Debris flows in the Eastern Italian Alps : seasonality and atmospheric circulation patterns

Abstract. The work examines the seasonality and large-scale atmospheric circulation patterns associated with debris-flow occurrence in the Trentino–Alto Adige region (eastern Italian Alps). Analysis is based on classification algorithms applied to a uniquely dense archive of debris flows and hourly rain gauge precipitation series covering the period 2000–2009. Results highlight the seasonal and synoptic forcing patterns linked to debris flows in the study area. Summer and fall season account for 92% of the debris flows in the record, while atmospheric circulation characterized by zonal west, mixed and meridional south and southeast (SE–S) patterns account for 80%. Both seasonal and circulation patterns exhibit geographical preference. In the case of seasonality, there is a strong north–south separation of summer–fall dominance, while spatial distribution of dominant circulation patterns exhibits clustering, with both zonal west and mixed patterns prevailing in the northwest and central east part of the region, while the southern part relates to meridional south and southeast pattern. Seasonal and synoptic pattern dependence is pronounced also on the debris-flow-triggering rainfall properties. Examination of rainfall intensity–duration thresholds derived for different data classes (according to season and synoptic pattern) revealed a distinct variability in estimated thresholds. These findings imply a certain control on debris-flow events and can therefore be used to improve existing alert systems.


Introduction
Debris flows are recognized as one of the most devastating natural disasters for mountainous regions at global scale (Dowling and Santi, 2014).The sudden occurrence combined with the high destructive power of debris flows pose a significant threat to human life and infrastructures (Petley, 2012).Therefore, developing early warning procedures for the mitigation of debris flows risk is of great economical and societal importance.Introduction

Conclusions References
Tables Figures

Back Close
Full Effective debris flows warning procedures require accurate knowledge on the relevant triggering mechanisms and their corresponding characteristics (Borga et al., 2014).Indisputably, rainfall is the predominant factor controlling debris flow triggering.Hence most of the work so far on the prediction of debris flow occurrence is focused on the identification of relevant rainfall conditions (Guzzetti et al., 2008 and references therein;Nikolopoulos et al., 2014).However, the vast majority of the literature on identification of debris flow triggering rainfall conditions deals primarily with the estimation of rainfall properties (e.g.rainfall duration, intensity or accumulation) leading to debris flows.Less attention has been paid to the seasonal and meteorological characteristics of the triggering rainfall events.Knowledge on the seasonality and meteorological patterns characterizing debris flow triggering rainfall events is important for two main reasons.First, classification of debris flow events according to these factors may be used for the development of a typology for debris flow rainfall events.This typology can subsequently be used for refining the rainfall triggering conditions according to different debris flow types and thus improve prediction.This hypothesis was examined by Govi et al. (1985) who analyzed the seasonality effect on the triggering of shallow landslides (soil slip -mud flow and soil slip -debris flow) in a sector of NW Italy.It is also justified from the recent works of Peruccacci et al. (2012) and Vennari et al. (2014) who demonstrated differences in debris flow triggering rainfall properties between warm and cold season for central and southern Italy, respectively.Furthermore, Toreti et al. (2013) showed that debris flow occurrence in southern Swiss Alps, exhibit a distinct pattern in large-scale atmospheric circulation and suggested that this information can be used to improve existing warning systems.On this line, Turkington et al. (2014), in a study centered on the southern French Alps, showed that empirical thresholds can be directly identified based on regional atmospheric patterns.The main objective of this work is to investigate the existence of distinct patterns in seasonality and large-scale atmospheric circulation associated with rainfall events that trigger debris flows.Furthermore, examination of debris flow rainfall properties with respect to seasonality and weather circulation patterns is investigated to evaluate the potential benefit of using such discriminant factors for the identification of debris flow triggering rainfall conditions.The work is focused over the region Trentino-Alto Adige in the eastern Italian Alps and the analysis is based on a 10 year record of debris flows and raingauge rainfall observations.Section 2 provides a description of the study area and the different data sources used in the analysis.Results from the analysis are presented and discussed in Sect.3. The main conclusions derived from this work are summarized in Sect. 4.

