The increasing frequency of severe weather phenomena is a typical sign of climate change at mid-latitudes. Such phenomena are generally very localised in space, so high precision in monitoring and forecasting is necessary to warn citizens and institutions in a timely manner. Undoubtedly, weather radars are among the most relevant tools for accurately monitoring precipitation events, therefore they are a key instrument for the development of nowcasting and early warning systems. In addition, radar systems have been and still are an indispensable tool for observing weather and increasing the knowledge on precipitation dynamics and hydrometeor statistics [1
], subjects of primary interest both in meteorology and in climatology.
A recent report of the World Meteorological Organisation (WMO) [5
] states that the number of X-band weather radars in use in WMO member countries has grown to almost 20% of the total counted radars. This increase is a clear signal of the power of an emerging technology that guarantees high observational skills as well as reduced installation and maintenance costs, and that has also been shown to be particularly suitable for mobile radar applications, compared to more widespread systems such as the conventional C- and S-band radars.
Conventional weather radars still offer higher scanning ranges and measurement accuracies than X-band systems, but the latter can provide much higher spatial and temporal resolutions, although over limited spatial domains. Indeed, the use of X-band radar systems for monitoring weather events at very local scales (for small basins or urban areas) has been tested in several sites and different countries [6
], and the benefits of rainfall characterisation has been clearly assessed [11
], even if for non-coherent systems. An additional capability comes from the small wavelength of the X-band that increases the sensitivity of the radar system to small raindrops, thus improving light rain detection. Conversely, the strong attenuation during the wave propagation in atmosphere limits the maximum scanning range to a typical distance of 70–80 km (in some cases up to 100 km). Such a limit can be somehow overcome through a network densification. X-band radars can also be suitable for addressing beam blockage problems of C- and S-band long-range radars in areas with complex terrain. For this purpose, it is interesting to mention a study of the US weather radar network [13
] that has investigated the possibility of replacing the Next Generation Weather Radar (NEXRAD) network with a much denser and properly organised X-band radar.
The integration of these data with other meteorological observing instruments can significantly improve nowcasting capabilities of operational weather services. In Tuscany these activities are in charge of the LaMMA Consortium (Laboratory of Monitoring and Environmental Modelling for the sustainable development).
LaMMA is a partnership between the regional government of Tuscany (Regione Toscana) and the Italian National Research Council (CNR) (www.lamma.rete.toscana.it
). It is the regional meteorological service for Tuscany, and includes the institutional duties of the daily operational meteorological surveillance. It is obvious that the LaMMA is interested in reliable instruments for monitoring weather events at high spatial and temporal resolution, primarily for supporting nowcasting and for the service of early warning alerts for the regional civilian protection.
The Italian operational weather radar network is currently composed of 21 systems, managed by a federation of national and regional bodies including the Department of Civil Protection (DPC), the Air Force, the regional weather services, and the National Aviation Authority. The network is composed of 19 C-band and 2 X-band pulse Doppler radars. In total, 14 of the 21 radars are dual-polarised. Composite products of the Constant Altitude Plan Position Indicator (CAPPI), Vertical Maximum Indicator (VMI), and Surface Rainfall Intensity (SRI) are generated in real time. DPC is responsible for the generation and dissemination of such products at the national level [14
In the framework of two European projects, ResMar (Réseau pour l’Environnement dans l’Espace Maritime) and PROTERINA-Due (La seconde étape pour la protection contre les risques naturels: les investissements sur le territoire), and spin-off activities (under the Cross-Border Cooperation Programme Italy-France “Maritime”), three new X-band radar systems have been installed in Tuscany that cover almost all the regional coasts and partially overlap and complement the existing national radar network. The choice of these systems derived from a balance between the available economic and financial resources and the need for qualitatively valuable observations, related to operational LaMMA nowcasting activities for civil protection purposes.
This paper describes the implementation of this regional X-band network, detailing each step of the radar signal processing chain that generates the final reflectivity products, starting from the acquisition of the signal power data. Section 2
describes the regional X-band radar network and its operational configuration. The signal processing chain is explained in Section 3
, with particular attention to the sea and ground clutter identification schemes, and to the 3D spatial data infrastructure and mosaicking procedures. The evaluation of the network performances is presented in Section 4
, for an application of the whole signal processing chain to three case studies. Conclusions and future developments are provided in the last section.
