Advances in Geosciences, 2, 59–63, 2005SRef-ID: 1680-7359/adgeo/2005-2-59European Geosciences Union© 2005 Author(s). This work is licensedunder a Creative Commons License.

Advances inGeosciences

Coupling a distributed grid based hydrological model and MM5meteorological model for flooding alert mapping

B. Tomassetti, E. Coppola, M. Verdecchia, and G. Visconti

Centro di Eccellenza CETEMPS, Università dell’Aquila, Italy

Received: 29 October 2004 – Revised: 8 February 2005 – Accepted: 16 February 2005 – Published: 29 March 2005

Abstract. The increased number of extreme rainfall eventsseems to be one of the common feature of climate changesignal all over the world (Easterlin et al., 2000; Meehl et al.,2000). In the last few years a large number of floods causedby extreme meteorological events has been observed over theriver basins of Mediterranean area and they mainly affectedsmall basins (few hundreds until few thousands of squarekilometres of drainage area) . A strategic goal of applied me-teorology is now to try to predict with high spatial resolutionthe segments of drainage network where floods may occur.A possible way to reach this aim is the coupling of meteo-rological mesoscale model with high resolution hydrologicalmodel. In this work few case studies of observed floods inthe Italian Mediterranean area will be presented. It is shownhow a distributed hydrological model, using the precipita-tion fields predicted by MM5 meteorological model, is ableto highlight the area where the major floods may occur.

1 Introduction

In the Center of Excellence for the Integration of RemoteSensing Techniques and Numerical Modelling for the Fore-cast of Severe Weather (CETEMPS) of the University ofL’Aquila the MM5 PSU/NCAR (mesoscale Model 5 ofPennsylvania State University/National Center for Atmo-spheric Research) has been made operational since 1999.The horizontal resolution of the model (3 Km) is compara-ble to typical resolution of hydrological model and then thepredicted rainfall fields are suitable to be used as input forhydrological modelling (Wilby et al., 1997).

Within the activities of the Center, a distributed hydrolog-ical model has also been developed; one of the main char-acteristics of the model is the possibility to rebuilt the rain-fall field on the model grid, acquiring and merging observedand simulated data. The final goal is to be able to predict

Correspondence to:M. Verdecchia(marco.verdecchia@aquila.infn.it)

flow discharge on a given domain, making simulation thatspan from few days in the past until few days in the future.For the past time step all the available observational data areused, for example rain gauge measurements, radar and satel-lite estimation, while for the future time step the precipitationfield is rebuilt on hourly base using the meteorological modelforecast.

Cetemps HYdrological Model (CHYM) is now being cali-brated in order to reproduce the observed flow discharge timeseries. Here we described a first application tool based on theflow-scheme calculated by the model that is able to producean alarm map of the simulated domain. We show as, in allthe simulated case studies, the tool is able to highlight thesegment of the drainage network where floods actually oc-curred.

In the next two sections a short description of MM5 andCHYM models are given and in Sect. 4 the proposed alarmindex is described. In the last section the practical applicationof the alarm hydrological mapping tool is described and it isshown how it allows to well detect the areas where importantfloods actually occurred.

2 MM5 model

For this study meteorological simulation are carried out us-ing the non hydrostatic version of MM5 Mesoscale Modelfrom NCAR/PSU (Dudhia, 1993 and Grell et al., 1994). Themodel explicitly solves the primitive equations and includesa multiple two-way nest capability. For all the simulationdiscussed here MM5 configuration is the same used for oper-ational activities of CETEMPS group: it uses 24 sigma lev-els not equally spaced, MRF parameterization for planetaryboundary layer (Troen and Mahrt, 1986) and Kain-Fritschcumulus convection parameterization associated to an ex-plicit configuration of cloud water and rain (Kain and Fritsch,1993). For the present study simulations are carried out usingthree nested domains with respectively 27, 9 and 3 km of res-olution and innermost higher resolution domain is centred for

60 B. Tomassetti et al.: Grid based hydrological model and MM5 meteorological model

Plinius Conf. 2004 – Mediterranean Sea

Draft paper 1

Coupling a distributed grid based hydrological model and MM5 meteorological model for flooding alert mapping

Barbara Tomassetti, Erika Coppola, Marco Verdecchia, and Guido Visconti

(1) Centro di Eccellenza CETEMPS, Università dell’Aquila, Italy Via Vetoio, Coppito, L’Aquila, Italy – E-mail barbara.tomassetti@aquila.infn.it

Submitted to EGU Advances in Geosciences

October 2004, in final form February 2005 (vers. 2.2)

