Bulgaria is a country with a high frequency of thunderstorms during the warm part of the year, from April to September. A pan-European study of thunderstorm climatology [1
] reported that the regions with the highest frequency are (1) the central Mediterranean, (2) the Alps, (3) the Balkan Peninsula, and (4) the Carpathians. The annual peak of thunderstorm activity is in July and August in northern, eastern, and central Europe, and in May and June for western and southeastern Europe. Trend analysis of the mean annual number of days with thunderstorms since 1979 indicates an increase in southeastern Europe. Stations in southeastern Europe, Belgrade, Bucharest, and Sofia, share similar features, with well-defined peaks in the summertime and a rapid increase in April/May, and a decrease in October. For Sofia, the highest in Europe, the mean annual number of days with thunderstorms and severe thunderstorms has been reported, respectively, at 44 and 13 days. Severe thunderstorms are defined in [1
], based on (1) radiosonde data: (a) mixed-layer convective available potential energy (CAPE > 150 J/kg) and convective inhibition (CIN > 275 J/kg), and (b) deep-layer shear combined with CAPE (WMAXSHEAR > 400 m2
); (2) Global Atmospheric reanalysis (ERA-Interim): (a) mixed-layer CAPE > 150 J/kg, (b) convective precipitation >0.075 mm/h, mixed-layer WMAXSHEAR > 400 m2
. Ref. [2
] reports that west Bulgaria is the region with the highest frequency of thunderstorms, but also severe weather events, including hailstorms, torrential precipitation, and severe convectively-induced wind storms. A recent study by [3
] reported on three supercell storms which developed over western Bulgaria on 8 July 2014, with hail stone sizes of up to 10 cm, strong wind gusts, and torrential rain. The Doppler radar data revealed the existence of a mesocyclone, meso-anticyclone, microburst, and a three-body scatter signature. Ref. [4
] conducted a study over 155 days with precipitation in southern Bulgaria from May to September 2002–2006, with separation into two samples: (1) days with frontal convective clouds, and (2) days with free convection. It was reported that for days with hail, the mean values and the thresholds of instability indices Convective Available Potential Energy (CAPE) and Lifted Index (LI) were similar to the values determined for thunderstorm development in other regions of Europe. However, it was concluded that neither the instability indices nor the environmental parameters (temperature and pressure at Lifted Condensation Level) were able to differentiate the precipitation type. Similar results are reported from other authors [5
]. Their studies showed that instability indices alone are not able to predict the probability of thunderstorm development satisfactorily. Furthermore, the universally determined threshold values are often unsatisfactory for different geographical areas with specific meteorological conditions. For these reasons, Refs. [7
] propose new local threshold values, and a combination of different instability indices and atmospheric parameters.
A review [10
], covering the application of ground-based Global Navigation Satellite Systems (GNSS) for studying water vapor field evolution, concluded that there is “clear evidence of the benefits that GNSS can bring to the monitoring of severe weather events”. A study in Portugal [11
] reported that: (1) the temporal variation of GNSS derived Integrated Water Vapor (IWV) correlated with rainfall, and (2) can be used for the detection of heavy rain. A recent study [12
] focused on the short-term forecast of intense rainfall using a neural network approach, and integration of IWV with meteorological data. Reported was an improvement in intense rainfall event detection and a reduction of the number of false-positive alarms, with a good classification score varying from 63% up to 72%, and a false positive rate of 21%. Reported was also a very high hit rate for the rain versus no rain detection and close to zero false alarms. Ref. [13