2. The X-Band Weather Radar Network in Tuscany
2.1. Set Up of the Radar Systems
The X-band radar system used for the network implementation is the WR10X produced by the ELDES s.r.l company. Three radar systems have been deployed along the Tuscan coastal area. Due to the compact size (90 × 130 cm2
) and limited weight (about 100 kg), the system can be installed almost everywhere and the installation requirements are practically reduced based on the availability of electric power (the consumption is less than 300 W on average) and the connectivity for data transfer. The radar transmitter is based on a magnetron cavity and on a solid-state modulator. The receiver is logarithmic in order to avoid amplifier saturation at intermediate frequencies and it has a dynamic range larger than 90 dB. Some general technical specifications of the radar system are shown in Table 1
With such technical characteristics, a large number of sites can be found that are suitable for the installation without the need of new infrastructure, but with just minor adaptation works.
In Tuscany, the implementation of the network started from the coverage of the coastal areas, because they are periodically lashed by storms coming from the sea, generally with a southwest flow.
The first weather radar has been installed at Cima di Monte, the top of a small isolated mountain of about 470 m high in the western part of Elba Island, funded by the ResMar project. The site is within the protected area of the Tuscan Archipelago National Park, and therefore a number of studies and permissions were mandatory for obtaining the final authorisation to install the radar. Despite some difficulties, the radar has been installed on a pre-existing antenna tower.
The second weather radar has been placed in the port of Livorno on a building of the Port Authority. The latter, which is the owner of the radar, in fact had the need of implementing a local precipitation monitoring system to support the management of the port operations, in particular to guarantee high levels of safety also during heavy rainfall events. In this context, the capability of a radar system to identify the precipitation dynamics is undoubtedly extremely useful. The system is installed on the top of a silo, at an elevation of about 72 m above sea level.
Finally, the third radar has been installed in Castiglione della Pescaia about 25 km north of Grosseto. It has been funded by the PROTERINA-Due project and it is positioned over the local municipality building, at a height of about 15 m.
The configuration of the network offers a good coverage of the Tuscan coastal area and archipelago, providing early detection of precipitations from the sea, which often cause great damage as a consequence of floods and landslides.
For a realistic characterisation of the system capability, the Plan Position Indicator (PPI) geometry has been simulated assuming standard atmospheric conditions [16
] and then interconnected with a Digital Terrain Model (DTM). This procedure has been implemented according to the 3D spatial data infrastructure described in Section 3.1
. Thus, a visibility percentage (with reference values of 16%, 33%, 50%, 66%, 83%, and with 0% and 100% corresponding respectively to null and full visibility) has been assigned to each volumetric cell, by considering the fraction of beam blockage. The merged visibility of the three radars is shown in Figure 1
, and it is characterised by an excellent coverage, also redundant, over the sea and inland.
Each single radar presents some partial or total shading areas: a southern limit due to hills and a number of northern sectors with low visibility towards the Apuan Alps are present for the Livorno radar; the Elba radar, positioned on a local top of an island, ensures a more homogeneous latitudinal coverage, with just two occultation sectors in the south and west directions; the Castiglione della Pescaia radar provides an optimal coverage of the Maremma region (southwestern Tuscany), but is completely blocked in the northerly direction due to orography.
2.2. Real Time Operativity
The operational configuration of the three radar systems has been set in order to complete a scan every 15 min, with synchronised start times. For each scan, the atmospheric scenario is reconstructed through 10 elevations from 0.5° to 5.0° in steps of 0.5°. The maximum selectable range is 108 km allowing the observation of eventual intense phenomena also at the range limit. The best nominal selectable range resolution is 90 m, but this is only allowed for very short ranges (21 km). Due to a technological limit of the system in the data transfer from the receiver to the processing unit (using a serial communication protocol), when the maximum range increases, not all of the acquired signal power samples can be transferred to the processing unit. At the maximum range of 108 km, the range resolution is degraded to 450 m despite a nominal resolution of 90 m, due to the forced subsampling of one sample every five. The available products used for monitoring real time phenomena are the Horizontal and Vertical Maximum Intensity (HVMI), VMI, PPI, and the Range Height Indicator (RHI). These products allow a continuous monitoring of the 3D structure of the atmosphere for weather observation and nowcasting purposes.
