Mesoscale organization and structure of orographic precipitation producing flash floods in the Lago Maggiore region

Transcription

1 Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 141: , January 2015 A DOI: /qj.2351 Mesoscale organization and structure of orographic precipitation producing flash floods in the Lago Maggiore region L. Panziera, a C. N. James b and U. Germann a * a MeteoSwiss, Locarno Monti, Switzerland b Embry-Riddle Aeronautical University, Prescott, AZ, USA *Correspondence to: U. Germann, Via Ai Monti 146, CH-6605 Locarno Monti, Switzerland. urs.germann@meteoswiss.ch This article investigates the mesoscale precipitation mechanisms affecting the severest flood events observed from 2005 to 2012 in the Maggia River, located in the Lago Maggiore region on the southern side of the European Alps. High-resolution volumetric radar data are used to describe the horizontal and vertical structure of precipitation, while sounding data depict the airflow conditions. The events causing the nine highest peak discharge rates in the river in the last eight years are characterized by the presence of convection, which is generally absent in the storms that produced lower peak flow rates. During major floods, convective cells repeatedly propagate over the watershed over time. At large temporal scales, precipitation patterns assume the form of an elongated band of precipitation. The main reason for the formation of the band is the different direction of the low-level and upper-level flow in relation to the orientation of the Alpine orography. When pre-frontal, moist and conditionally unstable air over the Po Valley is advected towards the Alpine chain by a southsoutheasterly low-level jet, convection develops over the slopes west of Lago Maggiore where the low-level flow is perpendicular to the terrain orientation. Convective cells are then transported towards the northeast by the upper-level southwesterly steering flow, reaching maturity over the Maggia watershed. This mesoscale process can trigger and maintain moist convection for a long time over the same region. Statistical analyses show that the strength and stability of the upstream flow at low levels (<1.5 km) determine to what extent precipitation will be convective. A comparison with the Mesoscale Alpine Program Intensive Observing Period (MAP IOP) precipitation cases shows that MAP IOP 2b and 3 are representative of the type of event that produces the biggest floods in the Lago Maggiore area. Key Words: Alpine radar; orographic convection; mesoscale flows; MAP; IMPRINTS Received 7 June 2013; Revised 13 February 2014; Accepted 14 February 2014; Published online in Wiley Online Library 3 April Introduction Flash floods caused by heavy orographic precipitation are often responsible for significant loss of life and property in a mountainous country like Switzerland (Hilker et al., 2009). Moreover, climate models indicate that the climatological frequency of their occurrence is likely to increase in the future (IPCC, 2007; C2SM, 2011). Reliable forecasts and the ability to more accurately identify the onset of a heavy precipitation episode in complex terrain would help local agencies take actions to mitigate their impacts. However, forecasting rainfall over mountainous regions is a very challenging task, because the physical mechanisms that produce or enhance precipitation over orography are difficult to Current address: Department of Civil, Environmental and Mechanical Engineering, University of Trento, Via Mesiano, Trento, Italy. represent in both numerical and heuristic forecasts (e.g. Richard et al., 2007; Mandapaka et al., 2012). Review of orographic precipitation processes can be found in Smith (1979), Barros and Lettenmaier (1994), Roe (2005), Smith (2006) and Houze (2012). The characteristics of the upstream flow in relation to the terrain largely determine the ways in which the flow interacts with the orography and, for opportune moisture contents, in which precipitation forms and falls out. One widely recognized variable to consider is the extent that the impinging flow is blocked by the terrain (Peterson et al., 1991; Houze et al., 2001; Neiman et al., 2002; Medina and Houze, 2003). The stability and the intensity of the incoming airflow, in relation to the average height of the orographic barrier, determine the magnitude of the blocking. When the air upstream from a mountain barrier is unstable or the component of the wind perpendicular to the barrier is strong, or the height of the mountain range is low, the airflow possesses enough kinetic energy to easily rise over the terrain, thus enhancing the c 2014 Royal Meteorological Society

2 Orographic Convection in the Lago Maggiore Area 225 rainfall over the first windward slopes. On the other hand, if the upstream airflow is stable or has a weak cross-barrier component, or the mountain is very high, the wind is blocked and it does not ascend over the terrain. In the latter case, the flow approaching the terrain may become deflected parallel to the mountain range, and therefore the upward motion and precipitation enhancement are observed ahead of the barrier. The correlation between the component of the low-level flow perpendicular to the barrier and the rainfall over the first peaks was found to be larger when the flow is unblocked than when it is blocked, for both the Coastal Mountains of California (Neiman et al., 2002) and the Alps (Panziera and Germann, 2010, hereafter PG2010). Microphysical processes determine the efficiency, the rate of the conversion of cloud water into precipitation, and the subsequent fall rate of precipitation through the air (e.g. Alpert, 1986; Smith et al., 2005). In the blocked case, the observed stratiform precipitation derives from the melting of ice particles which grow by vapour diffusion at higher altitudes. During unstable unblocked cases, heavy rainfall is observed over the first terrain peaks due to rapid coalescence in warm cloud at lower levels, and melted graupel particles formed by accretion above the freezing level. Additionally, if the upstream airflow is conditionally or even potentially unstable, the coalescence and riming processes may be favoured by the development of convective cells which produce high concentrations of liquid cloud water (Medina and Houze, 2003). When convective cells move repeatedly over the same location, a process commonly called echo training, large rainfall amounts are observed (e.g. Chappell, 1986; Doswell et al., 1996; Romero and Doswell, 2000; Davis, 2001; Schumacher and Johnson, 2005). The cells are generally embedded in quasi-stationary convective systems, and over the mountains they often organize in shallow or deep convective bands which create strong spatial gradients of precipitation (e.g. Maddox et al., 1978; Yoshizaki et al., 2000; Cosma et al., 2002; Anquetin et al., 2003; Kirshbaum and Durran, 2005; Borga et al., 2007). In fact, the orography can trigger and maintain convective development over specific regions through different mesoscale processes, such as stable or unstable upslope ascent, upstream, lee-side or thermal triggering of convection, or low-level convergence (e.g. Houze, 1993; Rotunno and Ferretti, 2001; Neiman et al., 2002; Rotunno and Houze, 2007). Deep convection over the orography has been recognized as responsible for a number of devastating flash floods throughout the world. The Big Thompson flood of 1976 in Colorado (e.g. Maddox et al., 1978; Caracena et al., 1979) and the Black Hills flood near Rapid City of 1972 in South Dakota (e.g. Maddox et al., 1978; Nair et al., 1997) are probably the two most notorious examples. At the mesoscale, they were both characterized by a low-level moist and conditionally unstable flow moving towards the orographic barrier, capped by a temperature inversion which suppressed convection over the plains. The instability was released through orographic lift once this flow impacted the mountains, and heavy rainfall occurred over middle elevations of the affected catchments without reaching the highest terrain. Even though limited remote-ensing observations hindered the detailed investigation of the precipitation structure of these two events, convective cells developing on the upstream flank of the orography and moving northnorthwestward could be observed. The stationary inflow, fixed topography, very light winds in the upper atmosphere and the absence of organized, convective-scale downdraught allowed the storms to remain quasi-stationary for a few hours, producing amounts of rainfall on the order of mm. Another destructive flash flood occurred in 1974 on the Hawaiian island of Oahu (Schroeder, 1977). Persistent rainfall was observed over the mountains as a strong, moist and potentially unstable easterly flow was forced to ascend over the orography, supported by strong divergence aloft. Rasmussen and Houze (2012) document a flash flood observed in a high-altitude cold desert valley in the Ladakh region, which destroyed the city of Leh, India. In contrast to the quasi-stationary floods mentioned above, this originated from a number of diurnally generated mesoscale convective systems forming upstream over the Tibetan Plateau and moving westward. Over the steep Himalayan barrier, such systems were fed by the southeasterly moisture-laden flow coming from the Bay of Bengal. Even though heavy rainfall is observed over mountain chains with different geographic and climatic characteristics throughout the world, some synoptic and mesoscale elements have been commonly observed (Lin et al., 2001): (i) conditionally or potentially unstable flow impinging on the terrain, (ii) a very moist low-level jet, (iii) steep mountain slopes, and (iv) a quasi-stationary synoptic-scale forcing to provide prolonged convection over a specific area. However, as highlighted by Lin et al. (2001), while major floods over Taiwan and Japan are favoured by high CAPE (Convective Available Potential Energy) and weak synoptic forcing, North American and European floods are less influenced by CAPE and more strongly affected by a deep short-wave trough. Lin et al. (2001) also noted that the concavity of a mountain range helps focus the convective triggering over specific locations in the European Alps and Taiwan. Figure 1. Topographic map of the Central Alps between Italy and Switzerland; the black box indicates the Lago Maggiore region. Detailed view of the Lago Maggiore region. The location of MeteoSwiss Monte Lema radar (1625 m above sea level) is indicated by the radar symbol. The Maggia catchment and the main valleys and lakes in the region are also indicated.

