This section describes the climatology of SC phenology and the observed change signals derived from the long-term trend analyses, the spatial differences in the temporal evolution of the SC parameters and the links to the seasonal climate and atmospheric circulation variability.
3.1 Climatology of snow cover phenology, snow duration and snow-free days
Figure 2 illustrates the long-term characteristics of the FSD, LSD, SCD, SCDmax and SFD over 1961–2020 for each Koeppen-Geiger climate region of the country.
Whilst the snow onset and snowmelt dates largely differ between stations (elevations), the long-term climatology of SC phenology reflects the characteristics of the major climatic zones of the country, resulting from complex interactions between atmospheric circulation and underlying topography. On average, the earliest FSC is characteristic to the ET climate (mid-September) and the latest to the Bsk climate (late-December) in relation to occurrence of low temperatures induced by high elevation. With a FSC of 263 JD (Julian Date), Vf. Omu station (2,504 m a.s.l.) is the site with the earliest first SC occurrence across the country. On the opposite, the latest FSC was recorded at Mangalia station (6 m a.s.l.), with a FSC of 364 JD located on the southern part of Romanian Black Sea coastline. In spring, LSC is the earliest in the areas with a Cfa climate (mid-January) and could advance notably until the second-half of June in the areas with ET climate (Fig. 3). The snow could cover the ground from mid-September in the areas above 1,000 m and up to one month later (mid-October) in those below 1,000 m. The opposite is found at the end of the snow season, when LSC may occur from mid-June and late-April, respectively. Both FSC and LSC show a considerable dependency on altitude. The FSC declines with 2.9 days 100 m− 1 in fall, suggesting an earlier onset of the snow season with elevation, whereas in spring, the LSC increases with about 3.3 days 100 m− 1, indicating a later end of the snow season with elevation.
SCD (and SCDmax) gradually increase with elevation from an average minimum value of 52 (7 days) in the lowlands to an average maximum of 277 (and 195 days) in the highlands. Both SCD and SCDmax show a high dependency on elevation, with an increase of about 8.7 and 10.4 days 100 m− 1, respectively. SFD shows an opposite vertical distribution pattern, reaching the average maximum of 310 in the areas below 500 m (especially in the Bsk) and the average minimum of 98 at above 2,000 m (in the ET). SFD decreases with elevation with 9 days 100 m− 1.
3.2 Trends in snow cover onset and melt dates
Figure 3: Spatial trends in SC onset (left arrow) and melt (right arrow) dates over the 1961–2020 period. The red/blue colours of the symbols show a retreating/advancing trend in the SC onset/melting dates, the grey circles show no trend, whereas the “x”-sign overlapping the symbols depicts the cases with no statistically significant trends (p > 0.05).
During fall, although the trends in FSC are mixed, the positive Theil-Sen slopes are dominant, indicating later onsets of SC at most WS (85%). These shifts are rarely statistically significant (p < 0.05), as observed at only 7% WS. The FSC advance is apparently slower in the highlands, with an average rate of 1.5 days decade− 1 (i.e., maximum of 4 days decade− 1 at Cuntu station, 1,456 m a.s.l., the Southern Carpathians) than in the lowlands, with about 1.7 days decade− 1 (i.e., maximum of 6.5 days decade− 1 at Focșani station, 57 m a.s.l., in the Romanian Plain). Negative trends, indicating a FSC retreat, were found at about 32% WS, including one mid-elevation mountain station (Băișoara, 1,360 m a.s.l., the Western Carpathians), and are not not statistically significant.
The end of the SC season is shifted to earlier times for most WS (96%), with an average retreating rate of 2 days decade− 1. This is a robust change signal in terms of trend direction at country scale, although the statistically significant trends are observed at only 20% of WS, mostly located in the lowlands, below 500 m. The LSC occurs slightly earlier at above 1,000 m, at rates of up to 5 days decade− 1 (i.e., Lăcăuți station, 1,776 m a.s.l., the Eastern Carpathians) than in the areas below 1,000 m, where rates are about 4 days decade− 1 (i.e., Focșani, 57 m a.s.l., the Romanian Plain). Contrasting trends, indicating later snowmelting dates, are underrepresented at country’s scale (below 3% WS) and have no statistical significance. Country-wide, the ‘no trend’ situations account for less than 5% for FSC and 1% for LSC.
