Surface-groundwater exchange between a wetland, sandur, and lava field in southeastern Iceland

Springs are discrete discharge points where groundwater is delivered to the Earth's surface (Woo 2012). Spring water is supplied by groundwater, which itself is supplied by melt water, surface waters, and precipitation. Due to this inter-connected relationship, studies of spring hydrology have hydrological (Levy et al. 2015), chemical (Morgenstern et al. 2015), and ecological (Jansson et al. 2007; Fattorini et al. 2016; Huryn et al. 2021) implications for the downstream systems they feed. In addition, well, an evaluation of springs can provide further insight into groundwater-recharge patterns.

Iceland offers a unique location for studies of groundwater-surface water exchange. Glacier melt waters feed rivers and lead to the formation of highly permeable glacial outwash plains (Sandur = singular; Sandar = plural). The island is volcanically active, and lava fields and older erupted lavas play a role in influencing flow paths and water storage. Springs can be found feeding rivers and wetlands along the edges of young, tertiary basalt lava fields (Kiernan et al. 2003; Scheffel & Young 2021) and discharging from groundwater seeps (springs in unconsolidated sediments) into sandar landscapes (Finger et al. 2013; Levy et al. 2015).

Long-term increases in precipitation, annual air temperature, and glacial melt are expected to bring about changes in the hydrology of Iceland's landscape (Jóhannesson et al. 2007). Icelandic glacial loss in 2018/2019 was one of the highest on record (mass change rate −15.0 ± 1.6 Gt a⁻¹), and almost 50% of the glacial loss in Iceland, since 1890, has occurred between 1994/1995 and 2018/2019 (Aðalgeirsdóttir et al. 2020). Understanding the impact of these changes on groundwater recharge and spring discharge will be valuable for a country where 95% of household water is supplied by groundwater (via springs, boreholes, and wells) (Kløve et al. 2017).

Spring-fed rivers in Iceland have a stable discharge and show little response to precipitation or snowmelt (Jónsdóttir & Uvo 2009). However, predicting the vulnerability of springs to climate change is complex, and many variables influence their longevity. While groundwater springs may exhibit little seasonal variation, long-term studies have found that if groundwater is not replenished, springs may be lost completely (Finger et al. 2013; Levy et al. 2015), or in aggrading permafrost zones, for instance, with glacier retreat, spring outlets can be frozen over, and cease to function, with groundwater being re-routed to other and new seepage outlets (Scheidegger et al. 2012).

A recent study from Scheffel & Young (2021) found that springs in Southeast Iceland may play a role in the maintenance of a coastal wetland. These springs are located at the border of the young-tertiary basalts of the Laki lava field and also occur, where the glacial Brunná River flows onto the northeast edge of Skeiðarársandur, a 1,300 km² outwash plain (Blauvelt et al. 2020). Whether these springs are fed by fast-flowing conduits in the lava field or through the sediments of the sandur has implications for the adjacent wetland ecosystem.

The flow and longevity of the spring-fed systems are tied to the forces that recharge the groundwater system connecting to the springs. Where groundwater is recharged by glacial-melt waters, the retreat of glacial margins can result in changing flow paths and water availability (Levy et al. 2015; Dochartaigh et al. 2019).

Both short- and long-term studies are needed to monitor the effects of climate variability and change on spring hydrology and the impacts on downstream systems in this region. Overall, this study aims to provide an improved understanding of springs feeding into coastal wetland ecosystems occurring in Southeast Iceland. As such, we present baseline hydrological data for springs in a sandur-lava field-wetland complex in Southeast Iceland. The location and structure of the springs are reported as are the water quantity and quality of a subset of springs. Stable isotope, hydrograph analysis, and water temperature changes are used to assess the impact of heavy rainfall amounts on spring discharge. Finally, we surmise on the possible response of these springs to future climate warming in this area of Iceland.

In the centre of the study site is a wetland meadow. Spring-fed streams flow into small ponds, then drain out onto the Skeiðarársandur glacial outwash plain.

Given that this coastal wetland has and continues to be threatened by erosion by an expanding sandur triggered during flooding from extreme rainfall and glacial floods (i.e., jökulhlaups) (Scheffel & Young 2021), the land owner has built berms of crushed basalt and compacted soil and vegetation to separate the wetland from the sandur. These berms lie on the western and southern edge of the property. They are effective at slowing groundwater water flow into the wetland (Scheffel & Young 2021) but overtopping of the southern berm, which is older and lower in elevation still occurs during large floods.