3 226 L. Panziera et al Orographic rainfall in the Lago Maggiore region The Lago Maggiore region (Figure 1), which is located in the Southern Alps between Piedmont (Italy) and Ticino (Switzerland), was chosen as the target area for studying precipitation during the Special Observing Period (SOP) of the Mesoscale Alpine Program (MAP; Bougeault et al., 2001; Rotunno and Houze, 2007) in autumn This region is climatologically prone to floods, with nearly the highest annual rainfall in the Alps, yet the frequency of precipitation is relatively low (Frei and Schär, 1998; Rudolph et al., 2011; Isotta et al., 2013). The local precipitation maximum is observed between the Maggia and Toce river watersheds. The large delta of the Maggia river, which protrudes into Lago Maggiore and forms a wide peninsula of about 10 km 2, indicates that floods have historically plagued this region. When rainfall occurs in the Lago Maggiore area, moist, unstable southeasterly flow is typically observed below 1.5 km above sea level (asl) upstream from the Alps, and southwesterly wind forms aloft (PG2010). The intensity and frequency of precipitation in the Lago Maggiore region are directly related to the upstream wind speed, while the direction of the wind determines the spatial distribution of precipitation (Houze et al., 2001, PG2010). Even though differences in the static stability of the air mass have less impact on the precipitation intensity than the wind speed and direction, unstable conditions cause more precipitation over the mountains than stable cases (PG2010). Most of the MAP SOP cases were characterized by statically stable atmospheric conditions; only MAP Intensive Observing Periods (IOPs) 2a and 2b exhibited unstable atmospheric profiles, and were characterized by convective precipitation. During IOP 2a, the windward slopes of the Lago Maggiore area, including the Maggia Valley, were not affected by heavy rain since the convective activity developed mainly south of the terrain over the Po Valley. On the other hand, IOP 2b was the most intense case observed in the Maggia catchment and Lago Maggiore region during MAP; thus, it has been widely studied in the literature, and it is commonly taken as representative of unstable unblocked precipitation events, while IOP 8 has been considered to be typical of stable blocked flow conditions (Medina and Houze, 2003). The Piedmont region, on the westerly edge of the Lago Maggiore area, also experiences intense rainfall (e.g. Isotta et al., 2013). Rotunno and Ferretti (2001) analyzed the flood that occurred in 1994 in this area, and they show that heavy rainfall is the result of the convergence between a southerly moist saturated airflow which flows over the Alps and a subsaturated easterly flow blocked by the eastern Alps and deflected westwards Nowcasting of orographic rainfall In small Alpine catchments with short response times, heavy rainfall combined with rapid runoff can result in flash floods. Forecasts of rainfall for just a few hours ahead can have a large value for these basins, since they are needed to prepare and issue timely warnings. Numerical models are intrinsically limited for the usage in the nowcasting time frame, since the assimilation cycles and the computational time are long, and the spatial resolution and the updating frequency are too coarse for applications. Radar-based heuristic systems are a valid alternative to numerical models for nowcasting rainfall in the mountains. Since orographic precipitation typically presents a large spatial and temporal variability, high-resolution rainfall measurements are needed for accurate predictions. The operational ground station networks can rarely provide high-resolution rainfall estimates; even in the European Alps, one of the regions with the densest rain-gauge network, typical spacing between stations is about 10 km, whereas the precipitation distribution varies at scales much smaller than 10 km (e.g. Frei and Schär, 1998; Germann and Joss, 2001). Ground-based radars, on the other hand, are designed to monitor precipitation over large areas with high spatial and temporal resolution. Although ground clutter and beam shielding strongly affect radar measurements, adequate quantitative precipitation estimates may be achieved when proper corrections are applied to the data (Joss and Waldvogel, 1990; Germann et al., 2006). The most commonly used radar-based nowcasting technique is Lagrangian extrapolation: the rainfall patterns are moved following the motion derived from the last radar images (e.g. Austin and Bellon, 1974; Mecklenburg et al., 2000; Germann and Zawadzki, 2002). Another promising nowcasting solution comes from analogues: the evolution of the rainfall field observed in the past with meteorological conditions similar to those observed in the immediate past are taken as forecast (Panziera et al., 2011). Both Lagrangian extrapolation and analogue-based methods outperform numerical model forecasts for the first few hours over complex orography (Mandapaka et al., 2012; Panziera et al., 2011). However, these methodologies have intrinsic limitations. Lagrangian extrapolation does not take into account the growth and the dissipation of precipitation which is typically observed in a mountainous region, and analogue systems have difficulties in forecasting rainfall patterns not already observed in the past Objective of this study The objective of this article is to answer the following questions: 1. What mechanisms create heavy precipitation in the Maggia Valley? 2. Is convection necessary for flash floods and, if so, what are the typical cell development and propagation characteristics? 3. What environmental conditions favour flash floods? 4. Are the widely studied MAP cases representative of those that produce floods in the Maggia Valley and in the Lago Maggiore area? To answer these questions, we present an observational study of the precipitation events that caused the greatest peak discharges in the river Maggia over the last eight years. We use weather radar observations to give a detailed description of the horizontal and vertical structure of precipitation patterns producing floods in the Maggia river. Special emphasis is given to the role that convective cells have in producing large rainfall amounts. In fact, even though the convective MAP IOP 2b case has been widely studied in the literature, previous studies have not described the detailed threedimensional convective cell structure and organization leading to heavy rain in the Lago Maggiore region. The detailed observational analysis presented here builds a conceptual model for orographic convection in the Lago Maggiore region which might be observed also over other mountain chains of the world. Moreover, it provides a reference for numerical model verification and for numerical simulations of precipitation and airflow in idealized mountains. Finally, a better understanding of the processes that create heavy orographic precipitation can help identify suitable predictors for improving analogue schemes or Lagrangian extrapolation nowcasting techniques Outline of this article This article is organized as follows. In section 2 we present a short description of the dataset employed in this study. The methodology adopted to describe the horizontal and vertical structure of the rainfall and to characterize the convective cells is illustrated in section 3. In section 4 we introduce the precipitation events which are the object of this study. In section 5 we give an overview of their synoptic and mesoscale conditions, while the structure of the storms is presented in detail in section 6. A conceptual model for the observed floods is proposed in section 7, followed by a comparison between the events presented in this study and the MAP cases in section 8. Concluding remarks are reported in section 9.