] proposes a method for retrieving two indices for the degree of inhomogeneity of water vapor using the GNSS carrier phase. The first index describes the spatial variation of Water Vapor Concentration (WVC), while the second indicates higher-order Water Vapor Inhomogeneity (WVI). The horizontal scales of the indices are about 60 km and 2–3 km, respectively. The indices were applied over Japan for August 2011, and their monthly averaged values indicated: (1) distinct diurnal variation in the mountainous region of central Honshu, and (2) coincidence with the diurnal variation in precipitation frequencies in the area. The relations between the indices and precipitation indicate that the WVI is strongly correlated with intense rainfall than IWV. IWV was found to be more strongly related to precipitation lower than 10 mm/h. The spatiotemporal variations of WVC and WVI were studied for a thunderstorm on 11 August 2011 for the Tokyo area, and both were found to increase ahead of the initiation of convective precipitation [13
]. Ref. [14
] investigated the relationships between intense rainfall and the convergence of surface winds and WVC for heavy-rainfall cases in July–August 2011–2013. It was reported that: (1) the peak of surface wind convergence was observed 10–30 min before the heavy rainfall, (2) convergence continued to increase for approximately 30 min prior to the convergence peak time, and (3) an increase of WVC coincided with an increase in wind convergence. Thus [14
] concluded that it would be possible to predict rapidly and accurately the occurrence of heavy rainfall by monitoring the temporal variations and the distributions of surface wind and WVC by a high-density observation network, like the one in the Tokyo region. Ref. [15
] studied the relation between the high-frequency IWV (1 min) and intense rainfall events in Brazil, in the November–December 2011 period. A sharp IWV increase, or “jump”, was found before intense rainfall and believed to be associated with water vapor convergence, the continued formation of cloud condensate, and precipitation particles. A wavelet correlation analysis showed oscillations in the IWV time series correlated with the intense rainfall events. These oscillations are on scales related to periods of approximately 32 to 64 min (associated with IWV jumps), and 16 to 34 min (associated with positive IWV pulses). The IWV time-derivative histogram before the rainfall event revealed different distributions influenced by positive IWV pulses (derivative >9.5 kg/m2
/hr) for higher intensity and extension events.
Reported in [16
] is a severe flood, caused by intense rainfall, in Colorado, USA, from 9 to 16 September 2013. Analysis of IWV for the 10-year period showed that in 2013: (1) the seasonal IWV maximum extended into early September, and (2) the September monthly mean IWV exceeded the 99th percentile of climatology with a value 25% higher than the 40-year climatology. Prior to the flood, a rapid increase in IWV was found from 22 to 32 kg/m2
, and remained around 30 kg/m2
for the entire event. The IWV frequency distribution in September was typically normal, while in 2013, a bimodal distribution was found with above-average IWV from 1 to 15 September, and much drier conditions from 16 to 30 September. The positive IWV anomaly during the flood was found to be a result of large-scale moisture transport from the Tropical Eastern Pacific and the Gulf of Mexico [16
]. The potential of high-frequency IWV (5 and 15 min) in combination with weather radar reflectivity was exploited by [17
] for a derecho event in Poland on 11 August 2017. Reported is a strong agreement between the IWV and the rate of IWV change spatial maps and radar reflectivity, with the maximum values of reflectivity and precipitation coinciding with the maximum IWV values. In addition, the GNSS derived gradients converged toward the maximum values of reflectivity. Ref. [18
] presented an H2
O alert system for Belgium using GNSS derived horizontal gradients of the water vapor content to detect small scale structures of the troposphere for a rainfall event on 28–29 June 2005. The alert was based on a dry/wet contrast in a 30 min time window before the initiation of a convective system. Validation of the alert system with precipitation by weather radar and satellite-derived cloud top temperature gave a score of about 80%. Ref. [19
] reports on an extensive observing period in May–June 2013, with heavy rain and floods in the Czech Republic and the catchment area of the Danube river in central Europe, and the northern part of the Alps. GNSS derived gradients were found to be significantly higher compared to the one derived from Numerical Weather Prediction (NWP) models. GNSS tropospheric products were found to provide more detailed structures in the atmosphere than the NWP models were able to capture.
