Trends in monthly temperatures
For the period considered, i.e. from 1980 to 2015, we calculated for each month and for all stations, the least-square linear trend of monthly average maximum temperatures (tx) and minimum temperatures (tn). Figure 1 plots the trends obtained for the Mont Aigoual station. Only trends in April, May and June were significant. For example, for tx the June one reached 0.110 (95% confidence intervals 0.026, 0.029) °C yr-1 or 1.1 °C per decade. For April, these trends reached 0.100 (0.025, 0.028) °C yr-1 and 0.088 (0.027, 0.030) °C yr-1 for tx and tn respectively. This pattern of change was observable for all stations. As a result, we aggregated months from April to June and analyzed the average temperatures over this period. For Mont Aigoual, tx and tn trends were then 0.103 (0.041, 0.044) °C yr-1 (p<0.001) and 0.076 (0.031, 0.040) °C yr-1 (p<0.001). Over the complete series of observations available for this station since 1901, Figure 2 shows that the starting point of our analysis, i.e. 1980, appeared to be in the change point region detected by .
These significant warming trends were not exclusive to the highest station. Larger values were observed, for example, at Nasbinal and Brenoux-Mende, with 0.100 (0.049, 0.037) and 0.109 (0.029, 0.024) °C yr-1 for tx and 0.130 (0.044, 0.033) and 0.096 (0.026, 0.025) °C yr-1 for tn respectively. These stations are located at altitudes of 1284 m and 1019 m respectively. One low elevation station did not have a significant trend for their tn: Arphy. Altitude played a highly significant role in tn trend (Fig 3a), which increased linearly with a slope of 0.0032±0.0008 °C yr-1 per 100 m and a y-intercept of 0.0418±0.0057 °C yr-1 (R2 = 0.36; p-value = 0.0003). For tx, no significant trends were observed (Fig 3b). We can consider this trend as spatially uniform and equal to 0.124±0.007 °C yr-1, corresponding to a +1.24 °C warming per decade. The effect of altitude was also tested by separating the stations according to their aspects. For tn and tx, the slopes and intercepts were not significantly different depending on whether the aspect was either north-facing or south-facing slopes.
The long monthly precipitation time series available, 1901-2014, for 13 of the 32 stations do not show significant trends for annual totals or seasonal ones from April to June, either for the whole period or only for the period ranging from 1980 to 2014. Graphs for the wettest station, Mont Aigoual (MAP = 1792±462 mm) and the driest one, Générargues (MAP = 1217±331 mm) are reported in Fig SI2. For these two stations we analyzed the co-variation between rain totals from April to June and the corresponding tx and tn. There is no significant effect of the precipitation amount on tn. For tx the effect is noticeable, but not significant at 5% threshold. For Mont Aigoual and Générargues, we observed declines of tx of 0.24±0.15 °C (p-value = 0.11) and 0.48±0.27 °C (p-value = 0.08) per 100 mm of precipitation. In 2008, the wettest year from 1980 to 2014, April to June precipitation was 907.8 mm at Mont Aigoual and 445 mm at Générargues. They caused theoretical declines in tx of 2.2°C and 2.1°C respectively. The observed tx were 10°C and 22.3°C for both stations.
Vegetation maps and land use/ land cover changes
Analyzing available maps, a significant and linear increase in the area of forests was observed, varying from 38% in 1955 to 68% in 2015 with a growth rate of 0.51±0.08% per year (Table 1). This forest expansion occurred at the expense of non-forest areas, which declined from 62% to 25% over the same period with a rate of -0.61±0.09% per year. For all available maps, the transition class was not well assessed. Absent in the 1955 map, the percentage of transition zones reached 7% in 2015. Its growth rate was only 0.11±0.05% per year.
For maps from the same source, i.e. CLC2000 and CLC2006, within the forest land cover type, broad leaved forests declined from 40% to 38%. Conversely, coniferous forests increased from 34 to 36%. This trend was ongoing until 2010 (CNP2010) since only 34% of broad leaved forests were present and coniferous forests reached 40%. Over the period 2000-2010, mixed forests accounted for a constant 26 percent value.
