Spatial variability of northern Iberian rainfall stable isotope values: Investigating climatic controls on daily and monthly timescales

This article presents for the first time a large dataset of rainfall isotopic measurements ( 18 Op and  2 Hp) sampled every day or every two days from seven sites in a west-to-east transect across northern Spain for 2010-2017. The main aim of this study is to: (1) characterize rainfall isotopic variability in northern Spain at daily and monthly time scales, and (2) assess the principal influencing factors determining rainfall isotopic variability. This comprehensive spatio-temporal 30 approach allows exploring the role of air mass source in determining the isotopic composition of rainfall in northern Iberia https://doi.org/10.5194/acp-2020-861 Preprint. Discussion started: 8 September 2020 c © Author(s) 2020. CC BY 4.0 License.

by using back-trajectories; Atlantic fronts are found to be the dominant source of northern Iberia rain events studied. The relative role of air temperature and rainfall amount in determining the stable isotope composition of precipitation changes along the west-to-east transect. Air temperature appears to be the most significant influence on  18 O p at daily and monthly time scales with the highest air temperature- 18 O p dependency found for the Pyrenean station while a few sites in the transect 35 show a significant negative correlation with precipitation amount. Distance from the coast, site elevation, and moisture source region (Atlantic versus Mediterranean) also significantly modulate the  18 O p values and ranges but the type of precipitation (convective vs frontal rainfall) plays a key control, with convective rainfall associated with higher  18 O p values.
This dataset of the rainfall isotopic composition represents another step forward towards developing a more detailed, mechanistic framework for interpreting stable isotopes in rainfall as a palaeoclimate and hydrological tracer. 40

Introduction
The oxygen isotopic composition of rainfall ( 18 O p ) is often considered as the dominant influence on the isotopic composition of terrestrial archives (ice cores, speleothems or authigenic lacustrine carbonates) used to reconstruct past climate (e.g., Leng, 2006). However, few palaeoclimate reconstructions are supported by an in-depth understanding of the regional climatic controls on modern precipitation  18 O ( 18 O p ) (e.g. Treble et al., 2005),. As a consequence, paleoclimate 45 proxies are often interpreted without a clear knowledge of the processes involved in modulating  18 O p at a particular region (López-Blanco et al., 2016;Moreno et al., 2017). It has long been established that  18 O p is an integrated product of air masses history , modulated by specific prevailing meteorological conditions (air temperature and amount of precipitation for example) (Craig, 1961;Dansgaard, 1964). This results in different dominant factors controlling  18 O p variability depending on the site location, i.e., latitude, continentality, elevation, seasonal distribution, local air temperature and amount and source 50 of precipitation (Rozanski et al., 1993). A detailed study of current  18 O p values and their variability in a given region is mandatory if one wishes to reconstruct past climate changes using  18 O in regional climate archives (Lachniet, 2009).
Long rainfall isotopic time series allow for comparison of the  18 O p signal with meteorological variables and calibration of proxy records. Unfortunately such long-term observational studies are scarce, and thus, only a few, although outstanding, examples of studies examining factors controlling  18 O p are available for continental Europe (Field, 2010;Genty et al., 2014;55 Meteorologia de Portugal, 2011;Martin-Vide and Olcina-Cantos, 2001) require further detailed and highly spatially-resolved studies.
A major advance in understanding the controls on  18 O p has been the proliferation of studies using daily-scale monitoring to 65 address the mechanisms behind isotopic signatures at daily timescales (Baldini et al., 2010;Fischer and Baldini, 2011), incorporating the complexity associated to the type of rainfall, for example (Aggarwal et al., 2016)Regrettably, the scarcity of Global Network of Isotopes in Precipitation (hereafter GNIP) sites in Iberia, particularly those recovering data at a daily scale, prevents a broader regional study of climate controls on  18 O p values. In the IP, only one study has analysed  18 O p variability at a daily basis covering a short 3-year period (2000)(2001)(2002) (Araguás-Araguás and Diaz Teijeiro, 2005) and, more 70 recently, a 3-year monitoring survey focused on the Iberian Range (Molinos, Teruel, NE Spain) (Moreno et al., 2014). That study revealed the importance of the source effect on  18 O p values, due to the alternating influence of two air masses trajectories with different isotopic ranges; basically, Atlantic fronts with more negative  18 O p values (from the west) and Mediterranean convective storms with more positive values (from the east) air trajectories (Moreno et al., 2014).
Additionally, another recent study based on back trajectories emphasized the role of recycled moisture uptake within the IP 75 in the final values of  18 O p in Central Spain (Eagle Cave) (Krklec and Domínguez-Villar, 2014). In addition, to date, the majority of empirical studies of the meteorological controls over δ 18 O P rely upon event scale, daily, or monthly time series from individual locations (Smith et al., 2016), which raises concerns about the spatial representativeness of the resulting statistical models and the mechanisms behind those relationships.
In this paper, we propose an alternative approach by analysing daily and monthly patterns of δ 18 O P from multiple stations 80 across northern IP, and the Balearic Islands, following a west-to-east transect (850-km in straight line) that extends from an area dominated by a typical Atlantic climate to fully Mediterranean sites. The overall aim is to characterize and quantify the dominant factors modulating  18 O p variations in time (daily and monthly) and space, to determine the causes of precipitation isotopic variations regionally. Additionally, this study will serve to improve the interpretation of oxygen isotope paleorecords from the region that depend on  18 O p (Bartolomé et al., 2015;Domínguez-Villar et al., 2017;López-Blanco et al., 85 2016;Pérez-Mejías et al., 2019;Sancho et al., 2018Sancho et al., , 2015.

