A RADAR-based climatology of convective activity in the Veneto region

Transcription

1 A RADAR-based climatology of convective activity in the Veneto region Michela CALZA, Alberto DALLA FONTANA, Francesco DOMENICHINI, Marco MONAI, Andrea M. ROSSA Regional Agency for Environmental Protection of Veneto, Meteorological Center of Teolo, Italy Contact:

2 INTERREG IIIB FORALPS 2

3 Technical Reports Abstract Convection is one of the most important meteorological phenomena during the warm season in northern Italy. Thermal convection is relevant close to mountains; moreover, potential instability along the Po Valley can generate several convective phenomena, sometimes associated with severe weather. The unique capability of weather RADARs to monitor precipitation with high spatial and temporal resolution is a well-known feature that allows a more detailed study of this kind of phenomena. A specific tool for weather RADAR data analyses was applied by ARPAV, focusing on the study of convection in the Veneto Region (north-eastern Italy). The Storm Cell Identification and Tracking (SCIT) algorithm has been exploited to construct a detailed climatology of convection over the domain of Mt. Grande RADAR, managed by the Meteorological Center of Teolo (CMT). Cells identified by the algorithm were catalogued and referenced in space and time; a cell density function was also derived. An off-line version of the SCIT algorithm has been implemented to collect and archive data in a systematic way. A flexible web-based analysis tool has been devised to inquire the SCIT database according to cell attributes. This tool allows the user to extract cells for selected periods of time and stratify them according to one or several of the about 40 parameters of SCIT. RADAR volumes for the warm seasons 2005, 2006 and 2007 were analyzed to document the convective activity in terms of the cell density, i.e. the number of cells per unit area. On the overall, more than cell identifications were recorded. Preferred times of the day, geographical distribution, dependence from the month and tracks of convective storms were identified by mean of the SCIT. For example, the province of Vicenza, north west of the RADAR, was identified as the area with the highest frequency of convective activity. This area was hit 3 cells/km2/3yr cells against a value of 1 cells/km2/3yr relatively to the entire RADAR domain. The area with the least convective activity turned out to be the province of Rovigo in the south. A relative maximum in the overall cell density map has been found west of the Lake of Garda with 1.5 cells/km2/3yr. This finding confirms that the Lake Garda FORALPS target area is a preferred region for convective activity; an overall maximum has been found for this area for hailproducing. It should also be noted that convective activity in the Garda region is likely to be underestimated, due to the beam-blocking exerted by the mountain barriers north of Verona and Brescia and to beam height increasing with distance from the RADAR. RADAR data were used to carry out a preliminary verification of the convective component of the COSMO LAMI quantitative precipitation forecast (QPF) for the warm seasons (May-Sep) 2005, 2006 and As SCIT does not record quantitative precipitation estimates (QPE), the reflectivity information associated with the cells was converted to rain using a convective Z-R relation. This should give a rough estimate of the rainfall amounts produced by convection and be indicative for the rainfall distribution. This SCIT-derived QPE was compared with the convective component of the LAMI QPF and, for reference, with the rain gauge accumulations of the CMT observing network. An eye-ball verification revealed that the QPF maxima are located mostly in the Alpine and pre-alpine areas of the Veneto region, whereas the minima are observed on the southern plains; this general distribution is in good agreement with the QPE derived from SCIT and the rain gauge network. SCIT-derived QPE maps agree with the rain gauge accumulations of the CMT observing network, at least regarding the geographical position of the maximum amounts of precipitation. 3

4 INTERREG IIIB FORALPS 4

5 Technical Reports Contents 1 Introduction 7 2 Methodology The SCIT Algorithm The Mt Grande RADAR archive Construction of a convection database The numerical weather prediction model COSMO LAMI 12 3 Operational results Structural overview of the analysis software Graphical user interface Products 17 4 Scientific results Sensitivity to and tuning of the reflectivity scale filter Verification of the SCIT detection capability Building a climatology of convective activity Stratification with respect to various parameters Evaluation of the COSMO LAMI convective precipitation 30 5 Summary and conclusions 33 6 References 44 7 Acronyms 44 5

6 INTERREG IIIB FORALPS 6

7 Technical Reports 1 Introduction An improvement of warning methods based on measurements from weather RADARs can provide an effective tool for a wise management of water resources in the Alpine Space, in particular for what concerns the identification of areas prone to meteorological and hydrological extremes. The occurrence of such extremes is often related to the onset of convective activity. In fact, convective activity is frequent during the summer season along the Po Valley, where the atmosphere is potentially unstable most of the time, also during the night. When a dynamical forcing is acting, typically a baroclinic wave, it is quite likely that convective overturning will occur. Whereas systematic studies exist that investigated the frequency of particular threats connected to severe convection, like hail and tornadoes (Giaiotti et a.,2007, Tuovinen et a., 2006), this is not the case for the observation of the single convective cells, either isolated or as part of a larger system. Several studies have been carried out that describe the areal distribution of rainfall in the region of Veneto, but the same cannot be stated concerning convective activity. It is well known to forecasters that, during the summer, convection is frequent especially over the mountains and the foothills. For instance, the target area of the lake of Garda is frequently hit by severe storms coming from the west and propagating eastward. Hail, sometime severe, is not unusual over this area as well as all along the foothills to the east. However, an objective identification of the areas more subject to storms and severe storms is still missing. Figure 1. Cell density map with the condition probability of hail 75% for the warm seasons 2005, 2006 and 2007 performed on a lat/lon grid with a mesh size of 0.1 x 0.1 (8x11 km). The color scale is in percent of the maximum number of cells detected in one grid point (Nmax, reported in the upper right). 7