Fig 4a&b present the within-year patterns of change in white sky albedo WSA for the four target areas: coniferous and broadleaf forest, low shrubland and natural grassland. These patterns cover the twelve years (2000-2011) of 8-day mean ± SD. For the two forested areas, winter albedos were very close with 0.085±0.005 for the coniferous canopy and 0.087 ± 0.006 for the broadleaved one. Then albedo increased to peak early May at 0.126±0.007 and 0.136±0.005 followed by slight summer declines and then finally went back to winter values (Fig 4a). The WSA temporal patterns for low shrublands and natural grasslands were affected by snow cover within the cold period. On average, snow began late November-early December for the two vegetation types and disappeared end of February in grasslands and later, end of March, in the low shrublands (Fig 4b). During the vegetative period, albedos reached their maximum values at the beginning of July for natural grasslands (0.186±0.011) and then mid-July for low shrublands (0.156 ± 0.005). Standard deviation was higher in natural grasslands because they were more sensitive to local variations of soil water limitations and erratic stormy rainfall. When pooling the three forest types in the forest land cover type, a significant linear relationship was observed between April to June WSA and percent forest cover (PFC) surrounding the 32 weather stations: WSA = -7 10-4 ± 7 10-5 PFC + 0.186 ± 0.005 (R2 = 0.75, p-value < 0.0001). This means that the albedo decreased by 7 10-3 when the percentage of forest increased by 10%. The albedo level corresponding to the y-intercept was equal to the one previously observed on the target zone of pure natural grassland plotted in Fig. 4b. This local albedo around the stations, which is assumed to be slightly constant over the observation period 2000-2011, played a significant role in warming trends. For tx, this warming trend increased linearly with WSA with a slope of 0.385 ± 0.182 °C yr-1 albedo unit-1 and a y-intercept of 0.063 ± 0.027 °C yr-1 (R2 = 0.15; p-value = 0.04) (Fig 5). This means that transition from natural grasslands with an albedo of 0.185 to a forest with an albedo of 0.125 dampened the warming trends by 0.23 °C per decade. For tn, no significant relationship between warming trend and WSA was observed. The slope was lower than the one obtained for tx trend, with a value of 0.189 ± 0.217 °C yr-1 albedo unit-1 (R2 = 0.03; p-value = 0.39). The dampening effect of forests on the warming trend for minimum temperatures was less than for maximum temperatures. This result tends to demonstrate the likely effect of biophysical variables such as albedo on changes in the regional field of air temperatures. However, this site-specific result does not contribute in explaining the slightly decreasing, statistically non-significant, elevation-dependent warming of spring maximum temperatures.
In our aforestation scenario, we applied the IBPM model. We examined the three components of change in surface temperature separately. In April, the applied albedos were respectively 0.180±0.002 and 0.110±0.006 for natural grasslands and forests. Those in July were 0.184±0.002 and 0.124±0.002. Over the 6-month period analyzed, changes in albedo were rather constant and weak, ranging from 0.059 ± 0.009 and 0.070 ± 0.003. As a consequence, changes in daily net shortwave radiation were limited ranging from +24.6 ± 9.8 W.m-2 in September to +37.7±10.1 W.m-2 in July. The shortwave radiative forcing effect due to albedo change yielded a warming between +0.48 ± 0.22°C in April and +1.06±0.39°C in July. The main effect was the result of changes in roughness. Figure SI3 plots the monthly field of wind velocity. Mean daily wind speed at 2-m height was the highest in April with a value of 2.64 ± 1.05 m s-1 and the lowest in August at 2.00±0.75 m s-1. The corresponding aerodynamic resistances for grasslands and forests were respectively 38.5± 15.2 s.m-1 and 2.5 ± 1.0 s.m-1 for April and those for August, 44.8 s.m-1 ± 19.2 and 2.9 ± 1.1 s.m-1. Monthly cooling induced by changes in the energy redistribution factor attributable to changes in surface roughness ranged from -1.98 ± 0.75°C in April to -4.03 ± 1.04°C in July. The role of energy redistribution associated with changes in Bowen ratio was limited. Figure SI4 represents the time courses of Bowen ratios of both ecosystems. During the well-watered month of April, Bowen ratio for natural grassland was 1.79±0.41 and 1.38±0.15 for forest. With the occurrence of the summer drought, this ratio increased for both ecosystems respectively to 7.48±1.09 and 5.25±0.70 in July and then 7.10±1.66 and 5.34±0.79 in August. The Bowen ratio cooling only reached -0.17 ± 0.07°C in April and -0.15 ± 0.04°C in July. Finally, by summing up the three terms of the surface temperature change, we obtained a decrease between -1.67 ± 0.78°C in April and -3.12±1.11°C in July (Fig. 6). We demonstrate here that the drivers of this unexpected spring warming are not local and surely not related to landscape changes, so we studied further atmospheric patterns known to have significant impact on the Mediterranean climate.
By considering Mont Aigoual as our reference station, we analyzed the correlation between the monthly temperatures tx and tn from 1980 to 2015 with the corresponding values of three teleconnection indices, i.e. WEMO, NAO and MO. The time courses of the nonparametric Spearman’s correlation have been reported on Table 2. For WEMO, Spearman r was highly significant (p<0.001) in April and May for both tx and tn. The correlation declined in June but with p-values remaining nevertheless <0.01. The second time-period for which we observed a significant statistical link between WEMO and temperatures was from September to November with a peak in September (p<0.001) followed by a gradual decrease in the significance. Correlations during the other months were not significant. For NAO index, highly significant correlations were observed for tx in January, February and September those of March and May having p-values <0.01. For tn a high significance was only observed for February and September, January having a p-value <0.05. With MO, only correlations for February was significant with p<0.001 for tx and p<0.01 for tn.
By averaging indices and temperatures for April to June, months with highly significant warming trends (see Fig 1), we found, as expected, highly significant correlations of WEMO with both tn (r=-0.618, p<0.001) and tx (r=-0.698, p<0.001). For the same period, the correlations were not significant for either tn or tx both with NAO and MO. For extending such analysis to the whole-data set, which is the 32 station data, we performed a principal component analysis (PCA). The first component of the PCA accounted for 43.7% and 32.8 % respectively of the total variance for tx and tn, and the second component only described 6.4% and 7.6%. The Spearman correlations between the three indices and the first eigenvectors were only significant for WEMO with r = -0.54 (p<0.001) and r= -0.67 (p<0.001) for tn et tx respectively. The link between climate indices and more specifically the WEMO and maximum and minimum temperatures in spring raises questions about the underlying mechanisms.