Prevailing climate regime and site description
Our study compares for the first time rainfall isotopic values and meteorological variables (temperature, precipitation, 90 moisture sources and type of rainfall) at seven sites in northern Iberia and Balearic Islands, covering an 850-km long westto-east transect from an area under typical Atlantic (Oviedo and El Pindal) to fully Mediterranean (Mallorca Island and Barcelona) climate. The west-to-east transect is completed with three additional sites in a transitional zone: two from the Iberian Range (Molinos and Ortigosa de Cameros) and one from the Pyrenees (Borrastre) (Figure 1a). At those seven locations rainfall was sampled daily covering different time periods except at El Pindal where rain was collected every 48h 95 https://doi.org/10.5194/acp-2020-861 Preprint. Discussion started: 8 September 2020 c Author(s) 2020. CC BY 4.0 License. (Table 1). Borrastre record is, to our knowledge, the most comprehensive dataset of daily  18 O p for Spain in terms of both the time span covered (2011)(2012)(2013)(2014)(2015)(2016) and number of samples (380 days) (Table S1).
In north-western and north-central Iberia, precipitation is mainly controlled by the presence of westerly winds and the passage of Atlantic fronts, mainly during November-April (Martín-Vide and Olcina Cantos, 2001). During the rest of the year, the subtropical Azores high-pressure system shifts northward, favouring stable conditions by blocking the westerly 100 circulation and moisture inflow (Archer and Caldeira, 2008), thus reducing precipitation. This wet winter/dry summer regime is quite different from that in the north-eastern Mediterranean region, where winters are generally dry (foehn effect) whereas warm season precipitation (from late spring to early autumn) is dominated by convective storms and also easterly advections over the Mediterranean Sea (backdoor cold fronts) (Millán et al., 2005). These local to mesoscale storms are primarily associated with frequent and persistent sea breezes (Azorin-Molina et al., 2011) which bring warm and moist air masses from 105 the Mediterranean sea inland (Azorin-Molina et al., 2009). During the summer season, this is typically the only source of precipitation in the north-eastern area, bringing an average of 100-125 mm yearly (Millán et al., 2005). Backdoor cold fronts from the Mediterranean Sea are sporadic events occurring in autumn (secondarily in winter-spring), but they cause heavy precipitation and flooding (Llasat et al., 2007). Figure 1B summarizes these three major precipitation regimes defined by Millán et al. (2005): (i) Atlantic frontal systems (westerly winds), (ii) convective-orographic storms, and (iii) Backdoor cold 110 fronts from the Mediterranean Sea (easterly winds).
Winter precipitation in large parts of the IP is strongly influenced by the North Atlantic Oscillation (NAO) at annual and interannual scales: higher precipitation occurs when the NAO index (NAOI) is negative and the storm tracks are shifted southwards, more directly influencing the IP (Trigo et al., 2002). Lower correlation values (r =-0.1-0.4) between the NAOI and winter rainfall, however, are observed in our region of interest, the northern IP (Goodess and Jones, 2002), which 115 encompasses both the wet western regions and the dry Mediterranean in the northeast. For the latter region, a significant relationship with the Western Mediterranean Oscillation (WeMO) in spring and autumn is attributed to fluctuations of warm moist inflow air from the east and its influence on Mediterranean cyclogenesis (Martin-Vide and Lopez-Bustins, 2006).
The rainfall influencing the seven stations included in the studied transect originates in two dominant source regions: the tropical-subtropical North Atlantic and the Western Mediterranean (Gimeno et al., 2010). Below, the four regions across 120 which the seven stations are distributed are described in terms of their climatology. Regional meteorological data are provided in Figure 4A.
The Cantabrian coast. The site of El Pindal in the Cantabrian coast ( Figure 1A) is characterized by a typical oceanic climate with mild summers and winters (Cfb, following Köppen and Geiger -KGC-classification) due to the proximity to the coast.
Rainfall mainly occurs in late autumn and early winter with a minimum in summer ( Figure 4A), and are associated with 125 Atlantic frontal systems (westerly winds). Additionally, rainfall samples from Oviedo (climate characteristics similar to those at El Pindal) were collected and are also included in this study.
The Iberian Range. Ortigosa de Cameros is located in the Encinedo Mountain area in the westernmost sector of the Cameros Range (Iberian Range, Figure 1A) and is dominated by a continental Mediterranean climate (Dsb, following KGC classification). Rainfall mostly occurs in autumn and spring, with some convective-orographic storms in summer 130 (climograph in Figure 4A). Also located in the Iberian Range and at similar elevation but further east, in the Maestrazgo basin, the Molinos site is characterized by similar climate (Dsb in KGC classification), with a highly pronounced seasonality and precipitation occurring mainly in spring and in autumn ( Figure 4A).
The Pyrenees. Borrastre village is located in the Central Pyrenees ( Figure 1A Cristo localities in the Mallorca island and by Barcelona ( Figure 1A). Precipitation is mostly distributed from October to April ( Figure 4A) mostly associated with backdoor cold fronts from the Mediterranean Sea (easterly winds) as the influence of Atlantic precipitation is weakened over this area. 140