8 INTERREG IIIB FORALPS At least partially, this gap can be filled by means of imagery from the RADAR of the Meteorological Centre of Teolo (hereafter CMT), whose operational installation dates back to the late eighties. In fact, CMT has recently adopted the Hydromet Decision Support System (HDSS) ( an advanced RADAR-based decision support system, conceived as an integrated tool to extract the maximum information from weather RADAR data. The SCIT (Storm Cell Identification and Tracking) algorithm, an important part of the HDSS system, is an advanced tool for real-time identification and tracking of storm cells. The main work of this study consisted in the construction of an off-line version of SCIT which permitted to reprocess archived RADAR volumes. In addition, a web-based analysis tool was set up to allow a systematic evaluation of numerous SCIT parameters. This tool has been used to extract the position of each cell, to compute the number of cells per unit area, and finally to build a cell-density function over the RADAR domain. The analysis has been applied to the years 2005/2007, obtaining a glimpse of a climatology of convective cells. Such a climatology can be stratified according to a set of criteria based on different cell attributes. Histograms showing the number of cells as function of the maximum reflectivity, probability of hail, probability of severe hail as well as of other parameters have been produced. This procedure allows to complement the information about the distribution of convective activity with an identification of the areas more prone to the severe threats associated with convection. Besides the geographic position, the direction and the speed of movement of each cell can also be extracted from the SCIT database. In this way the algorithm can serve as a monitoring tool as well as a predictor of the storm movement, making it an ideal companion of numerical models in performing nowcasting activity. Furthermore, precipitation maps based upon the cell density distribution can be computed and compared with outputs from the COSMO LAMI model (available at CMT) and with measurements from the rain gauges network. Comparison of the model output with RADARbased estimates can provide an alternative and, under some aspects, better way to verify the model output for convective precipitation, since small scale storms can at times be totally missed by the ground stations network, even if supplying a conspicuous amount of precipitation. The SCIT analysis tool is presented in this report. Also, a number of sample analysis are included based on the analysed data set which includes the warm seasons of 2005, 2006 and Methodology This section describes the methodology we applied to construct an inventory of convective activity in the domain under surveillance by the CMT RADAR of Mt. Grande. The inventory is essentially meant as a basis for the validation of quantitative precipitation forecasts (QPF) from Numerical Weather Prediction (NWP) models. A description of the Storm Cell Identification and Tracking Algorithm is first presented, followed by a description of the Mt. Grande historical RADAR data archive, and of the application used to construct a database of convective events. Finally, a short description of the NWM Model COSMO LAMI is provided. 8

9 Technical Reports 2.1 The SCIT Algorithm The SCIT algorithm, described in detail in Johnson et al (1998), implements a 3D centroid identification and tracking procedure. It processes volumetric reflectivity information from RADAR data, on a radial by radial basis. Three dimensional storm identification is performed in subsequent stages. First, storm segments are identified in the radial data. This process is repeated using seven different reflectivity thresholds (30, 35, 40, 45, 50, 55, 60 dbz). Then, individual segments are combined into 2D storm components, based on spatial proximity. Figure 2 shows an example of how the centroid delineation of the SCIT algorithm is performed. Figure 2. Example of Storm Cell Tracker processing. The figure shows that a search is made in the radial direction for gates of common reflectivity. These gates are then built azimuthally into 2D features (From Johnson et a., 1988). Figure 3. Example of Storm Cell Tracker processing. The figure shows how 2D features described in Figure 2 are correlated in the vertical to produce 3D storm centroid locations (From Johnson et a., 1988). 9

10 INTERREG IIIB FORALPS As shown in Figure 3, once all reflectivity segments are grouped in 2D in the horizontal plane, a vertical continuity search is performed. Given some reflectivity and distance constraints, the 2D features are correlated in the vertical to provide the location of storm centroid in a 3D reference system. Storm cells identified in two consecutive volume scans (10 minutes apart) are associated temporally to determine the cell track. Figure 4 shows an example of cell identification and tracking. Each circle along the track corresponds to the position of the cell at each time step. Tracking is performed by applying a weighted least squares fit once a storm is identified for two consecutive volume scans. A higher weight is given to the latest storm position in order to account for radical changes in the storm direction, while still not allowing for unreasonable storm motion. Since individual storms, rather than storm systems, are detected by the SCIT algorithm, individual storm cells may be tracked. In addition, individual storm cell characteristics, such as cell-based Vertically Integrated Liquid water equivalent (VIL), may be determined and trended. Thus, the evolution of cell-based characteristics is available to the forecaster. Short-term storm cell movement forecasts are also determined. As an example, in Figure 4 the circles marking the centroid positions and the SCIT tracks have a color code based on the value of the associated Maximum Reflectivity (white dbz, grey dbz, dark grey dbz, violet 45-50dBZ, red dbz, black dbz, yellow > 70 dbz). Figure 4. Example of cell identification and tracking. Each circle along the tracks corresponds to the position of the cells at each time step. The SCIT tracks have a colour based on the value of the associated Maximum Reflectivity (white dbz, grey dbz, dark grey dbz, violet 45-50dBZ, red dbz, black dbz, yellow > 70 dbz). 10

11 Technical Reports Table 1. List of the attributes available in the SCIT cell table. StationID RADAR name DateTime date and time CellID Id of the cell Latitude/Longitude lat/lon of storm (deg) 3DXLoc X location of 3D detection in km from RADAR (km) 3DYLoc Y location of 3D detection in km from RADAR (km) AgeHours/AgeMin/AgeSec how long the cell has been tracked AgeVS number of volume scans storm has been tracked with AlgRank rank of cell in cell table Azimuth azimuth from the RADAR center of the cell location (deg) CellBase lowest detection level of the cell (km) CellMass total reflectivity weighted mass of the cell (kilograms) CellTop highest detection level of the cell (km) CellVolume overall volumetric extent of the cell (cubic meters) ConvDepth total depth of convergence (km) CoreAspectRatio the ratio of depth to depth of the cell FcstError estimated forecast error of SCIT track (km) HailSizeEstimate hail size (in) HeightOfCenterOfMass height of the center of the cell mass (km) HeightOfMaxdBZ height of the max dbz found in the cell (km) LowLevelXLoc x location from the RADAR of lowest level reflectivity of the cell (km) LowLevelYLoc y location from the RADAR of lowest level reflectivity of the cell (km) MaxConvergence maximum detected convergence in storm (m/s) MaxdBZ value of maximum reflectivity with storm (dbz) MesoDetectionType TVS (Tornado Vortex Signature), non-tvs ProbOfHail probability of hail (%) ProbOfSeverehail probability of severe hail (%) Range range from RADAR center (km) RowName position in storm cell array of storm SevereHailIndex HDA index of storm (dimensionless) StormDir direction from which the storm is moving (deg) StormSpeed speed of the storm (m/s) StormTopDiv the amount of divergence of the storm (m/s) UMotion u component of storm motion (m/s) VIL highest column vertically integrated liquid water (kg/m**2) VMotion v component of the storm (m/s) 11