Over 50 springs were identified at the study site on 29 June 2018 and 18–30 May 2019 (Figure 1, Supplementary Table 1). A sub-sample of six springs: two at the edge of the lava field, two at the edge of the sandur, and two in the wetland were selected for more in-depth study (see Table 1). Wetland spring (W1) was also selected to examine its long-term water level and temperature regime.

Water level and water temperature are useful for evaluating groundwater movement in surface waters. Cold and continuous groundwater flow is an indicator of a deep aquifer with old groundwater. It can take days to years for surface waters to reach these aquifers, so their discharge is unaffected by local changes in temperature and precipitation (Kresic & Stevanovic 2010). In contrast, springs fed by shallow aquifers with low storage capacity would have a periodicity that is more-strongly linked with groundwater-recharge events such as rainfall and snowmelt (McDonnell et al. 2007; Kresic & Stevanovic 2010).

An ONSET HOBO U20 Water Level pressure transducer (± 0.075% per 0.3 cm water) was installed at the sandur spring S1 and the wetland spring W1 from 1 July 2018 to 18 May 2019, and reset again from October 2019 to October 2020. An ONSET HOBO U20 Water Level pressure transducer was tied 1-m above the ground, onto a hydro pole at the border of the sandur and wetland. It recorded air temperature and atmospheric pressure hourly, from 1 July 2018 to ∼ 1 October 2020. Atmospheric pressure data is required to correct water level data obtained with non-vented HOBO U20 Water level sensors into units of water depth (m) (Rosenberry & Hayashi 2013).

Air temperature would be less likely to influence spring water that comes from deep groundwater stores but would have an influence on older surface water and rainwater temperatures. Additionally, air temperature can provide an indication of the energy available for warming the waters (Woo 2012), and for glacial melt (Jónsdóttir & Uvo 2009).

When atmospheric pressure data were missing, air pressure data from the nearest weather station (Kirkjubæjarklaustur – Stjórnarsandur) was used to adjust for water levels. Water levels were compared with measured water levels and when needed, were corrected with measured data.

Water temperature was also recorded at the streams emerging from 12 springs across the study site using HOBO Pendant MX Temp MX2201 and Temp/Light MX2202 Loggers (±0.5 °C, −20–70 °C and ±10% light accuracy for direct sunlight). Metal corner braces were zip-tied to rocks to create a platform for the logger to be secured to. The temperature loggers were installed horizontally across the corner braces with the light sensor facing upward. Due to algae growth on the sensors, the light data were not used in this study.

An ONSET HOBO RG3-M tipping bucket rain gauge (±0.2 mm/tip) at the study site collected rainfall data from 19 May 2019 to 30 July 2019 until it was knocked over in a windstorm. On occasion when the weather data (precipitation, air temperature) were not available for the study site, they were obtained from the closest Icelandic Meteorological Office (IMO) at Kirkjubæjarklaustur – Stjórnarsandur, a distance of 20.5 km to the southwest.

The velocity-area method was used to estimate discharge from selected springs following methods described by Herschy (1993) where section discharge, ⁠, is calculated by multiplying the width, depth, and velocity measured at each stream section modified by spring flow.

Water velocity was measured using a SENSA-RC2 Water Velocity Meter (range 0.000–4.000 m/s, Resolution: 0.001–0.020 m/s, accuracy ± 0.5% ±5 mm/s). The velocity of each section was the average of two 1-min average velocity measurements. As the streams emerged from the springs and had no other apparent sources (besides rainfall), the discharge of the channel was taken to be the discharge of the spring. The exception was the water emerging from W2, W1, and L2 where multiple small springs were located close together, their waters merging to form a stream where the measurements were taken. It was assumed that the discharge measured from the stream should be representative of the discharge from the spring(s) if all discharged water flowed to the outlet stream, and the outlet stream did not have any interfering surface or subsurface inflows.

Stream discharge was measured twice daily, usually in the morning and evening (20–29 May 2019). Due to a technical malfunction of the velocity meter, measurements from the evening of 28 May 2019 and the morning of 29 May 2019 were omitted.

