During the last several years heavy precipitation events have produced severe damage in the Alpine region and the interest in improving the forecast of such cases has grown correspondingly. The main problem to be solved is the predictability of both the large-scale and the convective precipitation, with the latter usually being the main cause of the extensive damage. Convective precipitation plays a major role in the Mediterranean area, where both complex orography and the sea are the main forcings. The use of a mesoscale model to resolve such phenomena is imperative, although the model resolution is usually lower than the scale of the convective precipitation and the interaction between cumulus convection and the large-scale precipitation is still uncertain. Furthermore the vertical distribution of the latent heat released by cumulus convection depends on the model-resolved scale; as resolution increases, most of the processes associated with precipitation become convective and so does part of the flow; therefore the interaction between the resolved and unresolved scales becomes more complex. The explicit computation of cloud and rain processes associated with a cumulus convection parameterization may be necessary when the resolution falls between 10 and 50 km (Molinari and Dudek 1992).
In general, limited area models (LAMs) are good tools to forecast the heavy precipitation events but the reliability of the results depends on both the cumulus convection parameterization and the nonconvective precipitation scheme used.
Various studies have been reported to assess the role of the cumulus convection parameterization in forecasting heavy precipitation events, using LAMs. A case of explosive marine cyclogenesis, analyzed by Kuo and Low-Nam (1990), helps to isolate the key factors that determine such events; the authors show the relevant role played by the precipitation parameterization. A study by Kuo et al. (1996) compares the performance of several cumulus convection parameterizations and points out the role played by the latent heat released by both the resolvable and the subgrid scale in a case of a marine cyclone. Recently, Wang and Seaman (1997) compared the performances of several cumulus convection schemes for six different cases over the United States for both warm and cold seasons and established the skill of the Kain–Fritsch parameterization especially during the cold season and for heavy precipitation events.
Studies of heavy precipitation in northern Italy (Paccagnella et al. 1992; Buzzi et al. 1995, etc.) have shown difficulties in reproducing locally organized convective precipitation, but the overall model results show a fair agreement with the observed data. Uncertainties are also related to the complex orography of the Mediterranean area. Sensible heat and moisture fluxes, developed on the coastal region, rise along the mountain slopes, triggering convective precipitation.
The catastrophic events experienced in the last few years have drawn the attention of the Italian meteorological community, in an attempt to reproduce at least a few of them such as the Piedmont flood of November 1994. Encouraging results were obtained using either hydrostatic or nonhydrostatic models: Paccagnella et al. (1995) successfully reproduced this event using a hydrostatic model, with better results at higher resolution. Ferretti et al. (1996) achieved a good precipitation forecast using a nonhydrostatic model, while the hydrostatic version overestimated the rainfall on the Alps.
The purpose of this paper is to study the interaction between cumulus convection precipitation and resolved precipitation over complex topography (Fig. 1) under different meteorological forcings. This is reached by testing different cumulus convection parameterizations with the Pennsylvania State University–National Center for Atmospheric Research Fifth-Generation Mesoscale Model, version 1 (PSU–NCAR MM5V1). In particular, results are reported simulating several precipitation events that occurred in northern Italy during June 1990. The focus on these events is justified by the fact that results are available for the same period from an ad hoc observational campaign in this region [Monitoring Precipitation Activity in the Padana Region (MATREP), June 1990]. The data include precipitation (OBS) on a rather fine mesh grid.
Furthermore, during June 1990, precipitation was highly localized, especially for the events associated with a weak large-scale forcing, making the simulation of these cases more challenging.
A description of the meteorological situations of the cases analyzed during June 1990 is given in section 2. In section 3 the model and the cumulus convection parameterizations are briefly presented. The results of the model are discussed in section 4 and the conclusions are given in section 5.
During June 1990 several meteorological situations developed, and the authors selected those of 6–11, 15–16, and 20–21 June. The selection is based on the fact that all three events were more or less related to large-scale forcing but for two cases precipitation was triggered by large-scale forcing and modulated by local forcing (cases 6–11 and 19–21 June), while for case 14–16 June the situation was reversed. During 6–11 June a deep cyclonic system was located over Great Britain while a weak cyclonic circulation developed over northern Italy (Figs. 2a,b). In the afternoon of 7 June as the frontal air, coming from the northwest, crossed the Alps, a low-level warming and an upper-level cooling were observed (Figs. 2c,d), due to the barrier effect of the Alps. Over the Po Valley, heavy precipitation was reported between 7 June at 2000 and 2300 UTC (Fig. 2e), producing severe damage. Later during 8 June, precipitation was still reported on the east side of the Po region (Fig. 2f). Figures 2e,f show the 24- and 12-h accumulated precipitation, respectively. It should be noticed that the apparent discrepancy between the two is related to the fact that raingauge stations measure either 24-h averages or 12-h averages, and the 24-h intervals span across three different 12-h intervals. This is true for all the observed precipitation data. During 8 June, a second system approached the western side of the Alps and reached the central region at noon (Figs. 3a,b); the system moved rapidly eastward toward the Balkans (Figs. 3c,d), but local thunderstorms were detected (Figs. 3e,f). During 9 June a “cutoff low” (Figs. 3b,d), extending from south England to France, slowly moved toward the southwest; it reached the Po Valley during 10 June at 0000 UTC. The satellite imagery shows the clouds approaching from the west and moving rapidly toward the east (Figs. 4a,b).