Figure 4 illustrates the vertical distribution of the observed trends in the SC parameters investigated in the present analysis within the elevation range of the country. Country-wide, the correlation between elevation and FSC and LSC trend slopes is weak (r2 < 0.10) and not statistically significant. Furthermore, the differences between the aggregated trends of highlands and lowlands, in both fall and spring, are limited to about 1 day decade− 1. On average, the greatest FSC advance (1.6 days decade− 1) was observed between 1,500-2,000 m, as well as in the lowlands below 500 m. It is worth mentioning that at above 2,000 m (ET climate), the change signal is reversed, suggesting a FSC retreat or an earlier snow cover onset. Nonetheless, in fall, faster advances of FSC with elevation have been found only in the Dfc region, with averaged FSC increases of 1.58 days decade− 1 at 1,000–1,500 m, as compared to 1.63 days decade− 1 at 1,500-2,000 m.
In spring, the negative LSC trends, indicating earlier snowmelt with elevation, generally correspond to the Dfc (1,000–2,000 m) and Dfb (500-1,500 m) climate regions.
3.3 Trends in snow cover duration
SCD trends indicate a widespread decline across Romania (95% WS), although only 32% trends are statistically significant (Fig. 5). Notwithstanding the mixed SCD trends, the decline is prevalent both in the lowlands and highlands.
SCD shortens more rapidly in the areas below 1,000 m, where local negative trends indicate a maximum rate of 11 days decade− 1 (i.e., Focșani station, 57 m a.s.l., the Romanian Plain), as compared to that found in the mountains, at above 1,000 m, of about 8 days decade− 1 (i.e., Lăcăuți station, 1,776 m a.s.l., the Eastern Carpathians). The SCD trends do not reveal a clear elevation dependency. On average, the mountains between 1,000 and 2,000 m show large SCD declines with aggregated trends of 4–5 days decade− 1. Marked local decreases (over 8 days decade− 1) correspond to the mountain areas with a Dfc climate (i.e., Lăcăuți station, 1,776 m, Eastern Carpathians), as well as to some lowland areas, with a Cfa climate (i.e., Focșani station, 57 m a.s.l., Romanian Plain). The weakest trends were observed at above 2,000 m, with rates of about 2 days decade− 1. In the areas below 1,000 m, especially in those with a Cfa or Cfb climate, the trends are widely negative (Table 3). Slightly positive trends were sparsely observed at only 4 WS, located in the southern and eastern lowlands of the country (Bsk climate), and are not statistically significant.
SCDmax was also found to decline at the majority of WS (81%) across Romania, with a share of 19% of statistically significant trends. The WS showing significant negative trends are located in mid-elevation areas of the Dfc and Dfb (3 WS between 1,300-1,500 m) regions, but mostly to the low-elevations of Dfb, Cfa and Cfb (19 WS below 1,000 m) ones. On average, the 500-1,000 and 1,000–1,500 m elevation ranges show the most pronounced declines, estimated at about 2.5 days decade− 1. At a WS scale, the decrease in SCDmax could reach a maximum rate of about 8 days decade− 1 (e.g., Joseni station, 750 m a.s.l., the Eastern Carpathians). In the mountains, the negative change signal is prevalent and includes the highest elevations (− 1.0 days decade− 1), although not statistically significant. At country scale, ‘positive trend’ and ‘no trend’ cases are rare (7% and 10% WS, respectively) and are specific to the mid-elevations of Dfc (+ 3.4 to 3.6 days decade− 1), as well as to the lowlands with a Cfa and Cfb climate (+ 1.8 to 2.1 days decade− 1). The Bsk climate areas show a stable SCDmax regime. In these areas, the “no trend” cases could be largely explained by the very few days with persistent snow cover throughout the analysed snow years, especially in the last decades (e.g., in 1990, 2001, 2006 and 2007 SCDmax is between 1–2 days). The vertical distribution of SCDmax trends provides no consistent elevation-dependency evidence.
