Groundwater fluctuations during a debris flow event in Western Norway – triggered by rain and snowmelt

Pore pressure is crucial in triggering debris slides and flows. Here we present measurements of ground water pore 10 pressure and temperature recorded by a piezometer 1.6 m below the surface on a slope susceptible to debris flows in Western Norway. One of the largest oscillations in data collected over four years coincided with a debris flow event on the slope that occurred during storm Hilde on 15–16 November 2013. More than 100 landslides were registered during the storm. Rainfall totalled about 80–100 mm in 24 hours, locally up to 129 mm, and an additional trigger factor for the slides was a rapid rise in air temperature that caused snowmelt. On 15 November, the groundwater level in the hillslope rose by 10 cm per hour and 15 reached 44 cm below the surface. At the same time, air temperature rose from 0 °C to over 8 °C, and the groundwater temperature dropped by 1.5 °C. The debris flow probably occurred late in the evening of 15 November, when the groundwater level reached its peak. Measurements of the groundwater in the hillslope in the period 2010–2013 show that the event in 2013 was not exceptional. Storm Dagmar on 25–26 December 2011 caused a similar rise in groundwater level, but did not trigger any failures. The data suggest that during heavy rainstorms the slope is in a critical state for a slide to be triggered for a short 20 time – about 4–5 hours.


Introduction
It is well known that groundwater pore pressure is crucial in triggering shallow debris slides and flows (e.g. Iverson, 1997), but how exactly does pore pressure vary in a hillslope during a rainstorm? How much and how rapidly does the 25 groundwater level rise before a slide is triggered? And, for how long during a rainstorm is the slope landslide-prone? We try to answer these questions using groundwater level data logged by an automated piezometer installed in a borehole on a hillslope in western Norway, where a debris flow occurred during heavy rainfall and snowmelt in November 2013.
Such instrumental data are rare because it is difficult to predict which slope will fail. Several studies have measured 30 the hydrologic response to rainstorms in hillsides prone to shallow slides (e.g. Collins et al., 2012;Fannin and Jaakkola, 1999;Johnson and Sitar, 1990;Sidle, 1986), or measured the response to artificial sprinkling of water over a slope to force a slide to occur (e.g. Harp et al., 1990;Reid et al., 1997). Only a few studies provide instrumental data directly from a natural debris flow event (Montgomery et al., 2009;Reid et al., 1988). Here, we present continuous groundwater level measurements from a hillside susceptible to shallow debris slides and flows for the period 2010-2013. The data include the day when a debris flow 35 occurred in this particular slope during the storm named Hilde on 15-16 November 2013.
During this storm, the intense rainfall and snowmelt triggered more than 100 landslides in Western Norway. The maximum rainfall intensity was 80-100 mm in 24 hours, locally up to 129 mm. Most of the slides were debris slides and flows (114), but rockfalls (28) and snow avalanches (7) also occurred ( Fig. 1). Many roads were damaged, a bus got stuck, people 40 were evacuated, and houses and cars were damaged by slide material, but fortunately no one was killed (Fig. 2). The number of slides makes this one of the largest landslide events in Norway during the last two decades (Krøgli et al., 2018).
The storm was a classic extreme precipitation event on the west coast of Norway. A warm mature front, partly occluded, passed southern Norway on 15 November. The pressure configuration, with a very low-pressure system northeast of Jan Mayen and a high-pressure center southwest of the UK, generated a strong north-south pressure gradient and induced a 45 southwesterly flow of moist warm air towards western Norway (Fig. 3). It rained heavily as the moist air was lifted by the warm front. The precipitation further intensified as the flow ascended over the mountains of western Norway. This weather situation leads to the formation of narrow plumes of intense high-level moisture, often referred to as atmospheric rivers. These cause most of the extreme precipitation events on the west coast of Norway (Azad and Sorteberg, 2017).