The choice of such a scan configuration that includes ten elevations with the maximum range was made to meet two requirements: the necessity of monitoring the full area and of finding the optimal set of elevations to be used for the clutter signal filtering. The scan repetition time has been set to 15 min, according to the time needed for processing ten elevations.
The real time use of the network together with other meteorological instruments has demonstrated good performances in identifying and tracking the target phenomena, in some cases even at long distances, different to what often occurs with the X-band radars [3
]. The use of the HVMI product is crucial for the correct interpretation of convective phenomena and heavy rain events. However, some limitations exist due to strong signal attenuation in the presence of rain. In the proximity of the radar, a small water coverage on the radome can introduce a very strong attenuation and in extreme cases even affects the signal extinction. Moreover, at distance of some tens of kilometres, intense precipitation can hinder the detection of more distant phenomena. Sometimes stratiform rain is not detected, especially for very light precipitation in areas far from the radar.
The densification of the network with the synergic analysis of the radar signals can help to overcome these limitations and the radar composite can effectively help in the correct identification of the structure and dynamics of the precipitation systems.
A further problem of radar systems is the presence of sea and ground clutter that can mimic or overlap to precipitation signatures. In the operational phase, an animation of the next PPI images can help to identify the precipitation through the motion of a reflectivity field; nevertheless, for a more accurate assessment of the meteorological situation, it is essential to remove the unwanted signals.
An analysis has been made to improve the monitoring efficiency of the radar network. This study started from the power data received and collected by the radar systems. The complete signal processing chain has been implemented. It applies the radar equation to retrieve the reflectivity value for each observed volumetric cell and includes all the software modules for estimating and removing clutter signals. Moreover, a procedure has been developed for geolocating the volumetric cells in a non-conventional manner.
All these software modules are described in the following paragraphs, detailing major innovations and the final results.
4. Analysis of Three Case Studies
The signal processing chain was tested on three selected case studies, in order to assess the reliability of the regional radar network in monitoring the precipitation events with the spatial and temporal resolutions necessary for very short range forecasting activities.
The case study of 23 May 2016, 09:00 UTC, is characterised by a cyclogenesis originating between the regions of PACA (Provence-Alpes-Côte d'Azur, France) and Piedmont (Italy) with strong atmospheric instability brought to a sequence of train fronts, that developed several convective precipitation phenomena. The fronts reached the Tuscan coasts in the early morning while moving towards the inland areas. In the late-afternoon and early-evening, new showers and thunderstorms, initially located in the northwestern areas, were in gradual shift towards the inland areas. Peaks of 24 h-cumulated precipitation up to 40 mm were registered in the internal areas along the Apennine and in the eastern part of the region.
In Figure 5
a,c,e, no clutter removal is applied, and large ground clutter echoes are clearly visible near the Tuscan coasts and over Corsica, while sea clutter disturbances are also present around the radar sites of Livorno and Elba.
After the application of both (i.e., sea and ground) clutter removal schemes, most of the ground clutter is removed over the Corsican and Tuscan coasts (Figure 5
b,d), as well as in the inner areas of the southern part of the region (Figure 5
f), even if some disturbances still remain, mainly close to the radar sites of Elba and Livorno. The sea clutter is largely reduced around the Elba radar site (Figure 5
d) and also around Livorno (Figure 5
b), however with less effectiveness; in fact, some noisy echoes with low reflectivity values still remain due to the persistence of clutter. The same also occurs in the scans at higher elevations (up to 2.5°) for both radars. Note the semi-circular shape of the sea clutter with rather high values of reflectivity around the Livorno site (Figure 5
a) due to its low altitude above the sea level, which facilitates unwanted interactions of the radar signal with the marine surface. Additionally, around Elba Island (Figure 5
c) there is often the presence of circular clutter echoes, which, however, rapidly disappear at higher elevations thanks to the radar position at about 470 m above the sea level (a.s.l.).
After the clutter filtering, the principal precipitation patterns of the systems impacting the south-eastern part of Tuscany are retained, although slightly reduced in their extension. Undoubtedly, the algorithm has to properly balance the efficient removal of clutter echoes while keeping the relevant rainfall features.