Hydrographs and thermographs can be displayed with hyetographs (precipitation time series) to identify components of streamflow coming from these different sources. The lag between the peak in the recharge event (precipitation, snow, glacial melt) and change in spring discharge or chemistry is related to various factors such as recharge volume, intensity, flow paths, and residence times (McNamara et al. 1998). The height of the resulting peak can also be an indicator of the importance of runoff to the water supply. Streams with flow peaks occurring during spring melt and rainy seasons are runoff dominated. Glacial rivers can be identified by large peaks in discharge from spring to autumn when higher air temperature and radiation increase snow and ice melt. If the discharge is steady with a slight and delayed increase in discharge following a precipitation event, it is more likely to be a groundwater-dominated spring/stream (Jónsdóttir & Uvo 2009). The steadier the discharge, the deeper the groundwater aquifer is likely to be (McDonnell et al. 2010).

In this study, hydrographs, which plotted water level and rainfall across time, were compared to decipher the timing of peaks in water level in relation to precipitation events. Precipitation data were summarized using the programming language R (R Core Team 2019) in RStudio (RStudio Team 2020) to provide hourly and daily rainfall. To reduce skewing of daily averages and totals, days with more than four hours of missing data were removed from analyses. To determine if the amount of total precipitation received in a day had an impact on water levels or temperatures, daily total precipitation was grouped into four categories: dry (0 mm), negligible (0.1–1 mm), light (1.1–10 mm), and heavy (>10 mm). These light and heavy precipitation groupings are frequently used in precipitation studies (Sun et al. 2006; Vincent & Mekis 2006). Moreover, the 0.1–1 mm range was not grouped into the ‘dry’ category to cautiously ensure that the effects of negligible rainfall were not missed. In some locations in the Arctic, trace precipitation events can comprise a sizeable component of total precipitation (Woo & Steer 1979).

Water temperatures for each location were plotted against time, along with daily precipitation and air temperature. The timing of peaks in water temperature was noted in relation to precipitation events and air temperature. When temperatures measured by the logger matched with the air temperature, it was assumed that the water level was below the height of the logger, and the data points were removed (Young et al. 2010).

Water chemistry has been used to determine recharge locations with mixed results. The quality of water will change as it interacts with rock and surface waters, dissolving anions, cations, and trace elements. The isotopic composition of the water will also change as water travels along various pathways. Differences in isotopic composition between precipitation and surface waters were identified over 59 years ago (Craig 1961). Water isotopomers (¹H₂¹⁶O, ¹H²H¹⁶O, ¹H₂¹⁸O) are differentially evaporated, with lower molecular weights evaporating more easily. Fresh rainwater therefore has low ¹⁸O and ²H content compared to older surface water that has experienced evaporation. The chemistry and stable isotope composition will shift towards the chemistry and stable isotope composition of the largest contributing source of recharge (Jefferson et al. 2006).

Hydrograph separations are commonly supplemented by stable water isotope ¹⁸O and ²H data (McDonnell et al. 2007). This approach can be useful for determining the source (recharge area) of surface and groundwater. The difference in stable isotope composition was examined for water bodies under baseflow conditions and following rainfall events. The isotopic composition of rain has been found to vary across space and time (Lawrence & White 1991). Samples were obtained from surface water (the Brunná River), spring water (from the six main study springs), and rainwater. All samples were collected in 30 mL high-density polyethylene (HDPE) sample bottles.

The water sampled from springs was collected from inside the spring opening (pipe) before the water mixed with surface waters or could be affected by evaporation. Before the isotope samples were collected, the water quality of the spring's waters were assessed with a Hanna HI98194 Multimeter Probe (e.g., temperature, pH and mV, oxidation reduction potential, dissolved oxygen, conductivity, and total dissolved solids) to ensure that the sample was representative of the water in the spring. Water samples were not obtained until the water quality readings were stable. Surface water was obtained by grab samples from the edge of the Brunná River.

To obtain rainwater samples for water isotope testing, a beaker (1 L) with a plastic funnel attached to the top was placed in the wetland to capture rainwater. The collector was left to collect rain for 12 days during the May 2019 field season.