The Trentino-Alto Adige region
The Trentino-Alto Adige study region is located in the Eastern Italian Alps (Fig. 1); it covers 13 607 km 2 and is characterized by complex topography with elevation ranging from 65 to almost 4000 m a.s.l.(mean elevation is approximately 1600 m a.s.l.).Mean annual precipitation amounts exhibit strong spatial variability in the region, with annual sums of slightly above 500 mm in the north-western portion of the region (the Venosta Valley, located in the rain-shaded Inner Alps) and exceeding 1500 mm in the southeastern edge of the area.The features of the precipitation mean annual climatology exhibit characteristic seasonal variations (Norbiato et al., 2009;Parajka et al., 2010).The precipitation regime in the northern part of the study area is characterized as continental, with a unimodal cycle and the highest precipitation amount during the main convective period (May-September).The southern portion of the study area exhibits Introduction

Conclusions References
Tables Figures

Back Close
Full a bimodal regime, with maxima in spring-early summer and in autumn, which generally receives the most abundant precipitation.Typically the precipitation during cold months (October to April) is in the form of snow and widespread precipitation while mesoscale convective systems and localized thunderstorms dominate the precipitation regime during warm months (May to September) (Norbiato et al., 2009;Mei et al., 2014).
There are two important factors that make the area attractive for this study.First, the region is characterized by significant societal risk due to both the high frequency and the impact (in terms of casualties) of landslides in the area (Salvati et al., 2010).Second, the availability of a long-term record of precipitation and debris flows (Nikolopoulos et al., 2014), as described in detail in the following section.

Debris flow and rainfall database
Ten years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) of available precipitation observations and debris flow (DF, hereinafter) records are analyzed in this work.Hourly accumulation values of precipitation are obtained from a network of 205 rain gauges that cover the study region (Fig. 1).As can be seen from Fig. 1, rain gauge stations are spread quite uniformly over the region providing a spatial density approximately equal to 1/70 km −2 .Compilation of the DF events occurred in the region during the period 2000-2009 was based on two independent databases, covering the two administrative units: Trentino (377 events) and Alto Adige (444 events).The selected 821 were identified based on a larger number of events in order to get the same level of spatial and temporal occurrence accuracy.
Available information includes the location of the individual DF initiation point (shown in Fig. 1) with a 500 m spatial accuracy and the date of occurrence with a daily accuracy.

Weather circulation patterns
The Hess and Brezowsky Grosswetterlagen (GWL) classification system is used for the classification of large scale atmospheric flow and weather circulation patterns.The GWL classification system is based on the mean air pressure distribution (sea level Introduction

Conclusions References
Tables Figures

Back Close
Full and 500 hPa level) over the North Atlantic Ocean and Europe.The classification initially identifies three groups of circulation types (zonal, mixed and meridional), which are divided into 5 major types, which in turn are divided into 29 subtypes (Grosswetterlagen, GWL) (Gestengabe and Werner, 2005;James, 2007).This classification system is frequently used to characterize the atmospheric flow and weather patterns over the eastern North Atlantic and Europe (Gestengabe and Werner, 2005;Kyselý and Huth, 2006;Planchon et al., 2009).Following Gestengabe and Werner (2005) and Parajka et al. ( 2010), the original GWL classes were further grouped into six categories (Table 1) that were used for the description of the general weather regime during DF events.For more detailed information on the individual GWL weather types, the interested reader is referred to Gestengabe and Werner (2005) and James ( 2007).
The monthly frequency of the GWL groups (Table 1) is presented in Fig. 2, for the whole study period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009).As it is shown, the occurrence of Mixed weather pattern dominates the other patterns consistently over all months, with monthly occurrence being greater than 30 % in all cases.On the other hand, the Mixed CE type is associated with the minimum occurrence (less than 5 % for all months except ∼ 10 % in August), while the frequency for the rest of the weather patterns is within the same range and generally between 10 to 20 %.One noticeable feature is that during the winter period, apart from the Mixed type, the Zonal West pattern occurrence is significant and distinctly higher than the rest.

Spatial distribution of debris flow triggering rainfall properties
Characteristic properties, namely duration and accumulation, for each DF triggering rainfall event are estimated from the closest available gauge.Calculation of eventbased properties is based on the identification of individual events in the rainfall records by separating subsequent rainfall events according to an inter-event period of 24 Introduction

