrological years 2010–2011 and 2015–2016 was not especially low, and, despite the modest SWE available for melt, the snow-free day of the year (80 % of peak SWE melted) during these years was not significantly earlier than the other 5 years on record. Total rainfall was below average in only 2 (1997–1998 and 2010–2011) of the 4 hydrological years with large spring burned areas; the bulk of precipitation occurred in the summer for all 4 years (Table 2). Cumulative post-melt rainfall until 15 May averaged 25.5 ± 3.3(SE) mm between 1997 and 2016 (Fig. 3). A total of 3 of 4 hydrological years with high burned area in the spring were below the 20-year average rainfall, with 2001–2002 being the lowest and 1997–1998 just above average. In 2015–2016, only 8.5 mm of rain fell following snowmelt prior to ignition of the Horse river wildfire, and only 0.3 mm fell over the next 2 weeks leading up to the burning of the Poplar Fen watershed (Fig. 3). The hydrological year with the lowest early spring cumulative rainfall in the 20-year record was 2007–2008 (1 mm), which was a year of low burned area in the spring (Natural Resources Canada, 2017). However, during this year a total of 151 mm of snow fell in the area, which is 55 mm more than the 20-year average (Table 2), and snow-free conditions were not reached until 30 April.

Over the 2015–2016 winter (mid-October–mid-April), average air temperatures were -6.5 ∘C, which is 2.9 ∘C warmer than for the previous 19-hydrological-year (1996–2015) average (-9.4 ∘C). Periodic warm spells were observed in late January and February, when air temperatures rose above freezing for several consecutive days (Fig. 4a). Manual snow measurements yielded an area-weighted average peak SWE of 105 mm (Fig. 4d) on 21 March 2016, which is 2 mm higher than the peak measured at the Gordon Lake snow pillow. In spring 2016, the primary snowmelt period occurred between 21 March and 19 April. Air temperatures did not deviate far from the 20-year normal during this time, with the exception of 27–30 March, when daily maximum air temperatures in the area rose to over 9 ∘C. The strongest deviation prior to the fire was measured following snowmelt when maximum daily air temperatures reached 27 and 33 ∘C in April and May, respectively. At this time, average daily relative humidity decreased (Fig. 4b) and average daily wind speeds exceeded 20 km hr-1 prior to the fire's ignition (Fig. 4c).

The NW fen (Fig. 1) water table range was ∼ 0.79 m (+0.12 m to -0.66 m) between 8 June 2011 and 17 May 2016 (Fig. 5). Average NW margin water table was 0.32 m lower than NW fen between 2011 and 2015. Both NW fen and margin exhibited relatively low water tables (dry conditions) at the beginning, increased water table in the middle years (wetter conditions), and lower water tables in a drying period towards the end of the 5-year record. The late fall and early spring NW fen water table was near or above ground surface in hydrological years 2012–2013 and 2013–2014 (Fig. 5), which corresponded with delayed ground thawing at 0.1 and 0.2 m peat depths until mid-May. Conversely, the year 2014–2015 water table was ∼ 0.2 m b.g.s. (below ground surface) in the fall and at the surface in the early spring, which began to decline rapidly in June (Fig. 5). This hydrological year corresponded with delayed ground thawing until mid-May at only 0.2 m peat depth. Furthermore, between 2011 and 2015, NW fen water table underwent periods of decline over the summer in all years except 2013. By early fall, the NW fen water table in all 5 years reached an annual low and, in 2012–2014, rose in the late fall prior to freeze-up. Conversely, rainfall was not sufficient in 2011 and 2015 to raise the fall NW fen water table above the low levels observed in the summer (Fig. 5).

The 2015–2016 hydrological year began with water levels that were among the lowest in the 6-year record (Fig. 5). By the end of winter, all manually surveyed fen monitoring wells were empty of water (water tables > 0.8 m b.g.s.). The comparison of fall 2015 logged water levels to manual winter 2016 observations (before snowmelt recharge and before pressure transducers were installed for the 2016 field season) evidenced an additional 0.12–0.26 m water table decrease, demonstrating mid-winter water table decline and drying of overlying peat substrate. Ground thawing at 0.1–0.2 m depth occurred in mid-April (earlier than 2013–2015) toward the end of snowmelt, and at this time (16 April 2016) the NW fen water table had increased to 0.67 m b.g.s. The remaining snowmelt period initiated a water table rise of 0.46 to 0.21 m b.g.s. on 3 May which then decreased in the total absence of rainfall to 31 cm b.g.s on 17 May, the day that the Poplar Fen area burned over (Fig. 5).