Correspondence to Marco Verdecchia (marco.verdecchia@aquila.infn.it) ABSTRACT. The increased number of extreme rainfall events seems to be one of the common feature of climate change signal all over the world (Easterlin et Al., 2000, Meehl et Al., 2000). In the last few years a large number of floods caused by extreme meteorological events has been observed over the river basins of Mediterranean area and they mainly affected small basins (few hundreds until few thousands of square kilometres of drainage area) . A strategic goal of applied meteorology is now to try to predict with high spatial resolution the segments of drainage network where floods may occur. A possible way to reach this aim is the coupling of meteorological mesoscale model with high resolution hydrological model. In this work few case studies of observed floods in the Italian Mediterranean area will be presented. It is shown how a distributed hydrological model, using the precipitation fields predicted by MM5 meteorological model, is able to highlight the area where the major floods may occur. 1. Introduction In the Center of Excellence for the Integration of Remote Sensing Techniques and Numerical Modelling for the Forecast of Severe Weather (CETEMPS) of the University of L’Aquila the MM5 PSU/NCAR (mesoscale Model 5 of Pennsylvania State University/National Center for Atmospheric Research) has been made operational since 1999. The horizontal resolution of the model (3 Km) is comparable to typical resolution of hydrological model and then the predicted rainfall fields are suitable to be used as input for hydrological modelling (Wilby et Al., 1997). Within the activities of the Center, a distributed hydrological model has also been developed; one of the main characteristics of the model is the possibility to rebuilt the rainfall field on the model grid, acquiring and merging observed and simulated data. The final goal is to be able to predict flow discharge on a given domain, making simulation that span from few days in the past until few days in the future. For the past time step all the available observational data are used, for example rain gauge

measurements, radar and satellite estimation, while for the future time step the precipitation field is rebuilt on hourly base using the meteorological model forecast. Cetemps HYdrological Model (CHYM) is now being calibrated in order to reproduce the observed flow discharge time series. Here we described a first application tool based on the flow-scheme calculated by the model that is able to produce an alarm map of the simulated domain. We show as, in all the case studies simulated, the tool is able to highlight the segment of the drainage network where floods actually occurred. In the next two sections a short description of MM5 and CHYM models are given and in section 4 the proposed alarm index is described. In the last section the practical application of the alarm hydrological mapping tool is described and it is shown how it allows to well detect the areas where important floods actually occurred.

Figure 1: MM5 simulation for September 9-10, 2000: two-day accumulated precipitation (cm) for the Soverato case.

Fig. 1. MM5 simulation for 9–10 September 2000: two-day accu-mulated precipitation (cm) for the Soverato case.

all the case studies in the zone where flood events occurred.All the simulation are driven using ECMWF (European Cen-ter for Medium-range Weather Forecast) global analyses andforecast and Boundary Conditions are upgraded every 6 h.Figure 1 shows the 48 h accumulated rain as predicted by themodel for one of three case studies discussed in the last sec-tion.

3 CHYM model

The Cetemps HYdrological Model is a grid based distributedModel that can be run in any geographical domain and withany resolution up to the resolution of the implemented Digi-tal Elevation Model (DEM), i.e. about 300 m; for operationalpurpose the capability to simulate an arbitrary domain cor-responds to the need to run the model for those river basinsthat are more stressed by the current meteorological events.CHYM model simulates the hydrological processes using anapproach similar to other models (Todini et al., 2002; Borahet al., 2002) and the calibration of the model to realisticallysimulate the flow discharge for the Central Italy is still inprogress. A complete description of the model (Coppola etal., 2003; Tomassetti et al., 2003, Coppola et al., 20051) isbeyond the scope of this paper and we quickly describe hereonly a couple of aspects that are important for the applicationdescribed in this paper.

1 Coppola, E., Tomassetti, B., Verdecchia, M., Visconti G.:CHYM – A grid-based hydrological model: description of algo-rithms for determination of drainage structure and combination ofsimulated and observed rainfall data, in preparation, 2005.

Plinius Conf. 2004 – Mediterranean Sea

2

2. MM5 Model For this study meteorological simulation are carried out using the non hydrostatic version of MM5 Mesoscale Model from NCAR/PSU (Dudhia, 1993 and Grell et Al., 1994). The model explicitly solves the primitive equations and includes a multiple two-way nest capability. For all the simulation discussed here MM5 configuration is the same used for operational activities of CETEMPS group: it uses 24 sigma levels not equally spaced, MRF parameterization for planetary boundary layer (Troen and Mahrt, 1986) and Kain-Fritsch cumulus convection parameterization associated to an explicit configuration of cloud water and rain (Kain and Fritsch, 1993). For the present study simulations are carried out using three nested domains wth respectively 27, 9 and 3 Km of resolution and innermost higher resolution domain is centred for all the case studies in the zone where flood events occurred. All the simulation are driven using ECMWF (European Center for Medium-range Weather Forecast) global analyses and forecast and Boundary Conditions are upgraded every 6 hours. Fig.1 shows the 48 hours accumulated rain as predicted by the model for one of three case studies discussed in the last section.