Conclusions References
Tables Figures

Back Close
Full consecutive hours without rainfall (Nikolopoulos et al., 2014).This procedure results in the identification of a total of 128 individual rainfall events, which have triggered the 821 DF analyzed.To examine the spatial distribution of DF rainfall properties, results are grouped into different classes of duration and accumulation respectively and are mapped over the study region (Fig. 3).Geographical distribution of the DF rainfall properties, as shown in Fig. 3, allow us to investigate and potentially identify areas over the study region with distinct pattern in the characteristics of the triggering rainfall.In terms of duration, DF occurrence is dominated by long duration (> 24 h) events which account for ∼ 82 % of the DF occurrences, while in terms of rainfall accumulation DF events are distributed more uniformly with 25-30 % of cases corresponding to each of the three higher classes (> 100, 50-100 and 20-50 mm) and ∼ 14 % for the lowest class (< 20 mm).Visual interpretation of the spatial pattern of rainfall properties (Fig. 3) shows that classes of rainfall and duration are rather mixed without revealing clustering of specific rainfall properties.Perhaps the only noticeable feature from Fig. 3 is that most of the DF events located in the northwestern part of the study region are generally associated with relatively low (< 50 mm) rainfall accumulation.This is an indication of the link with the local climatic characteristics, with this portion of the study area being included in the dry internal alpine region, characterized by relatively low mean annual precipitation amounts.summer while at the southern part, DFs occur predominantly during the fall season.

Seasonality of DF events
In addition, examining the seasonal distribution of DF rainfall events (Table 2) shows that summer season is associated with 59 % of the rainfall events while spring and fall seasons correspond almost equally to 16 % each and winter to ∼ 9 %.Relating the seasonal distribution of the number of rainfall events with that of DF occurrences indicates clearly that rainfall events during fall season are associated with the highest (on average) DF numbers per event.Specifically, the ratio DF/event is 1.25 (winter), 2.2 (spring), 5.4 (summer) and 17.75 (fall).
To further investigate the relationship between seasonality and number of DFs triggered per rainfall event, we classified DF rainfall events according to the total number of DFs triggered and analyzed the seasonal distribution of each class.Five classes were considered that included rainfall events with total DF triggered equal to: 1 (class 1, 63 events), 1 < DF ≤ 5 (class 2, 35 events), 5 < DF ≤ 10 (class 3, 13 event), 10 < DF ≤ 20 (class 4, 11 events) and DF > 20 (class 5, 6 events).Examination of the results (Fig. 4b) shows that summer is clearly the dominant season in all classes.This is not surprising given that the greatest number of rainfall events is also associated with summer season.For the first class that involved events with only 1 DF occurrence, 60 % of occurrences is almost uniformly distributed among winter, spring and fall and the rest ∼ 40 % corresponds to summer.There are no winter events for classes higher than class 2 suggesting that all winter events are associated with low number of DF occurrence.On the contrary, spring events are apparent for class 4, suggesting that spring events can be associated with the triggering of several DFs.Interestingly, the highest class (DF > 20) is equally distributed between summer and fall season.Occurrence of a large number of DF-triggering rainfall events during summer is hypothesized as a result of mesoscale convective systems that although they have a relatively limited spatial extent, they are usually associated with high intensities.On the other hand, rainfall events during fall season are commonly widespread systems covering a large spatial extent and are associated with moderate intensities but long durations and rather wet antecedent soil moisture conditions.Introduction

Conclusions References
Tables Figures

Back Close
Full A further step in the seasonal analysis of DF is related to the seasonality of DF rainfall properties.Figure 5 reports the distribution (as boxplots) of DF rainfall accumulation and duration for the summer and fall seasons.The sample size for the DF cases of winter and spring (see Table 2) is considered rather limited (number of DF < 50), to be able to derive statistical properties of the underlying distribution with an adequate degree of robustness and therefore results for these cases were omitted.Results show that rainfall characteristics of DF triggering events are significantly different between summer and fall season.Events during the summer are associated with lower rainfall accumulation and shorter duration while during the fall events are characterized by higher accumulations and longer durations.Specifically, average rainfall accumulation (duration) for summer events is equal to 44 mm (48 h), while for fall events is equal to 128 mm (98 h).These results are in agreement with the explanation provided above regarding events of convective nature dominating summer rainfall while frontal systems associated with long-lived widespread systems occurring in the fall season.