A second event of heavy local precipitation was detected on 15 June, which was not directly related to a large-scale forcing, but more likely to local convection produced by a sea-breeze circulation on the western Apennines and by a strong convergence area on the eastern Po Valley. A low-level west-southwest wind carried moisture toward the eastern edge of the Gulf of Genoa during the entire period (Figs. 5a,b). North of the Alps, a northerly wind was forced to turn round the Alps because of the barrier effect, producing a westerly flow south of the Alps, which became northeasterly and southeasterly on the east side (Fig. 5c). On the same day at 1200 UTC south of the Alps, at low level a convergence zone (Fig. 5c) and an upper-level zonal flow were observed, while a northerly wind was observed in the north side of the Alps (Fig. 5d). The satellite imagery shows a circulation in the eastern Alps associated with a strong convergence zone and aligned clouds in the eastern Po Valley (Fig. 6); a cell in the southeast was also detected on 15 June at 1800 UTC. The 24-h accumulated precipitation showed highly localized, heavy rain events in the southeast and in the northwest of Italy (Fig. 5e), while light rainfall was detected all over the Po Valley and in the eastern Alps. The 12-h accumulated precipitation clearly shows that the two events of heavy rainfall were confined between 15 June at 1800 UTC and 16 June at 0000 UTC (Fig. 5f).
During 20 June a new cold front crossed the Alps (Figs. 7a,b), and produced heavy precipitation on the east side of the western Alps and in the Po Valley (Fig. 7e). A deep upper-level trough located over Great Britain was rapidly moving eastward; a large area of instability associated with the front slowly approached the western Alps, entering the Po Valley during the afternoon of 20 June (Figs. 7c,d). The front was clearly held back by the Alps, while on the northern side it was rapidly moving eastward. A surface depression (Fig. 7c) developed on the Po Valley associated with a deep convective cell. The satellite imagery (Fig. 8) shows the prefrontal cell on the Po Valley and the front approaching from the west. Precipitation was detected west of the Alps (Fig. 7e) and in the Po Valley (Fig. 7f).
In this study the hydrostatic version of the mesoscale model MM5V1 from PSU–NCAR is used; this is the first version of the new generation of the model originally described by Anthes and Warner (1978). It is a primitive equation model available in both a hydrostatic and nonhydrostatic version (Grell et al. 1994); it is written in a terrain-following coordinate σ = (p − pt)/(ps − pt), where p the pressure, pt is a constant pressure at the top of the model, and ps is the surface pressure. A finite difference scheme and a time-split explicit method for the basic equations are used. Several parameterizations for the boundary layer, the radiative transfer, and the cumulus convection are available; a boundary layer with multiple layers (Zhang and Anthes 1982) and a simple radiative transfer scheme (Anthes et al. 1987) are used for this study. Both explicit computation (EXP) of cloud water and rain (Hsie 1984) and a nonconvective precipitation scheme (nonexplicit—NEXP) associated with a cumulus convection parameterization are used; a comparison among Anthes–Kuo (AK), Grell (GR), and Kain–Fritsch (KF) cumulus convection parameterizations is reported.
Cumulus convection parameterization
The GR is a cumulus convection parameterization based on Lord (1978), but in this case the moist convective-scale downdraft is taken into account. This scheme is a cloud version (Grell 1993) of the Arakawa–Schubert (1974) parameterization; there is not direct mixing between the air in the cloud and the environment, but only at the top and the bottom of the cloud circulation, and only one deep cloud is considered. A quasi-equilibrium assumption, the one proposed by Arakawa–Schubert (1974), is used for the closure.
In this study a number of simulations are carried out, over a domain with 79 × 99 horizontal grid points and 20 unequally spaced vertical levels (σ = 0.0, 0.02, 0.04, 0.08, 0.14, 0.21, 0.3, 0.4, 0.5, 0.55, 0.69, 0.74, 0.79, 0.84, 0.88, 0.92, 0.96, 0.98, 0.99, 1). The center of the domain is at 44°00′N and 11°00′E, the grid size is ΔX = 27.8 km, and the time step is Δt = 80 s; the hydrostatic approximation is assumed. Data analyses from the European Centre for Medium-Range Weather Forecasts (ECMWF) are used to initialize the model and boundary conditions are updated every 12 h. A summary of the model simulations is given in Table 1. The table shows the start and end time of each run, the cumulus parameterization used, and whether it is associated with an explicit computation of cloud water and rain.
