High-resolution numerical models can be effective in monitoring and
predicting natural hazards, especially when dealing with Mediterranean
atmospheric and marine intense/severe events characterised by a wide range
of interacting scales. The understanding of the key factors associated to
these Mediterranean phenomena, and the usefulness of adopting
high-resolution numerical models in their simulation, are among the aims of
the international initiative HyMeX – HYdrological cycle in Mediterranean
EXperiment. At the turn of 2013, two monitoring campaigns (SOPs – Special
Observation Periods) were devoted to these issues. For this purpose, a new
high-resolution BOlogna Limited Area Model-MOdello LOCale (BOLAM-MOLOCH) suite was implemented in the Institute for Environmental Protection and Research
(ISPRA) hydro–meteo–marine forecasting system (SIMM –
Forecast verification is an essential activity of any operational
institution or centre (dealing with numerical predictions) that arise from
the need to constantly assess the skill and value of the forecasts provided
by numerical models (see, e.g., Jolliffe and Stephenson, 2011). A
statistical verification approach is mandatory to provide a robust and
reliable assessment through time of the system performance, and an
evaluation of the impact of any modification introduced in the system
(
Nevertheless, erroneous conclusions on forecast performance could be drawn from case-study verification if the predictability of the weather systems under investigation is not taken into consideration. In particular, focusing on cyclones producing high-impact weather over the Mediterranean (Jansa et al., 2014), highly predictable atmospheric processes, such as large synoptic disturbances undergoing Alpine lee cyclogenesis (see, e.g., Speranza et al., 1985), and much less predictable atmospheric processes, such as moist cyclogenesis in presence of complex orography (see, e.g., Romero, 2011), can give rise to similar patterns producing the same kind of ground effects. Numerical weather prediction (NWP) models provide for the former, and not for the latter, correct and timely early warning. This arises from the intrinsic physical properties of the involved weather systems, rather than the NWP model skill.
The necessity to enhance the knowledge (and to assess the predictability) of
high-impact atmospheric and marine events in the Mediterranean Basin, and to
properly resolve the scales involved that range from synoptic to the
meso-gamma scale, led to the establishment at the beginning of the 21st
century of the “HYdrological cycle in Mediterranean
EXperiment” (HyMeX,
An ad hoc forecasting activity (Ducrocq et al., 2014; Ferretti et al., 2014) based
on several numerical model forecasts was set up during each SOP to identify
in advance the Intense Observation Periods (IOPs) to be monitored. For this
specific activity, the Institute for Environmental Protection and Research (ISPRA) provided the meteorological products of its
operational hydro–meteo–marine forecasting system called
Nine out of sixteen IOPs monitored during the first SOP campaign affected
the three Italian HyMeX hydrometeorological sites (North-Eastern Italy
(NEI), Liguria–Tuscany (LT), and Central Italy (CI)). Among these, IOP16
(25–29 October 2012) and IOP18 (31 October–1 November 2012) deserved
particular attention. Although the weather systems associated with these
IOPs were rather different from a synoptic point of view (and, hence, in
terms of predictability), they displayed some similarity in terms of ground
effects, producing heavy precipitation over the NEI, LT and CI sites and
intense storm surge events over the northern Adriatic coastline and the
Venice Lagoon (the so-called
The paper is organised as follows. The description of the operational SIMM and of the HyMeX-based SIMM is depicted in Sect. 2. Section 3 provides the synoptic description of IOP16 and IOP18. Forecast verification results for both the meteorological and sea storm surge components of the two analysed configurations of SIMM are presented in Sect. 4. Conclusions and final remarks are reported in Sect. 5.
The first chain considered, the operational SIMM, is a quite recent upgraded configuration of the chain originally implemented in 2000 and later updated for the meteorological component in 2009 (Speranza et al., 2007; Mariani et al., 2014). This chain adopts the 2009 version of the hydrostatic BOLAM meteorological model and two new components for the wave and storm surge forecasts (see Fig. 1).