4 Orographic Convection in the Lago Maggiore Area Data 2.1. Radar data The MeteoSwiss weather radar located on the top of Monte Lema (1625 m asl; Figure 1), one of the southernmost mountains of the Alpine range in the Lago Maggiore region, was used to estimate the rainfall at the ground, to investigate the vertical structure of the storms and to estimate the wind field in particular regions of the radar domain. The analysis was not extended further into the past since data prior to the year 2005 had only a global bias adjustment; however, data from 2005 were adjusted for both local and global bias, giving more reliable precipitation measurements (Germann et al., 2006). In May 2011 the single-polarization Monte Lema radar was replaced with a state-of-the-art dualpolarization radar. The lack of data during the radar upgrade in May 2011 does not affect our study, since there were no heavy precipitation events that month. We made sure that the bias adjustment schemes and radar upgrades of the new radar did not contaminate the analysis of the precipitation events that occurred after May 2011 by applying an opportune adjustment factor. Information on rainfall at the ground is obtained by the operational MeteoSwiss radar product for quantitative precipitation estimation (Joss and Lee, 1995). This product represents the best radar estimate of precipitation at ground level, and is retrieved through a weighted mean of all the radar observations aloft. The horizontal spatial resolution of the precipitation map is 1 km 1 km, and the temporal resolution is 5 min. An evaluation of the radar performance in the Lago Maggiore region shows that rainfall estimates significantly improved in the last years, as a result of proper clutter elimination, visibility and profile correction, global and local bias adjustment (Germann et al., 2006, and section 2.1 of PG2010). The scan strategy of the MeteoSwiss radars was designed to obtain high-resolution radar measurements in the complex orography of the Alpine region, and it comprises 20 complete sweeps with elevation angles between 0.3 and 40 every 5 min. Polar data of reflectivity, which have a resolution of 1 in azimuth and 1 km in range, were used in this study for characterizing the precipitation vertical structure. In order to estimate the mesoscale wind field, the intensity and direction of the flow at different levels are estimated every 5 min by fitting a linear wind to all valid Doppler measurements in pre-defined regions. PG2010, and in particular their section 5 and Figure 10, provide details about the regions in which the wind is estimated. The performance of this wind estimation technique is good if there is a sufficient number of echoes and if the wind field within the region of interest is approximately linear. These flow estimates are more reliable than ground station measurements because they are not affected by local winds influenced by obstacles located close to the anemometer Soundings Atmospheric soundings from the Milano Linate Airport at Milan, Italy, are used in this study to characterize the mesoscale conditions observed upstream from the Alps. The sounding site is located about 50 km southeast of the Lago Maggiore region (Figure 1). The 1200 UTC (1300 or 1400 local time) sounding launched just before the most intense phase of precipitation of the events studied here was taken as representative of the mesoscale conditions upstream of the Alps for the given event. In this way, we consider soundings launched at similar times of the day, avoiding significant diurnal variations of meteorological quantities that might affect our results. However, for two precipitation events the 0600 UTC sounding was used, since the 1200 UTC measurements were not available Synoptic analyses The synoptic conditions are characterized by means of superposed epoch analysis, created by averaging synoptic fields for all of the convective and non-convective precipitation events. The synoptic data were obtained using National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis grids at horizontal resolution (Kalnay et al., 1996). 3. Methodology This section provides a description of the methodologies that were developed in order to describe the horizontal patterns of precipitation (section 3.1) and the vertical structure of the storms (section 3.2). The methods by which the convective cells are investigated are presented in section Surface precipitation analysis To characterize surface precipitation, we make use of the MeteoSwiss radar product that represents the best radar estimate of the precipitation at ground level (Germann et al., 2006), as mentioned in section 2.1. First, we computed maps of accumulated precipitation and mean rainfall rates for the Lago Maggiore area. Second, we characterize the rainfall rates along the parallel cross-section represented in Figure 2. The transect is 130 km long and 15 km wide; to facilitate computations, it was divided into 15 parallel sections separated by 1 km. It extends parallel to Lago Maggiore from southwest to northeast, running from the first Alpine slopes southwest of the lake through the lower part of the Maggia Valley to the crest of the Alps. The reason for the choice of this transect is our desire to investigate the evolution of the rainfall patterns affecting the Maggia Valley; in fact, the transect extends along the typical direction of motion of the precipitation affecting the catchment, which is parallel to the 3 km wind, i.e. from southwest to northeast (PG2010). The contribution to the total rainfall of different rain rate classes along the transect was also computed. To get this information, we followed two methodologies: the horizontal precipitation field over the transect for a given time instant, averaged in the direction of the shortest axis of the transects, was normalized by the total rainfall observed over the whole transect during the considered storms (transect normalization), or with the total rainfall observed at each gate of the transect during the analyzed events (gate normalization). The contribution to the total rainfall of a given rain rate class at a specific gate is then the sum of the normalized rainfall whose original rainfall rate was within the given rain rate class. In other words, the transect normalization analysis shows which rainfall rates and gates contribute more to the total rainfall amounts along the transect; the gate normalization analysis, on the other hand, shows which rain rates contribute more to the total rainfall measured at a given bin of the transect Vertical structure analysis Vertical sections of average reflectivity along the three crosssections of Figure 2, are computed in order to investigate the vertical structure of precipitation. The transect of Figure 2 was introduced in the previous section, whereas the transects of Figure 2 run from the Po Valley to the crest of the Alps; the motivation for their choice will be given in section Both the transects are 15 km wide, while their length is 108 and 123 km respectively. Polar reflectivity data up to a height of 12 km above each transect were taken into account to produce the cross-sections; the reflectivity observations from the polar radar volume for each of the sections composing the transect were remapped to a grid with a horizontal resolution of 1 km and a vertical resolution of 100 m. For clarity, the original sample

5 228 L. Panziera et al. (c) (d) (e) Figure 2. Transects along which the vertical structure of precipitation is studied: parallel transect, perpendicular and north south transects. (c) (e) represent the sample volumes of the radar bins along the central cross-section of each transect. The orography along the cross-sections is also shown in each panel: grey shading represents the terrain elevation averaged in the direction of the short axis of the transects, and black shading represents the maximum terrain elevation. volume size of the radar bins along the central cross-section of each transect is represented in Figure 2(c) to (d)). Thus, for each transect a three-dimensional matrix with the reflectivity data measured at a given time step is obtained. For example, a matrix km 3 was obtained for the parallel transect. The reflectivity data were then averaged in the direction of the shortest axis of the transects, so that a two-dimensional grid was obtained. Further averaging over all the time steps of a given storm gives the average reflectivity value at a given height and distance along the transect during the storm. The vertical structure of precipitation was analyzed also by means of contour frequency by altitude diagrams (CFADs; e.g. Steiner et al., 1995; Yuter and Houze, 1995), which concisely characterize the vertical structure of the storms over a specific region of the radar domain, the Maggia catchment in our case. In order to produce CFADs, the radar data were remapped following the same methodology used for vertical sections over the Maggia catchment. They were then arranged into predefined classes of reflectivity and altitude for all the time steps belonging to a given precipitation event, and the frequency of occurrence in each class with respect to the total number of occurrences was computed. The reflectivity range selected was 5 50 dbz, with a class interval of 1 dbz; the vertical resolution is 100 m Convective cells In order to identify convective cells in radar images of surface precipitation, we make use of the Thunderstorm Radar Tracking algorithm (TRT; Hering et al., 2004). A cell is defined as a connected zone of radar pixels larger than a given area and whose reflectivity exceeds an adaptive detection threshold. From a sequence of radar images containing convective cells, the algorithm creates trajectories. Splits and merges between cells are also taken into account. If merges among cells occur, trajectories are continued in the larger cell, whereas the trajectories of the smaller ones die. If cells split, then the trajectory of the original cell stops and the trajectories of the smaller split cells begin. Based on the displacement of the centre of mass, which is assumed to be the centre of the cell, the velocity of the cells is also estimated. In this article we use the threshold of 43 dbz (16 mm h 1 )to identify a convective cell, with a minimum area of 10 km 2 and a minimum duration of 30 min. These thresholds were selected

6 Orographic Convection in the Lago Maggiore Area 229 empirically in order to detect significant convective cells and, on the other hand, to prevent the detection of both weak and small cells as well as larger stratiform precipitation areas. 4. The precipitation events In this article we study the precipitation events that produced the highest peak discharge rates in eight years at the outlet of the Maggia river. We take the peak flow discharge rate at the outlet of the river as an indirect measure of rainfall in the Maggia catchment. The reason for this choice resides in the fact that we want to study the events which caused large river discharges, having the greatest impact on human lives and infrastructures. Moreover, river flow discharge rate is one of the most important quantities to be forecast in case of floods. Table 1 displays some characteristics of the events that produced the 25 highest peak discharge rates in the river. The first column gives the name which will be used to refer to each event. Since we show in this section that the precipitation structure of the nine most intense events is more convective than the following cases, we call those events convective (C1,..., C9); on the other hand, events from 10 to 25 are denoted as non-convective (NC1,..., NC16). The second column indicates the peak discharge rate; as a reference, consider that the peak discharge of 1000 m 3 s 1 has a return period of 1.5 years for the river Maggia ( accessed 28 February 2014). Subsequent columns of the table show the date of the peak discharge (column 3), the start time (column 4) and duration of the event (column 5) as determined by examination of the radar images over the period around the time of peak discharge in which continuous rainfall was observed over the Maggia catchment. Nine events approached or exceeded the peak discharge with a return period of 1.5 years (1000 m 3 s 1 ), and the remaining 16 events produced lower peak discharges. The duration of each precipitation event averaged about 1 day, and was in general inversely related to the measured peak runoff. The sixth column of Table 1 indicates the total rainfall per unit area estimated by the radar for each event averaged over the Maggia catchment (sections 2.1 and 3.1), and the seventh column shows the mean rainfall rate per unit area averaged through the duration of the event; these two columns therefore indicate the total amount of water and the mean intensity of the rainfall measured in the catchment. They show that the events with the largest peak discharge rates are in general associated with rain rates higher than the events with low discharges. However, events C6 and C7 had average rain rates lower than some of the other events with comparable or lower peak discharges. On the other hand, events NC4 and NC7 experienced mean rain rates that were comparable to the rain rate of the events that produced the largest peak discharge. These discrepancies probably result from the fact that the precipitation rates were averaged over the entire catchment and may not have been representative of the intensity of rainfall and the amount of runoff taking place in localized areas within the catchment. Moreover, each case may have had different initial soil conditions, which would also affect how much water runs off in the river. The eighth, ninth and tenth columns of Table 1 describe the convective cells observed over the Maggia catchment, identified by means of the TRT system (section 3.3). The eighth column indicates the average number of convective cells observed over the Maggia catchment per hour during each event. We see that the events that produced peak discharge approaching 1000 m 3 s 1 or more had generally more convective cells passing over the catchment per hour, whereas the remaining cases had less than 1cell h 1. The only exceptions were the C7 case, which had a cell frequency of only 1.6 h 1 despite a large peak discharge, and the NC2 and NC3 cases with lower discharge rates yet cell frequency slightly larger than 1 h 1. The ninth column indicates the median of the spatial extent of the cells; in general the events with the largest runoff display convective cells larger than the weakest events. The tenth column gives the percentage of the total rainfall over the Maggia watershed caused by convective cells for each event. Values larger than 10% are noted for the first nine cases, whereas the remaining cases have values significantly lower. Moreover, more than 50% of the total rainfall during the C8 case, and over 40% of the precipitation from four other events (C2, C3, C5 and C9) originated from convective cells. The last column of Table 1 contains the 95th percentile of the height of the radar echoes observed over the Maggia catchment for each event. This quantity was calculated in order to have Table 1. Characteristics of the precipitation events. See text (section 4) for details. Event Peak runoff Date of Start day/hour Duration Tot rainfall Mean RR Freq. cells Area cells Contrib. R Echo (m 3 s 1 ) peak runoff (UTC) (h) (mm) (mm h 1 ) (cells h 1 ) (km 2 ) cells (%) ht. (km) C September / C October / C June / C August / C July / C June / C October / C July / C June / NC November / NC May / NC November / NC June / NC May / NC May / NC April / NC October / NC October / NC March / NC May / NC December / NC September / NC September / NC November / NC June /