A mosaic of the three radars has been implemented for the above described case study, after the application of both the sea and ground clutter removal processes (Figure 6
a). The good performances of the clutter removal schemes in eliminating most of the sea and ground disturbances is clearly evident.
In order to assess the quality of this final product, a comparison was performed with the Italian national radar mosaic (Figure 6
b) and with hourly-cumulated precipitation fields, elaborated from measurements of the regional rain gauge network (Figure 6
c). The weather station network in Tuscany consists of 801 measurement sites equipped with rain gauges and other meteorological sensors. The rain gauge measurements are representative for an area of a few dm2
with a typical accuracy of 0.2 mm for the tipping bucket gauges.
The precipitation events are correctly detected in space and time and the higher spatial resolution of the X-band systems with respect to the C- and S-band radars allows for the non-upgraded spatial characterisation of the precipitation systems, with an improved definition of the rainfall field patterns, as noted for the cells in the south- and north-eastern part of the region.
On the other hand, due to the rapid attenuation of the signal with the distance, some reflectivity echoes far from the Livorno radar site were lost, observed by both the rain gauge and the national radar networks in the northern part of Tuscany. For this reason, it is important to proceed towards an integrated network approach that is capable to reducing the signal loss problems, improving attenuation correction, and increasing the spatial and temporal resolutions of the final products.
The second case study is for 13 September 2016, for which we show an example of a radar mosaic image superimposed to the Meteosat Second Generation (MSG) High Resolution Visible (HRV) channel at 12:00 UTC (Figure 7
a). In this case, only the Livorno and the Castiglione della Pescaia radars are processed, because the Elba system was out of order during the case study period. Over the central Mediterranean, the atmospheric pressure maintained medium to high values, with a weak flow of current from the northeast towards the Italian peninsula. Infiltration of cold air at high altitudes from the Balkans determined instability conditions in the Tuscan inland areas. In the late morning, cumulus clouds developed in the inner areas behind the reliefs and scattered thunderstorms developed in the hinterland in the afternoon. A maximum precipitation of about 50 mm cumulated in 24 h was registered in the proximity of the Apuan Alps and up to 40 mm in the hinterland of Grosseto.
Very localised precipitation cells that are widespread across Tuscany are well detected by the regional network, as shown from the overlapping with the MSG reflectance image, in which cloud systems are clearly visible in bright white color (Figure 7
a). In addition, the greater spatial resolution of the regional X-band radars allows the detection of the precipitation patterns more accurately than the national C- and S-band ones, and allows the observation of more and finer structures within the precipitation cells with respect to the National Mosaic (Figure 7
b). Finally, the comparison with the 1-h (12:00–13:00 UTC) cumulated precipitation obtained from the regional rain gauge network further confirms the reliability of the implemented signal processing chain in detecting the precipitation systems and their fine spatial structure (Figure 7
c), reminding us that we are comparing instantaneous radar imagery with precipitation fields obtained from rain gauge measurements cumulated for one hour.
Some clutter disturbances over the sea and near the coasts still remain: for this purpose, the development of additional filters which account, for example, for the range dependence of the sea clutter near the Livorno site is in progress, and it should further improve the performances of the present schemes for removing the clutter effects.
A third case study, characterised by local heavy precipitations, was further investigated. On 8 June 2016, the central Mediterranean was characterised by an increasing surface pressure with moderate conditions of atmospheric instability caused by the intrusion of fresh air at high altitudes. In the early morning, a few clouds associated with mist and fog developed in the inland plains and hilly areas. By the late morning, cumulus clouds generated in the internal areas gave rise in the afternoon to scattered showers and thunderstorms, locally heavy, in the Garfagnana and in the provinces of Florence, Siena, and Arezzo. A cumulated value of precipitation of about 40 mm in 1 h was registered in the hinterland of Florence.
c shows the MSG HRV reflectance image in which an extended cloud system developed in the inland areas of Tuscany is visible. Very bright, white clouds attest for a great vertical development of the systems. Also in this case, the precipitation cells, whose rainfall intensity was measured by the rain gauge network (Figure 8
d), were correctly localised in time and space, with valuable pattern details (see Figure 8
a). The precipitation structures are observed in a more detailed way by the regional network than by the national mosaic, which shows a reflectivity signal that is spread over most of the Tuscan territory (Figure 8
b). The finer spatial resolution of the X-band systems results in a better characterisation of the precipitation events. Conversely, the greater attenuation of the X-band wavelength with the distance sometimes causes the partial loss of reflectivity signal, as in the north-eastern areas of the region.