The operational SIMM is initialised on a daily basis by means of
0.5
A sketch of the general processing scheme for the
operational SIMM: from the ECMWF forcing to the storage, online publication
and use for research and dissemination of weather, marine and coastal
products. The operational configuration includes the BOLAM meteorological
model nested at 0.3 and 0.1
BOLAM is one-way nested over two domains (Fig. 2) covering the Mediterranean
basin, with horizontal spatial size equal to 0.3
The 0.1
Computational domains for the meteorological component of the
operational SIMM, namely the 0.3
Simultaneously, the 0.1
The SHYFEM low-resolution finite-element grid, with a zoom over
the Venice Lagoon. After
Two finite-element grids for the Mediterranean Sea are used in SHYFEM for
both the BOLAM and IFS initialisation: a low-resolution grid with 13 180
elements (Table 2 and Fig. 3); and a high-resolution grid with 50 409
elements (Table 2). For each grid, the operational system carries out a
first run to predict the storm surge contribution over the entire
Mediterranean, followed by a second run to calculate the total water level,
that is, the astronomical tide added to the storm surge contribution (or
tidal residual) The storm surge contribution includes here the
seiches, as well.
A data assimilation module based on the 4-D physical space assimilation system is also present (Bajo et al., 2012) for integrating the residual sea
level measurements from the tide gauges of the ISPRA observation network
located alongside the Italian northern Adriatic coastline (hereinafter
referred to as RMLV – see for more details
As in Fig. 1, but for the SIMM configuration (not yet
operational) designed for the HyMeX campaigns. This configuration includes
the BOLAM meteorological model nested at 0.07
Schematic description of the principal model settings of the operational SIMM and the HyMeX-based SIMM.
A newer configuration of the SIMM forecasting chain was designed for the HyMeX campaigns. The new chain, depicted in Fig. 4, differs from the operational one for using the higher-resolution BOLAM-MOLOCH suite in place of lower-resolution, two-domain nested BOLAM configuration.
Parameters of the low- and high-resolution finite-element grids of SHYFEM.
Synoptic analysis charts at 00:00 UTC. Upper panels for IOP16:
The BOLAM configuration adopted within this suite is based on a newer
version (dated 2012) of the numerical code and uses a wider domain, covering
an area 54–25
The non-hydrostatic MOLOCH model is then nested into the 0.07
The wind fields modelled by the BOLAM-MOLOCH suite are then used for
initialising Mc-WAF (see Casaioli et al., 2014), whilst only the
0.07
During the IOP16, a large Atlantic cyclone crossed the Iberian Peninsula on
25–26 October. Then, it moved rapidly eastwards, reaching on 27 October the
Genoa Gulf (Fig. 5a), where it was reinforced and made almost stationary by
the phenomenon of Alpine cyclogenesis during the next day (Fig. 5b).
Finally, it left Italy on 29 October. Observations of the weather system
development were provided by several operational and specifically deployed
instruments, including ground meteorological networks, radiosondes, and
ground- and satellite-based remote-sensing instruments. In the morning of 27
October, an instrumented SAFIRE Falcon 20 aircraft was additionally used to
obtain airborne measurements over the CI site. These data were collected
into the HyMeX database (
The event occurred during the IOP18 was characterised by the formation of a Mediterranean low-level cyclone over the Gulf of Lion on 31 October (Fig. 5c), embedded in the zonal flow connected to an upper-level trough (not shown). Subsequently, the cyclone moved along a similar path of the IOP16 one (Fig. 5d). The difference between the two events was mainly in the time and space scale of the dynamical and physical processes involved. This, in turn, reflected in a different predictability of the two weather systems. Also in this case, several detailed and specific observations were collected in order to monitor and study the key processes responsible for this event.
Both events produced intense precipitation over Italy, with some similarity in the rainfall patterns. The precipitation over the Italian HyMeX hydrometeorological sites, and also over southern Italy, was associated with southerly low-level advection of warm and moist air and subsequent frontal passage.
Over LT, strong advection of warm and moist air over steep and complex
orography acted in both cases to concentrate intense precipitation in very
small areas. Over CI, the key factor was the prediction of the development
of multiple squall lines over the Tyrrhenian Sea, embedded in the warm and
unstable south-westerly flux. These arose in several precipitation bands
over peninsular Italy, whose exact space-time location was hard to predict.