7 230 L. Panziera et al. an indication about the vertical development of the storms, and to detect the presence of convection; in fact, convective cells are expected to be more vertically developed than the average precipitation structure. We see that the first nine events are in general more vertically developed than the other cases, probably due to the larger number of convective cells passing through the catchment. However, there are a few exceptions: the C7 case is less vertically developed than cases with similar peak runoff rate, and events NC2, NC4 and NC11 have a larger vertical extents than cases with comparable peak discharges. A comprehensive analysis of Table 1 reveals that the first nine precipitation events causing the highest peak discharge in the river Maggia appear much more convective in nature than the others: they exhibit generally higher rain rates and in many cases more total rainfall over the Maggia catchment, larger and more frequent convective cells, and greater vertical development. This is the reason why we refer to the first nine cases of Table 1 as the convective events (C1, C2, etc.), and to the remaining cases as the non-convective events (NC1, NC2, etc.). In order to answer the scientific questions posed in section 1.3, in this study we compare and contrast the convective and non-convective events with respect to the synoptic setting, thermodynamics and dynamics of the upstream flow, and the resultant structure and organization of precipitation. For consistency, we compare the nine convective cases (events C1 to C9) with the nine non-convective cases producing the largest peak discharge (events NC1 to NC9). Our analysis, therefore, includes a total of 18 precipitation cases. Figure 3 gives a concise view of the wind field, rain rate and river discharge for two convective (C1, C2) and two non-convective (NC7, NC9) precipitation events. For each event, the 3 km wind (the mid-level flow introduced in PG2010 and in section 2.1), the average rain rate in the Maggia catchment and the average rain rate due to convective cells, and the river discharge are shown. We note that both wind speed and rain rate can vary significantly within the same precipitation event even in a very short time period. This variability illustrates that a single wind measurement, for example taken from a sounding, cannot be representative of a whole precipitation event. Moreover, peak river discharges can result from rainfall rates that are not necessarily extreme, but which occur after prolonged periods of rainfall. It is also evident that the response time of the river Maggia is between about 3 and 5 h. The wind estimates from the C1 event possess two large gaps, but it can be seen that during this case the wind varies by as much as 10 m s 1. A significant part of rainfall is due to convection, which contributes to the largest rain rates. During the C2 case there is also considerable variability in the wind and a significant increase of about 10 m s 1 within 3 h; convection is also significant in this event. Events NC7 and NC9, which are respectively the 16th and 18th most intense in terms of river discharge, tend to have more constant rain rates and were generally not convective in nature. 5. Synoptic and mesoscale characteristics of the events In section 5.1 we provide a short description of the synoptic conditions observed during the precipitation events. The mesoscale conditions are investigated in section 5.2; parameters derived from the Milano atmospheric sounding are presented in section 5.2.1, the structure of the wind in section 5.2.2, and the dependence of rainfall in the Maggia catchment on the intensity of the upstream flow at different heights in section Synoptic conditions Synoptic composites (or superposed epoch analyses) were produced for the convective and non-convective precipitation events. Figure 4, depict the 500 hpa geopotential height, mean sea level pressure and surface wind over central Europe for the convective and non-convective events respectively. As shown for both types of cases, troughing was prevalent in the upper levels over western Europe during these events; however, the troughing extends further towards the south during the non-convective cases. Thus, the upper-level synoptic lift is likely more pronounced during the non-convective events. Cyclonic circulation is observed at the surface, with southerly winds over northern Italy impinging on the Alps. Although the prevailing surface wind during the convective cases is from south, the nonconvective cases have southeasterly flow at the surface. Figure 4(c), (d) show the 850 hpa geopotential height, temperature and mixing ratio. We note that for the non-convective cases the trough is elongated from the British Isles to the Mediterranean Sea, whereas for the convective cases the trough is less meridionally developed. A cold frontal zone is clearly evident in the convective cases, which often stalls along the western side of the Alpine chain, suggesting more focused and prolonged frontal lifting or triggering, whereas frontal activity is less pronounced and located further south over the Mediterranean Sea during non-convective cases. Moreover, mixing ratios are also significantly higher for the convective cases than for the non-convective ones, suggesting greater rainfall potential in the Lago Maggiore region during convective events. Figure 4(e), (f) show the CAPE and convective inhibition (CIN) composites. Both CAPE and CIN are low over the Alps and almost zero for the non-convective cases. However, during the convective cases a maximum in CAPE is present over the Ligurian Sea, elongated towards the south to the central Mediterranean. The CAPE is not a maximum over the Alps, but its values are rather pronounced and average between 750 and 1000 J kg 1. CIN, on the other hand, is low over the Alps, but increases towards the south over Sicily where the circulation at the upper levels is more anticyclonic. The synoptic patterns and thermodynamic fields presented in Figure 4 imply that, during the convective events, enough CIN would exist over the Mediterranean Sea to prevent deep convection upstream from the barrier and use up the available CAPE. Therefore, moist southerly wind from the sea could flow relatively unimpeded until triggering convection where the CIN is weaker along the first Alpine foothills in the Lago Maggiore region Mesoscale conditions Sounding parameters Table 2 reports a number of parameters derived from the atmospheric soundings launched at Milano Linate (Figure 1) considered as representative of each precipitation event, and Figure 5 shows single and composite soundings for both the convective and non-convective cases. The thermodynamic profiles of Figure 5 look significantly different for the two types of precipitation events. First, convective cases are considerably warmer than non-convective ones, as can also be deduced from the higher surface temperatures (T bl ) and lower freezing-level heights in Table 2. Second, convective cases have a conditionally unstable stratification, in contrast with the non-convective cases, which generally exhibit a moist neutral profile. An inversion layer was present below 800 hpa in a few of the non-convective events, but was not apparent in the average sounding in Figure 5. Although there are no significant differences in the height of the level of free convection (LFC) between convective and non-convective cases, the height of the lifting condensation level (LCL) tends towards lower values for non-convective cases, in agreement with the higher relative humidity profiles in the composite sounding for these events. However, there was substantial variation of both LFC and LCL heights among the precipitation events. Table 2 also shows the values of moist Brunt Väisälä frequency (Nm 2 ) averaged in the lowest 3 km, and CAPE computed by lifting the air parcel starting from different 500 m layers. The convective cases generally have negative values of moist static stability, whereas most of the non-convective events are statically stable or have profiles close to neutral. The sounding profiles during the

8 Orographic Convection in the Lago Maggiore Area 231 (c) (d) Figure 3. Wind speed, rainfall rate and river discharge rate for (a, b) two convective and (c, d) two non-convective events. Within each panel are shown the 3 km wind speed (m s 1 ) estimated by Doppler velocity radar, the average rain rate (mm h 1 ) over the Maggia catchment and rain rate due to convection (thin line), and the discharge rate (m 3 s 1 ) of the Maggia river at the outlet. convective cases generally exhibit substantial CAPE, unlike nonconvective cases with the exception of NC5. The amount of CAPE generally decreases with increasing initial parcel altitude, since the stratification of upper levels is generally almost neutral. The only exceptions to this behaviour are observed for C3, C4, and C9. CIN is rather low for all the precipitation events in the Lago Maggiore region. The specific humidity was averaged in the lowest 5 km in order to have an estimate of the moisture content of the atmosphere, which is generally lower during non-convective cases than in convective cases.