The basic experiment simulates the days of 6–11 June 1990 (cases A, B, and C), 14–16 June 1990 (case D), and 19–21 June 1990 (case E). To study the influence of the initial conditions, the same events starting at a different time and for a shorter interval (cases F, G, and H) are run. As an example case F starts on 7 June at midnight and ends 24 h later; this interval is included in case A, which runs for 48 h starting at 1200 UTC 6 June. The same happens for case G, which covers part of the period of case D, and for case H, which includes partly the period of case E.
To highlight the role of convective and nonconvective precipitation for the weather systems approaching the Mediterranean area, the analysis of the simulations is roughly divided in three classes: the meteorological events associated with a strong large-scale forcing (i.e., a front), the events associated with a weak large-scale forcing or a “local” forcing, and the combinations of the above. We assume that the precipitation, for some events, is partially (case A) or totally (case E) driven by the large-scale forcing, which is either a strong frontal instability, as for case E, or a postfrontal instability as for cases A–C. By local forcing we mean that the precipitation is not directly produced by the front but more likely is produced by the interaction of the large-scale flow with the local forcing, the latter being mostly the orography, as for case D.
Impact of resolvable scale
The results of the explicit and nonexplicit simulations are analyzed for three events (cases A–C, D, E); these are chosen because they well represent the meteorological situations driving the cases analyzed in this study. Only KF is discussed here because GR shows similarities with KF and does not add relevant information to the following discussion, and AK shows the tendency to produce subgrid-scale precipitation only (except for case D). We refer to the next section for an analysis of the GR and AK schemes. We show the cumulus convective rain for the explicit simulation only, because there are no appreciable differences between the cumulus convective precipitation for the explicit and nonexplicit simulations.
The analyses are performed neglecting the first 12 h of the model results and using the following 24-h accumulated rain for the three cases (A–C, D, E). The differences between the results of the 24- and 48-h simulations will be analyzed later.
Cases A–C: Cases associated with both a large-scale forcing and a local instability, that is, “mixed” forcing
The model results show a similar areal extent of the precipitation for both EXP (Fig. 9a) and NEXP (Fig. 10b), but the rainfall amount for NEXP is approximately 1.5 times larger than for EXP almost everywhere. The overestimation for NEXP is large in the region where the precipitation is not directly produced by the front: to the far east the maximum reaches 150 mm for NEXP and 90 mm for EXP. The areal extent of the precipitation is fairly reproduced by both NEXP and EXP; the 24-h OBS (Fig. 2e) does not have any precipitation report over this area, while the 12-h OBS (Fig. 2f) shows widespread rain, and the satellite imagery (Fig. 4a) shows clouds covering the region. The precipitation over the western Alps is completely missed by both EXP and NEXP, while the precipitation east of Genoa is overestimated. This may be related to the postfrontal convective precipitation, and we will address this point in the next section.
The comparison of OBS, NEXP, and EXP allows the assessment of overestimation of the amount of precipitation for this case.
Case D: A case associated with local forcing
A considerable overestimation of the precipitation, with respect to the OBS (Fig. 5e), is found for case D. Both NEXP (Fig. 9b) and EXP (Fig. 11b) produce heavy precipitation, mostly nonconvective, on the northeast side of the Po Valley, but the NEXP amount is 30% larger than that of EXP. On the contrary, precipitation is produced by the convective scheme (Fig. 11a) in the center of Italy. Overestimation of precipitation is found in the Balkan region, also. In this area the data show accumulated precipitation in the 12-h period (Fig. 5f) while no rainfall is recorded in the 24-h period (Fig. 5e). The satellite imagery (Fig. 6) shows clouds covering the region.
The overestimation of the precipitation is larger in this case than in the previous ones. This may be explained by the absence of a front driving the precipitation. The convergence area in the northeastern Po Valley and the southerly advection of moisture are responsible for the precipitation in this area, while in central Italy the rain is associated with a westerly flow (Fig. 5d), advecting humid air toward the Apennines, resulting in a local instability.
Case E: A case associated with large-scale forcing
In this case the precipitation is overestimated by both NEXP (Fig. 9c) and EXP (Fig. 12b) with respect to OBS (Fig. 7e). Similarities with case A are found: indeed the two simulations show that most of the differences are produced by the nonconvective precipitation, and the cumulus convective scheme does not show appreciable differences between EXP and NEXP (Fig. 12a). The meteorological situation may explain this behavior: this event is characterized by frontal unstable air crossing the Alps producing precipitation on the eastern side of the northwestern Alps that is still well reproduced by both NEXP and EXP, but large differences are found between the two forecasts of precipitation. NEXP produces up to 150 mm in the frontal region, as compared with 30 mm predicted by EXP. OBS shows maximum values reaching 63 mm in the same area, while the precipitation over the sea south of Italy is strongly overestimated by NEXP.
The areal extent of the precipitation related to a large-scale forcing is fairly well reproduced by both EXP and NEXP. However EXP underestimates the precipitation, while NEXP overestimates it; in NEXP the rainfall is mostly produced by the nonconvective scheme.