The hillside and instrumentation
The hillside we instrumented is typical of Western Norway, and is prone to debris flows and shallow slides. The hillside is near Sogndal, east of the lake at Anestølen (Fig. 4). Recent slide events on this slope occurred on 18 May 2004, 22 September 2007(Tyssebotn and Velle, 2010 and 15 November 2013 (Olsen et al., 2015) (Fig. 4B). The lower part of the slope is covered by slide deposits and till and the average slope angle is 25 -26 . The sediment-covered slope tapers off upwards 85 into steeper and exposed bedrock and cliffs (Fig. 5A). From the outcrops of the slide scar of the 2007 slide event (Fig. 4B), we found that the thickness of the deposits on the slope varies, probably between 2 and 5 m.
We drilled through boulders and relatively firm deposits down to 2.4 m below the surface using a hammer drill powered by compressed air (Fig. 5B). We pushed and hammered threaded pipes, 1 m long and 32 mm in diameter, down into 90 the borehole. The pipes were subsequently screwed together. The lower pipe (piezometer) has twelve slits, 10 cm long and 2 mm in width, to allow water seepage. The data logger, a mini-diver (DI 501) manufactured by Van Essen instruments, was attached to a wire and inserted into the pipe. The lower end of the mini-diver is at 1.64 m below the ground surface (Fig. 5B).
The drilling and instrumentation are described in Tyssebotn and Velle (2010).

95
The weather station at Anestølen is operated by the Hydrological Department at the Norwegian Water Resources and Energy Directorate and measures air temperature, precipitation, wind, atmospheric pressure, snow depth and groundwater level ( Fig. 4B). Precipitation is measured every 10 minutes, but for periods the bucket that collects the rain was full, and the data are not reliable. However, we found the measurements of precipitation to be reliable around the time of storm Hilde (Olsen et al., 2015). The other precipitation data we present are 24-hour totals measured manually at Selseng (Fig. 4A), 3.5 km south of 100 Anestølen at about the same altitude (station 55730, data downloaded from http://eklima.met.no). Atmospheric pressure and air temperature (Fig. S1) at the weather station are measured every 15 minutes. All data from the weather station are available from http://sildre.nve.no (station no. 77.24.0).
The mini-diver we installed in the hillslope piezometer measure total pressure and temperature. Pressure is measured 105 in cm H2O to an accuracy of ± 0.5 cm H2O with a resolution of 0.2 cm H2O. Temperature is measured in degrees Celsius to an accuracy of ± 0.1 C, with a resolution of 0.01 C. The atmospheric pressure was subtracted from the measured pressure to obtain the true water pressure above the sensor in the mini-diver. We used the atmospheric pressure measured at the weather station from 2 October 2012 onwards (data downloaded from http://sildre.nve.no/Sildre/Station/77.24.0 ). Measurements from before 2 October 2012 were corrected using atmospheric pressure measured at the airport in Sogndal (station 55700, 110 downloaded from http://eklima.met.no). The difference in altitude between the weather station (442 meters above sea level) and the sensor in the borehole at the slope (484 m) is 42 m. This corresponds to a pressure difference of about 5 hPa, which we subtracted from the atmospheric measurements. The difference in altitude between the airport (497 m) and the borehole is https://doi.org/10.5194/hess-2020-264 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. only 13 m. For these measurements we ignored the altitude difference. The mini-diver was set to record measurements every 4 hours, and we obtained almost continuous measurements for the years 2010, 2011, 2012 and 2013. 115

The 201debris flow
The slide happened in the evening of 15 November or during the following night. It was first observed by a person driving an all-terrain vehicle to the mountain farm at Anestølen in the morning of 16 November; he found that the road was covered by wet debris, and was unable to proceed (Fig. S2). The driver thinks the flow happened sometime during the night, 120 but cannot exclude the possibility that it was the evening before, on 15 November (Olsen et al., 2015). The nearest slides to Anestølen (see three triangles in Fig. 1 immediately west of this site), happened on the evening of 15 November, and during the night and early morning of 16 November at 02:00 and 07:30.
The slide is a typical debris flow with levees and lobes. The flow had a point source at an altitude of 720 m, widened 125 downslope, and continued in a channel with levees on both sides (Figs. 2B, S3). The channel was eroded about 2 to 3 m into the deposits, and exposed the bedrock in certain places. A spillover lobe is evident on the southern side of the slide. Towards the lake, the flow split into three different paths ( Fig. 2B) (Olsen et al., 2015), and ended in the lake (441 m altitude). Here, a fan and a delta of fine-grained material, mostly sand and organic material, were deposited (Fig. S4). We inspected the slide on 17 November (Fig. S5) and found snow inside the very wet slide debris. The entire lake was colored brown by suspended 130 sediment from the slide that had entered the lake.