This manuscript aims to exploit the synergy between GNSS derived IWV and instability indices, to obtain a GNSS-based product tailored for thunderstorm analysis and nowcasting in the Sofia plain. In Section 2
the GNSS derived IWV dataset is presented, as well as the instability indices and statistical analysis. The classification function scores for thunderstorm forecasting in the May–September 2010–2018 period are discussed in Section 3.1
, and two case studies of supercell and multicell thunderstorms are presented in Section 3.2
. Discussion and conclusions are in Section 4
and Section 5
The forecasting of thunderstorm formation and development is a major challenge in operational meteorology, and is associated with large economic losses. The hail and thunderstorm on 8 July 2014 in Sofia was estimated to cost over 123 million EUR in insured losses over 2 hours [47
]. In this work, the added value of combining the instability indices with GNSS-IWV has been demonstrated for thunderstorm days in the warm part of the year from May to September, for the Sofia plains region. The results clearly indicate improvement of POD, FAR, CSI, and TSS scores for a classification function, combining instability indices and IWV for the 2010–2015 period. Furthermore, the performance of the classification function was tested on an independent sample and its effectiveness was confirmed. The results were confirmed by the F-test, which showed that there was a significant difference in the maximum IWV, for TH and NTH days. This is an encouraging result, and suggests that the proposed function can be used as guidance in diagnosing thunderstorms in the Sofia plains region. The limitation of this function is its local validity for the Sofia region. However, the obtained quantitative monthly threshold for thunderstorm development could be useful guidance for operational meteorology. In particular, the proposed statistical approach can be applied to the instability of indices and IWV computed from high-resolution mesoscale NWP models like WRF. However, despite the improved topography representation, the current state-of-the-art mesoscale models underperform in regions with complex topography, and during thunder and hailstorm events. One of the reasons is the lack of observations with high temporal and spatial resolution, representing the local environment. Despite the GNSS derived IWV having a temporal resolution of 15 minutes, its interpretation for the two case studies in 2014 was a challenge. As shown by other studies, IWV accumulation is one of the ingredients for thunderstorm formation. However, matching surface observations with high temporal and spatial resolution is a major limitation of this study. Thus, we were not able to offer a conclusive indicator for individual thunderstorm and IWV development, as found by studies in other regions and continents. However, a recently deployed network in Bulgaria is designed to provide collocated GNSS and surface observations with high temporal resolution, and will overcome this limitation. Since May 2019, 12 GNSS and meteorological stations have been processed as a part of the BalkanMed Real-time severe weather service (BeRTISS) project [48
]. The results of this study will be used to improve the thunderstorm forecast in the warm season, and will be included in the Bulgarian Integrated NowCAsting tool.
Thunderstorm climatology shows that Sofia has the highest mean annual number of days with thunderstorms and severe thunderstorms across Europe. The thunderstorm activity in the Sofia plain starts with a rapid increase in April/May and decreases in October; thus, this study covers the thunderstorm period from May to September 2010–2015. For the days with thunderstorms (TH) and no thunderstorms (NTH), monthly threshold values of IWV were computed, and a good separation between the two groups was found. Based on IWV alone, the highest probability of detection score obtained was 0.91 but it was associated with a high false alarm ratio of 0.45. Thus, as a next step, stepwise discriminant analysis was applied to derive classification functions using 1) instability indices (F1), and 2) instability indices in combination with IWV (F2). The F-test analysis showed that only the K index and IWV have statistically significant values for TH and NTH days. A comparison between F1 and F2 monthly probability of detection scores gave an advantage to F2, with the largest improvement of 10% in September, followed by June and August with 8%. The largest reduction in the false alarm ratio score was in September, followed by May and August. Analysis of both POD and FAR scores gave the best performance for May, followed by June and September 2010–2015. Evaluation of the monthly classification functions was carried out using an independent sample period of 2017–2018. Combined analysis of the probability of detection and false alarm ratio scores gave the best performance for September, followed by June and May. This result confirmed the finding from the first sample period, that for May, June and September, the best scopes are obtained by the classification function combining IWV and instability indices. The IWV and lightning flash rates for two case studies of a multicell thunderstorm on 18 June 2014 and a supercell thunderstorm on 8 July 2014, were analyzed. Both thunderstorms were classified as TH based on the monthly threshold values of IWV for June and July. IWV reached a monthly threshold of 14.5 and 3.5 hours before the thunderstorm started.