The comparison between EXP and NEXP for various meteorological events highlights the role played by the explicit computation of cloud water and rain, and it helps to evaluate the ability of the convective scheme to deal with the nonconvective precipitation. The nonconvective rain is reduced considerably for the cases associated with a well-defined large-scale forcing (E and part of A) if explicit computation of the cloud water and rain is used, allowing for both a better definition of water vapor and a better total water transport. On the contrary, the convective precipitation is only slightly reduced. The two cases associated with mixed and local forcing, respectively A and D, show a reduction in the amount of precipitation of the highly localized nonconvective cells. The events analyzed in this study are characterized by different meteorological situations: during event A, precipitation is partly related to postfrontal instability, and for case E to frontal instability. For case D, a low-level cold advection associated with a convergence area allowed for development of local thunderstorms. The 24-h accumulated total precipitation clearly shows a reduction of the rainfall if the explicit computation is used, but an overestimation is still found. The microphysical processes accounted for by the explicit computation of cloud water and rain may explain the different response to the various forcings: a front, associated with a pronounced temperature gradient, may result in change of phase with a conversion to ice or water vapor thus producing rain reduction; while the nonconvective parameterization (NEXP) only removes the excess of water over saturation as rain, and latent heat is added to the thermodynamic equation.
In summary, a fair estimation of the amount of precipitation and a good definition of the rain cell position, as compared with the observations, are obtained if the explicit computation of cloud water and rain is used, for cases related to frontal instability. For cases related to local forcing better results are obtained using EXP rather than NEXP, but the overestimation is larger than in the other cases. These results suggest that the explicit computation of cloud water and rain improves the precipitation forecast in the Alpine region. Threat scores and bias values supporting this finding will be presented in section 5.
Impact of cumulus convection parameterization
The following analysis is performed comparing the three schemes, for the same cases as in the previous section.
Cases A–C: Cases associated with mixed forcing
The total precipitation does not show strong differences for the three simulations (Figs. 10b,d,f), but the partition between subgrid and resolved precipitation is different for each scheme. The AK scheme handles most of the precipitation as subgrid (Fig. 10e), while GR treats it as resolved (Fig. 10c). Similarities between GR and KF are found, but in KF the precipitation on the south Balkan region is produced by the convective scheme only (Fig. 10a). In this region, the comparison of the model simulations (Figs. 10b,d,f) with the OBS (Fig. 2e) shows an overestimation of the precipitation by all the schemes. Different rainfall is found in the Apennines, near the Gulf of Genoa: the KF scheme is not activated, GR produces approximately 50% of the total precipitation as subgrid, and AK produces 100% as subgrid precipitation. Furthermore, none of them reproduce the precipitation on the western Alps, which is due to postfrontal instability associated with weak uplifting. The latter may explain the lack of activation of all the schemes. The wind field (not shown) clearly shows the flow deflected around the Alps instead of going over, suggesting the advection of a stable dry layer of air.
The different response of the schemes to the forcing may be related to the trigger function of the convective scheme itself: both GR and KF are sensitive to thermal forcing at the surface. The KF scheme is based on convective available potential energy, and similarly GR on buoyant energy by the grid field. Furthermore, in KF there is a term that allows the activation of the scheme in the absence of grid-scale saturation; the term takes the form of an additional temperature increment to a lifted parcel based on the resolved-scale vertical motion. The GR trigger function depends on the rate of destabilization based on the change of the available buoyant energy due to the large-scale or subgrid effects. On the other hand AK is activated when conditional instability exists and the integrated moisture convergence in a column (Mt) exceeds a threshold value in the column.
Therefore the increasing local destabilization, by a weak advection of wet air from the sea toward the Apennines, is able to trigger GR, but not KF, which needs a rising motion. This case, actually, is not associated with deep convection.
Case D: A case associated with local forcing
The results do not show large differences for event D either. Heavy precipitation is produced on the northeastern side of the Po Valley by KF (Fig. 11b), GR (Fig. 11d) and AK (Fig. 11f). All the parameterizations overestimate the precipitation in this area producing 159 mm for KF, 160 mm for GR, and 112 mm for AK, while strong underestimation is found in the western Alps. The OBS shows values not higher than 90 mm (Fig. 5f) on the eastern side, and 900 mm (Fig. 5e) on the western Alps. Therefore, the three convective schemes fail in reproducing the precipitation on the west side, most of which is produced by the explicit scheme for both GR and KF, and by the convective scheme for AK on the east side.
A different local feature develops during the afternoon of 15 June: in the northeastern Alps a low-level convergence is responsible for the activation of all the schemes, the low-level convergence produces Mt values higher than the threshold in AK, but both the rising motion and the instability are weak and produce only poor convective precipitation for KF and GR. On the northwestern side of the Alps the flow is mostly northwesterly and gives rise to a downslope wind on the lee side of the Alps that is not able to trigger the convective schemes, and thus the explicit scheme probably did not reach saturation.