9 232 L. Panziera et al. (c) (d) (e) (f) Figure 4. Synoptic composites for convective and non-convective cases. (a, b) 500 hpa geopotential height (dam; continuous line), mean sea level pressure (hpa; dashed line) and surface wind (symbols). (c, d) 850 hpa geopotential height (dam, continuous line), temperature ( C; dashed line) and mixing ratio (g kg 1 ; shading). (e, f) convective available potential energy (J kg 1 ; shading) and convective inhibition (J kg 1 ; dashed line) Structure of wind Figure 6 shows the composite vertical profiles of the wind speed from Milano soundings in the lowest 5 km. Figure 7 shows the distribution of radar-derived low-level flow (LLF) and upperlevel flow (ULF) speed and direction for both the convective and non-convective cases. LLF and ULF represent the wind over the first Alpine slopes at heights of about 1.5 and 5 km, respectively (section 2.1), and are available every 5 min, unlike sounding measurements. Thus the distributions in Figure 7 describe the variation of wind velocity throughout the entire duration of the precipitation events. From the surface to 2 km the non-convective cases have generally stronger mean wind, with wind speeds comparable to the convective cases between 2 and 3 km. Above 3 km, the convective cases show stronger winds. Both convective and nonconvective cases are characterized by the presence of a moist low-level jet (LLJ) over the Po Valley, as clearly indicated by the wind speed maxima near 1 km altitude in Figure 6. The intensity of the jet is about 4 m s 1 stronger and its altitude is slightly higher during non-convective cases than in convective events. Its altitude in the convective cases averaged about 800 m, as opposed to 1.1 km in the non-convective cases. The altitude of the LLJ for each case is reported in Table 2. We note that this height varies

10 Orographic Convection in the Lago Maggiore Area 233 Table 2. Quantities derived from the Milano sounding for the precipitation events. See text (section 5.2.1) for details. Event Represent sounding T bl 0 C ht LFC LCL CAPE CIN N 2 m Spec. hum. Wind speed Fr LLJ ht. WV flux layer (km) unit day/time, month ( C) (km) (m) (m) J kg 1 (J kg 1 ) 10 4 (s 2 ) (g kg 1 ) (m s 1 ) (km) (kg m 1 s 1 ) C1 26/1200, No jet C2 06/1200, C3 06/0600, C4 17/1200, , C5 17/1200, , C6 15/1200, No jet C7 03/1200, , C8 13/1200, C9 12/0600, NC1 04/1200, , NC2 30/1200, No jet NC3 04/1200, , NC4 03/1200, , NC5 27/1200, No jet 241 NC6 02/1200, , NC7 27/1200, NC8 04/1200, NC9 29/1200, , Figure 5. Spaghetti plot of soundings from Milano, Italy, for convective and non-convective events. The composite sounding is shown in bold. substantially between events, and that a second jet maximum is also observed for most of the events. Table 2 also lists the average wind speed from each sounding in the lowest 3 km for each event along with the moist Froude number, Fr = U/N m H (or inverse of the non-dimensional mountain height). Here U is the component of the wind perpendicular to the mountain barrier, N m is the moist Brunt Väisälä frequency and H is the height representative of the Alpine crest (Houze, 1993). Mean wind and static stability were obtained up to a height of 3 km, which was the assumed value of H. It should be noted that the wind observations were missing from the C1 event sounding and in this case U was obtained from the LLF radar estimate at around 1.5 km. The moist Froude number is undefined for most of the convective cases, which are statically unstable. On the other hand, it shows large variability for the non-convective cases, ranging from 0.6 to 2.5, being undefined for three events. It therefore holds that the convective events have environmental conditions typical of unstable unblocked cases, whereas the airflow during non-convective events does not present a typical regime. The distributions in Figure 7, obtained at 5 min temporal resolution, confirm that the LLF is generally stronger during the non-convective cases, yet the ULF does not differ significantly between the two sets of events. On the other hand, the distributions of wind direction differ notably between convective and non-convective cases. The LLF is mainly southerly during the convective cases, yet southeasterly during the non-convective cases. The ULF is southsouthwesterly in the convective cases, as opposed to a more southerly flow during non-convective cases. Moreover, the range of directions observed in the convective cases is narrower than for the non-convective cases, suggesting that the events with the highest peak discharges in the Maggia river occurr within a very limited range of wind directions. Table 2 also reports the value of the water vapour flux directed towards the Alpine chain integrated over the lowest 5 km (values are not shown for C1, C6 and NC2 due to missing wind velocities in this layer). We note that there is a large variability among the events, even within the convective and non-convective groups. However, important differences are noticed in the vertical profiles of water vapour flux for convective and non-convective cases (Figure 8). Convective cases have a larger moisture flux at elevations above km than the non-convective cases. At

11 234 L. Panziera et al. wind profilers located in coastal sites and the rain gauges in the mountains was km much smaller than the distance between Milano and the Maggia catchment in our study (70 90 km). Nevertheless, they found that near the barrier the height of the crest is indicative of the layer of greatest correlation between mean wind speed and rainfall, which does agree with Figure 9 of PG Structure of the storms In this section the structure of the orographic precipitation during the 18 Maggia river flood events is extensively analyzed using three-dimensional weather radar data. Surface precipitation estimates are shown in section 6.1, and the contribution of convective cells to the total rainfall and their preferred regions of development and disappearance are presented in section 6.2. Section 6.3 depicts the vertical structure of precipitation and the sensitivity of this structure to environmental variables. Figure 6. Composite vertical profile of wind speed (mean ± standard deviation at each level) from the Milano sounding for convective and non-convective cases. lower elevations, moisture fluxes for the two groups of events are quite similar Correlation analysis In this section we investigate the dependence of rainfall in the Maggia catchment on the speed of the upstream flow at different heights. The objective of this analysis is to identify the layer whose mean wind correlates best with the rainfall measured in the Maggia catchment. This work is based on the correlation coefficient analysis performed by Neiman et al. (2002), who investigated the relation between low-level upstream flow and orographic precipitation during winter rainfall events in the coastal mountains of California. Figure 9 shows the vertical profile of the correlation coefficient between the wind speed in the Milano sounding averaged over 250 m deep layers and the mean rain rates estimated by the radar over the Maggia Valley for 5 h after the sounding time. All the precipitation events considered in this study have a southerly wind component, and therefore Figure 9 represents the vertical profile of correlation between rainfall in the mountains and the wind impacting the Alpine barrier. Since rainfall intensity in the mountains is proportional to the moisture flux impinging on the orographic barrier (Alpert, 1986), the lifting of the layers that present maxima in the vertical profiles of Figure 9 can be considered the most efficient for producing orographic rainfall. The figure shows that upstream of the Alps the layer with mean wind velocity correlating best with rainfall is located at about km asl. A second maximum in the vertical profile appears at about 3 km asl. Both convective and non-convective cases show this general behaviour, although convective cases have slightly higher (lower) values of correlation coefficients at low (high) altitudes than non-convective cases. Similar results were obtained by PG2010 for a larger collection of 58 events of orographic precipitation in the Lago Maggiore region. In Figure 8 of PG2010, a maximum in correlation was found around 1.5 km asl for soundings neutral and unstable in the m layer. Since most of the soundings of the events presented here are all neutral or unstable in that same layer, we confirm the findings of PG2010, that the height of the maximum correlation upstream from the Alps is situated below the crest of the barrier (about 3 km). The secondary maximum at around 3 km in this study has not been identified by previous work. A comparison of our results with the findings of Neiman et al. (2002) is difficult, since the observing system used in that study was somewhat different from the one used here. In particular, the distance between their 6.1. Surface precipitation structure The total rainfall estimated by the Monte Lema radar for both convective and non-convective precipitation events is presented in Figure 10. For reference, the duration of each event and the location of the Maggia catchment are also shown in the figure. We note that in some of the convective and non-convective precipitation events rainfall accumulation exceeds 200 mm northwest of Lago Maggiore. In all the convective cases, total rainfall accumulations exceed 100 mm, whereas a few of the nonconvective events total less than 100 mm. The most noteworthy characteristic of the precipitation totals shown in Figure 10 is a pronounced rainfall maximum in all nine of the convective events covering portions of the Maggia catchment, forming a roughly elliptical pattern whose main axis is parallel to the lake. The elliptical maximum tends to begin near the mouth of the Toce Valley and extends northnortheastward as far as the Alpine crest (Figure 1 gives geographical references). The maximum is clearly observed in all the convective cases, whereas it tends to be broader and not as consistently focused over the Maggia catchment and have weaker spatial gradients during the non-convective cases. Over the Po Valley, larger accumulations are generally observed for the non-convective cases. The small spots of large rainfall accumulations observed for both convective and non-convective cases over specific places (for example the two spots just north of Valtellina in cases C3, C6, C7, NC3, NC5, NC6 and NC7) may be attributed to residual ground clutter contamination and should not be regarded as rainfall maxima. Figure 11 shows the mean rain rates for each case, computed by dividing the total accumulation by the number of hours for each pixel. Non-convective cases are in general longer than convective ones, producing lower mean rain rates over Maggia Valley and the Lago Maggiore area. A comparison between Figures 10 and 11, and between the total accumulations and mean rain rates reported in Table 1, suggests that in general it is the intensity of the rainfall, rather than the total storm accumulation, that produces the highest peak discharges in the Maggia catchment. Figures 12 and 13 depict the radar-derived rainfall rates at the surface along the parallel transect shown in Figure 2, following the methodology introduced in section 3.1. Figure 12 shows the joint contribution of each rain rate class at each gate to the total rainfall measured over the whole transect during the convective and non-convective storms (transect normalization). To simplify interpretation, a horizontal line at 10 mm h 1 was inserted in the figure. We note that most of the rainfall for both convective and non-convective cases occurs between 60 and 100 km along the transect, in a region which extends from the first Alpine foothills near the mouth of the Toce Valley northnortheastwards to the Verzasca Valley, which is located between Maggia and Leventina valleys (Figures 1 and 2). However, a significant difference is

