studied sea surface temperature in the Mediterranean for a 28-year long period, ranging from 1982 to 2009. SST data used for that study were obtained from the NASA/NOAA Pathfinder data bases. The Pathfinder project is dedicated to the production of global SST maps from 1982 to the present day with data measured by the AVHRR sensors aboard the NOAA satellites. AVHRR Pathfinder Version 5.0 data are available at a global scale and at 4km spatial resolution, and they are obtained twice daily for daytime and night-time satellite passes. This SST data set was validated for its use in the Mediterranean by . determined that Pathfinder data were a consistent data set that could be used in the detection of trends and SST variability. More information on Pathfinder SST data is available at and on the data set website

http://www.nodc.noaa.gov/SatelliteData/pathfinder4km/

A monthly climatology for the period January 1982 to December 2009 was built in to study SST in the Mediterranean. Mean monthly values were calculated from daily data in every grid point covering the Mediterranean, with the restriction of having at least 75 % of valid daily values, preferably night-time values but using daytime ones if they were not available. Additionally, clustering techniques were used to study SST spatial distribution patterns across the whole study period for monthly values and anomalies. From a temporal analysis two distinct, well-defined SST regimes were found for summer and winter with two transitional periods in spring and autumn. Usually, the spring transition from the winter to the summer regime is more abrupt and shorter in time than the autumn transition. The spring transition period lasts between 1 or 2 months in April and May, although in some years April shows more resemblance to the winter than to the summer regime. The autumn transition lasts longer, from 2 to 3 months, i.e. September to November, and runs more smoothly. It should be noted, to avoid confusion, that seasonal nomenclature for temperature regimes does not strictly coincide with the climatological seasons. In some cases climatic season and SST regime can be shifted so we can find a summer SST regime during climatological fall or even early winter.

From the clustering analysis on spatial SST distribution two main modes of areal distribution for winter and summer respectively, and two transitional periods in spring and autumn were found (Fig. ). For the winter mode a clear positive north-to-south gradient (Fig. a) was found for the whole Mediterranean with higher SST values in the southernmost part of the Eastern Mediterranean basin. The winter mode usually starts in late November and lasts until March, although in some years April retains some structure similar to winter months. After the spring transitional period (Fig. b), usually April and May, the summer mode (Fig. c) presents a completely different structure characterized by the presence of clearly distinct areas. The coldest areas in summer are the Alboran Sea, Gulf of Lion and the Aegean, especially its eastern part. Higher SST values are found in distant areas like the one between the central coast of the Iberian Peninsula (IP) and the Balearic sea, the Thyrrenian and Ionian seas and the area from the south of Crete to the Egyptian coast. The highest temperatures are found on the Gulf of Libya and in another area running from southern Turkey and Cyprus to the coast of the Middle East. This distribution can be conditioned by oceanic circulation and, at least in part, by meteorological causes: for example, the Gulf of Lyon and Aegean sea areas are affected by strong and persistent wind regimes, Mistral and Etesian winds respectively. The rest of the Western Mediterranean (WMED) basin wind regime in summer is dominated by breeze cycles and vertical air mass recirculation over the basin which may be the drivers of the clusters in the Valencia region-Balearic Islands area and in the Thyrrenian sea. In the Eastern Mediterranean basin , the Gulf of Libya and Middle East are areas where breeze cycles develop in summer. The prevailing light surface winds can favour the stagnation, or weak displacement, of surface waters, an exception is made of the Aegean sea where the Etesian (N-NE) winds are predominant in summer. The autumn transitional period (Fig. d) usually starts in October, although in some years it can comprise part of September, and it is not as steep as the spring transitional mode, usually ending in November and leading to the winter regime in December.

Regional Atmospheric Modelling System (RAMS) has been used in previous work at Fundación CEAM in the study of Mediterranean meteorology and pollutant dispersion from versions 4.x up to the most recent 6.0 version, the one used to run simulations shown in this paper. RAMS optimum configuration for mesometeorological studies in the Valencia region was investigated by . RAMS has also been used to study the effect of sea surface temperature on torrential rains , air pollution dispersion and on the implementation of a heatwave alert system in the Valencia region . RAMS has also been used in the study of heavy rains and floods in the Mediterranean area by other authors like , , and .