Figure 2 Accumulated rain field as assimilated by CHYM model for the Val Canale event.

3. CHYM Model The Cetemps HYdrological Model is a grid based distributed Model that can be run in any geographical domain and with any resolution up to the resolution of the implemented Digital Elevation Model (DEM), i.e. about 300 meters; for operational purpose the capability to simulate an arbitrary domain corresponds to the need to run the model for those river basins that are more stressed by

the current meteorological events. CHYM model simulates the hydrological processes using an approach similar to other models (Todini et Al., 2002, Borah et Al., 2002) and the calibration of the model to realistically simulate the flow discharge for the Central Italy is still in progress. A complete description of the model (Coppola et Al., 2003, Tomassetti et Al., 2003) is beyond the scope of this paper and we quickly describe here only a couple of aspects that are important for the application described in this paper. The architectural characteristics of model have been designed to assimilate different data sources for the estimation of rainfall field on the model grid. For the operational activities rainfall is estimated on CHYM grid at hourly time step using observed data for the past and meteorological model simulation for the future. The observed data are taken from a rain gauge network operating in the Central Italy and also rain estimation are obtained using radar measurements and infrared signal from geostationary satellite platform. When different data sources are available at same time step, these are assimilated with different priorities using a Cellular Automata based algorithm quickly described in the following. An example of assimilation of MM5 predicted rainfall is shown in Fig. 2.

Figura 3: CHYM alarm map for the Soverato flood, alarm index is calculated from September 9, 2000 h: 12 until September 11 2000 h: 12.

Starting from the DEM matrix for each cell is defined the flow direction. According to the minimum energy principle the flow direction is assumed that for which the slope is maximum, this corresponds to the assumption that flow occurs with a strong preferential direction. Due to the finite resolution of the DEM few singularities are usually found for those cells for which it is not possible to coherently establish the flow direction. In order to solve these singularities, a Cellular Automata based algorithm is used for smoothing the DEM: the height of those cells where flow direction has been already calculated is taken fixed

Fig. 2. Accumulated rain field as assimilated by CHYM model forthe Val Canale event.

The architectural characteristics of model have been de-signed to assimilate different data sources for the estimationof rainfall field on the model grid. For the operational activ-ities rainfall is estimated on CHYM grid at hourly time stepusing observed data for the past and meteorological modelsimulation for the future. The observed data are taken froma rain gauge network operating in the Central Italy and alsorain estimation are obtained using radar measurements andinfrared signal from geostationary satellite platform. Whendifferent data sources are available at same time step, theseare assimilated with different priorities using a Cellular Au-tomata based algorithm quickly described in the following.An example of assimilation of MM5 predicted rainfall isshown in Fig. 2.

Starting from the DEM matrix for each cell is defined theflow direction. According to the minimum energy principlethe flow direction is assumed that for which the slope is max-imum, this corresponds to the assumption that flow occurswith a strong preferential direction. Due to the finite resolu-tion of the DEM few singularities are usually found for thosecells for which it is not possible to coherently establish theflow direction. In order to solve these singularities, a CellularAutomata based algorithm is used for smoothing the DEM:the height of those cells where flow direction has been al-ready calculated is taken fixed and is slightly smoothed forthe other cells; the smoothing is established according to theaverage height of surrounding cells. The algorithm is iterateduntil the flow direction is established for each grid point. De-spite the final goal is the real time flood forecasting on thesimulated domain, a first operational version of the model al-lows to obtain a sort of flood alarm index using only the flowscheme and rainfall estimation. The alarm index is discussedin the next session.

B. Tomassetti et al.: Grid based hydrological model and MM5 meteorological model 61

4 The flood alarm index

An alarm index is estimated for each cell of CHYM hydro-logical model as the ratio of total drained rain in the last 48hours and total drain surface, this correspond to the averagerainfall in the upstream basin of each cell. Both the drainedrain and drained surface are calculated iterating the followingstep:

1. from each cell a “stone” rolls down to the sea followingthe flow direction map calculated from the DEM

2. each time a stone pass by a cell a counter is incrementedfor that cell

These steps are iterated until all the “stones” have reachedthe sea. At the end of the iteration we have a matrix con-taining the total number of cells drained by each grid pointand this matrix can be used for many purposes, for exampleto distinguish cells where runoff occur overland from cellswhere runoff occur along channel. In addition the algorithmcan be generalized to estimate different quantities: for exam-ple if the area of the starting cell is associated to the “stones”we can estimate the total area drained by each cell, while ifthe rainfall in the starting cell is associated to the ”stones”we can estimate the total drained rain.