The water isotope samples (n = 23) were analyzed by the University of Waterloo Environmental Isotope Laboratory using a Los Gatos Research T-LfWIA-45-EP Liquid Water Isotope Analyzer (LWAI). The accuracy of these analyses is 0.8% Vienna Standard Mean Ocean Water (VSMOW) for δ²H and 0.2% VSMOW for δ¹⁸O. Two-sample Student t-tests (Bluman 2006) were used to investigate differences in δ²H and δ¹⁸O composition between wet and dry conditions and between May and October 2019. The Local Meteoric Water Line (LMWL) for South Iceland was defined as δ²H = 6.5 δ¹⁸O – 3.5 by Sveinbjörnsdóttir et al. (1995) and was used as the baseline to compare δ²H and δ¹⁸O compositions obtained in the present study.

The Hvoll coastal wetland is rich in springs with over 50 springs being identified at the site, although the true number is likely much higher as safety concerns and nesting birds prevented investigation of some locations.

These events indicate that the wetland spring has a relatively stable discharge. Compared to S1, W1 showed a delayed and dampened response to precipitation events. The medium sandur spring, S1, had a more rapid response to precipitation, especially following precipitation events in the late summer and early autumn when glacial melt is elevated. Groundwater in Iceland can be replenished quickly following rain events as rain quickly flows over glaciers and penetrates highly permeable young lava fields (Jónsdóttir 2008). Waters from the S1 spring appear to be emerging from a hole in a tubular structure, possibly a lava tube (Figure 2(e)). These conduits have low resistance and are able to rapidly transport large volumes of water (Kiernan et al. 2003). They may be considered analogous to soil pipes developing in silty material underlining peat wetlands in Subarctic Canada. Woo & DiCenzo (1987) describe how these peatland pipes can transport considerable water from a wetland to nearby small streams, providing about 10% of the summer stream runoff, whereas Darcy's groundwater flow only provided about 1%. Jones (2010) provides additional details of soil pipes and their importance in hillslope drainage, streamflow, and water quality (acidity).

Increases in water levels and discharge may be expected for springs following precipitation-related jökulhlaup events, as more than 10 mm of precipitation caused significant increases in water level for the S1 and W1 springs. Jökulhlaups (glacial outburst floods – Björnsson & Pálsson 2008) at the Brunná River could not be identified with certainty over the course of the study; however, the nearby glacial-fed Djúpá River experienced discharge peaks that corresponded with those at S1 (Figure 3).

In contrast, the present study showed that W1, a spring closer to the centre of the wetland, experienced little variation in both diurnal and monthly water levels (Figure 4). Even when water levels were recorded over a prolonged period (October 2019 to October 2020), the daily average water level at W1 only varied from 8 to 19 cm, with daily water temperature from 3.1 to 5.5 °C (Supplementary Figure 1). Similarly, it has been observed that groundwater streams in Iceland often show little diurnal variation in flow (Crossman et al. 2011). Diurnal trends tend to be dampened for water stored or transported deeper into the vertical sediment profile (Tristram et al. 2015) or through groundwater aquifers (Jónsdóttir & Uvo 2009).

The pH of springs at the study site was slightly acidic, ranging from 6.1 to 6.6 (Table 2). Cold water springs across Iceland generally have pH values between 7.4 and 9.3 (Guðmundsdóttir et al. 2019). Studies have shown that water may have a more acidic pH if (non-explosive) volcanic ash and positive ion salts are present (Gíslason et al. 1996). However, the salinity of all springs was ∼0.02 ppt, indicating fresh water. Runoff rivers on tertiary basalt formations of the Eastfjords have a pH range of 6.1–7.2 (Gíslason et al. 1996).

If springs were fed by runoff, a change in water chemistry would be expected following precipitation events. From 20 to 30 May 2019, temperature, EC, and salinity were stable for the main study springs (Figure 6). The mode pH values for the small Wetland springs, W1 (6.1) and W2 (6.1), were lower than for rain (6.2), possibly attributed to water flow through the organic material (Mankasingh & Gísladóttir 2019). The two Sandur springs S1 and S2 had an increase in pH 24 h after the 28 May rainfall event. Higher pH values for water in Iceland can be attributed to groundwater flow through the basic basalt bedrock and fewer interactions with plants (Mankasingh & Gísladóttir 2019), and the atmosphere (Gíslason et al. 1996).