To examine how fen areas of varying topographic position were wetting and drying over the 2015 growing season, water table and hydraulic gradients were compared between the contrasting upper and lower fen areas (Fig. 6). Average water table depth below surface differed by 0.05 m between upper (0.22 ± 0.05(SD) m) and lower (0.17 ± 0.04(SD) m) fen nests. In both areas, the hydraulic head in underlying mineral layers mirrored the water table profile (Fig. 6). Vertical hydraulic gradients (a metric of groundwater recharge–discharge) in both upper and lower fen areas were strongest when water tables were highest and weakened (less groundwater recharge to the fen) during periods when rainfall was less abundant. Throughout the entire monitored period, the lower fen nests had the strongest average hydraulic gradients (0.021 ± 0.008(SE)), showing groundwater discharge that remained positive throughout the measurement period. Conversely, upper fen nests had weaker hydraulic gradients (-0.007 ± 0.004(SE)), which experienced flow reversals (downward), and were negative throughout most of the year. Margin areas exhibited the lowest water tables, as well as hydraulic gradients (-0.03 ± 0.03(SE)) (recharge), over the 2015 growing season (Fig. 6).

Between June and October 2015, duff and margin peat VWC (both at ∼ 0.2 m b.g.s.) averaged 0.33 and 0.41 m3 m-3, respectively, with a higher coefficient of variation for duff (0.21) compared to margin (0.06) peat (Fig. 7). The duff experienced extended drying periods in the summer–fall and, by freeze-up, reached the minimum VWC for 2015 (0.24 m3 m-3). Margin peat VWC had also reached a minimum by fall (0.39 m3 m-3); however, values were similar to late spring 2015 VWC (∼ 0.42 m3 m-3). During winter 2015–2016 (31 October–21 March), VWC in the duff and margin peat decreased an additional 0.06 and 0.03 m3 m-3, respectively, and, following snowmelt, increased from 0.19 to 0.32 m3 m-3 and 0.36 to 0.38 m3 m-3, respectively. Two weeks prior to the Horse river wildfire, upland duff and margin peat VWC were 0.31 and 0.37 m3 m-3, respectively (Fig. 7), and likely continued to decrease prior to the fire in the absence of rainfall.

Following the first month of startup in 2015, the DC illustrated an inverse relationship with upland duff VWC (R=-0.94; p<0.001) (Fig. 7); the dry conditions caused DC to increase from 18 to 496, between 12 April and 31 October. VWC on 31 October and the corresponding DC were chosen to represent the final fall moisture and DC equivalent values for the various overwintering DC calculations. The overwintering period ran from 31 October 2015 to 19 April 2016, the day following 3 consecutive days with noon air temperatures ≥12 ∘C. For scenario 1, the 2016 spring startup DC was predicted based on the relationship between upland duff DC and VWC in 2015, and a 2016 spring startup DC of 357 was estimated given a starting soil moisture of 0.37 m3 m-3 (Fig. 7). Scenarios 2 and 3 produced startup values using the overwintering procedure with standard carry-over and wetting-efficiency coefficients, resulting in startup DCs of 242 and 212, respectively. Scenario 4 used the overwintering procedure with no precipitation values or default coefficients, but rather with directly measured duff recharge from 31 October to 19 April. Snowmelt increased duff VWC by 0.13 m3 m-3 in the ∼ 0.25 m thick soil horizon, resulting in 32 mm of recharge (35 % of estimated upland snowfall), yielding a startup DC of 321. DC was then calculated from 19 April to 17 May 2016, and all starting DCs increased 131 units over that time period (Table 3).

The greatest depth of burn was measured in the margins (0.13 ± 0.01 m) with lower (0.10 ± 0.02 m) burn depths measured at upland locations (Table 4). Burn depth values in burned fen areas were 78–83 % lower than margin and upland areas. Estimated fuel consumption rates (depth of burn × average bulk density) generally echoed the trends in surface change with slight differences due to higher bulk density measured in margin peat. No surface changes or fuel consumption were observed in the lower fen area.

Within the Boreal Plain region of northeastern Alberta, average precipitation is less than potential evapotranspiration in most years (Bothe and Abraham, 1993). Consequently, water deficits are common in the WBP, relying on infrequent wet periods every 10–15 years to replenish storage deficits (Marshall et al., 1999; Devito et al., 2005). Historical precipitation data illustrate that rain and snow patterns are variable in the WPB (Table 2; Fig. 3). Total snowfall was near or below average in years during which spring wildfires burned large areas. Although modest snowfalls are a recurring influence, they do not necessarily dictate fire magnitude; a total of 5 years with spring wildfires of low burn area were identified, possessing similar (or lower) total snowfall values than large spring burn area years (Table 2). Earlier snowmelt can extend the dry WBP spring and drying of organic soils, which could therefore extend the period over which spring fires can be generated (Westerling et al., 2006). However, the timing of snowmelt in the WBP does not appear expedited in years of large burned area in the spring, with no significant patterns in the timing of snow-free conditions observed in the 7-year Gordon Lake snow pillow record (Fig. 3). However, years of high total annual snowfall all align with years with low burned area in the spring (Table 2). This suggests that large SWE can contribute to decreasing the total annual area burned in the spring. Low and infrequent early precipitation events occurred in 3 of 4 years with high burned area in the spring. However, due a large proportion of rainfall in continental western Canada generally occurring in summer (Smerdon et al., 2005), dry early spring is not exceptional and not restricted to years of high burned area in the spring. The year with the lowest early spring cumulative rainfall in the 20-year record was 2008; however, above-average snowfall and late snow-free conditions decreased wildfire susceptibility in the spring, further demonstrating the importance of a large snowmelt for reducing wildfire vulnerability (CFRC, 2001).