Results 150
The groundwater level on the slope varies according to season and precipitation. In winter the groundwater table is low, and in 2013 the water level was below the sensor in February and March, apparently because the ground was frozen (e.g. Ireson et al., 2013) (Fig. 6). During spring, from April to June, the groundwater is recharged by snowmelt, rain and thawing of the frozen ground (Du et al., 2019), which cause high peaks and rapid changes. In summer and fall, the groundwater level is controlled by precipitation events, and normally oscillates between 150 cm and 75 cm below the ground (Figs. 6, S6-S8). 155 A few times in the summer, the groundwater level is lower than the sensor. Most of the oscillations show a rapid, almost instantaneous rise, and a longer, slower, declinethe oscillations are clearly asymmetric (Fig. 7).
The peaks and troughs of the oscillations occur simultaneously upslope and in the valley bottom (piezometer at the weather station), but the peaks upslope are much sharper and vary in amplitude (Fig. 7). The valley bottom is saturated almost 160 every time there is a precipitation event. The upslope peaks reach different heights, depending on the amount of infiltration (Fig. 7). Another difference is in the shape of the peaks. The upslope maxima are very sharp, lasting less than 4 hours, while the maxima downslope are broad. The peaks dissipate more rapidly upslope than downslope, as has also been described elsewhere e.g. Alaska (Sidle, 1984(Sidle, , 1986. The groundwater level starts to decline within hours after reaching peak levels, while the valley bottom peaks last for half a day or more (Fig. 8). 165 Groundwater temperatures follow the season, but show small, irregular oscillations that reflect infiltration episodes (Figs. 6, S6-S8). The big picture is a steady, slow temperature decline through the winter to about 1.5  -2 C, an abrupt drop in the spring when temperature reaches its minimum (0.5  -1.2 C) with wiggles in the temperature curve reflecting meltwater reaching the sensor, followed by a large rise from May to August, up to 9.5 -10 C. In September the groundwater 170 temperatures start a steady decline towards the winter minimum. However, the temperature curves show a number of small anomalies that coincide with peaks in groundwater level. These anomalies are caused by infiltration of surface water that is either colder or warmer than the groundwater. Infiltration episodes between October and May cause the temperature to drop; the largest observed were drops of 1.5 C and 1.8 C related to snowmelt (Figs. 6, S7). Between May and October, surface water is warmer than the groundwater; infiltration episodes cause temperature rises, the largest recorded being about 1 C and 175 associated with summer rain events in July (Fig. S7) and August (Fig. 6).
One of the largest oscillations in both groundwater level and temperature occurred during storm Hilde on 15-16 November 2013, and triggered the slide event. In only 8 hours, the groundwater level rose by 46 cm and the temperature dropped by 1.5 C (Fig. 8). The precipitation at Anestølen began at 08:20 on 15 November and fell as snow until about 14:30, 180 in all 22 mm (3.3 mm/h), because the air temperature was between 0 and 0.6 C (blue line in Fig. 8). From about 15:00 the air temperature rose quickly and reached 9.5 C at midnight (Fig. 8), and precipitation was 4.2 mm/h until 22:30 p.m. The response in groundwater level was a rapid rise, 9.5 cm/h (Fig. 9), and a drop in temperature, from 4.9 C to 3.4 C between 15:00 and 19:00.

185
Prior to the storm event, there was a thin cover of snow on the ground. The snow pillow at the weather station measured 20 mm of snow water equivalent before the storm began (Fig. S9). Photos of the slide the morning after the storm show patches of snow on the ground (Figs. S2-S4). We also observed lumps of snow inside the slide debris. This suggests that there was about 10-20 cm of snow on the ground prior to the arrival of warm, moist air with the storm.

Discussion
One weakness of this study is that the measurements on the slope are from one piezometer only. The hydrological response to the rainfall depends on antecedent moisture conditions and the porosity, hydraulic conductivity and thickness of deposits on the slope (Johnson and Sitar, 1990;Montgomery et al., 2009). The slope is covered by till and colluvium, deposits that vary widely in composition and grain size. Thus, observations from one borehole may not be representative of other areas 225 on the same slope (Fannin and Jaakkola, 1999). The distance from the borehole to the 2013 slide is about 400 m (Fig. 4B).
Another weakness is that the piezometer in the hillslope was set to a recording interval of 4 hours, which is too long to capture details of the most rapid changes (Fannin and Jaakkola, 1999). The strength of the study is the continuous measurements of both pore pressure and groundwater temperature in conjunction with the nearby weather station over a period of 4 years that also covers a slide event. 230