Moreover, the regional radar network (Figure 8
a) identifies a small precipitation cell in central-eastern Corsica in an area not covered by the Italian weather radar network (Figure 8
c shows a cloud system originating in the Corsican mountains). Only a few pixels of the national network have low reflectivity values (<25 dBZ) in this area (Figure 8
Some ground clutter disturbances still remained along the Tuscan coasts and on Elba Island and Corsica: further methods for assessing the reflectivity thresholds are under development to remove these spurious signals.
Finally, a 3D preliminary product has been realised in a GIS environment (see Figure 9
) for the case study of 23 May 2016, 09:00 UTC. The shapefiles for the 3D contour reflectivity of the three radars with an elevation of 2° have been superimposed to an ASTER DTM (10 m resolution), in order to obtain a 3D view of the observed cloud systems.
Undoubtedly, the combination of homogeneous information coming from different layers in the atmosphere allows us to directly characterise the spatial distribution of the precipitation clouds. The partial overlapping of the radar footprints provides significant information about different heights and stages of cloud formation. Using this 3D approach, the reflectivity value is assigned to its effective height in the atmosphere, showing the centroid of each 3D cell.
In this paper, the setup of a regional X-band radar network is described, with focus on the reasons leading to the different development steps, both from the technical and scientific points of view. In Tuscany, most of the precipitation events originate in the southwest direction with orographic triggering. Indeed, the combined effect of mountains in proximity of the coast with the airmass flow over the sea ingesting humidity is one of the principal causes of heavy precipitations. Therefore, the implementation of the network in this initial phase mainly involved the coverage of coastal areas, including a large portion of the Tyrrhenian Sea, to improve the monitoring over this area and to fill some coverage gaps of the national weather radar network in Italy. Once the radar sites were defined and the installation procedures were completed, the network became operational. It is currently used for carrying out the institutional surveillance duties of the LaMMA Consortium, which is the weather service of the regional administration of Tuscany. The standard products are presently used as provided by the company that supplied the radar systems. At the same time, the development of a chain for processing the radar signals has been set up to exploit the whole information content of the volume scans. This work has involved the whole processing chain from the acquisition of the signal power to its conversion in reflectivity before addressing the issues of geolocation and signal filtering for clutter removal. Finally, a composite method has been implemented for merging all the information provided by the radars into an integrated weather monitoring system. A minimum set of four elevations for each scan has been identified, suitable for signal processing and clutter filtering, essentially without information loss. The signal processing results were shown, with particular focus on sea and ground clutter removal. The mosaic results were then compared with ancillary data for three case studies relating to different seasons and precipitation types.
The present paper describes what is still a preliminary product currently under testing, which is providing good results especially on monitoring intense events and their spatial and temporal characterisation (as validated during the LaMMA operational surveillance duties). It still has some limitations and critical aspects from which suggestions for future studies can be derived. For example, the information on the visibility field should be included in the composite generation, in order to avoid contamination of uncorrected clutter or signal power weakness due to partial beam blockage. Additionally, the vertical profile of reflectivity should be considered to improve the assessment of the weather patterns and to correct for the effects of the attenuation and bright band [43
]. Regarding the clutter removal algorithms, more sophisticated techniques should be investigated such as Bayesian approaches [44
] or adaptive filtering schemes [45
], to evaluate their impacts on the signal processing. In addition, as proposed by [45
], the range should be considered as a further parameter to discriminate clutter free cells, because large distances imply great sampling heights in the atmosphere and consequently a lower probability of finding clutter echoes.
Some additional efforts will be put to further investigate the best operational scanning strategy consisting of a proper combination of time and space resolutions. This configuration should better exploit the technical characteristics of the X-band radar systems for resolving and monitoring precipitation dynamics at very local scales.