12 Orographic Convection in the Lago Maggiore Area 235 (c) (d) (e) (f) (g) (h) Figure 7. Distribution of (d) low-level flow and (e) (h) upper-level flow (a, c, e, g) speed and (b, d, f, h) direction for (a, b, e, f) convective and (c, d, g, h) non-convective events, estimated by Doppler velocity radar data with a temporal resolution of 5 min over the first Alpine slopes. See text for details. noted between convective and non-convective events: the rain rates having the greatest contribution to the total rainfall along the transect during convective cases are larger than 10 mm h 1, whereas for non-convective events they are smaller than 10 mm h 1. Moreover, the rainfall extends more towards the southwest in the non-convective cases than in convective ones, which indicates more widespread precipitation, as was noted also in Figure 11. Figure 13 shows the contribution of each rain rate class to the total rainfall measured at each gate along the transect during the convective and non-convective storms (gate normalization). Since the contributions to the total rainfall in Figure 13 are independent of the total rainfall measured at other locations

13 236 L. Panziera et al Analysis of convective cells Figure 8. Composite vertical profile of moisture flux (mean ± standard deviation at each level) from the Milano sounding for convective and non-convective cases. Figure 9. Vertical profile of correlation coefficient between the wind measured by Milano radiosounding and the radar-derived rain rate in the Maggia catchment for both convective and non-convective events. The wind is averaged over 250 m deep layers, and rain rate over the whole Maggia Valley area is averaged over 5 h following the launch of each sonde. along the transect, the gate normalization analysis enables a more detailed examination of the distribution of rain rates in regions where the transect normalization does not show any particular differences at a given location. The large contributions to total rainfall observed between 0 and 10 km along the transect for convective cases are due to a few thunderstorms that developed in that area, and, as shown in Figure 12, they do not significantly contribute to the total rainfall along the transect over all the convective cases. From 10 to 60 km and 100 to 130 km along the transect, small rain rates contribute most to the total rainfall for both convective and non-convective cases; however, in the convective cases a significant portion of rainfall comes also from large rain rates. Moreover, between 60 and 100 km distance, where most of the rainfall is observed, precipitation during convective cases is characterized by substantially larger rain rates than in non-convective events. The figures presented in this section show that the climatological maximum of precipitation observed in the elliptical region which extends from the Toce Valley to the Verzasca Valley appears in both convective and non-convective precipitation events. The aim of this section is to give a detailed description of the behaviour of convective cells observed during the convective events. Figure 14 depicts the contribution of both convective and non-convective rainfall to the total rainfall of the convective precipitation events. In this analysis the non-convective precipitation events are not taken into account since convective cells are almost absent in those events. In this study, convective (non-convective) rainfall is defined as precipitation (not) generated by a convective cell. Figure 14 shows the total surface rainfall accumulation from the convective precipitation events. For comparison, the total estimated rainfall from convective cells identified during the convective precipitation events is shown in Figure 14. Figure 14(c) is the difference between the total rainfall and convective rainfall. Thus, Figure 14 provides additional spatial detail to complement the percent contributions of convective cells for the Maggia catchment given in Table 1. Here, we see that there are significant differences in the character of precipitation over the Maggia Valley. In the upper part of the catchment, most of the rainfall is non-convective, yet in the lower part the largest contribution to the total rainfall comes from convective cells. However, when integrating over the whole basin, the convective contribution is smaller than the non-convective one for all the precipitation events, except for C8 (Table 1). Figure 14 shows also that the maximum accumulation of non-convective rainfall lies in the lee of the maximum accumulation of convective rainfall, considering that the wind at mid levels is generally from the south and southwest during convective events. Moreover, the convective contribution is more elongated and linearly shaped than the non-convective contribution, which is more widespread and less characterized by gradients of the rainfall field. For a given sequence of radar images, the TRT algorithm is able to reconstruct the trajectories of convective cells, assuming that the centre of mass is the centre of the cell. Thus, the TRT algorithm gives us the exact location where each cell is identified for the first and the last time. Figure 15 shows these two locations for all convective cells occurring over the Lago Maggiore region during the nine convective cases. Since the threshold used to detect convective cells is 43 dbz, we call these points the development and disappearance locations, rather than trigger and dissipation locations. In fact, convective cells are already well organized once they reach the 43 dbz (16 mm h 1 ) threshold, and they have not completely dissipated after they decrease below that threshold. Thus, by using the 43 dbz threshold we identify strong and mature convective cells. The coloured dots in Figure 15 indicate the amount of rainfall generated by each cell, derived from the distribution of total rainfall generated by each cell throughout its mature life cycle over all the convective cases. An analysis of the cell development and disappearance points in Figure 15 reveals many interesting characteristics of convection in the Lago Maggiore region. Most of the convective cells first develop over the first windward slopes of the Alps, and rarely over the interior region of the Alps. A few cells also develop over the Po Valley upstream from the barrier. The convective cells producing the most rainfall, represented by the red dots, develop more randomly along the first windward slopes of the Alps and the Po Valley, whereas the weak and to some extent the moderate rain-producing cells (blue and yellow dots, respectively) develop more commonly near the first steep Alpine slopes. Therefore, orographic triggering seems to play a major role in the development of weak to moderate cells, but the strongest cells are apparently not well correlated with the slope of the underlying orography. The first steep windward slopes of the Alps on the west side of the lake are oriented from roughly southwest to northeast (from the bottom-left corner of the figure northeastward to the western shores of Lago Maggiore). In this region of sloping terrain there is a dense cluster of many weak and moderate cell development locations, extending from southwest to northeast.

14 Orographic Convection in the Lago Maggiore Area 237 Figure 10. Total rainfall accumulation (mm) for each of the convective and non-convective cases. The Maggia catchment is indicated by the grey line, the black line represents the 800 m orographic contour, and the white lines indicate the lakes. Coincidentally, the upper-level steering flow is also oriented from southwest to northeast during these events (section 5.2.2), parallel to the orientation of the Alpine foothills that are initiating the convective cells. Thus, the convective cells become naturally aligned into roughly southwest northeast echo trains as they propagate parallel to the Alpine foothills and then across the Maggia catchment. This behaviour of repeated convective development and propagation across the same area creates the c 2014 Royal Meteorological Society large elliptically shaped precipitation maxima observed during the convective events as shown in Figures 10 and Vertical structure of precipitation The aim of this section is to investigate the vertical structure of radar reflectivity. In sections and respectively, we characterize the vertical structure of precipitation above the Q. J. R. Meteorol. Soc. 141: (2015)

15 238 L. Panziera et al. Figure 11. As Figure 10, but for the mean rain rate (mm h 1 ). parallel cross-section and above the Maggia catchment. Then, the sensitivity of precipitation vertical structure to low-level wind and stability is analysed in section Vertical structure parallel to cell propagation Figure 16 depicts the average reflectivity with height observed along the parallel cross-section shown in Figure 2 for c 2014 Royal Meteorological Society each convective and non-convective precipitation event. The orientation of the cross-section is parallel to the prevailing direction of convective cell motion, from the southwest towards the northeast as illustrated in section 6.2. For a correct interpretation of the reflectivity structure seen in Figure 16, we must consider the backscattering effects related to radar measurements, in addition to the dynamical, thermodynamical and microphysical processes producing the precipitation. The Q. J. R. Meteorol. Soc. 141: (2015)