The storm Hilde landslide event, 15-16 November 2013
The groundwater level in the piezometer on the slope peaked between 19:00 and 23:00, and this is probably the time window when the debris flow was triggered. We measured the peak at 23:00 at 44 cm below surface, but because the piezometer only took measurements every 4 hours, and the rise of the groundwater declined between 19:00 and 23:00 without 235 any change in precipitation or air temperature, it is likely that the actual peak was reached earlier than 23:00 and was higher than 44 cm (see Fig. 6). Extrapolations of the groundwater level curve from 19:00 and 23:00 indicates that the peak might well have been at a depth of 30 cm and sometime around 20:30-21.30 (Fig. 8). From a study in Oregon, USA, Montgomery et al. (2009) concluded that most of their piezometers recorded that groundwater was still rising at the time of a debris flow failure, and this suggests that the debris flow at Anestølen occurred earlier than 23:00. 240 The substantial drop in groundwater temperature suggests that part of the infiltration was from snowmelt. There was already some snow on the ground on 15 November (Fig. S9), and the precipitation in the morning and until about 15:00 fell as snow because of the low air temperature (Fig. 8). During this time there was no response in the groundwater level or temperature. However, after 15:00, when the warm, moist air of the storm reached Anestølen, the groundwater rose by 9.5 245 cm/h and at the same time, the temperature of the groundwater dropped by 0.2 C/h. The rapid snowmelt added extra water to the slope and augmented the infiltration rate and groundwater rise, as seen from rainfall on snow on slopes in Alaska (Sidle, 1984) .
The groundwater level in the hillslope continued to rise as long as the infiltration rate was higher than the downslope 250 discharge of groundwater (Fig. 10, middle panel). The heavy rain ceased at 22:30, and after 23:00 the groundwater level fell by 3 cm/h (Fig. 10, lower panel). The groundwater level was still rising just after 19:00, so the peak itself must have been https://doi.org/10.5194/hess-2020-264 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
shorter than 4 hours. Such a short peak suggests that very little groundwater from higher up the slope or from bedrock fractures (Johnson and Sitar, 1990) could have contributed to the rise in groundwater level. The entire groundwater rise can be explained by vertical infiltration through the surface. The full amplitude of the groundwater peak was 54 cm during storm Hilde (Fig. 8), 255 and this is an order of magnitude larger than the total amount of rainfall and snow (53 mm). A similar relationship has been found in other studies (e.g. Sidle, 1984).

Other groundwater episodes
Melting of the snow cover on the slope leads to high groundwater levels. In 2013 we recorded the year's highest peaks 260 in April and May (Fig. 6), when the ground was still covered by substantial amounts of snow (Fig. S9). The high peaks correspond to lows in the groundwater temperature curve, reflecting the infiltration of cold meltwater. The small wiggles in the temperature curve end on 26 Maythe same day the snow pillow at the weather station was free of snow (Fig. S9). The highest peaks were on 15 April and 16 May, 33 cm and 28 cm below the ground respectively, and coincided with rain (Fig. 6) and high air temperatures (> 8C) (Fig. S1). This high pore pressure did not cause any sliding, perhaps because there was still 265 considerable snow cover, estimated at around 1 m (Fig. S9).
The fluctuations that occurred during storm Dagmar on 26-27 December 2011 were exceptional and similar to those during storm Hilde event. During storm Dagmar, groundwater rose to 28 cm below the surface, and the groundwater temperature dropped by 1.8C (Fig. S7). The groundwater rose by 13.5 cm/h between 19:00 and 23:00 on 26 December (Fig.  270 S11). A major cause was a considerable increase in air temperature that melted a lot of snow, in combination with some precipitation (37.5 mm; Fig. S7). In spite of this, there were no observations of any sliding on this particular slope.
Another episode when there was a very rapid rise in groundwater level occurred on 18-19 August in 2012. Here, the groundwater rose from below the sensor and up to 97 cm below the surface (Fig. S6) at a rate of more than 15.7 cm per hour 275 (Fig. S10). This was the highest rate measured during the four years of recordings (Figs. 10, S10-S12). Nevertheless, no slide was observed, probably because the groundwater level was too low when the rain started, and so the peak did not reach higher than 97 cm below the surface. This shows that it is not the rise itself, but the level that the groundwater reaches, that is important. conductive layers are thinning out or are blocked by less permeable deposits (Johnson and Sitar, 1990;Sidle, 1984). In order for the pore pressure to reach such a high level during a rainstorm, the groundwater level before the onset of the storm has to be sufficiently high, not much deeper than 1 m below the ground. In addition, our data indicate that the groundwater level will only be above 50 cm for a very short time, probably less than 5-6 hours, and that these conditions only persist as long as infiltration is higher than the downslope discharge of groundwater. This pattern is in contrast to groundwater measurements from the nearby weather station in the valley bottom, which show much broader peaks. 290