Debris flows distribution and weather circulation patterns
Following the same methodological framework of Sect.3.2, we examine in this section the relationship of DF occurrence and corresponding triggering rainfall properties, with the weather circulation patterns.As it may be observed in Fig. 6, weather circulation patterns corresponding to Mixed, Zonal West and Meridional Southeast and South (SE, E) groups dominate DF occurrence in the region.In addition, visual inspection of permittingstatisticallysignificantinterpretation of results, nevertheless it is interesting to note that rainfall events that triggered the highest number of DF in the region and occurred during summer and fall season (see Fig. 4b) are predominantly associated with weather circulation patterns (50 % SE, E and ∼ 17 % N) that are much less frequent than Mixed (which corresponds to 33 %) and Zonal West (0 % occurrence) according to the climatology presented in Fig. 2. Examination of rainfall characteristics as a function of the weather circulation patterns (Fig. 7) shows the variability of both rainfall accumulation and duration with weather type.On average, accumulation and duration increases consistently moving from Zonal West (38 mm, 42 h) and Mixed (37 mm, 42 h), to N (75 mm, 59 h), to SE,E (128 mm, 99 h).Note that due to sample size limitations, results in Fig. 7 are presented for the four most dominant weather type groups.

DF rainfall thresholds
Results obtained from previous sections revealed strong dependencies between DF triggering rainfall properties with (a) season and (b) weather circulation patterns.This allow us to hypothesize that there is merit in classifying DF events according to these factors and identifying the DF rainfall thresholds separately for each case.This could potentially result in the development of a set of thresholds that can be used according to different conditions (e.g.depending on the weather type) thus providing a more accurate prediction in comparison to using a universal threshold.
Rainfall thresholds, used for predicting possible debris flow occurrence, identify critical rainfall condition by linking rainfall intensity (or accumulation) and duration (for a review see Guzzetti et al., 2007Guzzetti et al., , 2008)).In this study we considered a widely used model for the definition of the rainfall thresholds, which is the intensity-duration (ID) threshold commonly adopting the power-law form (1) Introduction

Conclusions References
Tables Figures

Back Close
Full where I (mm h −1 ) is the mean intensity and D (h) is the duration of the DF triggering rainfall.The multiplier (α) and exponent (β) parameters are constants and are estimated by fitting the power-law model to the empirical data.For the estimation of parameters α and β we applied the frequentist approach proposed by Brunetti et al. (2010).Note that the frequentist method allows identifying ID thresholds at different levels of exceedance probabilities (see for example Brunetti et al., 2010 andPeruccacci et al., 2012).In this work we adopted a 5 % exceedance level, which means that the probability of a debris flow triggering rainfall event (I, D pair) to be under the estimated ID threshold is less than 5 %.
To evaluate the significance of seasonal and weather type dependence of ID thresholds, we estimated the ID parameters from each DF sample corresponding to different season and weather circulation type.Note that in both cases, only the dominant seasons and weather types were examined.Also, I, D pairs used for the estimation of thresholds were filtered to remove values associated with I < 1 mm h −1 and/or D < 2 h, which are considered to fall within a range associated with high estimation uncertainty.Marra et al. (2014) has shown that using I, D pairs below these thresholds may lead to unrealistic values of ID parameters.Filtering and splitting the original DF sample (821 data points) according to season and weather type group introduces uncertainty in the ID estimation procedure due to the size of individual samples considered.To account for this effect, we applied the methodology developed by Peruccacci et al. (2012).
Specifically, for each case, ID estimation was repeated 1000 times on samples derived from the original population following the bootstrapping resampling technique (see Peruccacci et al., 2012 for more details).Results on the estimation of parameters α and β and the uncertainty associated to sample size are summarized in Table 3 and are visualized in Fig. 8 (for different seasons) and Fig. 9 (for different weather types).Several conclusions can be drawn from these results.The estimated ID parameters for summer and fall differ mainly on parameter α, with ID for fall being associated with higher value.The sampling uncertainty appears significant mostly for the α parameter Figures  parison with the other cases but also with reference to other thresholds reported in literature (see Guzzetti et al., 2007).Despite the associated uncertainties due to sampling size, overall the results presented (Table 3, Figs.8 and 9) show that classification of DF according to season or weather type can lead to considerably different thresholds.This has important implications for the operational use of the thresholds.