Over NEI, the two IOP events are typical examples of the so-called “dark
Bora” phenomenon, that is, a cyclonic Bora wind with cloudy sky and
precipitation (Jurčec, 1981). A warm south-easterly Scirocco wind
flowing over the Adriatic Sea swerved westwards (following the classical
Bora NE–SO wind direction), due to the barrier effect jointly produced by
the Dalmatian orography and the cold, stable air present over the
north-eastern Italian region during the passage of the occluded front. This
phenomenon, matching the astronomical tide, was also responsible for the
The evolution of these tidal events was monitored through the national RMLV
network. At the Punta della Salute tide gauge, the sea level exceeded twice
the warning level during IOP16 (more than 120 cm), whilst it exceeded once
the alarm level during IOP18 (143 cm at 00:40 UTC of 1 November). According
to historical records, the latter peak represents the 16th maximum
tidal level observed in Venice at Punta della Salute since 1872. The same
meteorological event was responsible of a tide peak Value
expressed with respect to the mean sea level computed by averaging the
observations of sea level in Punta della Salute carried out over 25 years,
from 1885 to 1909, and setting the central year (1897) as a reference
value.
Each operational modelling component of SIMM has been objectively evaluated
during the entire lifetime of the system (see, e.g.,
The 0.07
The assessment of the Mc-WAF component during the two HyMeX SOP campaigns
was presented by Casaioli et al. (2014) in a distinct paper due to the
complexity of this modelling component that spans from the Mediterranean to
local coastal scales using, for its initialisation, the meteorological
fields provided by both the currently operational (0.1
Concerning SHYFEM, Cordella (2013), Ferla (2013), and Coraci (2014) computed, for the northern Adriatic Sea, the impact on forecast performance of the different meteorological forcing (BOLAM vs. IFS) used within the operational SIMM. These studies considered 1 year of simulations – from October 2012 to October 2013 – that included mostly of the two SOPs. Although the performances are generally good and quite similar over the considered period, the BOLAM-forced SHYFEM overperforms the IFS-forced SHYFEM when considering high tide events (> 80 cm) and longer lead times. The following Sect. 4.2 provides a comparison of the two SHYFEM configurations that belong, respectively, to the operational SIMM and the HyMeX-based SIMM. Since the latter configuration is not yet fully operational, the comparison is here available only for the period covering the IOP16 and IOP18. In addition, the comparison does not consider the simulations obtained by deploying the data assimilation scheme that is currently under development.
A quantitative precipitation comparison for a period covering both IOPs (25 October–1 November 2012; see Fig. 6) is suitable to assess the overall forecast performance of the three NWP models. For this purpose, attention is focused on the three Italian HyMeX hydrometeorological sites.
First, hourly rainfall data provided by Italy to the HyMeX database were
considered and accumulated on a daily basis starting from 00:00 UTC.
Similarly, hourly precipitation forecast fields were accumulated from 00:00 to 24:00 UTC
(i.e.,
Comparison between the daily rainfall observations and the daily
precipitation fields generated by the tree inter-compared meteorological
models. The time series are averaged over three selected areas:
In the majority of cases, the 0.1
In addition, a statistically robust verification was performed for MOLOCH by assessing the quantitative precipitation forecast (QPF) performance over the period September–December 2012 that includes the first SOP devoted to the monitoring of high precipitation events. The QPF comparison focused on the entire MOLOCH domain (see Fig. 2). Hence, observations from rain gauge networks of Austria, Bosnia and Herzegovina, Croatia, France, Italy, Slovenia, and Swiss were considered.
Categorical scores calculated over the period September–December 2012
for the 24 h MOLOCH forecasts.
Several categorical scores were calculated (Fig. 7): the frequency bias
(BIAS), the probability of detection (POD), the probability of false
detection (POFD), the false alarm ratio (FAR), the equitable threat score
(ETS), a bias-adjusted ETS (ETSA
Before populating the contingency tables, both observations and forecasts
were accumulated on a daily basis from 06:00 to 06:00 UTC. This time
period was chosen to include into the verification observational analysis
those Croatian and Slovenian rainfall data that were already 24 h
accumulated over this time slot when stored in the HyMeX database. The
observed and forecast fields were up-scaled over a common 0.05
A detailed description of the verification procedure and of the various tools used can be found in Mariani et al. (2014).