16 Orographic Convection in the Lago Maggiore Area 239 (c) Figure 12. Joint contribution of each gate of the parallel cross-section of Figure 2 and of each rain rate class to the total rainfall measured by radar at ground level over the whole parallel cross-section during the convective and non-convective storms. (c) shows the terrain elevation along the cross-section: grey shading represents the elevation averaged in the direction of the short axis of the transect, black shading represents the maximum elevation. (c) Figure 13. Contribution of each rain rate class to the total rainfall measured by radar at ground level at each gate of the parallel cross-section of Figure 2 during the convective and non-convective storms. (c) The terrain elevation along the cross-section. lower reflectivity values seen at the bottom of the cross-sections below about 2 km are due to partial shielding of the lowest radar beams. Moreover, the thin, horizontally aligned peak reflectivity seen at around 3 km altitude especially in most of the nonconvective cases can be partly attributed to enhancement of reflectivity due to melting snow (bright band contamination). On the other hand, the convective cases in Figure 16 have less of a bright-band signature. A clear difference in the vertical structure of the reflectivity is apparent between convective and non-convective cases in the figure. Convective events are much more vertically developed and the reflectivity is generally stronger than non-convective cases at all altitudes. In a few convective cases, echo tops rise as high as km (C3, C5 and C8), although the remainder of the events have echo tops generally below 8 km. A clear maximum in reflectivity is seen at low levels between about 50 and 110 km along the cross-section (i.e. between the Toce Valley and Leventina Valley), consistent with the high surface rainfall rates observed in this region during convective events (Figures 10 13). This maximum is centred around 80 km along the cross-section and is due to the repeated development of convective cells over the steep terrain slopes west of Lago Maggiore which are then transported northeastwards by the upper-level wind across the Maggia Valley until weakening near the Leventina Valley. The reflectivity maximum during the non-convective events is also observed at around 80 km in the cross-sections, yet the horizontal extent of the echoes varies substantially among the events and extends further upwind over the Po Valley than the

17 240 L. Panziera et al. (c) Figure 15. Development and disappearance locations of all convective cells observed during convective precipitation events. The coloured dots represent the intensity of each cell in terms of the produced rainfall: weak cells (those producing less than the median rainfall) are shown in blue, moderate cells (with rainfall between the median and the 90th percentile) in yellow, strong cells (with more than the 90th percentile) in red. Section 6.2 gives details. Figure 14. Total, convective, and (c) non-convective contributions to the total rainfall accumulation of the convective precipitation events. convective events. The average reflectivity patterns in Figure 16 correspond well with cross-sections of the frequency of occurrence of precipitation (not shown), demonstrating that the locations where the largest reflectivities are observed are also the locations where precipitation is more prolonged and frequent. From Figure 16 it can be deduced that the convective cells are generally shallow in all events, with mean reflectivity dramatically decreasing with height above about 4 5 km. In fact, above this height the maximum in reflectivity between 50 and 110 km in the cross-sections is generally weak or not observed. Since the height of the freezing level in convective cases is around km, and the largest reflectivities are measured below this height, the high rain rates at the ground observed in Figures 12 and 13 between 50 and 110 km are principally due to raindrops which rapidly grow by coalescence. Such a low-level reflectivity maximum was observed also by Medina and Houze (2003) for the same region, and by Caracena et al. (1979) for the Big Thompson flood in Colorado Vertical structure over the Maggia catchment In this section, the vertical structure of the storms over the Maggia catchment is investigated by making use of contour frequency by altitude diagrams (CFADs). The focus is now only on the reflectivity distribution with altitude, rather than on both horizontal and vertical structure of reflectivity as in the previous section. Figure 17 shows CFADs calculated for each of the convective and non-convective events. The frequencies below 2 km should be interpreted with caution as they are affected by radar visibility issues. The CFADs of convective events generally have higher reflectivity values at every height than the non-convective cases, and they reveal the higher vertical development of precipitation than the non-convective ones. Moreover, the sharp spike in each of the CFADs between 2 and 3 km altitude, characteristic of bright-band contamination, are much more pronounced in the non-convective cases, indicating the prevalence of stratiform echoes. These spikes are also evident in the convective events, despite the greater vertical development of the echoes. Thus, stratiform precipitation is also a significant contributor to the precipitation during the convective cases. Since the CFADs in Figure 17 include reflectivity measurements over the entire Maggia catchment, they capture many of the stratiform echoes that have been shown to be more prevalent in the upper part of the basin (Figure 14). Consequently, the CFADs of the convective cases likely would have been more convective in nature if they had been limited to just the lower part of the catchment. The CFADs of the convective (non-convective) events shown in Figure 17 resemble the summer (autumn and spring) season CFADs of Rudolph and Friedrich (2013).

18 Orographic Convection in the Lago Maggiore Area 241 Figure 16. Average reflectivity along the parallel cross-section of Figure 2 for each convective and non-convective precipitation event. The orography along the cross-section is also shown: grey shading represents the terrain elevation averaged in the direction of the short axis of the transect, and black shading represents the maximum elevation Sensitivity of precipitation vertical structure to environmental variables The sensitivity of precipitation vertical structure to both dynamic and thermodynamic conditions of the impinging flow is investigated in this section. Cross-sections of average reflectivity for given intervals of LLF intensity are presented in Figure 18. The orientation of the cross-sections is in the direction parallel to the predominant direction of the LLF. Since the dominant LLF in convective cases is from the south, as shown in Figure 7, we present the convective storm vertical structure depending on LLF intensity along the north south cross-section in Figure 2, which extends from the Po Valley south of Lago Maggiore to the upper part of the Maggia catchment. For the non-convective cases, we make use of the perpendicular cross-section of Figure 2, since the LLF is mainly from the southeast in those precipitation events. This cross-section extends from the area of Milano to the Alpine crest west of the Toce Valley, yet it does not enter into the Maggia catchment for two reasons. First, in order to be oriented from c 2014 Royal Meteorological Society the southeast to the northwest, it would need to begin too far east, from outside the Lago Maggiore area. Second, its location is very similar to the cross-sections adopted to study MAP SOP precipitation cases in previous articles, such as Medina and Houze (2003) and Bousquet and Smull (2003). We remind the reader that the LLF is estimated every 5 min by Doppler velocity radar measurements, and it represents the wind between 1.5 and 2 km over the first slopes of the Alps south of the Maggia catchment (Figure 10 in PG2010). In the text boxes on Figure 18 we report the number of 5 min radar volumes and convective cells populating each cross-section, the mean rain rate and the average velocity of the cells. Figure 18 shows clearly that average reflectivity increases at all altitudes with increasing LLF intensity. However, convective cases have some exceptions: when increasing from moderate to strong LLF, the reflectivity increases mainly at altitudes below about 6 km. Moreover, in the first 20 km of the transect over the Po Valley, it actually decreases slightly with increasing LLF intensity. For all flow speeds, the altitude where the maximum mean reflectivity occurs is km for the convective cases, below Q. J. R. Meteorol. Soc. 141: (2015)

19 242 L. Panziera et al. Figure 17. Contour frequency by altitude diagrams (CFADs) for each precipitation event during convective and non-convective cases. the freezing-level height. Figure 18 also reveals that the number and the intensity of convective cells increases with increasing LLF intensity, for both convective and non-convective cases. For convective cases, the ratio of observed convective cells to the number of samples is about 0.2 with weak LLF, increasing to 0.85 for moderate LLF, and up to 1.25 with strong LLF. For non-convective cases, these ratios are much lower, ranging from about 0.10 for weak and moderate flow to 0.17 for strong flow. The intensity of the convective cells, which is represented by the average surface rain rate for each cell, also tends to increase with increasing wind speed, except for convective cases with moderate winds where the mean rain rate per cell is slightly lower than for weak winds. One factor that probably contributes to the direct relationship between reflectivity and upstream flow