145
Rainwater was collected using a similar procedure to that recommended by the International Atomic Energy Agency (IAEA) for daily sampling (http://www-naweb.iaea.org/napc/ih/IHS_resources_gnip.html) for six of the seven stations (Oviedo, Ortigosa de Cameros, Molinos, Borrastre, Mallorca and Barcelona). Thus, precipitation events greater or equal to 1 mm were sampled from a rain gauge which allows measuring the amount of rain fallen and sample it manually taking out the water from the rain gauge with a syringe. The collected water was then homogenized and filtered at the time of sampling, later a 150 5ml aliquot was stored in polypropylene tubes sealed with screwcup without air inside and kept cold in a refrigerator until isotopic analysis. Rainfall samples were collected at the end of each precipitation event, immediately afterwards whenever possible or after a few hours, with the total event precipitation homogenized. At El Pindal site the procedure was different: rainfall was collected every 48h for several months (November 2006 to April 2009) using an automated sampler (Table 1) located on the roof of the San Emeterio lighthouse located <10 m from the modern sea cliff and 200 m from the 155 cave. Thus, since the samples were automatically collected and remained in the lighthouse for several days, a film of paraffin oil was used to prevent evaporation.
The observation staff in charge of each location collected a sample directly following every rainfall event, except in El Pindal that the system was automatic and in Mallorca, where several events were missed during the first two years of the collection period, preventing the calculation of monthly averages for some intervals (monthly and annual averages and 160 standard deviations in Table 2). In addition, seven rainfall events were collected at two different localities in Mallorca https://doi.org/10.5194/acp-2020-861 Preprint. Discussion started: 8 September 2020 c Author(s) 2020. CC BY 4.0 License.
(Manacor and Porto Cristo) obtaining similar  18 O p results. For those events, a weighted-average value using the two localities was calculated (see Table S1). Thus, 47 rainfall samples were collected from Oviedo manually in 2015. In Ortigosa de Cameros, rainfall was manually collected daily between September 2010 and December 2014 by the staff (guides) of the La Viña and La Paz show caves, with an interruption from December 2012 -January 2014. In Molinos, rainfall was 165 manually collected by the staff of the Grutas de Cristal cave every day for just over five years (March 2010-May 2015. The first 2.5 years rainfall data from that survey was previously published (Moreno et al., 2014;Pérez-Mejías et al., 2018). In Borrastre, rainfall was manually collected daily using a Hellman rain gauge daily from April 2011 to May 2016 (380 events).
In Barcelona, rainfall samples were obtained from the weather station on the roof of the School of Physics of the University of Barcelona using a standard rain gauge. 170