The choice of 48 h as time interval is justified both by em-pirical observations discussed in the next section and by thefact that this corresponds to a typical runoff time for the riverbasins simulated in this work, namely a sort of typical timeinterval that a rain drop would take to arrive at the mouth ofthe river; we also notice how the total drained area is pro-portional, as a first approximation, to the river cross section(see as an example Borah et al., 2002) and then the proposedflood alarm index is a sort of measure of river stress.

What is more important for operational activities and foractual use of the model for flood alert mapping is that theproposed alarm index seems to be suitable to highlight thesection of the drainage network that was actually interestedby flood events at least for all the case-studies reported in thisstudy and discussed in the next session. An example of alarmmapping for one of the case studies is shown in Fig. 3.

5 Case studies

Three case studies have been selected for different region ofItaly:

1. The Soverato flood – 9–10 September 2000 (Figs. 1 and3). Soverato torrent drains an area of few hundreds ofsquare kilometres and is situated in Southern Italy’s Cal-abria Region. Flooding occurred between the towns ofSoverato and Montepaone after a prolonged and intenseprecipitation event, causing landslides, great disruptionand losses of life. Rainfall peaks above 300 mm in 24 hwas recorded close to the village of Soverato (Montaniet al., 2003).

Plinius Conf. 2004 – Mediterranean Sea

2

2. MM5 Model For this study meteorological simulation are carried out using the non hydrostatic version of MM5 Mesoscale Model from NCAR/PSU (Dudhia, 1993 and Grell et Al., 1994). The model explicitly solves the primitive equations and includes a multiple two-way nest capability. For all the simulation discussed here MM5 configuration is the same used for operational activities of CETEMPS group: it uses 24 sigma levels not equally spaced, MRF parameterization for planetary boundary layer (Troen and Mahrt, 1986) and Kain-Fritsch cumulus convection parameterization associated to an explicit configuration of cloud water and rain (Kain and Fritsch, 1993). For the present study simulations are carried out using three nested domains wth respectively 27, 9 and 3 Km of resolution and innermost higher resolution domain is centred for all the case studies in the zone where flood events occurred. All the simulation are driven using ECMWF (European Center for Medium-range Weather Forecast) global analyses and forecast and Boundary Conditions are upgraded every 6 hours. Fig.1 shows the 48 hours accumulated rain as predicted by the model for one of three case studies discussed in the last section.

Figure 2 Accumulated rain field as assimilated by CHYM model for the Val Canale event.

3. CHYM Model The Cetemps HYdrological Model is a grid based distributed Model that can be run in any geographical domain and with any resolution up to the resolution of the implemented Digital Elevation Model (DEM), i.e. about 300 meters; for operational purpose the capability to simulate an arbitrary domain corresponds to the need to run the model for those river basins that are more stressed by

the current meteorological events. CHYM model simulates the hydrological processes using an approach similar to other models (Todini et Al., 2002, Borah et Al., 2002) and the calibration of the model to realistically simulate the flow discharge for the Central Italy is still in progress. A complete description of the model (Coppola et Al., 2003, Tomassetti et Al., 2003) is beyond the scope of this paper and we quickly describe here only a couple of aspects that are important for the application described in this paper. The architectural characteristics of model have been designed to assimilate different data sources for the estimation of rainfall field on the model grid. For the operational activities rainfall is estimated on CHYM grid at hourly time step using observed data for the past and meteorological model simulation for the future. The observed data are taken from a rain gauge network operating in the Central Italy and also rain estimation are obtained using radar measurements and infrared signal from geostationary satellite platform. When different data sources are available at same time step, these are assimilated with different priorities using a Cellular Automata based algorithm quickly described in the following. An example of assimilation of MM5 predicted rainfall is shown in Fig. 2.

Figura 3: CHYM alarm map for the Soverato flood, alarm index is calculated from September 9, 2000 h: 12 until September 11 2000 h: 12.

Starting from the DEM matrix for each cell is defined the flow direction. According to the minimum energy principle the flow direction is assumed that for which the slope is maximum, this corresponds to the assumption that flow occurs with a strong preferential direction. Due to the finite resolution of the DEM few singularities are usually found for those cells for which it is not possible to coherently establish the flow direction. In order to solve these singularities, a Cellular Automata based algorithm is used for smoothing the DEM: the height of those cells where flow direction has been already calculated is taken fixed

Fig. 3. CHYM alarm map for the Soverato flood, alarm index iscalculated from 9 September 2000, h: 12 until 11 September 2000,h: 12.