20 Orographic Convection in the Lago Maggiore Area 243 Figure 18. Cross-sections of average reflectivity for three classes of low-level flow intensity for convective cases along the north south cross-section in Figure 2 and non-convective cases along the parallel cross-section in Figure 2. The orography along the cross-section is also shown: grey shading represents the terrain elevation averaged in the direction of the short axis of the transect, and black shading represents maximum elevation. The number of samples, as well as the number of convective cells, their average rain rate and mean velocity are also shown for each cross-section. speed observed in the vertical sections is the presence of a larger number of more intense convective cells when the low-level flow is stronger. The average velocity of the cells, which is also reported in Figure 18, is generally proportional to the LLF intensity, even if it is much lower than that. Composites of average rain rate at the ground depending on the LLF speed were also produced. However, they are not shown here since they corroborate the findings of PG2010. Specifically, both of these studies find that the mean rain rates increase significantly over the Lago Maggiore region with LLF intensity, whereas the direction of the upstream flow regulates the spatial distribution of the rainfall over the orography. This occurs for both convective and non-convective precipitation events. Composites of average rain rates and CFADs depending on wind shear direction and intensity did not show particular signals. In this study the vertical structure of the storms was also investigated in relation to the static stability upstream from the Alps. The latter was characterized by moist Brunt Väisälä frequency and CAPE derived from Milano soundings. For given intervals of moist static stability and CAPE, we produced CFADs of reflectivity measured over the Maggia catchment in the 5 h following the launch of the sounding. Moist static stability was averaged over different 500 m deep layers, from the ground to 3 km, and CFADs were created for stable, unstable and neutral stratification. The results obtained when computing the mean static stability in the layer km are shown in Figure 19. We see that CFADs for unstable conditions depict storms of a more convective nature with respect to neutral and stable stratification: reflectivity values are observed at higher and higher altitudes as static stability decreases. When computing mean static stability in other layers, the signal is not so clear. Mean rain rates at the ground corresponding to the different stability classes were also computed (not shown), and it clearly appeared that precipitation tends to be more intense with unstable conditions than with neutral and stable ones. A similar analysis was conducted by PG2010 for a much larger dataset, comprising 106 days of orographic rainfall. They found that precipitation intensity over the Lago Maggiore area is slightly more intense with unstable than with stable conditions, a result that is confirmed by this study. However, the effect of moist static stability on precipitation is weaker than that given by upstream flow intensity and direction. The stability of the atmosphere was also estimated by CAPE, computed by lifting the air parcel from different 500 m deep layers (as shown in Table 2). Figure 20 presents the CFADs obtained for three different ranges of CAPE, computed by lifting the air parcel originating in the km layer, which look very similar to those obtained when computing CAPE by lifting the air parcel from the km layer. Figure 20 illustrates the deepening of the precipitation structure as CAPE increases from low (<100 J kg 1 )tohigh(>1000 J kg 1 ). The moderate and high CAPE values depict a more convective precipitation structure than with low CAPE, as evidenced by a higher frequency of occurrence of high reflectivity at all levels and greater vertical development. It is interesting to note that CFADs corresponding to CAPE calculated for parcels originating at the km layer (not shown) produced only a weak signal. Mean rainfall rates at the ground for the same CAPE classes were computed (not shown), and it was found that precipitation is more intense with large values of CAPE tahn with low and moderate values. In this section we showed that precipitation vertical structure over the Maggia catchment is sensitive to the moist static stability and CAPE of the lower layers, in particular the km layer. This is consistent with the results of the correlation coefficient analysis presented in section 5.2.3, where it was shown that the layer whose mean wind velocity correlates best with rainfall in the Maggia catchment is km. The lifting of this layer was thought to be the most efficient in producing precipitation over the Maggia catchment. This section confirms this finding, since the average stability of this layer, and CAPE computed by lifting the air parcel from this layer, give a clear indication of the potential for convection over the Maggia catchment. 7. Conceptual model Figure 21 proposes a conceptual model for the orographic precipitation mechanisms which lead to the most intense flash floods in

21 244 L. Panziera et al. (c) Figure 19. Contour frequency by altitude diagrams for both convective and non-convective precipitation events for different classes of moist static stability derived from the Milano sounding. The mean squared moist Brunt Väisälä frequency (Nm 2 ) of the layer m was used to classify the soundings as stable (Nm 2 > s 2 ), neutral ( 0.2 < Nm 2 < s 2 ) and (c) unstable (Nm 2 < s 2 ). (c) Figure 20. Contour frequency by altitude diagrams for both convective and non-convective precipitation events for low (< 100 J kg 1 ), moderate (100 to 1000 J kg 1 ), and (c) high (> 1000 J kg 1 ) CAPE derived from the Milano sounding. CAPE was computed by lifting an air parcel from 500 to 1000 m. the Maggia catchment and more generally in the Lago Maggiore region. The observational study on the structure of the convective storms presented here constitutes the basis for the figure. The precipitation events that cause the largest peak discharge rates of the Maggia river are characterized by the presence of convective precipitation, as shown in Table 1 and throughout the article. These events occur when an upper-level trough approaches the Alps and creates surface troughing. A surface cold frontal zone develops in association with the upper trough, and a strong temperature gradient sets up with cold air on the western and northern side of the Alps and warm, moist air over the Po Valley (Figure 4). The red streamline in Figure 21 indicates the southsoutheasterly low-level flow observed over the Po Valley (Figures 5 and 7), which is frequently associated with a LLJ (Figure 6 and Table 2). The blue streamline represents the southwesterly flow observed at mid and upper levels (Figures 5 and 7). Thus, the potentially and conditionally unstable air which is observed at low levels in the Po Valley (Figure 5 and Table 2) is advected towards the northnorthwest into the Lago Maggiore region, where it is forced to ascend over the first Alpine peaks and reaches saturation. The sufficient CAPE, low CIN, and strong upslope orographic lifting create efficient convective cell development over the terrain just west of Lago Maggiore (Figure 15). The convective cells are then transported towards the northeast by the upper-level southwesterly wind, reaching maturity over the Maggia catchment and then disappearing over the interior of the Alps northeast of Lago Maggiore. Since the upper-level steering flow coincides with the southwest to northeast orientation of the orographic feature that most frequently initiates convection, the convective cells are naturally aligned into echo trains that repeatedly advect across the Maggia catchment, where they contribute greatly to the total rainfall, as shown in Figure 14. Raindrops in convective cells grow principally by coalescence, as shown in the vertical cross-sections of Figure 16 where the reflectivity is a maximum at low levels generally below the freezing-level height. The continuous advection of convective cells over the same area is the main reason for the peak rainfall accumulations and the high mean rain rates in the elliptical region

22 Orographic Convection in the Lago Maggiore Area 245 Figure 21. Conceptual model of the echo training process by which mountains trigger and maintain prolonged moist convection over the same region. The red streamline indicates warm, moist, convectively unstable low-level flow. The blue streamline indicates the upper-level steering flow. This flow configuration frequently results in flash floods in the Maggia catchment and in the Lago Maggiore area. observed for all the convective events, which extends roughly from the Toce Valley to the Verzasca Valley northwest of Lago Maggiore as shown in Figures 10 and 11. The convective origin of the heavy rainfall observed in this area is clearly apparent in Figure 14. To the lee of this region precipitation amounts are also quite large, and they are mostly produced by dissipating convective cells. The vertical development of the storms is directly related to the magnitude of the low-level flow impinging on the Alps and to the CAPE and moist static stability of the lower atmospheric layers upstream (Figures 18 20). The number and the intensity of convective cells are also proportional to the low-level flow speed. The schematic diagram (Figure 21) represents a process by which mountains can trigger and maintain moist convection over the same region. This process is due to the advection of warm, moist potentially unstable air towards the barrier at low levels, and to the different directions of the lower- and upperlevel wind, with the steering flow being parallel to the barrier. With respect to conceptual models that explain orographic convection over other mountain ranges, the mechanism depicted in Figure 21 emphasizes the fundamental role of the steering flow in determining the continuous advection of convective cells over the same area. The schematic of Figure 21 shows that flash flooding over the orography may also occur with strong upperlevel flow, in addition to the weak upper-level flow cases observed, for example, in both the Big Thompson and Black Hills storms (e.g. Maddox et al., 1978; Caracena et al., 1979). The extent to which the flow configuration depicted in Figure 21 resembles the mechanisms presented in Figure 16 of Rotunno and Ferretti (2001) for the Piedmont flood remains to be investigated. However, the precipitation events in this study seem to be characterized by a more southerly, moist and unstable lowlevel flow compared to the easterly blocked airflow presented in Rotunno and Ferretti (2001). The mechanisms represented by Figure 21 complement and add details to the conceptual model for unstable unblocked flow proposed in Figure 17 of Medina and Houze (2003) for the Lago Maggiore area. 8. Comparison with MAP cases Since the Lago Maggiore region was chosen as the target area for studying precipitation during the MAP (Bougeault et al., 2001; Rotunno and Houze, 2007), in this section we compare the precipitation events of this study with the MAP SOP cases, in order to understand whether the MAP cases are representative of the rainfall events that produce floods in the Maggia Valley and, more generally, over the entire Lago Maggiore area. Table 3 reports the peak discharge rates of the Maggia river for the events of this study and those of the MAP SOP. As shown in the table, MAP Table 3. Peak discharge rates of the Maggia river for the precipitation events presented in this study and for the MAP SOP cases. Peak discharge rate (m 3 s 1 ) Event 2258 C C C C MAP IOP 2b 1216 C C C7 997 C8 992 MAP IOP C9 775 NC1 750 NC2 708 NC3 684 NC4 605 NC5 555 MAP IOP NC6 533 NC7 516 NC8 516 NC9... Other MAP IOPs 90 MAP IOP 8