Analytical methods
The isotopic composition of oxygen and hydrogen in rainfall samples, expressed as  18 O and  2 H, reported in ‰ relative to Vienna Standard Mean Ocean Water (VSMOW). Molinos, Borrastre and most of Ortigosa de Cameros samples (143 samples) were analysed using a Finningan Delta Plus XL mass spectrometer at the IACT-CSIC in Granada. Water samples were equilibrated with CO 2 for the analysis of  18 O values (Epstein and Mayeda, 1953), while the hydrogen isotopic ratios 175 were measured on H 2 produced by the reaction of 10 μL of water with metallic zinc at 500°C, following the analytical method of Coleman et al. (1982). The analytical error for  18 O and  2 H was ±0.1 and ±1 ‰, respectively. The Mallorca and Barcelona samples and the remaining samples from Ortigosa de Cameros (50 samples) were analysed at the Scientific and Technological Centre from the University of Barcelona,  2 H via TCEA pyrolysis coupled to Thermo Delta Plus XP mass spectrometer and  18 O with a MAT 253 Thermofisher spectrometer coupled with a gas bench. The analytical error for  18 O 180 and  2 H was ±0.2 and ±1.5 ‰. El Pindal samples were measured at three different laboratories (see Stoll et al., 2015, for more details). Rainfall collected from November 2006 through the end of February 2007 was analysed at the University of Barcelona using the procedure described above. Rainfall collected from June 2007 to May 2008 was analysed in the Marine Biological Laboratories of the University of Oviedo, using equilibration with CO 2 on GV Multiflow-Bio unit coupled to a GV ISOPRIME CF mass spectrometer. Rainfall collected from June 2008 to April 2009 and samples from 2015 were 185 analysed using equilibration with CO 2 on Gas Prep unit coupled to a Nu Instruments Horizon mass spectrometer at the University of Oviedo. Uncertainties are ±0.1‰ (1s) for  18 O and ±1 %.for  2 H, based on replicate analyses. Unfortunately, no comparison was made between the different involved laboratories and thus the study does not account for possible offsets between them.
Additionally, 18 samples of potentially evaporated water with abnormally high values in  18 O p -and that occurred in summer 190 months when maximum daily air temperatures exceed 30°C -were classified as outliers and removed from the database.

Meteorological data 195
Meteorological data to investigate the statistical relationship between isotopic values (at daily and monthly time scales) and main climate variables (air temperature and precipitation) were obtained from the closest meteorological stations over the sampling periods as indicated in Table 1. For Oviedo, meteorological data are obtained from Oviedo AEMET station and for El Pindal (120 km from Oviedo; 70 km from Santander) since there was not good data from nearby stations, we decided to 200 use ERA-Interim re-analysis of the European Center for Medium-range Weather Forecasts (ECMWF) that provides gridded weather data (Berrisford et al., 2009). For Ortigosa site, meteorological data are obtained from Villoslada de Cameros meteorological station (http://www.larioja.org/ emergencias-112/es/meteorologia), at 6.5 km from the rainfall collection site.
The Borrastre sampling site has its own meteorological station (http://borrastre.dyndns.org/MeteoBorrastre) (Table 1), except for the first 22 events that were derived from ERA-Interim since the station was not yet operative. Finally, for 205 Mallorca we used data from Sant Llorenç station (8 km) while Barcelona meteorological data are obtained from Zona Universitaria station (www.meteo4u.com).