Again, the three cumulus convection schemes overestimate the precipitation on the west side of the Apennines near the Gulf of Genoa. This is mostly produced by the convective scheme for GR (Fig. 11c) and AK (Fig. 11e), and by both the convective and explicit schemes for KF (Figs. 11a,b). In this case, the precipitation is produced by a sea breeze that developed within the large-scale zonal flow (Fig. 5c). In the early afternoon, at low level, the westerly wind turns southwesterly, advecting humid air toward the Apennines ridge. This produces a local instability and rising motion, which triggers all the cumulus schemes. The overestimation of the precipitation is not as high as on the eastern edge of the Po Valley (Fig. 5e), suggesting that a different response is obtained for the different forcing;the mesoscale convergence is handled mostly at the resolved scale by GR and KF, but at the subgrid scale by AK. For the local sea breeze, all cumulus schemes are activated: AK handles the phenomena at subgrid precipitation only, and GR and KF at both resolved and subgrid-scale precipitation.
In the Balkan region, an overestimation of the precipitation is also found: the AK produces very localized heavy precipitation reaching values of 128 mm in the far east (Fig. 11f), while KF and GR (Figs. 11b,d) are closer to OBS (Fig. 5e). All of them correctly reproduce the position of the heavy and rapid precipitation event in southeastern Italy but underestimate the OBS, which records 156 mm (Fig. 6e).
Case E: A case associated with a large-scale forcing
The model simulations succeed in reproducing the precipitation pattern related to a front reaching the western Alps, on 20 June. Most of the precipitation is restrained to the western side of the Alps for all the schemes, and areas of large overestimation are shown. The KF scheme handles the precipitation both at subgrid and resolved scale (Figs. 12a,b) showing a good agreement with the OBS (Fig. 7e), but most of the precipitation is limited to the northwestern Alps. The GR scheme (Figs. 12c,d) produces highly localized precipitation over the Pyrenees reaching 61 mm, which is mostly produced by the resolved scale; the observed precipitation (Fig. 7e) reaches 7 mm in the same area. The AK scheme overestimates the precipitation in the same region (Figs. 12e,f), and farther east; furthermore most of the precipitation is produced by the subgrid scale (Fig. 12e). Both GR and KF fail in producing the prefrontal precipitation in the Po Valley: GR does not produce precipitation at all while AK produces aligned cells near the Gulf of Genoa and the northeastern Po Valley. This suggests a low sensitivity of GR and KF to a weak unstable environment and a weak uplifting, as in case A. The three schemes show a different response to the front: AK handles the precipitation mostly at subgrid scale, GR at resolved scale, and KF at both.
Therefore in this case, the frontal instability is capable of triggering AK, but not GR, while KF is triggered by both the instability and the rising motion, which is supposed to exist on the east side of the upper-level trough (e.g., Bluestein 1992).
The comparison among the model simulations using different cumulus convection parameterizations suggests a poor performance of the AK, while good results are achieved by both GR and KF.
It should be noticed that under the particular condition of zonal flow and advection of humid air, there is the tendency to produce precipitation on the Apennines near the Gulf of Genoa as for cases A and D (both cases without a well-defined large-scale forcing), by all the schemes. We may conclude that cases associated with a well-organized frontal system, such as case E, are able to trigger AK, which shows a tendency to handle the precipitation at subgrid scale. The GR and KF schemes are activated by the instability and the rising motion, respectively, but they give similar results. Localized deep convection, whether associated with a defined-scale forcing or not, is well reproduced by either KF or GR, as for case D; shallow convection is well reproduced by GR, as for case A.
The overall analyses suggest that both GR and KF, associated with explicit computation of the resolvable scale, are good parameterizations to forecast precipitation in the Alpine region. Similar results are obtained by Kuo et al.(1996) for a marine cyclone case, and by Wang and Seaman (1997) analyzing several different cases.
The different response of the cumulus schemes may be related to the trigger function (as already pointed out) and also to the vertical profile of the latent heat release: a prescribed vertical distribution of the convective heating is used by AK while ODEDP is used by KF. Several studies (Anthes et al. 1983; Kuo and Reed 1988; Reed et al. 1988) show that the cumulus convection plays an important role in deepening low pressure and in determining the precipitation rate. Indeed Kuo and Low-Nam (1990) found that the AK scheme produces a weaker deepening of the low pressure than the others and a time delay in the precipitation. Similar results are obtained in this study, but the meteorological characteristics of these experiments are different from those in Kuo and Low-Nam (1990).
Sensitivity to the initial conditions
An analysis is carried out on the sensitivity of the model results to the setting time of the initial conditions. Several simulations are performed varying the initial conditions and using explicit computation of cloud water and rain with KF (Table 1).
The initial conditions for case A (6 June at 1200 UTC) are changed setting them to 12 and 36h later than the reference time: one is a 24-h simulation starting 7 June at 0000 UTC (case F) and another is a 48-h simulation starting 8 June at 0000 UTC (case B). A third 48-h simulation starting 9 June, at 0000 UTC (case C) is performed to complete the analysis of this event. The 24-h simulations are performed for the other events (cases D and E), starting 12h later than reference times: case G starting 15 June, at 0000 UTC, and case H starting 20 June, at 0000 UTC.