Plinius Conf. 2004 – Mediterranean Sea

3

and is slightly smoothed for the other cells; the smoothing is established according to the average height of surrounding cells. The algorithm is iterated until the flow direction is established for each grid point. Despite the final goal is the real time flood forecasting on the simulated domain, a first operational version of the model allows to obtain a sort of flood alarm index using only the flow scheme and rainfall estimation. The alarm index is discussed in the next session 4. The Flood Alarm Index An alarm index is estimated for each cell of CHYM hydrological model as the ratio of total drained rain in the last 48 hours and total drain surface, this correspond to the average rainfall in the upstream basin of each cell. Both the drained rain and drained surface are calculated iterating the following step: 1) from each cell a "stone" rolls down to the sea

following the flow direction map calculated from the DEM

2) each time a stone pass by a cell a counter is incremented for that cell

These steps are iterated until all the "stones" have reached the sea. At the end of the iteration we have a matrix containing the total number of cells drained by each grid point and this matrix can be used for many purpose, for example to distinguish cells where runoff occur overland from cells where runoff occur along channel. In addition the algorithm can be generalized to estimate different quantities: for example if the area of the starting cell is associated to the "stones" we can estimate the total area drained by each cell, while if the rainfall in the starting cell is associated to the "stones" we can estimate the total drained rain.

Figura 4: CHYM alarm map for the Val Canale event, alarm index is calculated from August 28, 2003 h: 00 until August 30 , 2003 : 00.

The choice of 48 hours as time interval is justified both by empirical observations discussed in the next section and by the fact that this corresponds to a typical runoff time for the river basins simulated in this work, namely a sort of typical time interval that a rain drop would take to arrive at the mouth of the river; we also notice how the total drained area is proportional, as a first approximation, to the river cross section (see as an example Borah et Al., 2002) and then the proposed flood alarm index is a sort measure of river stress. What is more important for operational activities and for actual use of the model for flood alert mapping is that the proposed alarm index seems to be suitable to highlight the section of the drainage network that was actually interested by flood events at least for all the case-studies reported in this study and discussed in the next session. An example of alarm mapping for one of the case studies is shown in Fig. 3.

Figure 5: MM5 simulation for January 24-26 2003: two-day accumulated precipitation (cm).. The river beds Trigno and Sangro are highlighted.

5. Case studies Three case studies have been selected for different region of Italy: 1. The Soverato flood - September 9 -10, 2000 (Figures 1

and 3). Soverato torrent drains an area of few hundreds of square kilometres and is situated in Southern Italy’s Calabria Region. Flooding occurred between the towns of Soverato and Montepaone after a prolonged and intense precipitation event, causing landslides, great disruption and losses of life. Rainfall peaks above 300 mm in 24 hours was recorded close to the village of Soverato (Montani et Al., 2003).

2. The Val Canale flood - August, 29 2003 (Figures 2 and 4). A flash flood occurred in the Valcanale valley in the Friuli Region in the North East of Italy causing

Fig. 4. CHYM alarm map for the Val Canale event, alarm index iscalculated from 28 August 2003, h: 00 until 30 August 2003, h: 00.

2. The Val Canale flood – 29 August 2003 (Figs. 2 and4). A flash flood occurred in the Valcanale valley in theFriuli Region in the North East of Italy causing manydamages to buildings and loss of three people. Morethan 400 mm of precipitation in 24 h were recorded inthe villages of Grauzaria di Moggio and Pontebba.

62 B. Tomassetti et al.: Grid based hydrological model and MM5 meteorological model

Plinius Conf. 2004 – Mediterranean Sea

3

and is slightly smoothed for the other cells; the smoothing is established according to the average height of surrounding cells. The algorithm is iterated until the flow direction is established for each grid point. Despite the final goal is the real time flood forecasting on the simulated domain, a first operational version of the model allows to obtain a sort of flood alarm index using only the flow scheme and rainfall estimation. The alarm index is discussed in the next session 4. The Flood Alarm Index An alarm index is estimated for each cell of CHYM hydrological model as the ratio of total drained rain in the last 48 hours and total drain surface, this correspond to the average rainfall in the upstream basin of each cell. Both the drained rain and drained surface are calculated iterating the following step: 1) from each cell a "stone" rolls down to the sea

following the flow direction map calculated from the DEM

2) each time a stone pass by a cell a counter is incremented for that cell

These steps are iterated until all the "stones" have reached the sea. At the end of the iteration we have a matrix containing the total number of cells drained by each grid point and this matrix can be used for many purpose, for example to distinguish cells where runoff occur overland from cells where runoff occur along channel. In addition the algorithm can be generalized to estimate different quantities: for example if the area of the starting cell is associated to the "stones" we can estimate the total area drained by each cell, while if the rainfall in the starting cell is associated to the "stones" we can estimate the total drained rain.

Figura 4: CHYM alarm map for the Val Canale event, alarm index is calculated from August 28, 2003 h: 00 until August 30 , 2003 : 00.