The 2015–2016 hydrological year experienced the second warmest winter temperatures over the past 20 years. Periodic rises in air temperature above freezing conditions throughout the winter (Fig. 4a) supplied energy for mid-winter snowmelt and sublimation (Pomeroy et al., 1998), potentially decreasing available peak SWE for the spring snowmelt period. The modest snowpack melted over a 31-day period. Immediately following snowmelt, high air temperatures, low relative humidity, and high wind speeds (Fig. 4b, c) created weather conditions optimal for the spread of wildfire (Van Wagner, 1977). Similar mild winter temperatures and warm, dry spring conditions were present in previous years of high spring time burned area in 1968, 1998, 2002, and 2011. These years produced fires of a similar magnitude and total area burned to the Horse river wildfire of 2016 (Hirsch and de Groot, 1999; Tymstra et al., 2005; FTCWRC, 2012).

A 5-year (2011–2016) water table record illustrated the susceptibility of Poplar Fen to extended drying periods, with years of high spring (2011 and 2016) and summer (2015) burned area corresponding with low water table position (Fig. 5). At Poplar Fen, water tables also decreased over winter periods in the absence of precipitation-driven recharge. These prolonged periods of water table decline were evidenced by logged water table and mineral piezometer observations from the lower and upper fen areas (Fig. 6). In these areas, the hydraulic head in the underlying mineral substratum (∼ 1.5 m b.g.s.) closely mimicked the pattern of the water table, suggesting that the underlying groundwater at Poplar Fen is derived mostly from local recharge, rather than from regional groundwater, which would have a more stable hydraulic head (Siegel and Glaser, 1987). Therefore, peatlands that are supplied mainly by local groundwater (such as Poplar Fen) become particularly vulnerable to wildfire during high-risk fire weather conditions (Lukenbach et al., 2017).

Spring NW fen water table position was also related to the persistence of a frozen upper saturated zone. For example, near-surface water tables in fall of 2012 and 2013 (Fig. 5) allowed for relatively homogenous overwinter freezing of the upper saturated zone (Price, 1983), which reduced the permeability of the peat (Roulet and Woo, 1986; Quinton et al., 2009) and helped store subsurface water over the winter periods (Price and FitzGibbon, 1987). Ground ice persisted into mid-late May in 2013 and 2014, thus limiting snowmelt water infiltration (Roulet and Woo, 1986) and subsurface water loss to the underlying silty sand and outwash layers (Price and FitzGibbon, 1987). Conversely, the shallow (0–0.2 m) peat had reached above freezing temperatures by the end of snowmelt (mid-April) in 2016, suggesting that low (∼ 0.55 m b.g.s.) fall 2015 water tables had prevented the near-surface ground ice. Consequently, the entire saturated zone was free to recharge the underlying mineral layers over the 2015–2016 winter, and, during the 2016 snowmelt period, meltwater infiltrated readily to recharge the relatively deep water table. Thus, high antecedent fen water levels provide an important mechanism for overwinter storage and maintaining higher spring water levels.

Post-snowmelt 2016, the NW fen water table (0.3 m b.g.s.) was ∼ 0.3 m lower than the water table observed mid-June 2011 (Fig. 5), a period without rainfall and with high burned area in the spring when the 2011 Richardson Fire reached a size similar to the Horse river wildfire (Pinno et al., 2013). Surprisingly, spring 2016 water tables were more comparable to levels measured in the spring of 2012 (Fig. 5), a year of low burned area in the spring. The lower burned area was likely attributed to larger and more frequent rainfall events (an additional 14 mm) recorded in the region during the 2012 spring season. Peatland water table position, therefore, likely cannot serve as a stand-alone metric for estimating fen wildfire susceptibility in the region without considering the moisture deficits that can accumulate above the water table in the absence of precipitation.

