Droughts in Spain have very different dimensions, affecting agriculture, forests and the frequency of wild fires, but currently the most important impacts relate to the hydrological dimension, which may cause heavy losses in highly productive irrigated agriculture, reduce hydropower production, and cause problems for industry and the water supply (Jerez et al. 2013; Lorenzo-Lacruz et al. 2010, 2013). Due to dry summers and strong interannual variability of the climate in Spain, there is a very dense network of hydraulic infrastructures. Thus, after China, Spain has the second highest number of dams in the world, but the total surface area of the country is 5% that of China. The purpose of this dense network is to guarantee the water supply during the frequent drought periods. For this reason, hydrological droughts in Spain are not affected by short term droughts, usually identified on short SPI/SPEI timescales. Thus, the dense reservoir network even allows a noticeable reduction in the impact of annual drought events (12-month) on hydrological drought conditions. For example, the one affecting Spain in 2015 was the most severe since records began, but it did not cause hydrological droughts due to the large amount of hydraulic infrastructures. On the contrary, climate drought conditions recorded during 2–3 years, limit the capacity of infrastructures to cope with drought events, and cause problems for irrigation and water supply, as widely observed between 1992 and 1995. For these reasons, this paper focuses on long-term drought indices (12, 24 and 36 months), since they are really useful to assess the severity of hydrological droughts and determine how annual drought conditions can persist over longer timescales. Lorenzo-Lacruz et al. (2010) clearly illustrated how the multiannual large reservoirs, which are frequent and the basis for water management in Spain, are responding to very long drought timescales > 24-month.
Defining a drought period involves deciding which thresholds to use, and we focused on episodes covering the largest surface areas under defined drought thresholds. This meant that some minor drought periods (regional to local) were not identified in this study, but there is evidence proving that they are frequent on the Spanish mainland (Vicente-Serrano et al. 2014; Peña-Gallardo et al. 2016). We hope in future to be able to analyze all drought periods affecting the Spanish mainland to develop a drought catalog. In the meantime, this research shows that when identifying outstanding global drought events, the use of thresholds based on the drought index combined with the surface area affected by droughts are an interesting tool, even though they are insufficient to determine the real severity of these events at sub-regional scales. The use of different timescales is equally important for this purpose. Nevertheless, this is a difficult task, since a simple assumption of temporal propagation of the drought conditions across different timescales is not so simple, as illustrated in this paper, for events between 1989 and 1995. In summary, individual drought events must be put in a temporal context because in many circumstances they are promoted or exacerbated by past conditions.
In this study, we identified “simple” major drought episodes on a timescale of 12 months. These are the less complex droughts analyzed, and in general, they were found at the beginning of the study period. This became more complex when seeking to extend 12-month drought episodes to 24- and 36-month scales, when we detected overlapping effects at times. In these episodes, the eventual propagation from shorter to longer timescales is much more difficult to detect, since the drought conditions are not clearly identified at shorter timescales. This would be much more complex if shorter (< 12-month) and longer (> 36-month) scales were incorporated. In short, a drought episode must be understood according to its temporal context and never analyzed as an isolated feature.
Differences between drought indices
There is a progressively divergent evolution of the percentage of land affected by drought conditions defined by two different drought indices for the Spanish mainland. Given the diverse nature of the indices, it suggests that the triggering factors driving droughts on the Spanish mainland may have changed during the last few decades.
Various studies have suggested a possible role of global warming on drought severity on the global scale (e.g., Trenberth et al. 2014) following intensification of the hydrological cycle as a consequence of increased AED (Huntington 2006). Nevertheless, the uncertain availability of data is a serious drawback, and prevents a general conclusion being drawn on this issue on a global scale (Seneviratne et al. 2012). In this paper, we sought to avoid these constraints in our regional analyses of drought on the Spanish mainland using a high-resolution spatial dataset, and the main results suggest that the surface affected by droughts in the last two decades may be caused by different mechanisms from previous ones.