12 INTERREG IIIB FORALPS 2.2 The Mt Grande RADAR archive The Meteorological Centre of Teolo has a relatively long tradition in using RADAR for monitoring precipitation. Its RADAR data archive dates back to It is important to point out that only exploiting the CMT s historical RADAR data archive, it is possible to obtain a significant climatology along with a probability of occurence of storm cells. Unfortunately, the format of the RADAR volumes experienced a change through the years due to a renewal of the RADAR software and hardware. The main issues affecting the RADAR data archive can be summarized as follow: RADAR data older than May 2003 have a different format with respect to the current format. A software to convert RADAR data of past seasons to the SCIT format was developed and must be tested. The corrected reflectivity, i.e. the reflectivity after processing by clutter-filtering algorithms, on RADAR data of the year 2003 and 2004 has a bug due to a wrong calibration of the Doppler chain. This implies that the RADAR volumes are not suitable for SCIT processing. Processing the uncorrected reflectivity instead of the corrected reflectivity could be a solution. As a consequence of these two limitations, our storm cell climatology is currently limited to the years 2005/ Construction of a convection database The SCIT capability to identify individual storm cells and to determine the storm cells characteristics has been used to construct a database of convective activity in the RADAR domain. More precisely, the records of the SCIT algorithm, updated in real time as the RADAR system of Mt. Grande monitors the weather, have been organized in a SCIT database. Numerous storm attributes have been derived and tracked with the SCIT algorithm, as Maximum Reflectivity, Probability of Hail, Circulation Type etc. The fields of CMT s convection database include all the attributes available in the cell table. These attributes are reported in Table The numerical weather prediction model COSMO LAMI The NWP QPFs used in this study are taken from the COSMO LAMI, the Italian version of a model developed within the Consortium for Small Scale Modelling (COSMO) and deployed operationally at five European National Weather Services. COSMO LAMI is operationally run by ARPA-SIM on a rotated latitude-longitude grid with a mesh size of 7km and 35 vertical levels. The model solves the three-dimensional, fully elastic and nonhydrostatic atmospheric equations on an Arakawa-C grid using the split-explicit technique described by Klemp and Wilhelmson (1978). Prognostic variables include the three Cartesian velocity components (u, v and w), temperature (T), pressure perturbation (p') and mass fractions of water vapour (q v ) and cloud water (q c ). Vertical subgrid turbulence is parameterised following Mellor and Yamada (1982) and the surface flux formulation is based on bulk relationships using specifying a roughness length and drag coefficients for turbulent momentum and heat exchange between the atmosphere the ground. For a more complete description of the model, refer to Doms and Schättler (2002) and to Steppeler et al. (2003). 12

13 Technical Reports The parameterisation for grid-scale stratiform precipitation accounts for four categories of water (water vapour, cloud water, rain and snow), while deep convection is parameterised following the Tiedtke mass flux scheme. The latter is used to remove convective instability from the model atmosphere, because the spatial resolution of the model does not allow the explicit simulation of the corresponding circulation. Therefore, precipitation forecasts from the NWP model pertain to four different categories, namely stratiform (grid-scale) and convective (sub-grid-scale) rain and snow respectively. The resulting precipitation should therefore be an indicator for the amount and, especially, the distribution of the rainfall produced by deep convection. 3 Operational results This section describes the operational results achieved during the implementation of the SCIT algorithm. A flow diagram describing the subsequent stages of the analysis software is presented. A section is devoted to the detailed description of the graphical user interface implemented to inquire the database of cells attributes. The final section presents the graphical outputs available, including interactive density and trajectories maps, and histograms useful for a systematic analysis of convective activity in the RADAR domain. 3.1 Structural overview of the analysis software A stuctural overview of the analysis software is shown in Figure 5. As already mentioned, while the real time RADAR system of Mt. Grande monitors the weather the volumetric reflectivity is processed by the SCIT algorithm; moreover, an offline version of the SCIT algorithm allows to exploit CMT s RADAR data archive. The records of the SCIT output were organized in a mysql database. The set of attributes derived and tracked with the SCIT algorithm (about 40) characterizes the cell in terms of, storm speed and direction, maximum reflectivity and height of the maximum, probability of hail and severe hail, circulation type, a mass, volume, top and base of the cell, etc. (see section 2.3 for a short description of all the parameters managed by the SCIT). This allows to stratify all the cells, according to a number of criteria. A web interface for data Volumetric Polar Data Realtime HDSS / SCIT Offline SYSTEMATIC THUNDERSTORM CLIMATOLOGY Histogram Web Interface Contour Map SCIT DB for SCIT db queries Density Map Trajectories Map ARCHIVE Figure 5. General flow diagram of the developed analysis software. 13

14 INTERREG IIIB FORALPS extraction realized in PHP has been created to allow the user to select cells identified by the SCIT with specific characteristics for selected periods of time. The graphical outputs include interactive density and trajectories maps, and histograms useful for a systematic analysis of convective activity in the RADAR domain (see section 3.2 for a description of all the functionalities of the data extraction form developed). All graphical products are displayed in the HDSS WxScope web interface. The XML-based WxScope Plugin uses the WeatherObjects Markup Language (WxML) as the basis for its map and graph products. The main features of this plugin include zooming and panning capabilities, as well as other common GIS functionalities like overlapping of high resolution customized geographical layers. All these features are inherited for the display of the SCIT Database-derived products. 3.2 Graphical user interface A SCIT web interface for data extraction has been created to enable the user to select the convective cells identified by the SCIT, based on user-specified characteristics and periods of time. The data extraction form contains five main sections: 1) TIME, 2) LOCATION, 3) PARAMETER, DURATION AND PERIOD, 4) GRAPHIC OUTPUT. A detailed description of each section is provided below: 1. TIME: in this section the user can select the period of time to be analyzed. It is possible to specify the initial and final date and time (hour and minutes) of the period, and/or make a single or multiple selection of the months and years. A select element is also provided here that gives the user the possibility to select a specific decade of the month rather than the entire month. 2. LOCATION: in this section the user can specify the spatial range to consider for selection of the convective cells. More precisely, it is possible to select the entire RADAR range or a specific circular sub area. Latitudine and longitude of the circle center in decimal degrees or in degrees/minutes/seconds, as well as the circle radius in kilometres can be set. The check box is used to select the cells generated within or moving through the selected area. 3. PARAMETER, DURATION AND PERIOD: the purpose of this section is to allow the user to stratify all the cells according to one or several of the 40 parameters managed by SCIT (see section 3.3 for a complete list and description of all the SCIT parameters). For instance, one can select for the entire month of July 2005 the cells which at some stage of their lifetime exhibit a maximum reflectivity larger than, say, 55dBZ. Moreover, this section gives the possibility to stratify all the cells according to other citeria, like the hour of genesis and the lifetimes. It is possible to select one or more of the predefined cell lifetimes (less than 1 hour, between 1 and 3 hours, greater than 3 hours) and time of genesis (night, from 00:00 to 05:00 UTC, morning, from 06:00 to 11:00 UTC, afternoon, from 12:00 to 17:00 UTC and evening, from 18:00 to 23:00 UTC). Alternatively, the user has the chance to insert specified lifetimes and time of genesis of the cells. 4. GRAPHIC OUTPUT: this includes two subsections: GRAPH TYPE and OBJECT. In the first one the user has to choose the type of graphic output to display: point maps, density 14