210
Prior to conducting correlation analysis at daily scale, we removed the seasonal component of the variables by subtracting their monthly averages to avoid sympathetic seasonal correlations (e.g. (Kawale et al., 2011;Rozanski et al., 1993) (Table   3A). To establish correlations at monthly scale with meteorological variables (Table 3B),  18 O p monthly averages weighted by the amount of precipitation were calculated using the following formula ( Figure 4B): with Q = rainfall quantities for the day i (in mm). Daily values were not averaged since there was only one rainfall sample per day resulting from the homogenization of all the event samples of that day. Spearman's rank correlation analysis, a nonparametric alternative to Pearson correlation analysis was preferred to account for non-linear relation, with r as the correlation coefficient (PAST software, Hammer et al, 2001) (Table 3).The analyses were conducted at daily (Table 3A) and monthly (Table 3B) time scales. Bonferroni test was applied to prevent data from incorrectly appearing to be statistically 220 significant by making an adjustment during comparison testing. Additionally, to integrate the temperature effect and the amount effect, a multiple regression model for  18 O was carried out using PAST software for every studied site (Table 3C). all the 120 vectors were produced, they were averaged, and one unique vector was assigned to each rainfall event. After that, all the averaged vectors associated with each different location studied, are presented in a compass rose using 10° intervals, together with  18 O values and rainfall amount of each event (mm) provided by the closest weather stations to each analyzed location ( Figure 5). 230 Lastly, to better explore the role of the type of precipitation in controlling the isotopic composition of rainfall across northern Iberia, we applied a disaggregation procedure of precipitation series following the same methodology described in Millan et al. (2005). This novel method classifies each precipitation event on the basis of its characteristics, distinguishing between three categories ( Figure 1B, Table 4): (i) frontal systems associated with passing cold fronts from the west, (ii) convectiveorographic storms driven by differential heating, sea breezes and local winds (Azorin- Molina et al., 2009) and (iii) easterly 235 advection from the Mediterranean Sea (backdoor cold fronts). The Kruskal-Wallis H test (sometimes also called the "oneway ANOVA on ranks") is a rank-based nonparametric test (Hammer et al., 2001) that was applied to the three rainfall categories to determine if there were statistically significant differences on their  18 O p distributions (Table 5).

Daily scale rainfall isotopic variability in northern Iberia
The rainfall samples for the studied stations at a daily scale, define local meteoric water lines (LMWL) that are roughly parallel for all sites with similar offset from the Global Meteoric Water Line (GMWL, δ 2 H = 8*δ 18 O + 10) ( Figure 2). All 245 the slopes and the intercepts are lower than the GMWL, with slopes ranging from 6.9 to 7.2 and intercepts from 1.05 to 6.4  (Table S1). At the three sites, this period was among the rainiest of our record with some of the lowest  18 O p values recorded.

260
Seasonality in  18 O p in northern Iberia is further explored in Figure 4B (data in Table 2, Table S2). Al stations exhibit a clear seasonal pattern in temperature with a peak in July/August and minimum values in December/January. The seasonal signal in  18 O p roughly follows this pattern with peak  18 O p values in summer and minimum  18 O p in winter. It is worth noting that precipitation exhibits a bi-modal pattern which is not reflected in  18 O p . The average seasonal differences between July-August and minimum  18 O p in January-February are quite large: 5.8‰ at Borrastre, 4.6‰ at Ortigosa de Cameros, 6.2‰ at 265 Molinos and about 4‰ at Mallorca-Barcelona. Interestingly, the Oviedo-El Pindal samples reveal a very different pattern, with a marked reduction in seasonality compared to the other sites (2 ‰  18 O p difference between winter and summer) ( Figure 4B).

Discussion 270
This discussion section is focused on analysing the main factors controlling  18 O p in the studied transect in northern Spain at daily and monthly time scales. Sect. 5.1 is dedicated to the influence of geographical parameters, such as distance to coast or elevation of the studied sites. Sect. 5.2 deals with the role of meteorological parameters, in particular, local air temperature and precipitation amount. Sect. 5.3 investigates the role of moisture origin on  18 O p variability while Sect. 5.4 explores the 275 role of rainfall type (convective, frontal) in determining  18 O p .