Over the 3-month period and the up-scaled verification domain, the MOLOCH configuration tends to overpredict the occurrence of precipitation events since a BIAS value slightly greater than one is obtained for all thresholds (Fig. 7a). In other words, the number of false alarms – that is, the rainfall events predicted but not observed – is not negligible and it is greater than the number of observed rainfall events not detected (i.e., the misses). This is a signal that MOLOCH, although providing more realistic precipitation patterns, tends to produce an excess of rainfall. It is also remarkable that BIAS lightly decreases for increasing thresholds.
In Fig. 7a the FAR values provide a quantification of the fraction of events forecast but not observed. Since for a perfect forecast system FAR should be equal to zero, it is even clearer that, for this configuration of MOLOCH, the number of false alarms is not negligible. It could be surprising that the trend of FAR does not follow that of BIAS, but this is the consequence of the fact that, unlike the latter, the former does not take into account the number of misses in the verification sample. The other two scores present in Fig. 7a are POD and POFD that measure, respectively, the fraction of events observed that were correctly forecast and the fraction of no-events (observations below the threshold) that were incorrectly forecast. For a perfect forecast system, POD (POFD) should be equal to one (zero). MOLOCH shows higher POD values at the lower thresholds. For the higher thresholds, there is a decrease of the number of events observed that were properly predicted. The values of POFD are very low and quite close to zero, due to the magnitude of the no-events (i.e., observations and forecasts both below the considered rainfall threshold) with respect to the false alarms.
24 h accumulated precipitation on 26 October 2012. Focus over
Italy:
The MOLOCH skill scores are reported in Fig. 7b. A perfect forecast system
has all skill scores equal to one. Each one of these scores assesses a
different attribute of the forecast quality and some of them are more (e.g.,
ETS) or less sensitive (e.g., ETSA
Finally, EDI and SEDI calculated over the re-calibrated forecasts (i.e., bias removed), show that there is no significant difference in the model performance between the thresholds associated with low-base rate rainfall events and those associated with high-base rate.
At this stage, it is not possible to judge how satisfying these scores are since no other MOLOCH configurations were previously implemented in SIMM. Moreover, a direct comparison with the two hydrostatic BOLAM configurations is unfair due to the significant differences in the spatial scales resolved. Higher-resolution NWP models produce detailed forecast fields that contain differing degrees of small-scale details. Such forecasts could score worse than smoother fields modelled by lower-resolution NWP models due to the sensitivity of categorical scores to small displacement errors (“double penalty” – see, e.g., Mass et al., 2002; Weygandt et al., 2004; Lanciani et al., 2008; Gilleland, 2013).
In the framework of the HyMeX Science Team “Studies of IOPs (SOP1) – precipitation events”, an intercomparison study is
ongoing for evaluating the forecast performance of the different NWP models
used during the first SOP. This includes the assessment of the ISPRA 2.5 km
MOLOCH against the two MOLOCH configurations operational at CNR-ISAC, having
horizontal grid spacing of 1.5 and 2.3 km (see Table 4 in Ferretti et al.,
2014). According to preliminary results (not shown), no big differences in
score were found among the three MOLOCH configurations, except for a
(slightly) larger tendency of the ISPRA configuration to overpredict the
occurrence of rainfall events (i.e., BIAS
During IOP16, intense precipitation over Italy occurred in two phases. On 26
October, warm and moist south-westerly advection towards Liguria and CI
created the favourable conditions to the development of pre-frontal
convective cells (Fig. 8a). Particularly intense rainfall occurred over Liguria due to
orographic triggering. The rain gauge in Piana di Battolla (see red dots in
Fig. 8a) registered 245 mm 24 h
Contour plot of the CAPE forecast at 15:00 UTC of 26 October 2012
over the LT hydrometeorological site and 600–925 hPa shear for the
0.1
An overall comparison of the 24 h accumulated precipitation fields predicted
on 26 October by the three meteorological models (initialised at 12:00 UTC
of 25 October; Fig. 8b–d) evidences a strong similarity in the
rain/no-rain patterns with relevant differences in the small-scale details,
including the maximum precipitation amount. In particular, the rainfall peak
observed in Liguria (Fig. 8a) is misplaced by the 0.1
As in Fig. 8, but for the 48 h accumulated precipitation starting at 00:00 UTC of 31 October 2012. The forecasts were initialised at 12:00 UTC on 30 October.