15 Technical Reports maps, and contour maps. Regarding the last ones, it is possible to specify the mesh size of the lat/lon grid used to perform the count of SCIT detections and also the color bar limits. The checkbox list provided in the subsection OBJECT enables the user to choose between three different ways to inquire and display the data: Trajectory (it considers the entire cell trajectory), Starting Point (it considers only the starting point of the cell trajectory), and Ending Point (it considers only the ending point of the cell trajectory); All the graphical outputs are available on a web browser and necessitate of the WxScope Plugin of the HDSS system. This makes the graphical output completely interactive, in that it is possible to turn on and turn off the different information available, just like the functionalities of a GIS. For example, you can highlight the cell characteristics that you prefer, like the probability of hail along the tracks, or the maximum reflectivity of a cell. 5. HISTOGRAM OUTPUT: in this section it is possible to realize some histograms, useful for a systematic analysis of convective activity in the RADAR domain. The user can select the period of interest and stratify all the cells according to one of the available parameters. The checkboxes allow the user to choose the type of histograms to be displayed on the web browser. The available histograms are: the number of SCIT first-time detections of cells as a function of the hour of the day; the number of SCIT detections as a function of the month; the number of SCIT detections as a function of the day; the number of SCIT detections as a function of a SCIT parameter. For each parameter a set of classes can be defined, and the number of cells within each class can be counted. Figure 6. Part of the extraction form of the SCIT web interface (sections TIME and LOCATION). In this example, cells from the whole June 2006 detected on the entire RADAR domain were selected. 15

16 INTERREG IIIB FORALPS Figure 7. Part of the extraction form of the SCIT web interface (section PARAMETER, DURATION AND PERIOD). In this example, cells which at some stage of their life time exhibit a maximum reflectivity greater than 50dBZ with duration less than 1 hour and generated in the morning were selected. Figure 8. Part of the extraction form of the SCIT web interface (section GRAPHIC OUTPUT). In this example, a density map of the entire cells trajectories was selected. 16

17 Technical Reports Figure 9. Part of the extraction form of the SCIT web interface (section HISTOGRAM OUTPUT). In this example, a histogram showing the number of SCIT detections with a probability of hail 60% as a function of the days for the sample period August 2007 was selected. 3.3 Products This section presents the graphical outputs available on the analysis software. including interactive density and trajectory maps, as well as histograms useful for a systematic analysis of convective activity in the RADAR domain Density maps of convective activity The SCIT cell identification capability can be used to obtain the storm cell density all over the RADAR domain. The panels in Figure 10 show the density maps of the number of first-time detections (i.e. the number of convective cells that are born) in specified six-hours intervals for the period May-September of the three years From top left to bottom right: from 0 to 6 AM, from 6 to 12 AM, from 12 to 18 PM and from 18 to 12 PM, respectively. The color scale is expressed in percent of the maximum number of cells detected in one grid point (Nmax, reported in the lower right). In accordance with the general knowledge we have of convective activity, the preferred areas of genesis appear to be the reliefs west and north-west of the RADAR site. Nmax reaches its maximum in the afternoon and its minimum in the first hours of the day. 17

18 INTERREG IIIB FORALPS Figure 10. Cell density map for the warm seasons 2005, 2006 and 2007 performed on a lat/lon grid with a mesh size of (8 11 km) for cells generated from 0 to 6 UTC, 6 to 12 UTC, 12 to 18 UTC, and 18 to 24 UTC (top left to bottom right). Details are provided in the text. Figure 11 (left) shows the SCIT cell density on a lat/lon grid for the month of August 2005, while Figure 11 (right) shows the same data but with a lat/lon grid. The impact of the two different mesh size values is visible: the high resolution grid allows to depict more precisely the peaks of convective activity. For each period of reference the cell density has been normalized, i.e.divided by the maximum number of cells detected into a grid element. The color scale represents the percentage of the maximum number of cells detected in one grid point (Nmax, reported in the lower right).this allows an easier visual comparison among maps of different cell densities. 18

19 Technical Reports Figure 11. Left: Cell density map for Aug 2005 performed on a lat/lon grid with a mesh size of (16 22 km). Right: same day, grid with a mesh size of (8 11km). Details are explained in the text Histograms The histogram in Figure 12 shows the number of first-time detections as a function of the hour of the day, integrated over the whole period (May-September ). Again in agreement with general observations about the onset of convection, a maximum is reached in the afternoon and a minimum in the first hours of the day. Only cells with lifetime greater than 60 minutes have been considered, to avoid short-lived cells that often arise in connection with organized convective systems. The histogram in Figure 13 shows instead the number of cells as function of the lifetime (age) estimated by the algorithm. Figure 12. Number of SCIT first-time detections of cells with duration 1hr as a function of the hour of the day for the sample period May-Sep

20 INTERREG IIIB FORALPS Figure 13. Number of SCIT detections as a function of the lifetime (age) for the sample period May-Sep Other examples This section shows some other examples of graphical outputs. More in detail, Figure 14 shows the cell trajectories (blue line) of cells with duration 2hrs, with starting points marked in green and ending points in red. With the zoom functionalities it is possible to obtain a map in which the single municipalities interested by the cell trajectory are shown. Further, details about the cell characteristics (time in blue, maximum reflectivity in red and probability of hail in yellow) are displayed along the track (Figure 15). Figure 14. Map of cell trajectories (blue line) with starting (green) and ending (red) points of 29 June 2005 for cells with duration 2hrs. 20

21 Technical Reports As a further example, a density map for cells generated within a range of 50km with Probability of Hail > 75% for the sample period May-Sep is shown in Figure 16. Finally, Figures 17 and 18 display histograms related to two different SCIT parameters. The first shows the number of SCIT detections as a function of the Vertically Integrated Liquid water equivalent (VIL) for the sample period May-Sep , while the second shows the number of SCIT detections as a function of the highest detection level of the cell (CellTop in kilometers) for the same period. Figure 15. Zoom on a cell trajectory (blue line) of 29 June 2005 with time (in blue), maximum reflectivity (in red) and probability of hail (in yellow) displayed along the track. Figure 16. Local density map for cells generated within a range of 50km with the conditions POH > 75% for the sample period May-Sep The color scale is in percent of the maximum number of cells detected in one grid point (Nmax, reported in the lower right). 21