The comparison of the overlapping periods for cases A, B, C, F shows the sensitivity of the model to the variation of the initial conditions. The 12-h accumulated precipitation ending on 7 June at 1800 UTC shows that case F (Fig. 13a) reproduces the frontal precipitation in northeastern France better than does case A (Fig. 13b) while underestimating it in the Balkan area; the OBS to be compared is shown in Fig. 2f. Furthermore the rainfall near the Gulf of Genoa is missed completely by case F and largely overestimated by case A; both cases miss the precipitation on the eastern side of the western Alps. On the contrary the rainfall is well reproduced in central Italy in both cases. Case A after 12h of simulation shows a surface pressure disturbed by the topography, which tends to lag behind the observed one. This produces a wrong flow in the western Alps for both cases A and F, with the latter showing a better agreement (not shown) with the analyses than the former. Furthermore both cases show difficulties in reproducing the analyzed pattern of the surface pressure in the Balkan region.
The precipitation associated with a front entering the Po Valley from the northwest is well reproduced by case C. The comparison of the 12-h accumulated precipitation ending on 9 June at 1800 UTC shows that case C (Fig. 13c) reproduces the OBS (Fig. 3f) better than does case B (Fig. 13d). On the contrary, the front leaving the Alps and approaching the Balkan area is better reproduced by case B than by C. Furthermore case C completely misses the precipitation in the Po Valley, and in northeast Italy, while case B succeeds in reproducing the precipitation only in the northeast. The barrier effect of the Alps is driving these simulations: a strong surface pressure gradient intensifies producing a deep low in northeastern Europe, which is not shown in the analyses of 9 June at 0000 UTC; case C does not produce the strong surface pressure gradient on the Alps because it starts with a weaker signal than case B, and the frontal system crossing central France is well defined.
In comparing cases D and G with the OBS in the overlapping period of 12-h accumulated precipitation ending on 15 June, at 1800 UTC, considerable differences are found in the Balkan region and in the north of Italy. On the northwest side of the Balkan region, in the Adriatic sea, and in the east and northeast the precipitation produced by case G (Fig. 14a) shows good agreement with OBS (Fig. 5f). Case D (Fig. 14b) produces precipitation farther inside the Balkan area than case G. In the northwestern Alps, both cases fail in reproducing the precipitation showing independence from the initial conditions and suggesting a strong localized instability that is not reproduced by the model at this resolution. In the northeast the precipitation related to the convergence area is well reproduced by both, but case G perfectly reproduces the OBS precipitation (90 mm). The differences between cases D and G are related to the dynamics reproduced by the two runs. This meteorological event is driven by a mesoscale circulation produced by a convergence flow (Figs. 5a,c) that developed on 15 June between 0000 and 1200 UTC in the northeast Alps. The initial conditions do not clearly represent this feature for case D (not shown), while the signal is present in the initial condition for case G (Fig. 5a). Similarly for the differences in the Balkan region, a cyclone developing in that region on 15 June at 0000 UTC (Fig. 5a) is well represented by the initial conditions for case G, but not for case D (not shown). Therefore the mesoscale phenomena are well reproduced by the model if the signal is well defined in the initial conditions, otherwise the model may tend to end the phenomena.
Finally the comparison of the cases E and H with OBS (Fig. 7f) for the overlapping period of 12-h accumulated precipitation ending 20 June at 1800 UTC shows that the precipitation, associated with a front is better reproduced by case H (Fig. 14d) than by case E (Fig. 14c). However, these cases show fewer differences than cases A–F. Indeed both cases correctly reproduce the frontal rainfall, although in both cases a small time lag is present.
Several other simulations for cases A, B, C, and F are performed to understand better the sensitivity of the model to the initial conditions. The same simulations were aimed at studying the apparent improvement in forecasting the frontal precipitation when the run was started only a few hours before the event. The results (not shown) suggest that a front approaching the Alps from the northwest is better reproduced by the model when the simulation starts only 6h before the event while a front leaving the Po Valley and approaching the Balkans is better reproduced when the simulation starts earlier. Furthermore the sensitivity of the model to the initial conditions seems to depend on the strength of the meteorological forcing although a different response is obtained for cases A–F and E and H. Cases A–F are related to a frontal system approaching the north side of the Alps and entering the Po Valley only in the late stage as postfrontal air (case B); cases E and H are related to a strong front entering the same area in the early stage. Therefore the initial conditions include a strong signal for cases E and H but not for cases A–F. Finally cases D–G are related to a mesoscale circulation that is locally strong: the simulations performed using the initial conditions, which include the signal (case G), correctly reproduce the precipitation; case D, on the other hand, which starts with different initial conditions, gives quite dissimilar results.