On 27 October, all the three models are able to correctly forecast the
location of the rainfall patterns over the HyMeX hydrometeorological sites
(not shown). In this case, the 0.1
A further analysis of the LT event allows the aforementioned forecast
differences to be clarified. The CAPE forecast fields at 15:00 UTC of 26
October (Fig. 9) illustrate the situation leading to heavy precipitation on
a very small area in the subsequent hours (westernmost red dots on
Liguria
in Fig. 8a). There, a persistent southerly low-level jet on the eastern
Genoa Gulf advects moist, unstable air on the steep orography, generating a
long-lasting stationary convective cell. The unstable low-level jet is
evident as a high-CAPE area in all three forecasts. A low-CAPE area is
present west of Corsica in the 0.07
During IOP18, rainfall affected the same Italian areas (Fig. 10a). On 31 October,
south-westerly and southerly moist-advection-induced orographic
precipitation over the Lazio Region and LT with isolated peaks of more than
140 and 120 mm 24 h
Most of the features observed in the complex precipitation pattern of Fig. 10a
are caught by the three models for this IOP (initialised at 12:00 UTC of
30 October, Fig. 10b–d). In particular, with respect to the 0.1
Details added in the rainfall pattern by the MOLOCH forecast (Fig. 10d) seem
to be realistic, even if their verification requires a deeper inspection of
the NEI event alone, in view of its relationship with the severe
It must be noted that the IOP18 forecast depends critically on ICs. For
instance, if the forecast is initialised one day earlier, that is, at 12:00 UTC
of 29 October, the 24 h accumulated precipitation on 31 October turns
out to be largely overestimated, and the position of the maximum rainfall
over NEI turns out to differ in the two BOLAM and in the MOLOCH simulations
(not shown). More specifically, this manifestation of chaos affects the
evolution of the mesoscale cyclone (and the associated surface winds)
developing over the Mediterranean Sea, with relevant implications on the
prediction of the sea storm surge in the Venice Lagoon and the northern
Adriatic Sea with more than 2 days in advance (as shown later in Sect. 4.2).
Figure 12 illustrates this point through the comparison of the
EUMETSAT Meteosat Second Generation (MSG) 6.2
Contour plot of the 925 hPa (
Synoptic-scale verification of BOLAM forecasts using MSG WV
imagery at 00:00 UTC of 1 November 2012; the red ellipse represents the
cyclone vorticity centre.
Two subsequent 0.1
Further elements on the performance of the two SIMM configurations in
predicting low-probability/exceptional storm surge events can be provided by
assessing the SHYFEM forecasts for IOP16 and IOP18. The verification study
presented in this section deals with the SHYFEM model forced using the
meteorological fields provided by the 0.1
The SHYFEM simulations considered here refer to the period that goes from the beginning of IOP16 to the end of IOP18. Since each daily simulation spanned for several days (see “forecast range” in Table 1), observations were compared against SHYFEM forecasts with different lead times (i.e., the time from the run initialisation) to study the forecast sensitivity to ICs. In the rest of this section, the SHYFEM simulations are then referred with respect to the delivery date, and plotted accordingly in Figs. 14–17 (i.e., different colours correspond to different delivery date).
The use of MOLOCH in combination with SHYFEM will be the subject of a subsequent verification study. For this purpose, the new domain covering the entirety of Italy (see Sect. 2.2) must be considered; the MOLOCH domain used in the HyMeX configuration is too small to correctly drive the surge model.
The simulations of the three model configurations were checked against observations taken at six different tide-gauge stations. Five of them, from the ISPRA RMLV network, are located in the Venice Lagoon: namely Burano, Punta della Salute, Torson di Sotto, Faro Rocchetta, and Chioggia Vigo. The sixth one is the Venice Municipality tide gauge placed at the CNR Piattaforma. These locations have been chosen among the 23 measuring points available in the Venice Lagoon – together with the open-sea CNR platform – in order to provide a cross-shaped sample of the variability of both observed and forecast sea levels along and across the lagoon (Fig. 13). Since it was found that the grid resolution had a negligible effect on the SHYFEM performance for these events (not shown), only the results obtained with the high-resolution SHYFEM configurations are discussed here.