22 INTERREG IIIB FORALPS Figure 17. Number of SCIT detections as a function of the Vertically Integrated Liquid water equivalent (VIL in kg/m 2 ) for the sample period May-Sep Figure 18. Number of SCIT detections as a function of the of the highest detection level of the cell (CellTop in Km) for the sample period May-Sep Scientific results While the previous chapters summarized the operational aspects of the implementation and use of the SCIT algorithm, this section describes the scientific results achieved through its use. First, a reflectivity scale filter, introduced to smooth out the fine resolution reflectivity peaks at close 22

23 Technical Reports ranges, is described. The impact of two different values of filter size is discussed. Then, the main outc omes of a verification of the SCIT detection capability in term of probability of detection (POD) are presented. Finally, the climatology of convective cells of the warm seasons 2005, 2006 and 2007 is illustrated by means of diagrams showing the distribution in space and time of the number of cells and of histograms showing the number of cells as function of the principal attributes as maximum reflectivity, probability of hail and of severe hail. 4.1 Sensitivity to and tuning of the reflectivity scale filter Raw RADAR data have a decreasing resolution along a radial. This means that a storm close to the RADAR is sampled with a higher detail than far away, leading to a greater number of cell detections by the SCIT algorithm at a close range. Thu s, before being passed on to SCIT, the data are filtered such that the fine resolution reflectivity peaks at close ranges are smoothed out while the peaks at farther ranges are retained. This task is accomplished using of a kernel of constant size (the "filter size"), centered on each data bin. Consequently, the number of azimuth bins within the kernel increases with decreasing range. The impact of two different values of "filter size" (5 and 3 km) has been evaluated in individual cases and on the entire data set. Figure 19 shows that the 3 km filter is able to identify many more cells than the 5 km filter, particularly in the mountainous region and at longer ranges (see white arrows in the right panel). Figure 20 confirms this idea and reveals that the 3 km filter detects roughly twice as many cells as the 5 km filter. Figure 19. Left panel shows SCIT analysis for 24 Aug :30 with filter size parameter value 3 km, whereas the right panel is for the same time with filter size 5 km. 23

24 INTERREG IIIB FORALPS Figure 20. Left panel shows SCIT cell density on a lat/lon grid for the month of July 2006 with a 3 km filter size, whereas the right panel is for the same month with a 5 km filter size. The color scale is in percent of the maximum number of cells detected in one grid point (Nmax, reported in the lower right). 4.2 Verification of the SCIT detection capability A brief validation of the SCIT algorithm in the Veneto region was performed overlaying the trajectory of the cells to the reflectivity images. RADAR volumes from the summer season 2005 (June-August) were used to verify the SCIT cell identification capability in detail. The Probability of Detection, defined as : POD = hits/(hits + missed alarms) has been computed as function of the cell intensity. Hits indicates the number of cells correctly identified by the algorithm and missed alarms indicates the number of cells missed by the algorithm but recognizable by a human operator. It has been found that, increasing the lowest reflectivity threshold for cell identification, the POD also increases since weak, low-topped cells can easily be missed by the algorithm. In fact, after setting the threshold to the lowest value, 30 dbz, the verification yielded a POD around 60%, whereas for severe hail-producing cells, identified by mean of the refunds paid by the National Solidarity Fund to Agriculture, a POD around 90% has been found. The cell tracking capability has not been examined in detail and with a systematic evaluation. However, the algorithm appears able to track organized convective systems, also for several hours. This point is important in a climatological study like ours, as it provides information about the preferred genesis regions for convective activity and about its growth and decay. Indeed, if a cell is not consistently tracked SCIT classifies a 'lost' cell as a newly generated cell. 24

25 Technical Reports 4.3 Building a climatology of convective activity SCIT detection counts were performed on a lat/lon grid with a mesh size of For each grid element the count of the total number of SCIT detections, i.e. the cell density, was performed for the warm season months (May-Sep) of years 2005, 2006 and For each period of reference the cell density has been normalized, i.e. divided by the maximum number of cells detected in one grid point (Nmax, reported in the lower right). The color scale of all figures is a function of the fraction of this number. This allows an easier visual comparison among maps of different cell densities Monthly and warm season density maps of convective activity The cell density maps for each month of the warm seasons are reported in Figure 38. The main remarks concerning the monthly convective activity can be summarized as follows: in the year 2005 convective activity spreads out in the whole region in August, but the Nmax reaches its maximum in September and is located on the foot hills north west of the RADAR. In July convection is confined to the mountains, while in the other months it influences extensively the plain; in the year 2006 convective activity occurs in the whole region in August, when Nmax also reaches its maximum, located close to the RADAR site. In June and July storm cells are mostly confined to the mountains. May and September show a prevalence of convection over the plains; in the year 2007 convective activity is widespread in the region during May and June. Nmax reaches its maximum in August on the mountains north west of the RADAR. May, June, July and August exhibit a prevalence of convection over the mountains, whereas in September storm cells interest predominantly the plains close to the sea. In Figure 21 the cell density maps for the period May-September of the years 2005, 2006 and 2007 are shown. The maxima of the distributions are located prevalently on the reliefs west and north west of the RADAR. In 2006 a maximum emerges over the plains very close to the RADAR. In the map from 2007 a large prominent maximum over the mountains north-west of the RADAR site appears and a small distinct peak also emerges on the northern foothills. The overall density map, also shown in Figure 21, highlights the province of Vicenza, north west of the RADAR, as the area with the highest frequency of convective activity. In terms of the number of cells per square kilometre this area was hit, in the three years period, by 3 cells against a value of 1cell/km 2 /3yr relative to the entire RADAR domain. The area with the least convective activity turned out to be the province of Rovigo in the south. A relative maximum in the overall cell density map has been found west of the Lake of Garda (highlighted by an orange circle) with 1.51cell/km 2 /3yr. This finding confirms that the Lake Garda FORALPS target area is a preferred region for convective activity: a global maximum has been found for this area for cells which were likely to produce hail (see Figures 20 and 21). However, this value is likely to underestimated due to the beam-blocking exerted by the mountain barriers north of Verona and Brescia and to the beam height increasing with distance from the RADAR. 25