The initial and boundary atmospheric conditions used for the simulations in this paper come from the National Centre for Environmental Prediction (NCEP) reanalysis, obtained from the National Center for Atmospheric Research (NCAR) . Reanalysis data are available every 6 h at 2.5∘ × 2.5∘ resolution and 17 pressure levels. These data are used in a four-dimensional data assimilation scheme to define forcing at the lateral boundaries of the outermost five grid cells of the largest simulation domain. For the surface boundary conditions we have used land cover data sets from the US Geological Survey . SST initial data for the simulation was obtained for the model domain from the NASA/NOAA Pathfinder data bases.

RAMS own soil-vegetation surface scheme (LEAF-3) is applied to evaluate sensible and latent heat flux exchanges with the atmosphere, using prognostic equations for temperature and soil moisture. LEAF-3 has been prescribed with a homogeneous soil texture of the clay-loam type. Soil column holds 11 layers down to a depth of 2 m with moisture initialized with a uniform profile at a value of 0.38 m3 of water per cubic metre of total volume. The initial soil temperature profile is obtained by subtracting 2.3 ∘C from the surface air temperature in the top soil layer. Temperature linearly decreases by 1 ∘C in the bottom soil .

Simulation runs in this paper have been designed with four domains of decreasing size at increasing spatial resolution. Four, two-way interactive, nested domains have been used whose horizontal resolution is shown in Table . All four grids hold 45 vertical levels, starting from a 30 m thick level near the surface that gradually increases to a maximum of 1000 m thickness near the model top, located at about 17 000 m. The cloud and precipitation microphysics scheme has been applied in all the domains. From the work of the Kuo convective parameterization scheme has been activated in RAMS model for the three outer grids as it showed the best results in the modelling of torrential rain events. For model grid 4 (1.5 km horizontal resolution) no convective parameterization is activated so that the model is left free to generate its own small-scale features and convective precipitation.

To achieve the objective of this work, to determine SST contribution to the development/intensification of torrential rains in the Valencia region, we have employed a new approach for our simulation experiments with RAMS model. As a first step, we have studied the air mass trajectories leading to the rain event in order to determine the Mediterranean areas that can act as moisture/heat sources for that specific rainfall episode. Air mass backward trajectories in the days previous to the event have been computed with the HYSPLIT model available online on the NOAA Air Resources Laboratory website

Rolph, G. D.: Real-time Environmental Applications and Display sYstem (READY) Website http://www.ready.noaa.gov, NOAA Air Resources Laboratory, College Park, MD.

From 4 to 7 September heavy rains were recorded in the Valencia region, mostly in its southern-central areas. Accumulated precipitation values reached 500 mm during the whole event in some stations, with daily values greater than 200 mm . This rain event was characterized at surface levels by the presence of a high pressure anticyclonic area over central Europe and relatively low pressures over Northern Africa (Fig. a), driving the air mass along an easterly/northeasterly flow over the Mediterranean (Fig. a), while at upper levels (Fig. b) a cold air pool was located over IP.

Figure  shows SST fields used in the different simulations of the September 1989 rain event, corresponding to a summer SST spatial distribution, according to the work of shown in Fig. 1. As the rain event occurred during the first week of the month, monthly August data for SST have been used. Upper left map shows SST for the control run, original Pathfinder SST data while the rest of maps show the perturbed SST field, prescribing a value of 10 ∘C for the different simulations in Table .

Simulations for this event, with a duration of 120 h, begin on 3 September at 00:00 UTC and end at 00:00 UTC on 8 September. RAMS model accumulated precipitation total for the whole set of simulations is shown in Fig. . Control simulation (CtrlA) for the September 1989 event shows a wide precipitation area with amounts higher than 100 mm where two areas with maximum precipitation can be distinguished. The northernmost, and smaller, area reaches accumulated values of 400 mm while the greater one surpasses 600 mm, mostly standing over the sea. From these results, most of the precipitation, and also the highest values, occurred over the sea and coastal areas.