Concerning IOP16, all the model configurations provide a good tidal forecast, both in open-sea (Fig. 14a, c, e) and in the lagoon (Fig. 14b, d, f, for Punta della Salute), with a correlation coefficient varying from 0.89 to 0.97 depending on lead time, meteorological forcing and measuring station.
Position of the six tide-gauge instruments considered for the evaluation of the SHYFEM performance. Five stations are situated in the inner of the Venice Lagoon, whilst one is located in the Adriatic Sea.
For the ECMWF IFS-forced SHYFEM configuration, the latest delivered forecast
of the 28 November peak is very good. In all the other cases, this
configuration tends to underestimate the elevation peaks of an amount that,
as it could be expected, is minimum for the shortest-lead time run (see
Fig. 14a and b). This error is mostly absent in the forecast forced by the
0.1
Comparison for IOP16 between the SHYFEM forecasts and the sea
level observations registered at Piattaforma and Punta della Salute. Time in
CET (UTC
Comparison for IOP18 between the ECMWF IFS-forced SHYFEM
forecasts and the sea level observations registered at the six considered
tide-gauge instruments, with meteorological forecast forcing from 28 to 31 October:
As in Fig. 15, but with the meteorological forecast forcing
provided by the 0.1
The behaviour of the SHYFEM configurations during the IOP18 event is by far
more complex. Let us start with the description of the results obtained with
the forcing provided by the ECMWF-IFS meteorological fields (Fig. 15). Along
with an overall tendency of the SHYFEM runs to underestimate the main peaks,
a sort of anomalous behaviour is found in relation with the forecast quality
when increasing the lead time (hereinafter referred as to the
The results from the BOLAM-driven SHYFEM runs (Figs. 16 and 17 for the low-
and high-resolution case, respectively) display a general increase of
forecast quality. However, they are prone to lead time anomaly and
north/south lagoon unbalance, although at a different extent depending on
model resolution and initialisation date. The lead time anomaly is still
found: the forecast delivered on 29 October is much more accurate than that
delivered on 30 October, except for Burano (Fig. 16a) and Punta della Salute
(Fig. 16b). In these two locations (especially in Burano), the forecast
delivered on 29 October strongly overestimates the peak due to heavy
north/south lagoon unbalance. The forecast delivered on 30 October
underestimates the sea surge everywhere, producing the maximum error in
Chioggia (
When using the 0.07
The reason of the unbalance effect can be easily found in the observed,
strong north-northeasterly wind blowing over the lagoon in the late night of
31 October. Wind pushed the lagoon water southwards increasing the sea level
difference in the north–south direction. Thus, the behaviour of the SHYFEM
error can be easily related to the error associated with the forecast of
surface winds provided by the different BOLAM simulations. To further
clarify this issue, the 10 m wind field observed over NEI at 00:00 UTC of 1 November
– calculated by the Environmental Agency of the Veneto Region
(ARPA Veneto) using the CALifornia METeorological model (CALMET; Sansone et al., 2005) – is compared
against the corresponding BOLAM simulations obtained with different
initialisations (see Fig. 18). The 0.1
Figure 18 is also very effective in showing how much the dynamically relevant differences in the mesoscale cyclone shown in Fig. 12 can affect key features in forecasting both the rainfall and the sea surge event, namely: the veering of Scirocco flow over the northern Adriatic Sea (and its timing, not shown); the front sharpness; the wind speed and direction inside and outside the lagoon (which depend on the trajectory of the surface pressure minimum in the area); and the fine structure of the cyclone.