To verify our finding, the bias and the threat scores for the precipitation are computed. The scores, for both NEXP and EXP for the three cumulus convections and for all the meteorological cases, are given in Tables 2 and 3. The mean values of the scores, for 6-h rainfall simulation between 6- and 48-h forecasts, are computed accounting for the stations closest to the grid points (within 15 km). We analyzed the bias and the threat scores for different thresholds, but no considerable differences are found with respect to the one at 0.25 cm. Therefore the following discussion will focus only on this value. The bias (Table 3) shows an overestimation of the areal extent of the precipitation, while the previous analyses of the maps show an overestimation of the amount of the rainfall. Therefore the bias confirms our previous finding that NEXP overestimates the precipitation for all the convective schemes regardless of the meteorological situation, with AK showing higher values than the other schemes for each case. The threat score does not show strong differences from the bias and reaches the highest value (0.3), with the bias reaching 1.6, for case B using AK. Generally for all the schemes and for both the bias and the threat score, EXP shows better values than NEXP, and EXP associated with AK reaches higher values than EXP associated with GR and KF. The large overestimation of the amount of precipitation all over the domain obtained using AK explains the high values of the threat score. Furthermore, the threat score does not show much signal. Therefore the following discussion will focus on EXP for GR and KF only. The highest value of the threat score related to a good value of the bias is achieved for case B by both GR and KF. Case D shows poor values for both the threat score and the bias for every cumulus scheme, suggesting model difficulties in reproducing cases associated with mesoscale phenomena or local forcing. Although case E is driven by a strong large-scale forcing, the bias shows an overestimation for both GR and KF.
Overall GR and KF show good skills, depending on the meteorological situation: GR seems better to reproduce cases associated with a well-defined forcing as for cases B and C. Based on the mean value of the bias, KF is the scheme that better reproduces the precipitation in the Alpine region.
The analyses of simulations of meteorological events in the Alpine region, using several cumulus convective schemes with and without the explicit computation of cloud water and rain, help to evaluate the ability of the various schemes to reproduce the nonconvective precipitation under different forcings. Furthermore, they confirm previous findings (Wang and Seaman 1997; Kuo et al. 1996, etc.) that hybrid schemes should be used at this grid resolution, as stated by Molinari and Dudek (1992). The results clearly show a reduction of the amount of precipitation using EXP for meteorological events driven mainly by large-scale forcing. If the front is not well defined, the differences between the two schemes are reduced. The precipitation produced by events associated with a local forcing is poorly reproduced by both approaches, but NEXP is worse. Apparently, this is related to the microphysical processes: with the explicit scheme, the excess of water vapor over saturation is condensed into cloud water and partially converted to rain and part of it may undergo subcloud-layer evaporation, but for NEXP all the excess water over saturation is converted to rain. The reduction of the amount of precipitation is linked to the meteorological parameters and to the strength of the forcing. In any case, the model shows difficulties in handling the precipitation related to mesoscale phenomena.
The comparison of the three cumulus convective schemes shows a poor performance of AK, which has the tendency to handle the precipitation at subgrid scale producing a strong overestimation, regardless of the meteorological event. Apparently, this effect is due both to the trigger function and to the assumption of a given latent heat vertical profile. On the contrary, both GR and KF show good skill in reproducing most of the precipitation events related to several meteorological phenomena, but differences are found depending on the forcing. The results highlight the tendency of KF to handle the precipitation at the subgrid scale for cases associated with rising motion (such as strong frontal instability or deep local convection), while GR shows the same results for cases of local convection or weak frontal instability.
The bias and the threat scores confirm these findings, showing higher values for EXP than NEXP, and for GR and KF than AK.
The sensitivity to the initial conditions suggests, as expected, that the model precipitation is linked to the signal included in the initial condition itself. The strength of the forcing is crucial for the skill of the model to reproduce the precipitation: cases associated with a strong front (E and H) or strong local forcing are always well reproduced, while the ability of the model to reproduce the precipitation depends critically on the initial conditions for cases associated with a weak signal (e.g., cases A–F). This suggests a delicate equilibrium between the strength of the signal and the topography: if a strong pressure gradient is present in the initial conditions, as for case B, the model shows the tendency to enhance the barrier effect. Therefore either a better preprocessing of the data analyses or an improvement of the model to deal with steep mountains may help to solve the problem.
Improvement in the forecast of precipitation is obtained by performing simulations at higher resolution than that used in this work and using the nonhydrostatic version of the model. A simulation performed for case D using the nonhydrostatic approximation and 10-km grid size shows good agreement with OBS (Paolucci et al., 1997), strongly reducing the precipitation on the northeastern side of the Alps.
Therefore, because the low resolution may be a limiting factor for reproducing local convection in an area with complex orography, model simulations using higher resolution than the one used for this study may be a good direction for future work. In conclusion, for forecasting precipitation in the Alpine region the use of the explicit scheme associated with KF for the case of strong local convection and that associated with GR for the case driven by large-scale forcing, and a resolution higher than the one used in this work, are recommended. Furthermore, the strong sensitivity to the initial conditions requires an improvement of data analyses that may be obtained by assimilating local data.
This research was partly supported by the Italian Electric Power Agency (ENEL) who made available the database MATREP. NCAR is also acknowledged for the MM5 model.
Anthes, R. A., 1977: A cumulus parameterization scheme utilizing a one-dimensional cloud model. Mon. Wea. Rev.,105, 270–286.