In the A1 simulation (Fig. b) significant changes were observed with respect to the control simulation. The 100 mm area is now greater over land than in the control simulation but the most noticeable difference is the displacement of the maximum precipitation area some distance inland, although not far from the coast, with values higher than 500 mm. In this case, the maximum simulated precipitation is located over land on coastal areas and rapidly decreases when going offshore, contrary to the control simulation. A dramatic precipitation decrease appears in A2 (Fig. c) simulation with respect to the control one, with precipitation under 100 mm over the whole simulation area and its maximum located far to the south from the control simulation peak. For A3 experiment (Fig. d) the model shows a rainfall spatial distribution that resembles that of the control simulation but with lower values across the whole modelling area. Again, the precipitation maximum extends across the coastal areas and over the sea but with a more confined spatial extent than in the control simulation, especially to the north of the simulation area where a maximum precipitation area from the control simulation is not present in this case. Finally, A0 experiment (Fig. e) shows similar results to those from A2 simulation but with still lower values and almost without precipitation over land except from a reduced area over the coast.

It is remarkable that A1 simulation modifies precipitation field and increases its values, especially the higher ones, with respect to control simulation while the rest of the simulations clearly decrease precipitation values and its spatial extent. The analysis of vertical cross sections of equivalent potential temperature and vertical velocity (see Supplement) present some differences between the analyzed simulations. For control A1 and A3 simulations, a moist surface air flux towards the Valencian coast is found (see RAMS model trajectories in Fig. ), with higher moisture content for the CtrlA and slightly decreasing values for A1 and A3, while A2 and A0 surface fluxes present lower moisture contents than in the rest of the experiments. A notable difference comes for the vertical velocity field with strong vertical circulation present on CtrlA and A1 simulations being weak in the rest of experiments. In the case of the control simulation this vertical ascent is present just off the coast while in A1 it is displaced some distance inland, and probably enhanced by orographic lifting. These changes could explain the displacement of the maximum precipitation area from its offshore location on CtrlA to the new location over land on A1.

It is also noticeable that RAMS computed trajectories (Fig. b) for all the simulations are quite similar, and also similar to the ones computed from reanalysis data, concluding that no significant changes arise in atmosphere dynamics for the different experiments.

A long-lasting rain episode took place in the Valencia region between 22 and 26 October 2000. Rainfall also had a large spatial extent, affecting most of the Valencia region, with daily accumulated values greater than 300 mm in a noticeable number of stations with total episode accumulation over 500 mm (see Supplement). Regarding synoptic conditions (Fig. ), a cold trough at upper levels extended across western IP until an isolated cold air pool formed over Southwest IP and Northwest Africa on 22 October. For this event, a long east to southeasterly air flow travelled across the whole Western Mediterranean basin (Figs.  and a) heading towards the eastern coast of the IP. In the trajectories computed from simulation experiments (Fig. b), the cases with perturbed SST field move slightly to the north with respect to the control simulation in the area between Sicily and the coast of Libya to converge again near the Balearic Islands. Sea surface temperature field (Fig. ) for this event showed a transitional autumn distribution (see Fig. 1d). A more comprehensive description of the event can be found in .

In this case simulations ran for 120 h, starting on 21 October at 00:00 UTC and ending at 00:00 UTC on 26 October; model precipitation total is shown in Fig. . The precipitation event recorded in October 2000 (see Supplement) affected a wide area to the centre and north of the Valencia region; RAMS model accumulated precipitation results for the control simulation, CtrlB (Fig. a), show relatively good agreement with recorded precipitation regarding spatial distribution but fails in the maximum rainfall values. Model results show a large rain band extending parallel to the coast, but some distance inland, with values higher than 100 mm inland and weak precipitation over the sea. Regarding maximum values, some reduced areas over 200 mm are found and a maximum accumulated precipitation over 300 mm is located to the north of the modelled area. In the first simulation with perturbed SST field B1 (Fig. b), RAMS model shows similar results to those in control simulation but with slightly higher values over land. A greater precipitation area, with values over 100 mm, than in control run is found over land to the south of the model grid while similar or slightly higher values are found on the rest of the rain areas.