Finally, forecast verification of the storm surge contribution was also performed in Piattaforma (not shown). No tidal residual comparison is possible at the lagoon stations since only the total water level is available there. During IOP16, the observed tidal residual oscillates between 35 and 65 cm, with a period of 20–22 h (seiches). For each NWP model, the forecast error is mainly due to a phase shift in the prediction of this fluctuation. However, since the astronomical tide contribution is dominant in this event, the tidal residual forecast error is relevant only when it occurs in correspondence to the astronomical tide peak on 27 October. The results for the IOP18 event show, instead, that the storm surge is dominant with respect to the astronomical tide contribution. The observed storm surge displays a high (ca. 110 cm), long-lasting (ca. 36 h) single peak, and the model errors are mostly due to the underestimation of this peak. These results confirm the above-discussed outcomes of the tidal forecast verification.
The verification results can be summarised as follows. On IOP16, the SHYFEM performance improves by increasing the resolution of the meteorological forcing. As expected, the most recent forecast is always the best one. On IOP18, even if the BOLAM-driven runs are more skilful than the IFS-driven ones, the SHYFEM accuracy in predicting the sea level peak depends in a nontrivial way on meteorological input resolution, initial date, and geographical location. This is the combined effect of the low predictability of the IOP18 cyclone and the high sensitivity of tidal forecast to details of the meteorological fields.
As in Fig. 15, but with the meteorological forecast forcing
provided by the 0.07
Observed and forecast 10 m wind field over NEI at 00:00 UTC of 1 November 2012.
Evaluating the performance of an integrated meteo-marine modelling system, such as SIMM, is a demanding task since forecasts have to be verified at any model stage, and the results should be inter-related to provide a whole picture of the system skill and value. In addition, the assessment of such a system in forecasting high impact events is essential for any operational centre involved in predicting and monitoring natural hazards. In this context, forecast verification is also relevant to determine in advance whether modifications of the system configuration lead to performance improvements.
The first HyMeX SOP experience gives the opportunity (and the necessary observations) to perform an intercomparison study of two different configurations of SIMM, namely the operational and the ad hoc implemented for HyMeX. Despite an apparent similarity of weather patterns and surface effects, the IOP16 and IOP18 events are characterised by different predictability: the results indicate that the cyclone responsible for the latter event is much less predictable than that involved in the former one.
The QPF intercomparison displays an objective added value of MOLOCH forced
by the 0.07
The study is completed by a statistical verification of the MOLOCH QPFs during 3 months, including the first HyMeX SOP devoted to the monitoring of high precipitation events. This represents the first objective evaluation of the ISPRA version of the non-hydrostatic MOLOCH model. This is the starting point to assess the impact (and improvement, if any) on forecast performance of some planned modifications of model configurations – in particular the extension of the model domain to include Italy and the conterminous seas. At a first glance, the categorical scores are comparable with those obtained for the other two versions of MOLOCH used by the CNR-ISAC for the HyMeX SOPs.
The tidal forecast verification clearly displays, for both events, better results when SHYFEM is forced with the forecasts provided by the two BOLAM configurations, rather than ECMWF IFS fields. Concerning the intercomparison between the two different BOLAM initialisations, the different predictability of the two examined weather systems still emerges as a crucial issue. For IOP16, both initialisations provide very good results, with some small differences. For IOP18, the differences in the tide peak forecast due to the use of the two different BOLAM versions are of the same order of magnitude as the differences among runs with different initial dates (the most recent forecast being not always the best one) or the small-scale, latitude-dependent systematic errors. A wrong surface wind prediction over the Venice Lagoon introduces latitude-dependent systematic errors that do not depend from the overall sea elevation error (that can be assumed to be represented by the error at the CNR Piattaforma).
IOP18 cannot be regarded as a negligible exception. An association between
relatively high
Finally, the precipitation and tide forecast verification results display
the added value provided by high-resolution BOLAM-MOLOCH suite into the
SIMM, encouraging us to make it fully operational. The outcome about
predictability of intense
The authors are grateful to Marco Cordella and Devis Canesso (ISPRA) and
Marco Bajo (CNR-ISMAR) for providing useful information on the SHYFEM
simulations and for making available data from the RMLV tide gauges and the
Venice Municipality tide gauge.
Special thanks go to Maria Sansone and
Massimo Enrico Ferrario (ARPAV) for the CALMET analysis. The Italian Air
Force Meteorological Service is acknowledged for providing the ECMWF-IFS
data used to initialise the two SIMM configurations. KNMI and EUTMESAT are
acknowledged for the synoptic analysis charts and the MSG 6.2