Anthes, R. A., and T. T. Warner, 1978: Development of hydrostatic models suitable for air pollution and other mesometeorological studies. Mon. Wea. Rev.,106, 1045–1078.
Anthes, R. A., Y.-H. Kuo, and J. R. Gyakum, 1983: Numerical simulations of a case of explosive cyclogenesis. Mon. Wea. Rev.,111, 1174–1188.
Anthes, R. A., E.-Y. Hsie, and Y.-H. Kuo, 1987: Description of the Penn State/NCAR Mesoscale Model Version 4 (MM4). NCAR Tech. Note NCAR/TN-282+STR, 66 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307.].
Arakawa, A., and W. H. Schubert, 1974: Interaction of a cumulus cloud ensemble with the large-scale environment, Part I. J. Atmos. Sci.,31, 674–701.
Bluestein, H. B., 1992: Synoptic–Dynamic Meteorology in Midlatitudes. Volume 1: Principles of Kinematics and Dynamics. Oxford University Press, 431 pp.
Buzzi, A., C. Cacciamani, T. Paccagnella, P. Patruno, and N. Tartaglione, 1995: Preliminary meteorological analysis of the Piedmont flood of November 1994. MAP Newsl.,2, 2–6.
Ferretti, R., P. Antonelli, and G. Visconti, 1996: Preliminary results of the Piedmont flood of 4–6 November 1994. MAP Newsl.,5, 50–51.
Fritsch, J. M., and C. F. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci.,37, 1722–1733.
Grell, G. A., 1993: Prognostic evaluation of assumptions used by cumulus parameterization. Mon. Wea. Rev.,121, 764–787.
Grell, G. A., J. Dudhia, and D. R. Stauffer, 1994: A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 122 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307.].
Hsie, E. Y., 1984: Simulation of frontogenesis in a moist atmosphere using alternative parameterization of condensation and precipitation. J. Atmos. Sci.,41, 2701–2716.
Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci.,47, 2784–2802.
Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large-scale flow. J. Atmos. Sci.,31, 1232–1240.
Kuo, Y.-H., and R. A. Anthes, 1984: Semiprognostic tests of Kuo-type cumulus parameterization schemes in an extratropical convective system. Mon. Wea. Rev.,112, 1498–1509.
Kuo, Y.-H., and R. J. Reed, 1988: Numerical simulation of an explosively deepening cyclone in the eastern Pacific. Mon. Wea. Rev.,116, 2082–2105.
Kuo, Y.-H., and S. Low-Nam, 1990: Prediction of nine explosive cyclones over the western Atlantic Ocean with a regional model. Mon. Wea. Rev.,118, 3–25.
Kuo, Y.-H., R. J. Reed, and Y. Liu, 1996: The ERICA IOP 5 storm. Part III: Mesoscale cyclogenesis and precipitation parameterization. Mon. Wea. Rev.,124, 1409–1434.
Lord, S., 1978: Development and observational verification of a cumulus cloud parameterization. Ph.D. dissertation, University of California, Los Angeles, 359 pp.
Molinari, J., and M. Dudek, 1992: Parameterization in mesoscale numerical models: A critical review. Mon. Wea. Rev.,120, 326–344.
Paccagnella, T., S. Tibaldi, R. Buizza, and S. Scoccianti, 1992: High-resolution numerical modelling of convective precipitation over northern Italy. Meteor. Atmos. Phys.,50, 143–163.
Paccagnella, T., P. Patruno, and C. Cacciamani, 1995: Operational quantitative precipitation forecast of the Piedmont flood event at the Regional Meteorological Service of Emilia-Romagna. MAP Newsl.,2, 7–11.
Paolucci, T., R. Ferretti, and G. Visconti, 1997: Numerical simulations of summer convective precipitation cases in the Alpine region using different parameterizations: Preliminary results. Preprints, Seventh PSU/NCAR Mesoscale Model User’s Workshop, MMM Division, NCAR, 64–67.
Reed, R. J., M. D. Albright, A. J. Simmons, and P. Unden, 1988: The role of latent heat release in explosive cyclogenesis: Three examples based on ECMWF forecast. Wea. Forecasting,3, 217–229.
Wang, W., and N. L. Seaman, 1997: A comparison study of convective parameterization schemes in a mesoscale model. Mon. Wea. Rev.,1275, 252–278.
Zhang, D. L., and R. A. Anthes, 1982: A high-resolution model of the planetary boundary layer—Sensitivity tests and comparisons with SESAME-79 data. J. Appl. Meteor.,21, 1594–1609.
Summary of the model simulations for each case (first column) is given: the starting and ending time, day and hour (second column); the cumulus convective schemes (third column); and the non-convective one (last column).
Mean value of the threat score, for 6-h rainfall simulation at the closest grid point, for the three cumulus schemes, and for both EXP and NEXP for each case is given.
Mean value of the bias score, for 6-h rainfall simulation at the closest grid point, for the three cumulus schemes, and for both EXP and NEXP for each case is given.