The choice of 48 hours as time interval is justified both by empirical observations discussed in the next section and by the fact that this corresponds to a typical runoff time for the river basins simulated in this work, namely a sort of typical time interval that a rain drop would take to arrive at the mouth of the river; we also notice how the total drained area is proportional, as a first approximation, to the river cross section (see as an example Borah et Al., 2002) and then the proposed flood alarm index is a sort measure of river stress. What is more important for operational activities and for actual use of the model for flood alert mapping is that the proposed alarm index seems to be suitable to highlight the section of the drainage network that was actually interested by flood events at least for all the case-studies reported in this study and discussed in the next session. An example of alarm mapping for one of the case studies is shown in Fig. 3.

Figure 5: MM5 simulation for January 24-26 2003: two-day accumulated precipitation (cm).. The river beds Trigno and Sangro are highlighted.

5. Case studies Three case studies have been selected for different region of Italy: 1. The Soverato flood - September 9 -10, 2000 (Figures 1

and 3). Soverato torrent drains an area of few hundreds of square kilometres and is situated in Southern Italy’s Calabria Region. Flooding occurred between the towns of Soverato and Montepaone after a prolonged and intense precipitation event, causing landslides, great disruption and losses of life. Rainfall peaks above 300 mm in 24 hours was recorded close to the village of Soverato (Montani et Al., 2003).

2. The Val Canale flood - August, 29 2003 (Figures 2 and 4). A flash flood occurred in the Valcanale valley in the Friuli Region in the North East of Italy causing

Fig. 5. MM5 simulation for 24–26 January 2003: two-day accu-mulated precipitation (cm). The river beds Trigno and Sangro arehighlighted.

Plinius Conf. 2004 – Mediterranean Sea

4

many damages to buildings and loss of three people. More than 400 mm of precipitation in 24 hours were recorded in the villages of Grauzaria di Moggio and Pontebba.

3. The Sangro river flood 24-25 January 2003 (Figures 5 and 6). In this case a typical winter frontal system produced an intense precipitation that lasted for about 30 hours; many relatively large floods have been observed in the zone close to Maiella Mountains. The Sangro River has a drainage area of about 2000 square kilometres and is located in the Central East part of Italy in Abruzzo Region, the flood occurred in the higher part, about 40 kilometres from the spring causing damages to many buildings and a bridge.

For all the case studies a run with MM5 mesoscale meteorological model has been carried out and the model has been forced with ECMWF analyses and forecast, the configuration of the model is similar to the operative version (Paolucci et Al., 1999). The innermost domain of simulation, running with a 3 Km of spatial resolution has been set in the zone where major severe events occur. Fig 1 and 4 shows the accumulated rain field as predicted by the model for case studies 1 and 3. In fig 5 the field of accumulated precipitation in northern eastern Italy for 48 hours is reported. During the precipitation event a maximum of 300-400 mm of rain has been registered in 10 hours.

Figure 6: CHYM alarm map for the Sangr event, alarm index is calculated from January 23, 2002 h: 18 until January 25, 2002 h: 18.

In the Figures 3, 4 and 6 the alarm map calculated by CHYM model is reported and in all the cases the alarm index seems to well localize the segments of drainage network where more critical events actually occurred. In particular comparing maps 1 and 3 it can be seen how the

zone where flood event occurred is not the same of the maximum in the accumulated precipitation as we expect because of the complex geomorphology of the domain. All the alarm map has been computed using a 48 hours time interval for the reason explained in section 4, but also because this interval seems the more suitable to highlight the more stressed segments of drainage network. It is useful to remind that CHYM model rebuilt the rainfall field on the simulated domain at hourly base using observed and simulated precipitation data, therefore for operational purpose an effective alarm mapping could be obtained using the observed rainfall data in the last 24 hours and the first 24 hours of the forecast leading to the possibility to have flood warning about one day before the event. In the Fig. 7 the alarm index is plotted along the river bed of Sangro river and the two area with major flood risk can easily be seen.

Figure 7: The Alarm Index along the bed of Sangro river (0 of x axis corresponds to the spring) that has been interested by the flood in the region close to the spring.

6. Summary and future developments A grid based distributed hydrological model has been coupled with a mesoscale model for flood alarm index computation. Simulations have been carried out for three different case studies occurring in different zones of north, central and south Italy. In all the simulated case studies the segments of the drainage network where the flooding events have actually been observed are correctly located by CHYM flood alert tool. In all the domains of northern, central and southern Italy a critical threshold of 150-200 mm of alarm index has been found to perform correctly. Although these are only preliminary results they are encouraging to actually use this approach for operational

Fig. 6. CHYM alarm map for the Sangro event, alarm index iscalculated from 23 January 2002, h: 18 until 25 January 2002, h:18.