Conclusions
In this work, the seasonal and atmospheric circulation patterns of debris flows in the Eastern Italian Alps was examined.The study was focused on the Trentino-Alto Adige region and analysis was carried out over a ten years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) period, for which a unique catalog of debris flow occurrences and hourly rain gauge precipitation was available.The principal conclusions derived are summarized below.
-The vast majority (92 %) of debris flows occur during summer and fall season.Furthermore, the two dominant seasons exhibit a clear geographical preference, with summer and fall season dominating the northern and southern part respectively.
-Rainfall properties (accumulation and duration) derived for each individual debris flow location and from the closest available raingauge, exhibit a seasonal pattern as well.On average, summer events are associated with lower rainfall accumulation and shorter duration than fall events but with higher intensity.
- both duration and accumulation increase on average moving from Zonal West, to Mixed, to N and finally to SE, E.
Variability of rainfall properties with season and weather type was further examined in the context of ID thresholds used for the prediction of debris flow occurrence.Results revealed that there are indeed apparent differences in the ID thresholds estimated for each case (season or weather type).Although sampling size limitations introduce a considerable amount of uncertainty in the estimated thresholds, this alone cannot fully explain the observed differences.Therefore, results indicate that there is potentially merit in the application of a classification scheme (according to season and/or circulation type) on debris flow event for improving accuracy of threshold-based prediction systems.
An important note that should be kept in mind when considering the results reported in this work regarding the derived rainfall properties and ID thresholds is that those depend on (a) the identification of individual rainfall events and (b) on the accuracy of rainfall estimates obtained from closest available gauges.Regarding the first point, we acknowledge that the use of an inter-event period of 24 h may not always represent well the properties of the debris-flow triggering storms, especially in the case of short rainstorms spaced by few hours, and can therefore impact at some degree the derived values.We adopted the 24 h threshold as a realistic value of a minimum period between consecutive events according to our experience and also following previous work (Nikolopoulos et al., 2014) to allow our results to be comparable with similar studies.On the second issue, which relates to the DF rainfall estimation from closest gauges, recent work on this topic (Nikolopoulos et al., 2014;Marra et al., 2014) has demonstrated clearly that gauge-based estimates of DF rainfall are largely underestimated and this results also in underestimation of ID thresholds.However, the current work focuses mostly on highlighting the relative differences of rainfall properties and subsequently of ID thresholds among different seasons/weather types and thus the absolute accuracy is not of focus.Therefore, we feel that the patterns portrayed regarding the Introduction

Conclusions References
Tables Figures

Back Close
Full  Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Second, linking debris flow occurrence with seasonal and meteorological characteristics may provide indications on the potential impact of climate change on debris flow activity (Stoffel et al., 2014).As an example, Schneuwly-Bollschweiler and Stoffel (2012) concluded that the observed seasonal shift in debris flow activity in the Zermatt Figures Back Close Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | valley (Switzerland) is attributed to changes in precipitation and temperature regime in Swiss Alps over the last century.
Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Relationship between DF events and season is examined to analyze the importance of seasonality in (a) the occurrence of DF events and (b) the corresponding DF rainfall properties.Geographical distribution of the season of occurrence of the DF events is shown in Fig. 4a, which reveals two important features.First, summer and fall season dominate DF occurrence in the region of study.Specifically, 49 % of the DF occurred during summer and 43 % during fall season (see Table 2 for more details).Second, there is a very distinct geographical separation between the two dominant seasons of DF occurrence.DFs in the northern part of the area are mainly occurring during the Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 6a reveals
Fig. 6a reveals the clustering (i.e.geographical preference) of specific types.For example, the southern part of the study regions is associated to SE, E group while northwest and central east is associated to Mixed and Zonal West groups.Results regarding the connection between weather type and the different classes of rainfall events (see Sect. 3.2) are reported in Fig. 6b.Again, results show clearly that Zonal West, Mixed and SE, E circulation patterns are the most dominant ones, with Meridional North (N) and Northeast, East (NE, E) having an apparent but significantly less percentage of occurrence.Although the number of events included in class 5 is only six, thus not 7205 Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Weather circulation groups of Mixed, Zonal West and Meridional Southeast and South patterns dominate debris flow occurrences in the region.Debris flow triggering rainfall properties vary considerably with weather type and specifically, Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Figure 1 .Figure 4 .Figure 5 .Figure 6 .
Figure 1.Map of the Trentino-Alto Adige region.Shades of color show terrain elevation.Black triangles and grey circles show respectively the location of rain gauges and debris flows involved in current study.
and is higher for fall.However, it can only partially explain the seasonal difference in parameter α.Comparison of ID thresholds for different weather types shows that DF cases associated with Meridional North pattern, are characterized by significantly higher (than in all other weather types) values of both ID parameters.Results for Zonal West and Mixed patterns are associated with very low values for parameter β in com-

Table 3 .
Estimated ID parameters (α, β)from original sample and their corresponding mean (µ) and SD (σ) derived from the resampling exercise.