Some springs were found to be more sensitive to precipitation than others, and the effect of different categories of total daily precipitation (0 mm, 0.1–1 mm, 1.1–10 mm, and >10 mm) on the daily average water temperature was analyzed for the six main study springs. For instance, there was a significant effect of rainfall on daily average water temperature at the wetland springs: W1 (F_(1,188) = 7.1, p < 0.01) and W2 (F_(3,129) = 3.4, p = 0.02). A Tukey HSD test revealed that days with over 10 mm of precipitation resulted in a change in water temperature for both W1 and W2. The average water temperature for W1 increased from 4.9 ± 0.4 °C on dry days to 5.3 ± 0.2 °C when more than 10 mm of daily total precipitation fell and at W2, temperatures increased from 4.9 ± 0.3 °C to 5.2 ± 0.2 °C. Changes in water temperature following rainfall events are an indication of inputs from rainfall-generated runoff (Hamdan et al. 2016).

On 22 September 2019, there was a spike in temperature measured at the logger downstream of the Oxbow spring, but not upstream of it. The Oxbow spring (see Figure 1) is located at the bottom of a gentle slope at the edge of the lava field. These springs may receive surface runoff during heavy rainfall but this could not be confirmed during this study.

Water temperatures for small springs in the wetland located closer to the lava field did not show any significant change in average temperature with rainfall (L2: F_(3,129) = 0.83, p = 0.48; L3: F_(3,129) = 2.66, p = 0.051). Water temperature for the large spring, L1, near the lava field was affected by rainfall (F_(3,128) = 7.1, p < 0.01). For rain categories; 0.01–1 mm (Mean = 4.9 ± 0.3 °C), 1.01–10 mm (Mean = 4.9 ± 0.3 °C) and >10 mm (Mean = 4.9 ± 0.2 °C) daily total rainfall resulted in cooler average water temperatures compared to days with no rain (Mean = 5.1 ± 0.3 °C). At the beginning of a rain event, raindrops are several degrees cooler than the ambient air but are within 1 °C of the ambient air after the peak in rainfall (Byers et al. 1949).

At Hvoll, hourly time periods with rainfall averaged 11.4 ± 4.6 °C between 19 May and 1 October 2019. Therefore, inputs from local rainfall and runoff might be expected to cause an increase in water temperatures. However, no significant change was seen in average air temperatures measured at Hvoll when compared across precipitation groupings (F_(3,129) = 1.68, p = 0.18). L1's lower water temperatures on days with rainfall may therefore be an indication of older groundwater being pushed out following aquifer recharge (Zhang et al. 2020).

The six study springs and Brunná River plotted lower on the LMWL than the rainwater from the study, suggesting inputs from groundwater or surface waters. The springs had a narrow range in isotopic signatures (δ¹⁸O: −66.47 to −64.83, − δ²H: −9.80 to −9.54). In comparison to other studies of stable isotopic signatures in Iceland, these values were most like groundwater and spring water from the lower sandur of the Virkisá catchment in Southeast Iceland, although the springs in the present study plotted higher above the LMWL (Macdonald et al. 2016).

The Brunná River, Lava field springs L1 and L2, as well as the Wetland spring W1 plotted below the LMWL on 27 May 2019 (Figure 8). Groundwater stored and transported through basaltic rock can experience chemical reactions with the rocks that elevate δ¹⁸O but not δ²H (although elevated temperatures may be required for this reaction to occur) (Kristmannsdóttir & Ármannsson 2004). Many of the points also fell slightly above the LMWL. This may be caused by δ²H enrichment or δ¹⁸O depletion.

A two-sample Student t-test of spring stable isotope signatures indicated that precipitation causes a significant increase in δ²H but not in δ¹⁸O (δ²H: p = 0.002, δ¹⁸O: p = 0.30). The δ²H signature for the study springs shifted from an average of −66.9 ± 0.4 under dry conditions to slightly more enriched: −65.2 ± 0.4 on the day, following rainfall. There was a significant (p = 0.002) depletion in δ²H from May (Mean = −65.5 ± 0.5) to October (Mean = −66.1 ± 0.3). Increased δ²H can be caused by H₂S exchange or the hydration of silicates (Serno et al. 2017).

Climate models show varied predictions of future precipitation patterns across Iceland but generally agree that precipitation will increase with increases in air temperature (Aðalgeirsdóttir et al. 2006; Gosseling 2017). We surmise that this may result in higher water levels in the wetland triggered by spring-fed streams whose aquifers are recharged by rainfall. An increased probability of days with over 10 mm of precipitation is also expected (Gosseling 2017). During some days in this study with more than 10 mm of precipitation, we observed significant increases in water level at the wetland spring W1 and the Sandur spring S1. In the future, a higher number of days receiving heavy rainfall may result in an overall higher baseflow and spikes in wetland water levels.