Geographical controls on rainfall isotopic variability
The combination of the various isotope effects results in consistent and spatially coherent variation in  18 O p values that are 280 primarily related to latitude, elevation, moisture source and air masses history (Rozanski et al., 1993;Bowen, 2008). The LMWLs determined with daily data for each of the studied sites reveal a broadly similar regional signal and are consistent with previous studies using GNIP data from southern France (Genty et al., 2014), even considering that study is made with monthly  18 O p data. The slopes obtained are slightly lower in our study compared to a previous analysis from the IP  Schmidt (2006) in their global study of oxygen isotopic composition in seawater, is not what we expected. This pattern may be explained by two processes.
First, the fact that Oviedo and El Pindal rainfall samples show enriched  18 O p values (Table 2, Figure 4B) is consistent with 295 their location in the Cantabrian coast, very close to the Atlantic Ocean, with oceanic climatological conditions characterized by high mean temperatures (Table 1). Thus, Oviedo (and El Pindal) are the stations that receive the first precipitation produced by Atlantic air masses; therefore they are the stations in the transect least affected by the "continental effect": when clouds move inland from the Atlantic Ocean and become gradually isotopically depleted due to progressive rainout (Dansgaard, 1964). Thus, as we follow the typical movement of an Atlantic front, from west to east, we find progressively 300 more negative winter  18 O p values (considering an average of January-February-March, Table 2) going from El Pindal (-6.0‰) to Ortigosa de Cameros (-8.1‰), to Borrastre (-9.8‰) and, finally, to Molinos (-10.0‰). This pattern is not so evident in other seasons where the entrance of Atlantic fronts is not the only synoptic pattern that generates rainfall in the transect. However, the large observed differences cannot be explained only by this effect that accounts for a very small variation (about 0.002‰ per km in Europe as described in Rozanski et al., 1993).  Schmidt, 2006). An additional effect, probably more important than 0.5 -1 ‰ of difference, is the higher annual mean air temperature at those stations (Mallorca and Barcelona together 310 with Oviedo and El Pindal) compare to the other ones (Table 1). The effect of temperature to produce the less negative  18 O p values recorded will be explained below (Sect. 5.2).
In addition, the three stations with more negative monthly  18 O p values (Ortigosa de Cameros, Borrastre and Molinos) are at higher elevation than the other stations that are all located close to sea-level. Therefore, the "elevation effect" (Siegenthaler and Oeschger, 1980) likely also plays a role in explaining the more negative  18 O p values at those stations. Considering the 315  18 O p annual averages (Table 2)  per 100 m of elevation was observed (Ambach et al., 1968;Siegenthaler and Oeschger, 1980). However, in spite the difference in elevation, we need to consider that the sites are very distant and separated by the Mediterranean Sea. Therefore, 320 the altitude cannot be the only parameter controlling the differences between the studied sites.
Finally, the geographical factors reviewed in this section (distance to the coast or continental effect, elevation effect, and  18 O composition of the sea waters) exert a small direct influence on the observed spatial distribution of rainfall  18 O p at the studied sites but contribute to the effects of other, controlling factors: air temperature, rainfall amount, air mass trajectory and rainfall type, which will be described in following Sect. 5.2, 5.3 and 5.4. 325 https://doi.org/10.5194/acp-2020-861 Preprint. Discussion started: 8 September 2020 c Author(s) 2020. CC BY 4.0 License.