We found that the area detected under drought conditions prior to c. 1990 usually has a higher SPI than SPEI indicator. This result suggests that prior to 1990, droughts were mostly related to changes in precipitation, and after 1990 that, in addition to the role of precipitation variability, the atmospheric evaporative demand (AED) has acted as drought triggering factor. These results are coherent with temperature and precipitation evolution on the Spanish mainland, where the most recent studies have detected a general temperature increase from 1951 to 2010, particularly between 1970 and 1990, mostly affecting the spring and summer minimum records (Gonzalez-Hidalgo et al. 2015, 2017), while no significant trend in precipitation was found except for March and June (negative and significant) between 1946 and 2005, and particularly from 1970 (Gonzalez-Hidalgo et al. 2011); the results have been confirmed and updated to 2015 by Notivoli (2017). Furthermore, recent evaluation of AED on the Spanish mainland suggested a global increase of about 125 mm since 1960, which has been attributed to the increase in vapor pressure deficit (Vicente-Serrano et al. 2014b). These findings agree with the results of higher drought detection by SPEI than SPI in the last two decades, and suggest that the water supply to the atmosphere has not been sufficient to cope with the water demand, in line with the warming observed (Vicente-Serrano et al. 2014c). In brief, the global drought framework has recently been driven by new temperature conditions that control AED, irrespective of the evolution of precipitation.
Drought event propagation processes
Spatial variability of drought events is highly complex on a regional scale. This has been demonstrated in several regions of the world (e.g., Skaggs 1975; Song et al. 2014; Capra and Scicolone 2012), and also in previous analyses in the Iberian Peninsula (Vicente-Serrano 2006b) where it has been suggested that spatial drought patterns may differ as a function of the timescale of the drought index, and longer timescales may give rise to more diverse patterns, given the longer impact of the specific local precipitation events.
In the present study, we confirmed these previous findings at high-resolution level, but the results go further and show that there are different spatial gradients of drought propagation.
On mainland Spain, we found three spatial drought propagation gradients that could be linked to several factors controlling droughts. Various explanations have been suggested for precipitation and temperature evolution linked to global and local factors (e.g., Gonzalez-Hidalgo et al. 2011, 2017), in which different hemispheric circulation mechanisms, or more local ones, such as land use changes, have been suggested. Regardless of the exact reason, which is not the objective of this paper, we found that the onset of main drought episodes followed a coast-inland gradient, and no episodes were identified from inland to the coast. It is highly probable that this relates to the main sources of moisture from the Atlantic or Mediterranean water masses (Gimeno et al. 2010), and suggests that the low value of convective rainfalls on annual total precipitation, mostly produced in summer and highly irregular, thus with little effect on both drought indices.
The spatial propagation of droughts also shows the effect caused by the main relief features in the spatial distribution of affected areas, and resembles the well-known spatial areas affected by the most prominent tele-connection patterns. Thus, under west–east propagation episodes, the mountains act as a frontier separating the central-western Spanish conterminous land under drought conditions from the Mediterranean coastland that may be under moist conditions. These features closely resemble the spatial distribution of the North Atlantic Oscillation effects on precipitation and temperature, particularly during the Winter months. Under east–west propagation episodes, the same is true in the opposite direction, and resemble the areas affected by the Western Mediterranean Oscillation. This mountain border effect is more evident when the affected areas are less than 25% of total land (not analyzed in the present paper), but clearly recognized in the case of 1979–1980 episodes in the Mediterranean coastland (not presented).
The third most significant results concern the onset of drought. Apparently, drought episodes start at two different times: (i) the cold months from autumn (1971, 1981, 1994 and 1995), and (ii) the end of winter or beginning of spring (1965, 1992, 2005 and 2012). The differences in the onset of droughts would be able to explain the behavior of the event, and the differences found in the SPI and SPEI. In the first case, when drought starts at the end of autumn it seems to suggest that the previous summer avoided drought conditions by spring–summer precipitation delayed to the beginning of autumn, notwithstanding the positive significant trend of precipitation in October, this should be taken into account in future detailed research. In the second case, when drought starts at the end of winter, it seems to be related with scant winter precipitation. In brief, we can conclude that there are different types of drought, depending on their origin, sometimes promoted by scarce rainfall (the first case), or by means of a more complex mechanism combining low rainfall and the different factors controlling AED.
These findings could be the reason lying behind divergence in surface detection by SPI and SPEI, and also how drought spreads differently according to different temporal scales.
To conclude, the high-resolution analyses of drought on both scales, spatial and temporal, in the Spanish conterminous land indicates the extreme complexity of the phenomenon, the need for further research, and caution when extrapolating from general conclusions.