26 INTERREG IIIB FORALPS Figure 21. Cell count on a lat/lon grid with a mesh size of The color scale is in percent of the maximum number of cells detected in one grid point (Nmax, reported in the lower right). Upper left: warm season (May-Sep) 2005; upper right: warm season 2006; lower left: warm season 2007; lower right: total counts Daily convective activity The daily convective activity for each month of the period May-September of the three years is reported in Figure Each diagram reports the number of detections per day in any part of the RADAR domain. June 2005 aside, unusually poor of storm cells, the convective activity is quite frequent and periods of prolonged activity are also common. 26

27 Technical Reports 4.4 Stratification with respect to various parameters In this section the climatology has been stratified according to different months of the warm season and to various cell attributes. This allows to obtain not only a distribution of convective activity but also an estimate of the risk of severe threats associated with convection. Histograms showing the number of cells as a function of the cell lifetime, maximum reflectivity, probability of hail and severe hail have been produced. The examples we provide here confirm that the capability to make a diversified query and a versatile graphic representation of the cell attributes is the real force of the analysis software we devised Months The histogram in Figure 22 shows the monthly variation of the convective activity for all the cells and for different duration of cells as tracked by SCIT (duration < 1h, 1-3 h, > 3h) for the convective months of the three years 2005,2006 and August turns out to be the month with the greater number of storm cells Lifetimes The histogram in Figure 23 shows the number of cells as function of the lifetime (age) estimated by the algorithm. The first class includes the cells detected by a single RADAR scan. They have been considered apart since they are due, at least partially, to wrong time associations by the algorithm. Remarkably, a cell with a lifetime exceeding 9 hours has been tracked (highlighted by an orange circle). Figure 22. Number of SCIT detections as a function of the individual warm season months and of different cell lifetime ( all duration in blue, < 1h in red, 1-3h in grey and > 3h in yellow) for the sample period May-Sep

28 INTERREG IIIB FORALPS Intensity The histogram in Figure 24 shows the number of cells as function of the maximum reflectivity for the whole period (May-September 2005,2006 and 2007). The reflectivity bins are 5 dbz wide. The first bin starts from 30 dbz since this is the first threshold used by the algorithm for cell identification. The distribution looks normal since most of the cells are in the central part, the most populated class being between 40 and 45 dbz. Figure 23. Number of SCIT detections as a function of the lifetime (age) for the sample period May-Sep Note that a cell with a lifetime exceeding 9 hours has been tracked (highlighted by an orange circle). Figure 24. Number of SCIT detections as a function of the maximum reflectivity for the sample period May-Sep

29 Technical Reports Probability of hail and severe hail The histogram in Figure 25 shows the number of cells as function of the Probability of Hail (POH) estimated by the algorithm. The POH is based on the well known the Waldvogel (1979) criterion: the higher the 45 dbz echo level above the freezing level, the higher the POH. The first bin includes cells with probability strictly equal to 0. Only cells satisfying the condition max{dbz} > 50 dbz have been considered, to take into account only the cells for which the occurrence of hail can be suspected. Without imposing this condition, the number of counts into the lower bins would be overwhelming. The final distribution looks very sharp, i.e. values close to the extremes of the range are more probable. However, it must be remarked that the POH is underestimated in the cone of silence that surrounds the RADAR site, where the height of the cells is also underestimated. This effect has not been quantified so far. Figure 25. Number of SCIT detections as a function of the probability of hail with the conditions Zmax 50 dbz for the sample period May-Sep Figure 26. Number of SCIT detections as a function of the probability of severe hail (>2 cm) with the conditions Zmax 55 dbz for the sample period May-Sep

30 INTERREG IIIB FORALPS The histogram in Figure 26 shows the number of cells as function of the Probability of Severe Hail (POSH) estimated by the algorithm. The POSH is defined for hail>2 cm in diameter and is found computing the hail kinetic energy parameter and using it in a weighting function that take into account the heights of the 0 and -20 C temperature levels. The first bin includes cells with probability strictly equal to 0. As before, the condition max{dbz} > 55 dbz has been imposed to take into account only cells for which severe hail can be suspected. First bin apart, most of the times the probability ranges between 25 and 50%. Only in very small percentage of cases the POSH exceeds the 75%. Like the POH and for the same reason the POSH is underestimated in the cone of silence that surrounds the RADAR site. 4.5 Evaluation of the COSMO LAMI convective precipitation A first comparison of the convective component of the COSMO LAMI quantitative precipitation forecast (QPF) for the warm seasons (May-Sep) 2005, 2006 and 2007 was performed. As SCIT does not record quantitative precipitation estimates (QPE), reflectivity information was converted to rainfall rates using a Marshall-Palmer convective Z-R relation (Z=aR^b, a=300, b=1.4). This should give a rough estimate of the rainfall amounts produced by convection and be indicative for the rainfall distribution. This was compared with the convective component of the LAMI QPF and, for reference, with the rain gauge accumulations of the CMT observing network. The main purpose of this comparison is first to evaluate the skill of the COSMO LAMI in the forecast of the convective component of precipitations, with a specific attention to the geographical distributions of seasonal maxima, and then a coarse evaluation of the overall accumulations. An eye-ball verification was performed between the precipitation maps derived from the SCIT algorithm, the COSMO LAMI convective QPF, and the rain gauges network. For the present work, the accumulations for each month of the warm season (May-September) of the years 2005, 2006 and 2007, the accumulations for each warm season and the total accumulation of the three warm seasons together (15 months) were analyzed. All these precipitation maps are reported in Figures The comparison between the SCIT-derived QPE maps with the rain gauge accumulations of the CMT observing network reveals a good agreement in the geographical position of the maximum amounts of precipitation. On the other hand, these amounts are often different. This problem can descend from differences between the two QPE methods, or from limits of the RADAR measurements, or again from the scattered distribution of rain gauges. In the case of a small storm, for instance, it is possible that the most severe part of it wasn t recorded by any of the rain gauges. On the other hand, the rain gauges record also stratiform precipitation, which is not identified by SCIT. In the periods we consider (summer season) stratiform precipitation events are a very small part of total amounts, except for the month of May; actually the comparison of SCIT-derived QPE with the rain gauge network underlines the different nature of events in the month of May. Moreover, the SCIT identification capability, giving information on a lat/lon grid, allows to map specific features in the spatial distribution of the mean convective activity, at a scale smaller than what is possible with the rain gauge network. An example is the channel of relatively low amount of convective rain, corresponding to low cell density, over the southern part of the Valsugana (see Figure 27), and not visible in the maps from the rain gauges. This effect is likely due to the specific morphology of the area. 30

