3.2.2. 18 November 2018

During the selected wet period event, Greece was affected by a deep surface low-pressure system. As shown in the sea-level pressure and 850 hPa equivalent potential temperature and wind fields (Figure 13b), the surface low is located over the Southwest Ionian Sea at 0000 UTC on 18 November, resulting in very strong Southerly winds. During the next hours, as the surface low is moving Northeast, the Southwesterly flow advects warm and moist air over Greece (Figure A3a–b). As in the case of the dry season event (see Section 3.2.1), both CTL and ZTD experiments simulate the same atmospheric conditions with slight differences in the 500 hPa geopotential height and sea level pressure gradients (Figure A4) at the ZTD experiment initialization time (0000 UTC). Again, the impact introduced by the ZTD data assimilation is clearly evident in the initial PW field at 0000 UTC (Figure 13c). The PW differences at 0000 UTC between the two experiments arise again over locations where GNSS stations are found (Figure 1b), but a more complicated pattern is evident in this case study.

**Figure 13.** (**a**) WRF CTL simulated 500 hPa geopotential height (shading and contours) and wind barbs, (**b**) sea level pressure (contours) and 850 hPa equivalent potential temperature (shading) and wind barbs, and (**c**) PW differences between the CTL and ZTD experiments on 18 November 2018 at 0000 UTC.

Similarly, the 24 h modeled precipitation differences between the CTL and ZTD simulations arising from the modification of the initial conditions of the ZTD experiment due to the ZTD assimilation are more complex and widespread during this event (Figure 14d). The most noticeable differences are found over Western Greece, where the highest rainfall amounts were observed (Figure 14a). In particular, as shown in Figure 14d with black circles, the ZTD experiment produces lower (higher) precipitation over Kefalonia Island, the Aitolokarnania region, and central Peloponnese (the center of Zakynthos island, and Southwest and Southeast Peloponnese) compared to the CTL simulation, more accurately representing the observed rainfall. A more accurate reproduction of the precipitation observations by the ZTD experiment is also evident in Northwest Peloponnese, while in contrast, the ZTD assimilation leads to a more pronounced overestimation of the observed rainfall over central–North Peloponnese and Kythera Island, as indicated by the red circles in Figure 14d. Mixed results are found over the Epirus region, where the ZTD (CTL) experiment forecasts a rainfall pattern that is closer to the observed one in the central and Northern (Southern) part of the area. Overall, the model lacks the ability to capture the magnitude of observations, especially for intense precipitation, as shown by the notable underestimation (overestimation) of the observed rainfall over the area denoted by the orange circle in Chalkidiki (Magnesia) region in Figure 14b–c. However, the spatial distribution of the observed precipitation is well captured by both CTL and ZTD simulations. The above findings are also evident when examining the diurnal precipitation cycle in 6 h intervals and highlight the good performance of the WRF model in reproducing the temporal distribution of the observed rainfall, especially during

the ZTD experiment. Indicatively, both simulations (Figure A5b–c) adequately replicate the observed spatial pattern of the precipitation between 0600 and 1200 UTC (Figure A5a), with the ZTD experiment improving the rainfall amounts forecasts over North Epirus, Aitolokarnania and Peloponnese regions, and Kefalonia and Zakynthos Islands (Figure A5d).

**Figure 14.** Daily (24 h) precipitation from (**a**) observations, (**b**) CTL and (**c**) ZTD simulation, and (**d**) differences between the two experiments for the 18 November 2018 event.

### **4. Discussion**

The results presented above demonstrate a beneficial impact of ZTD assimilation on the daily precipitation forecasts over Greece during both dry and wet periods (Figures 3 and 4). This impact is more profound for heavy precipitation (>20 mm), for which statistically significant (at least at the 90% confidence level) improvements are provided by the ZTD assimilation concerning the probability of detection, the false alarm ratio, the quality of forecasts (ETS), and the magnitude of errors (MAE). This finding is of great importance as higher rainfall amounts are associated with more severe impacts, especially over the Mediterranean regions [20]. In general, the positive impacts of ZTD assimilation are more evident during the dry period. However, notable and statistically significant enhancements in the WRF model's performance are also found during the ZTD experiment for both 24 h (Figures 3 and 4) and 6 h precipitation (Figures 7–10). This fact reflects a significant added value of ZTD assimilation during wet season rainfall events, in contrast with previous studies, which concluded that when the atmospheric state is well described during large synoptic-scale cases, no further improvement of the forecast accuracy can be achieved by assimilating ZTD observations (e.g., Boniface et al. [24]). Moreover, the ZTD assimilation substantially advances the overall model performance in terms of qualitative and quantitative rainfall forecasting between 0600 and 1800 UTC in the dry season (Figures 5 and 6). Especially from 1200–1800 UTC, statistically significant (at the 95% confidence interval) higher POD, ETS, and MAE values, compared to the CTL experiment, are found across all precipitation thresholds. This is a key outcome considering that convective precipitation in the dry season mainly occurs during the examined time window and its forecasting is challenging, as NWP models lack the ability to resolve small-scale convective circulation [68,69]. In contrast, NWP models are more capable in dealing with large-scale dynamically-driven precipitation systems during the wet season [70]. This is evident in the current study because the WRF model, overall, performs better in the wet compared to the dry period. The case study analysis shows that ZTD assimilation provides a more accurate representation of the observed precipitation geographical extent in terms of 24 h and 6 h rainfall (Figures 12 and A2 in the dry period and Figures 14 and A5 in the wet period). It also highlights that the overall WRF performance improvement in rainfall forecasting under the ZTD experiment is due to the modification of the model's initial conditions at the time of the experiment's initialization through the assimilation of ZTD observations. The introduction of ZTD data assimilation especially affects the initial PW field, leading to noticeable differences between the conducted experiments, which emerge in locations where GNSS stations are found. This is evident in Figures 11c and 13c concerning the selected case studies, as well as in Figures A6–A13, illustrating the PW differences between the CTL and ZTD simulations at the ZTD experiment initialization time for each day of the rest of the rainfall events examined in the present study.