Tabel 3 summarizes the average long-term trends in analysed SC parameters across Romania. Furthermore, in Supplementary Materials, Table 1 allows the exploration of the spatial patterns of long-term changes in SC phenology, duration and number of snow-free days in Romania, based on the synthesis of trend analysis results performed at each of the 114 WS.
Table 3
Long-term average trends (with minimum and maximum in parentheses) in snow phenology parameters (days decade− 1) by climate regions of Romania. For comparison, different levels of statistical significance of trends are provided (p < 0.001***, < 0.01** and < 0.05*) and marked in bold
Climate regions | Elevation bands | FSC | LSC | SCD | SCDmax | SFD |
ET | > 2,000 m | −1.62 (-2.75, -0.75) | −2.85 * (-3.33, -2.38) | −2.15 (-2.22, -2.07) | −0.64 (-0.10, -1.17) | + 1.88 (+ 1.81, + 1.96) |
Dfc | 1,501–2,000 m | + 1.63 (+ 0.71,+2.89) | −3.39 (− 4.67, − 1.25) | −4.99 (− 8.21, − 1.23) | −0.64 (− 4.00, + 3.39) | + 2.16 (− 0.33, + 4.19) |
1,001–1,500 m | + 1.58 (+ 0.32, + 4.07) | −2.49 * (− 2.73,−2.14) | −4.91 *** (− 7.00, − 3.33) | −1.22 * (− 5.46, + 3.64) | + 2.80 (+ 0.48, + 4.58) |
Dfb | 1,001–1,500 m | −0.32 (− 2.78, + 1.15) | −3.24 * (− 3.89, − 2.27) | −3.11 (− 4.75, − 0.77) | −3.81* (− 5.00, − 1.43) | + 2.80* (+ 1.43, + 3.62) |
500–1,000 m | + 0.36 (− 1.77, + 2.47) | −2.45 (− 3.48, − 0.27) | −3.60 * (− 6.00, − 1.60) | −3.30 * (− 8.37, − 0.77) | + 3.95* (+ 1,67,+6.56) |
≤ 500 m | + 1.25 | −4.25 ** | −6.15 ** | −4.86 ** | + 6.18*** |
Cfb | 501–1,000 m | + 0.27 (− 0.53, + 3.11) | −1.81 (− 3.33, − 0.47) | −2.21 (− 4.44, + 0.30) | −1.91 (− 4.00, 0.00) | + 3.36 (+ 0.87,+5.4) |
≤ 500 m | + 1.26 (− 3.09, + 3.33) | −1.62 (− 4.00, + 0.67) | −3.19 (− 5.88, + 1.18) | −1.99 (− 5.11, + 2.11) | + 4.25* (0.0, + 7.69) |
Cfa | ≤ 500 m | + 1.81 (− 0.74, + 6.47) | −2.10 (− 4.40, + 0.53) | −3.93 (− 11.33, + 1.37) | −1.10 (− 3.75, + 1.81) | + 2.70 (-1.65,5.31) |
BSk | ≤ 500 m | + 0.56 (− 0.46, + 1.58) | −3.11* (− 3.52, + 2.71) | −3.3 (− 3.54, − 3.06) | No trends | + 0.45 (0.0,+0.91) |
The relationship between snow cover duration (SCD and SCDmax) and its timing (FSC and LSC) has been further analysed through Pearson's correlation analysis, at both country and regional (highlands versus lowlands) scales. Overall, the SCD variability is largely and statistically significant explained by the variations in both FSC (negative correlation, r = 0.84, p < 0.05) and LSC (positive correlation, r = 0.87, p < 0.05). Strongest relationships have been observed in the mountains (r ranging between 0.82 and 0.83) (Fig. 6). The correlations between SCDmax and FSC and LSC are weakly positive for both parameters, but still statistically significant. The metrics of this correlation are significantly higher at country level (r = 0.75, p < 0.05) and lower, depicting a moderate correlation across both highlands and lowlands, especially in the latter (r = 0.40, p < 0.05).