31 Technical Reports Figure 27. Total SCIT-derived QPE for the warm seasons ( May-Sep) The red ellipse highlights a channel of low precipitation quantitative in the southern part of Valsugana. To summarize, the comparison with the accumulated values from the rain gauge network confirms the goodness of SCIT-derived convective precipitation, in the limits given by the difference of estimation methods. Cases of convective precipitations in the area of good visibility of RADAR give the best agreement. For the same periods (May-Sep , total amount for each warm season, total amount for the 3 warm seasons together) the COSMO LAMI maps of accumulated convective rain were considered. The 24-hours convective rain for each daily run of the model was considered. The QPE and QPF maps for the total period (sum of 3 warm seasons) are shown in Figure 28 (other maps are available in the Figures 36-38). May-September COSMO LAMI convective QPF SCIT-derived QPE Rain Gauge QPE Figure 28. Left panel show s the total convective component of COSMO LAMI QPF for the warm seasons (May-Sep) , the middle panel shows the SCIT- derived QPE and the right panel shows the rain gauge accumulation of the CMT rain gauge network for the same period 31

32 INTERREG IIIB FORALPS The main findings for the COSMO LAMI model convective rain are summarized in the following: The QPF maxima from the model are located prevalently in the Alpine and Prealpine areas of the Veneto region, whereas the minima are observed on the southern plains; this general distribution is in good agreement with QPE derived from SCIT data and from the rain gauge network. Most of the areas in which SCIT shows frequent convective activity present a high seasonal cumulation in the convective rain forecast of the model (see Figure 29). The orographic forcing over the Alpine areas appears quite clearly in the convective component of the COSMO LAMI model. This kind of forcing generates specific features in the distributions of the convective precipitation, showing maxima (over the total period of 15 months) up to mm, against mm in the vicinity. This kind of features is not well confirmed in the maps of rain gauge or SCIT-derived QPE (Figure 30). The Adriatic sea seems to give a further contribution to convective forcing in the model parametrization. This is clearly visible by observing the relative maximum in the convective component of rain located above the northern Adriatic sea (about mm in the whole period). Such frequent convective activity is not confirmed by the SCIT detections, while the rain gauges in this area are not available. In comparison with the SCIT-derived QPE and rain gauge network QPE maps, the COSMO LAMI model presents a flatter distribution of convective precipitation in the plain. This difference is quite evident in the central part of Veneto region, where the SCIT-derived and the rain gauge network QPEs indicate a more frequent incidence of convective phenomena with respect to the rest of the plains (total amount of mm), while the COSMO LAMI total amount is around to mm, very similar to the rest of the plain of Veneto region. Figure The left panel shows convective component of the COSMO LAMI QPF for the warm seasons (May-Sep) The right panel is the SCIT-derived QPE for the same period. The circles highlight the area of Prealps of Brescia (red) and Vicenza (brown).

33 Technical Reports Figure 30. The left panel shows the orography of the model COSMO LAMI for northern Veneto, with isolines every 50 meters. The right panel shows the same area with the COSMO LAMI convective QPF for the warm seasons May-Sep overlayed; the color scale starts from 1400 mm to allow a visual reference of the highest accumulations. 5 Summary and conclusions It is well-known that even a dense network of ground stations could not offer the opportunity of studying convection in detail. Only high resolution RADAR data can give proper information on such kind of phenomena, including 3-D features. Provided a significant (multi-seasonal) set of data is available and a user-friendly tool for elaboration is applied, a unique climatology of convection and related phenomena can be set up. The Storm Cell Identification and Tracking (SCIT) algorithm has been exploited to construct a detailed climatology of convection over the Teolo-Mt.Grande RADAR domain, roughly corresponding to the Veneto region in north-eastern Italy. Cells identified by the algorithm were accounted and referenced in space and time, and a cell density function was derived. An off-line version of the SCIT algorithm has been implemented to collect and archive data in a systematic way. A flexible web-based analysis tool has been devised to inquire the SCIT database according to cell attributes. This tool allows the user to select cells for selected periods of time and stratify them according to one or several of the about 40 parameters. RADAR volumes for the summer seasons 2005, 2006 and 2007 were analyzed to document the convective activity in terms of the presence of cells over the CMT Mt.Grande RADAR domain. In more detail, the preferred times of the day, geographical distribution, dependence from the month and tracks of convective storms were identified by mean of the SCIT. The main findings are summarized below. 33

34 INTERREG IIIB FORALPS 34 The developed analysis software is a valuable tool to construct a multi-year climatology of convective activity over the RADAR domain of Mt.Grande. The data extraction form is a versatile interface that allows to inquire the database of cells attributes, assigned by the SCIT algorithm. Several parameters can be conveniently displayed on easily understandable diagrams, and a detailed characterization of convection can be derived. A systematic verification of the cell identification capability performed over the period June-August 2005 led to results dependent on the choice of the algorithm (lowest reflectivity threshold for cell identification). Without posing restrictions to the cell intensity, the probability of detection (POD) results 60%; however for hail producing cells a score around 90% has been obtained. Hence, SCIT appears as a valuable tool to construct a climatology of convective cells in the RADAR domain. The high resolution of RADAR data, along with the multi-reflectivity threshold-based rule we used for cell identification, depicts the SCIT algorithm as an ideal tool to map the distribution of the single convective cells over the RADAR domain. The convective activity in Veneto was documented for the warm seasons 2005, 2006 and The overall density map of the warm seasons highlights the province of Vicenza, north west of the RADAR, as the area with the highest frequency of convective activity. In terms of the number of cells per square kilometre this area was hit, in the three years period, by 3 cells, against a value of 1cell/km 2 /3yr relatively to the entire RADAR domain. The area with the least frequent convective activity turned out to be the province of Rovigo, in the south. A relative maximum in the overall cell density map has been found west of the Lake of Garda with 1.5cell/km 2 /3yr. This finding confirms that the Lake Garda FORALPS target area is a preferred region for convective activity. Significant month-to-month variability can be detected. In particular, months with a predominance of thermal convection which affects mainly the mountainous parts of Veneto (e.g. Jul 2006) are distinctly different from months where a synoptic influence prevails (e.g. Sep 2006). August appears to be the month with highest convective activity. The time of the day with most pronounced convective activity seems to be in the late afternoon / evening. The minimum, on the other hand, corresponds to the morning hours. In order to overcome a cell detection bias in regions close to the RADAR, a spatial filter must be used. The size of the reflectivity scale filter used by SCIT needs to be chosen as a trade off between efficiency of detecting cells at longer ranges and over-counting cells close to the RADAR. For the present study a 3 km filter size was preferred to a 5 km filter size, because it performed more reliably on a number of cases examined and missed less cells at longer ranges (e.g. over the mountains to the northwest). However, the cell count with the 3 km filter size is roughly twice as high as with 5 km, hence it may still suffer from over-detection at closer ranges. This is consistent with the rather bell-shaped distribution of the two-warm season data set. A first comparison with the convective component of the COSMO LAMI quantitative precipitation forecast (QPF) for the warm seasons (May-Sep) 2005, 2006 and 2007 was performed. As SCIT does not record quantitative precipitation estimates (QPE), reflectivity information associated with the cells was converted to rain using a convective Z-R relation. This SCIT-derived QPE was compared with the convective component of the LAMI QPF and, for reference, with the rain gauge accumulations of the CMT