3. The Sangro river flood – 24–25 January 2003 (Figs. 5and 6). In this case a typical winter frontal systemproduced an intense precipitation that lasted for about30 h; many relatively large floods have been observedin the zone close to Maiella Mountains. The SangroRiver has a drainage area of about 2000 square kilome-tres and is located in the Central East part of Italy in

Plinius Conf. 2004 – Mediterranean Sea

4

many damages to buildings and loss of three people. More than 400 mm of precipitation in 24 hours were recorded in the villages of Grauzaria di Moggio and Pontebba.

3. The Sangro river flood 24-25 January 2003 (Figures 5 and 6). In this case a typical winter frontal system produced an intense precipitation that lasted for about 30 hours; many relatively large floods have been observed in the zone close to Maiella Mountains. The Sangro River has a drainage area of about 2000 square kilometres and is located in the Central East part of Italy in Abruzzo Region, the flood occurred in the higher part, about 40 kilometres from the spring causing damages to many buildings and a bridge.

For all the case studies a run with MM5 mesoscale meteorological model has been carried out and the model has been forced with ECMWF analyses and forecast, the configuration of the model is similar to the operative version (Paolucci et Al., 1999). The innermost domain of simulation, running with a 3 Km of spatial resolution has been set in the zone where major severe events occur. Fig 1 and 4 shows the accumulated rain field as predicted by the model for case studies 1 and 3. In fig 5 the field of accumulated precipitation in northern eastern Italy for 48 hours is reported. During the precipitation event a maximum of 300-400 mm of rain has been registered in 10 hours.

Figure 6: CHYM alarm map for the Sangr event, alarm index is calculated from January 23, 2002 h: 18 until January 25, 2002 h: 18.

In the Figures 3, 4 and 6 the alarm map calculated by CHYM model is reported and in all the cases the alarm index seems to well localize the segments of drainage network where more critical events actually occurred. In particular comparing maps 1 and 3 it can be seen how the

zone where flood event occurred is not the same of the maximum in the accumulated precipitation as we expect because of the complex geomorphology of the domain. All the alarm map has been computed using a 48 hours time interval for the reason explained in section 4, but also because this interval seems the more suitable to highlight the more stressed segments of drainage network. It is useful to remind that CHYM model rebuilt the rainfall field on the simulated domain at hourly base using observed and simulated precipitation data, therefore for operational purpose an effective alarm mapping could be obtained using the observed rainfall data in the last 24 hours and the first 24 hours of the forecast leading to the possibility to have flood warning about one day before the event. In the Fig. 7 the alarm index is plotted along the river bed of Sangro river and the two area with major flood risk can easily be seen.

Figure 7: The Alarm Index along the bed of Sangro river (0 of x axis corresponds to the spring) that has been interested by the flood in the region close to the spring.

6. Summary and future developments A grid based distributed hydrological model has been coupled with a mesoscale model for flood alarm index computation. Simulations have been carried out for three different case studies occurring in different zones of north, central and south Italy. In all the simulated case studies the segments of the drainage network where the flooding events have actually been observed are correctly located by CHYM flood alert tool. In all the domains of northern, central and southern Italy a critical threshold of 150-200 mm of alarm index has been found to perform correctly. Although these are only preliminary results they are encouraging to actually use this approach for operational

Fig. 7. The Alarm Index along the bed of Sangro river (0 of x-axiscorresponds to the spring) that has been interested by the flood inthe region close to the spring.

Abruzzo Region, the flood occurred in the higher part,about 40 km from the spring causing damages to manybuildings and a bridge.

For all the case studies a run with MM5 mesoscale me-teorological model has been carried out and the model hasbeen forced with ECMWF analyses and forecast, the con-figuration of the model is similar to the operative version(Paolucci et Al., 1999). The innermost domain of simula-tion, running with a 3 km of spatial resolution has been setin the zone where major severe events occur. Figures 1 and4 show the accumulated rain field as predicted by the modelfor case studies 1 and 3.

In Fig. 5 the field of accumulated precipitation in northerneastern Italy for 48 h is reported. During the precipitationevent a maximum of 300–400 mm of rain has been registeredin 10 h.

In the Figs. 3, 4 and 6 the alarm map calculated by CHYMmodel is reported and in all the cases the alarm index seemsto well localize the segments of drainage network wheremore critical events actually occurred. In particular compar-ing maps 1 and 3 it can be seen how the zone where floodevent occurred is not the same of the maximum in the accu-mulated precipitation as we expect because of the complexgeomorphology of the domain. All the alarm map has beencomputed using a 48 h time interval for the reason explainedin Sect. 4, but also because this interval seems the more suit-able to highlight the more stressed segments of drainage net-work. It is useful to remind that CHYM model rebuilt therainfall field on the simulated domain at hourly base using

B. Tomassetti et al.: Grid based hydrological model and MM5 meteorological model 63

observed and simulated precipitation data, therefore for op-erational purpose an effective alarm mapping could be ob-tained using the observed rainfall data in the last 24 h andthe first 24 h of the forecast leading to the possibility to haveflood warning about one day before the event.