Groundwater-fed springs in Iceland will become important sources of water as air temperatures increase because they are less vulnerable to the forces of evaporation. Water from a groundwater spring like W2 would not be influenced by evaporation, as water is stored and transported beneath the ground surface. However, springs also fed by surface runoff (Supplementary Figure 2) may be more vulnerable to rising temperatures and corresponding evaporation loss.

Models of glacial mass loss due to climate change predict that Vatnajökull will lose 20–30% of its mass by the end of the century though variability in melting will likely occur due to cool springs delaying glacier melt, to dust and tephra from volcanic eruptions being blown onto the glacier from the uplands accelerating melt (Aðalgeirsdóttir et al. 2020). Glacial-fed rivers such as the Brunná River will have an initial increase in discharge as summer melt increases with rising air temperatures. At a certain point, the discharge will begin to drop, as the glaciers become depleted (Jónsdóttir 2008; Milner et al. 2009). In contrast, spring-fed systems, supplied largely by rainfall runoff, may become/remain an important source of water to hydrological systems in Iceland. The wetlands at Hvoll were shown to maintain a positive water balance, with inputs coming from precipitation, subsurface flows, and spring-fed streams in 2017 (Scheffel & Young 2021). During 24 h of a jökulhlaup, and heavy precipitation events in 2015 and 2016, sizeable spikes in water level at the Brunná River were observed (Scheffel & Young 2021). However, an increase in water levels in the wetland was not observed until the following day, and it was inferred that subsurface flows from the sandur to the wetland likely played a role in this delayed response. In the present study, the time of water level peaks at S1 occurred up to 2 days following days with more than 1 mm of precipitation. Hence, it is suggested that it could take up to 2 days for rainwater to travel through groundwater flow paths and reach the spring or for the increased pressure in the aquifer to propagate to the spring. Subsurface flow from the sandur to the wetland is supported by hydraulic gradient data, where water levels were higher in the sandur than in the wetland (Scheffel & Young 2021). During our study in 2019, increases in sandur spring water levels did not correspond with increases in water temperature at any of the medium-sized Sandur springs, except for on 23 September 2019 when all Sandur springs except S1 had a spike in water temperature following a period of heavy precipitation and possibly elevated flood waters from the Brunná River. Inflows from the warmer Brunná River occurred closer to the surface, and with limited mixing, this prevented temperature changes at S1 from being detected at −0.5 m water depth. If air temperatures increase in the future, shifts in glacial melt may influence the frequency of flood waters moving down the Brunná River and its impact on near-bank springs.

In this study, water temperature, discharge, and chemistry along with stable isotope data were used to investigate the hydrological behaviour of springs in a sandur-lava field-wetland complex. The springs showed stable water levels, temperatures (4–5 °C), and chemistry, as is expected of groundwater-fed springs. The stable water isotope signatures of the springs were indicative of groundwater as well. Heavier rains (>10 mm) resulted in increases in water level and/or temperatures in some cases. Following rainfall, occasional spikes in water level measured at the sandur spring corresponded with peaks in discharge at glacial-fed river Djúpá. This, along with the increasing diurnal amplitude as summer progressed suggests that the sandur springs may also receive shallow inflows and floodwaters from the Brunná River. To obtain a full picture of the mode of transport of water to the springs, the permeability and structure of the lava fields in Southeast Iceland should be investigated. Long-term studies of groundwater springs in Southeast Iceland will be required to better understand the seasonal and annual variability in spring hydrology, and their vulnerability to climate change. Finally, this study adds to our understanding of pipeflow hydrology in cold regions.

We would like to thank York University for research funds to Dr Kathy L. Young and Aiesha Aggarwal. The Government of Canada provided logistical funds to Aiesha Aggarwal through the Northern Scientific Training Program. We are especially grateful to Mr Hannes Jónsson and his wife Guðný M. Ström Óskarsdóttir, who allowed us to undertake this study on their property in Southeast Iceland. We would like to thank Marc Isabelle and D'Moi Keen for their assistance in the field. Finally, we appreciate the useful comments provided by two anonymous reviewers.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.