Future changes in precipitation and snowmelt?
Many of the landslides triggered during storm Hilde were caused by simultaneous heavy rainfall and strong snowmelt.
Such rain-on-snow events in Western Norway may become more frequent in the future. Storm Hilde was a typical atmospheric river storm, a type that has caused 57 of 60 extreme daily precipitation events in Western Norway since 1900. Of these, 62 % 295 occurred in the months November, December and Januarywhile none occurred in April, May, June and July (Azad and Sorteberg, 2017). The frequency of extreme precipitation events over Norway has increased by 25-35 % over the last 120 years (Sorteberg et al., 2018) and modelling indicates that it will increase further in the future (Hanssen-Bauer et al., 2017).
Since most of these events occur in the winter months, it is very likely that we will see more rain-on-snow events in Western Norway in the near future that could increase the risk of slide events and floods. 300 https://doi.org/10.5194/hess-2020-264 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. 1. The storm Hilde event in western Norway produced relatively large amounts of precipitation on the western slopes of the mountain range near the coast. Farther inland, precipitation was high but not extreme. The strong warm front 320 in the storm gave rise to a rapid temperature increase of 8-9 C, initiating snowmelt that supplemented the rainfall and lead to a rapid rise in groundwater. This situation triggered over hundred slides.
2. We measured groundwater levels in a hillslope that failed in the storm. The groundwater responded rapidly to the rainfall and increase in air temperature during the storm and rose by 9.5 cm/h, simultaneously with a pronounced 325 drop in groundwater temperature of 0.2C/h. The groundwater peak reached at least 44 cm below the surface and the temperature dropped by 1.6 C.
3. The slope remained saturated or near saturated for a short time. Critical conditions of the slope lasted only for 4-5 hours during the Hilde storm. 330 4. Two episodes stand out from the data collected over four years (2010, 2011, 2012 and 2013) in the hillslope piezometer;the groundwater peaks during storm Hilde in 2013 and storm Dagmar in 2011, both storms had ground water levels that reached higher than 50 cm below the surface. Normally the groundwater level fluctuated between 150 cm and 75 cm below the surface. The groundwater level in the hillslope piezometer was below 50 cm 335 for 99.87 % of the four years of data.
5. The infiltration of surface water is clearly recorded in the groundwater temperature curve as a short-lived anomaly.
The anomaly depends on the amount of infiltration and the temperature difference between the ground and the air.
The largest negative anomalies are related to snowmelt in the fall, 1.5-1.8 C. Summer rain caused at most a 1 C 340 positive temperature rise.

Author contribution
SB conceived the project, did the fieldwork, collected the data and wrote the first draft. AS described the weather 345 situation during the storm (Fig. 3), was responsible for the precipitation map in Fig. 2 and wrote the first draft of chapter 5.3 "Future changes in precipitation and snowmelt." https://doi.org/10.5194/hess-2020-264 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.

Declaration of competing interest 350
The authors declare that they have no conflict of interest.

Acknowledgments 355
Ola Olsen, Kevin Saurin and Sondre Wenaas mapped the 2013 debris flow event at Anestølen and collected information about the slide event. Geir Magne Tyssebotn and Julia Heggdal Velle drilled the borehole on the hillslope and installed the piezometer with the first mini-diver. Knut Møen (Norwegian Water Resources and Energy Directorate) answered questions about data from the weather station. Ottar Husum provided local information. Jan Helge Aalbu allowed us to use his photos in Fig. 2A and 2C. We inspected and plotted the data using the program Highcharts (https://www.highcharts.com/). Denise Christina 360 Rüther and Helge Henriksen provided critical comments that improved the paper, and Alison Coulthard provided assistance with language editing. https://doi.org/10.5194/hess-2020-264 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.