In the Fig. 7 the alarm index is plotted along the river bedof Sangro river and the two area with major flood risk caneasily be seen.

6 Summary and future developments

A grid based distributed hydrological model has been cou-pled with a mesoscale model for flood alarm index compu-tation. Simulations have been carried out for three differentcase studies occurring in different zones of north, central andsouth Italy.

In all the simulated case studies the segments of thedrainage network where the flooding events have actuallybeen observed are correctly located by CHYM flood alerttool.

In all the domains of northern, central and southern Italya critical threshold of 150–200 mm of alarm index has beenfound to perform correctly.

Although these are only preliminary results theyare encouraging to actually use this approach for op-erational activity. A first application for flood alertmapping daily updated is now available at URL:http://cetemps.aquila.infn.it/chymop.

Edited by: L. FerrarisReviewed by: anonymous referees

References

Borah, D. K., Xia, R., Bera, M.: DWSM – A Dynamic Water-shed Simulation Model, Mathematical Model for Small Water-shed Hydrology, WRP Edition, 2002.

Coppola, E., Verdecchia, M., Tomassetti, B., and Visconti, G.:CHYM: A Grid based Hydrological Model, Proceedings of Inter-national Symposium on Remote Sensing of Environment, Hon-olulu, Hawaii, November 2003.

Borah, D. K., Xia, R., and Bera, M.: DWSM – A dynamical wa-tershed simulation model, Mathematical Model for Small Water-shed Hydrology, WRP Edition, 2002.

Easterling, D. R., Evans, J. L., Groismann, P. Ya., Karl., T. R.,Kunkel, K. E., and Ambenje, P.: Observed Variability and Trendsin Extreme Climate Events: a Brief, Review Bullettin of Am.Met. Soc., 81, 3, 417–425, March 2000.

Dudhia J.: A nonhydrostatic version of the Pen State/NCARmesoscale moldel: validation test and simulation of an AtlanticCyclon and cold front, Mon. Wea. Rev., 121, 493–513, 1993.

Grell G. A., Dudhia, J., and Stauffer, D. R.: A description of the fifthgeneration of the Pen State/NCAR mesoscale model (MM5),NCAR Technical Note, NCAR/TN-398+STR, 121pp, 1994.

Kain, J. S. and Fritsch, J. M.: A one-dimensional entrain-ing/detraining plume model and its applicationin convective pa-rameterization, J. Atmos. Sci., 47, 2784–2802, 1990.

Meehl, G. A., Zwiers, F., Evans, J., Knutson, T., Mearns, L., andWhetton, P.: Trends in Extreme Wheater and Climate Events:Issues related to Modeling Extremes in Projection of Future Cli-mate Change, Bullettin of Am. Met. Soc., 81, 3, 427–436, March2000.

Montani, A., Marsigli, C., Nerozzi, F., Paccagnella, T., Ribaldi, S.,and Buizza, R.: The Noverato Flood in Southern Italy: Perfor-mance of Global and Limited Area Ensemble Forecasts, Nonlin.Proc. Geophys., 10, 261–274, 2003,SRef-ID: 1607-7946/npg/2003-10-261.

Paolucci, T., Bernardini, L., Ferretti, R., and Visconti, G.: MM5Real-Time Forecast of a Catastrophic Event on May, 5 1998, IlNuovo Cimento, 12, 727–736, 1999.

Tomassetti, B., Coppola, E., Cimini, D., Verdecchia, M., Marzano,F. S., and Visconti, G.: Assimilating satellite rainfall estimationinto hydrological model for flooding alert mapping, Actif Work-shop 1: Combination of data from remote sensing technologiesfor flood forecasting, 24–25 November 2003, Bologna, 2003.

Todini, E. and Ciarrapica, L.: The TOPKAPI Model, MathematicalModel for Large Watershed Hydrology, WRP Edition, 2002.

Troen, I. and Mahrt, L.: A simple model of the atmospheric bound-ary layer: Sensivity to surface evaporation, Boundary Layer Me-teorology, 37, 129–148, 1986.

Wilby, R. L. and Wigley, T. M. L.: Downscaling General Circula-tion Model Output: a Review of Methods and Limitation, Prog.Phys. Geo., 21, 530–548, 1997.

http://cetemps.aquila.infn.it/chymophttp://direct.sref.org/1607-7946/npg/2003-10-261