35 Technical Reports observing network. An eye-ball verification revealed that the QPF maxima are located prevalently in the Alpine and Prealpine areas of the Veneto region, whereas the minima are observed on the southern plains; this general distribution is in good agreement with QPE derived from SCIT and the rain gauge network. SCIT-derived QPE maps agree reasonably well with the rain gauge accumulations of the CMT observing network, at least about the geographical position of the maximum amounts of precipitation. The cell identification method we used suffers from some limitations. These be traced back to the limits of the RADAR measurements, i.e.: decreasing resolution with range: storms close to the RADAR tend to be detected as multiple distinct cells, as opposed to those that are far away. Smaller cells far from the RADAR can instead be missed. This positive bias is partially overcome by filtering the data in order to reduce resolution close to the RADAR; increasing beam height with range: low topped cells far from the RADAR can be missed; the cell densities have a negative bias increasing with range; beam blocking over the mountains: partially screened cells may not match the algorithm s rules for identification; this may result in a negative bias over the mountains. The cell tracking method still needs to be examined in detail; as mentioned before, cell tracking would be an important part of a climatological study of convection, and therefore needs to be studied in depth in the future. The next steps of this analysis also include: exploiting CMT s complete RADAR data archive (provided a quality control procedure is applied) to obtain a significant climatology along with a probability of occurrence of storm cells; evaluating the SCIT algorithm on cases stratified into categories as: isolated storms, Mesoscale Convective Systems (MCS), supercells and squall lines. 35

36 INTERREG IIIB FORALPS Figure

37 Technical Reports Figure 32.. Number of SCIT detections as a function of the day for the sample period of June 2005 (upper row), June 2006 (middle row), and June 2007 (third row). 37

38 INTERREG IIIB FORALPS Figure 33.. Number of SCIT detections as a function of the day for the sample period of July 2005 (upper row), July 2006 (middle row), and July 2007 (third row). 38

39 Technical Reports Figure 34.. Number of SCIT detections as a function of the day for the sample period of August 2005 (upper row), August 2006 (middle row), and August 2007 (third row). 39

40 INTERREG IIIB FORALPS Figure 35. Number of SCIT detections as a function of the day for the sample period of Sepetember 2005 (upper row), September 2006 (middle row), and September 2007 (third row). 40

41 Technical Reports Figure

42 INTERREG IIIB FORALPS Figure

43 Technical Reports Figure

44 INTERREG IIIB FORALPS 6 References Giaiotti, D. B., M. Giovannoni, A. Pucillo, and F. Stel, 2007: The Climatology of tornadoes and waterspouts in Italy. Atm. Res., 83, Tuovinen, J., J. Teittinen, A.-J.Punkka, and H. Hohti, 2006: A Climatology of large hail in Finland. Preprints, 32 nd Conf. on RADAR Met., AMS, Albuquerque. Johnson, J. T., P. L. MachKeen, A. Witt, E. E. Mitchell, G. J. Stumpf, M. E. Eilts, and K. W. Thomas, 1998: The Storm Cell Identification and Tracking algorithm: an enhanced WSR-88D algorithm. Weather and Forecasting, 13, Waldovogel A., Federer B. and Grimm P., 1979: Criteria for the detection of hail cells. Journal of Applied Meteorology, 18, Acronyms CMT COSMO-LAMI DB GIS HDSS POD QPE QPF SCIT VIL WXML XML Meteorological Centre of Teolo Consortium for Small-scale Modeling Limited Area Model Italy Database Geographic Information System HydroMet Decision Support System Probability of Detection Quantitative Precipitation Estimates Quantitative Precipitation Forecast Storm Cell Identification and Tracking Algorithm Vertically Integrated Liquid water equivalent Weather Objects Markup Language Extensible Markup Language 44

45 May 2005 Jun 2005 Jul 2005 Aug 2005 Sep 2005 Nmax=15 Nmax=16 Nmax=32 Nmax=44 Nmax=61 May 2006 Jun 2006 Jul 2006 Aug 2006 Sep 2006 Nmax=28 Nmax=26 Nmax=43 Nmax=51 Nmax=32 May 2007 Jun 2007 Jul 2007 Aug 2007 Sep 2007 Nmax=44 Nmax=42 Nmax=35 Nmax=49 Nmax=38 Figure 31. Cell density maps for the individual warm season months 2005 (upper row), 2006 (middle row) and 2007 (third row) performed on a lat/lon grid with a mesh size of 0.1 x 0.1 (8x11 km). The color scale is in percent of the maximum number of cells detected in one grid point (Nmax, reported in the upper right).

46 May 2005 June 2005 July 2005 August 2005 September 2005 Rain gauge QPE SCIT-derived QPE LAMI convective QPF Figure 36. Convective component of COSMO LAMI QPF (upper row), SCIT-derived QPE (middle row) and rain gauge accumulation of the CMT observing network (third row) for the individual warm season months 2005.

47 May 2007 June 2007 July 2007 August 2007 September 2007 Rain gauge QPE SCIT-derived QPE LAMI convective QPF Figure 37. Convective component of COSMO LAMI QPF (upper row), SCIT-derived QPE (middle row) and rain gauge accumulation of the CMT observing network (third row) for the individual warm season months 2007.

48 May Sep 2005 May Sep 2006 May Sep 2007 May Sep Rain gauge QPE SCIT-d erived QPE LAMI convective QPF Figure 38. As in Figure 36 for the entire warm season (May-Sep) 2005, 2006 and Figure 39. As in Fig. 36 for the warm seasons May-Sep combined.