Soil moisture in upland duff and margin peat followed a drying trend throughout 2015. Following snowmelt in 2016, water content in the upland duff and margin peat were not sufficiently higher than values observed in fall of 2015 (Fig. 7). These data suggest that there was no net wetting to the organic near-surface soils in the upland or margins at Poplar Fen from snowmelt infiltration. This soil moisture deficit was further enhanced by the lack of spring precipitation and increased evaporative demand (Hayward and Clymo, 1983) driven by the low humidity, high temperatures, and winds at the time of the Horse river wildfire (Fig. 4). This deficit would have increased the available fuels for the wildfire allowing for significant combustion of these organic layers (Table 4).

The historical meteorological and field hydrological data illustrate the susceptibility of regionally abundant WBP peatland watersheds to wildfire during extended dry periods. Results suggest that the wildfire at Poplar Fen, and the greater Horse river wildfire, was not simply a consequence of anomalous drought climate conditions, but rather interconnected hydrometeorological factors not uncommon to the Western Boreal Plain, occurring at least twice in the 5-year instrument record. These factors included low autumn soil moisture and water tables, modest snowfall, overwinter drainage, insufficient spring rainfall, and high spring air temperatures and winds. The synchronicity of these factors, occurring in the same hydrological year, combined with mature tree stands with high accumulated fuels ubiquitous to the region, likely contributed to the large magnitude Horse river wildfire. The similarities of the hydrometeorological events preceding the Horse river wildfire with previous years (1968, 1998, 2002, and 2011) of similar burned area in the spring (Hirsch and de Groot, 1999; Tymstra et al., 2005; FTCWRC, 2012) suggest that the mild and/or dry fall, winter, and spring conditions conducive for spring fire occur frequently in the region with a recent recurrence interval of 5 years. Moreover, conditions favouring spring wildfire may be enhanced by climate change, given the responsiveness of forest fuel moisture to changes in temperature and precipitation (Weber and Flannigan, 1997; Flannigan et al., 2016).

During summer 2015, vertical hydraulic gradients decreased in all fen and margin wells over periods of low precipitation. In lower fen these remained positive throughout the 2015 sampling period (Fig. 6), indicating upward groundwater discharge into the basal peat (water gain to peatland) from the underlying silty sand and outwash layers. In upper fen regions, these values were always lower and eventually became negative over time in the absence of rainfall, suggesting a flow reversal (downward) and loss of water from the basal peat to the underlying silt layer. Margin areas, located at a higher topographic position between fen and upland, exhibited the strongest negative hydraulic gradients, suggesting that these areas were recharging the underlying mineral layers throughout the entire year. These subtle differences in topographic position therefore played a large role in the observed differences in burn severity between these areas (Table 4). Hence, treed headwater moderate-rich fens and fen margin tracts in the WBP may be particularly vulnerable to wildfire.

The 2015 moisture conditions observed in the upland duff of Poplar Fen were illustrated reasonably well with the DC. The DC was overwintered for 2016 using a range of startup values from different methods (Table 3). Scenarios 2 and 3 produced DCs that were lower than the expected DC (scenario 1), since carry-over and wetting-efficiency coefficients overestimated the recharge to the duff layer by 15–21 %. These default coefficients may not have accounted for the high sublimation rates caused by low relative humidity and high solar radiation, common to the western boreal forests of Canada (Burles and Boon, 2011). The lower recharge values measured at Poplar Fen (35 % of melt water) may also be due to moisture deficits that accumulated since the summer of 2015, as a high proportion of the available meltwater went towards recharging the unsaturated mineral soil underlying the duff. The startup DC that was calculated using the directly measured duff recharge (scenario 4) was much closer to the expected DC, suggesting that the overwintering calculation is suitable for the duff layer at Poplar Fen when VWC is directly measured.

Due to differences in soil bulk density and depth of burn, average duff fuel consumption was ∼ 50 % less than the consumption observed in margins (Table 4). The observed duff fuel consumption at Poplar Fen (∼ 7.0 kg m-2), along with the VWC-inferred expected (488) and overwintered (452) final duff DC values, were both in line with fuel consumption and DC estimates from interior Alaska (Kane et al., 2007) and are on the higher end of DCs measured from other burned boreal forest fires throughout continental western Canada (de Groot et al., 2009). Thus, the overwintering procedures that were calculated using default wetting-efficiency coefficients produced lower final DC values that did not reflect the fuel consumption rates measured at Poplar Fen. The observed range in overwintering DC calculations in Table 3 highlights the difficulties in determining a proper startup DC for watersheds that experience periods of prolonged drying prior to snowmelt. These overwintering calculations have a substantial impact on DC values calculated for the following growing season, predominantly in the early spring. Estimations based on VWC measurements may therefore produce more accurate and conservative spring DC values, given that the selected coefficients may not properly represent the hydrological and meteorological processes occurring in the Western Boreal Plain during the snowmelt period.