3.4. Trends in the frequency of snow-free days
SFD exhibits positive changes throughout most regions of the country, with a high share of statistically significant trends (p < 0.05) of 95% (Fig. 7). Regionally, SFD increase is faster in the lowlands, especially in the central, southern, north-eastern and western parts of the country, with an average rate of 3.4 days decade− 1. In the mountains, SFD is on a slower increase (2.4 days decade− 1) and mostly not statistically significant. There are no elevation-dependency patterns found in the vertical distribution of SFD trends across the country (Fig. 5 and Table 3). Country-wide, SFD has increased with an average rate of 3.2 days decade− 1.
3.5. Breakpoints in snow cover variability
Regarding the temporal variations, the extreme snow years having the highest advances/retreats of FSC/LSC, or the shortest/longest SCD and SCDmax are not fully synchronised across the climate regions and elevation bands of the country, showing a large spatial variability of the snow cover regime (Fig. 8). The long-term variability of snow cover parameters shows the existence of multiple breakpoints which mark the moments when the observed shifts intensified. The Pettitt change point test revealed a more evident clustering of statistically significant breakpoints (6% of WS) after 1990 for LSC, with the earlier melting trend intensified especially during the 1990s (e.g. 1990, 1991, 1996). For this parameter, a few WS (about 2%) located in the eastern lowlands of the country, the significant breakpoints belong to the 1980s.
For SCD, breakpoints were identified mainly during the 1990s, especially between 1996–1999, with a share of about 17% of significant cases. Comparatively, for SCDmax, the share of significant breakpoints during the 1990s decreases drastically below 1%, with most cases found before 1990 (about 4% WS). For SFD, most statistically significant breakpoints were detected before 1990 (about 11% WS), between 1970–1973, but especially from 1986 to 1988. The FSC parameter shows multiple breakpoints, mostly not statistically significant, with an even clustering of statistically significant breakpoints (about 1% WS) both before and after 1990.
Figure 8: The annual variability of FSC, LSC, SCD, SCDmax and SFD for the period 1961–2020. The julian days are considered for a snow year (1st August until 31st July). The vertical lines depict the breaking points according to Pettitt’s test for single change-point detection.
3.6. Links to large-scale atmospheric circulation
The underlying atmospheric conditions and their modes, such as the North Atlantic Oscillation, may influence the duration of SC through their linkages to air temperature and precipitation (e.g., Scherrer et al., 2004). This section investigates to which extent the variability of SC parameters is controlled by the seasonal atmospheric circulation patterns. Generally, in winter, the NAO influence is stronger in the Central and Eastern Europe(Sfîcă and Voiculescu 2014). The correlation between FSC and NAO is very weak or absent for most WS. Exceptionally, for only two WS a significant positive correlation was found (Predeal + 0.33, Pitesti + 0.28). The LSC and NAO negatively correlate (< -0.3) at 40% of the WS. A stronger negative correlation (< -0.4) is depicted for most WS located in the north-east (i.e., Iasi − 0.52, Botosani − 0.45, Focsani − 0.43, Roman − 0.41, Cotnari − 0.40) and southern-east (i.e., Urziceni − 0.50, Constanta − 0.45, Medgidia − 0.44, Bucuresti-Afumati − 0.42) of Romania. The SCD shows a good negative correlation with NAO at only 14 WS (12.2% of total WS), whereas for the SCDmax, the number of WS having a good negative correlation increases to 49 WS (42.9%). For both parameters the correlation coefficients are lower − 0.4, but statistically significant. NAO was found strongly and negatively correlated (< -0.5) with SCDmax at 8 WS, located in south-western part of the country (Drobeta Turnu Severin − 0.51, Calafat − 0.52), but mostly in the north-east (Suceava − 0.52, Tg Ocna − 0.53, Roman − 0.53, Cotnari − 0.53, Botosani − 0.55 and Radauti − 0.61). NAO shows a good and significant positive correlation with SFD at only 13 WS (11.4% of the total sample), mostly located in the north-eastern part of Romania (Fig. 10).