The rest of the simulations for this event (B2, B3, B0) show similar results regarding spatial rainfall distribution while they differ in the amounts of precipitation calculated by the model. B2 simulation (Fig. c) shows a remarkable shift towards the coast line for simulated rainfall where a rain band over 100 mm is found. Precipitation inland, in the western half of the model grid, is clearly lower than in control simulation, more remarkably for the maximum areas. The third simulation for the October 2000 event, B3 (Fig. d), shows the same spatial structure for precipitation seen in B2 run with a rain band along the coast but with generally higher precipitation values than in B2 case. Finally, B0 simulation results (Fig. e) are fairly similar to those in B2 in spatial distribution just showing slightly lower accumulated precipitation values.

Summarizing the model results, B1 simulation shows similar results to the control simulation for precipitation spatial distribution but with some changes for accumulated rainfall while the rest of the numerical experiments shift precipitation rain band to the east just over the coast line. Regarding modelled precipitation, B2, B3 and B0 differ in accumulated value, B0 being the one with lower values and B3 the one with the highest ones. It is remarkable that B0 simulation gives very similar precipitation values as in B2.

Regarding Mediterranean moisture fluxes, RAMS model trajectories in Fig. b show the long-range marine advection towards the Valencia coast. In this event, control and B1 simulations show similar equivalent potential fields (see Supplement) in a cross vertical section at the latitude of the rain area. In the rest of the simulations, B3, B2 and B0 moisture flux is clearly weaker than in the first ones. For these latter modelling experiments, a decrease in equivalent potential temperature (see Supplement) is found over the coastal rain areas with respect to the CtrlB and B1 simulations.

The October 2007 rain event is the shortest of the three episodes studied in this paper. Rain started on the second half of 11 October and lasted for about 16 to 20 h depending on the location. In addition to its shorter duration, the rain had a high spatial concentration and intensity, focusing on the coastal area to the south of the Valencia and northern Alicante provinces. Records over 300 mm were obtained in this area with a few points over 400 mm .

Similar to the other two events, strong instability was present at upper levels with a cold trough arriving to the IP between 9 and 10 October. Finally, the trough evolved to an isolated cold pool which was located in the vertical of the Valencia region on 11 and 12 October (Fig. ). At surface levels a northeasterly wind flow (Fig. a) advected a humid air mass across the northwestern Mediterranean towards the Valencia region; it can be seen (Fig. b) that for the perturbed SST simulations the path of the air mass over the sea is longer than in control simulation but is still travelling over the same areas of the Mediterranean. This wind flux was driven across the southern edge of a strong anticyclone located over France and southern England, with high pressures extending across central Europe. Figure  shows SST distribution for this event, corresponding to winter mode.

The latter modelled rain event, October 2007, presents similarities regarding recorded precipitation spatial distribution with the one from September 1989. Rainfall was located in the area with highest torrentiality in the Valencia region and model results show roughly the same spatial distribution although the 2007 case has a lower spatial extent. Control simulation (Fig. a) shows a maximum precipitation accumulated value (above 240 mm) located on the centre of the model grid, some distance inland, with a secondary maximum area over the sea with a clearly lower value. A relatively large area with precipitation values exceeding 100 mm lays around the first cited maximum area.

The first simulation with perturbed SST, C1 (Fig. b), shows a notable decrease of about 100 mm in the maximum accumulated values. Concerning spatial distribution, rainfall structure over land present the same features as in control simulation; the main differences appear over the sea where the secondary maximum located over the sea in control simulation completely disappears in this case while a larger rain band appears to the south of the model grid. The simulation results for C2 (Fig. c) and CO (Fig. d) experiments present quite similar results. Both simulations show a dramatic decrease on model-accumulated precipitation that disappears over most of the model grid, only some residual precipitation is found inland far away from the control simulation precipitation area.

As in the previous experiments, air mass trajectories (Fig. b) and vertical cross section (see Supplement) of equivalent potential temperature plus vertical velocity from RAMS model results have been analysed. Air mass advection from northern Mediterranean brought marine air at surface levels towards the Valencia region. For the control and C1 simulations, clear surface moisture flux from the Mediterranean to the precipitation area is found while some differences appear in the location and intensity of vertical ascent. Again, C2 and CO simulations show very similar results with a drastic decline of surface moisture flux towards the coast, leading to reduced precipitation.