The influence of air temperature and rain amount on the spatial distribution of rainfall  18 O p values at daily and monthly time scales
Spearman's rank correlation analysis (Table 3) reveals that  18 O p does not correlate with air temperature or amount of 330 precipitation in a similar way at each station, neither at daily or monthly scales, thus reinforcing the need for conducting such studies on a local basis particularly when conducting paleoclimate reconstructions (Leng, 2006). Air temperature appears as the most robust influence across the west-to-east transect, with low but statistically significant correlations (daily scale) with  18 O p at all sites (red numbers in Table 3A) (Table 3A). Regarding monthly values, air temperature is significantly correlated with  18 O p values at eastern stations, with the highest coefficients associated with higher altitude sites (e.g., in Molinos with r s = 0.76 and p=3.36E-10 or in Borrastre with r s = 0.61 and p=1.44E-05) (Table 3B). 340 The dependence of  18 O p on air temperature has been extensively studied, yielding an average slope for mid-latitude continental stations of 0.58‰/°C (Rozanski et al., 1993). However, that value is highly variable in time and space. The  (Table 3A). Besides, there is a 350 significant correlation at the two sites of the Iberian Range (r s = -0.32; p=1.05E-05 in Ortigosa and r s = -0.19; p=0.005 in Molinos). Interestingly, the westernmost stations (El Pindal and Oviedo) do not show a significant  18 O p -precipitation correlation at the daily or monthly scale. This lack of correlation in the Atlantic sites (El Pindal and Oviedo) contrasts with a previous study carried out in northern Spain and also characterized by an Atlantic climate (Matienzo depression) where there is found a significant  18 O p -precipitation monthly correlation (r = -0.51; p < 0.01) (Smith et al., 2016). In our study, the 355  18 O p -precipitation correlation at monthly scale is only significant in Molinos, in the Iberian Range (r s = -0.4; p=0.018) while no correlation is observed in the other sites (Table 3B).
To further assess the relative role of temperature and amount effects, a multiple regression model for  18 O p was carried out for the seven studied sites in which the temperature effect exerted a clear dominant control (Table 3C. Still, both influences together account for less than 20 % of the variability of  18 O p in the study transect. Since the origin of rainfall and type of 360 rainfall (i.e., convective vs. frontal) is also spatially dependent in northern Iberia, these variables and their influence on the observed  18 O p variability are investigated in Sect. 5.3 and 5.4 below.

365
The source effect describes how air masses derived from different moisture sources have distinct  18 O p values (e.g., Friedman, 2002). The source effect results from varying air mass histories, different conditions of the moisture source (temperature, relative humidity and wind speed) and regional differences in the  18 O of the surface ocean (LeGrande and Schmidt, 2006). In the case of northern IP, it is necessary to consider the effect of both the Atlantic Ocean and Mediterranean Sea as important sources of atmospheric moisture (Gimeno et al., 2010) whose relative influence on regional IP  18 O p could 370 be very different because of the complex regional topography of the area. General  18 O values of seawater reconstructions (LeGrande and Schmidt, 2006) indicate different values for the Atlantic Ocean and the Mediterranean Sea due to temperature and salinity differences. Source  18 O values range from 1 to 1.5‰ in the subtropical Atlantic to 2‰ in the Mediterranean (Schmidt et al., 1999). Although these differences (about 0.5 -1 ‰) are small, since they are further modulated by the air mass history, we expect to see a change in the relative influence of moisture source on  18 O p along the 375 west-to-east transect.
Evaluation of monthly  18 O p patterns represented in Figure 4B  To evaluate the role of moisture source in determining  18 O p values at a daily scale in northern Iberia and Balearic islands, back trajectories were calculated for all the rainy days and subsequently averaged into wind rose diagrams, following 390 representation applied in previous studies (Smith et al., 2016), for three stations along our northern Iberia transect: Oviedo and Mallorca, the two extreme locations of the studied transect, and Borrastre, situated at an intermediate location, representing a total number of 519 events ( Figure 5). To facilitate statistical comparison of the mean trajectory paths and moisture uptake regions with the oxygen isotope signature of sampled rain events, the vector angle between every site (Oviedo-Borrastre-Mallorca) and each hourly position along 120-h back trajectories (at 700 and 850 hPa) for each event was 395 estimated, following the methodology presented in Baldini et al. (2010) (Figure 5). Once all the vectors were produced for each sampled event, they were averaged, and presented in a compass rose using 10° intervals, together with  18 O p values and rainfall amount of each daily sample (mm) provided by weather stations closed to each location analyzed ( Figure 5). This analysis reveals the dominance of western trajectories in the three studied sites, with very few episodes associated with other directions (Figure 5). Only some episodes from SW (e.g., Borrastre) or SE (e.g., Mallorca) trajectories are found and, 400 interestingly, they have distinct  18 O p value (see below). This low, almost negligible, presence of trajectories associated with Mediterranean air mass advections, does not inhibit the possibility of a moisture uptake over the Mediterranean or moisture recycling with altitude in the mountain region surrounding Borrastre since meteorological processes connected to convection (e.g., orographic, dynamic, thermal) can produce moisture uptake in less than 6h (Romero et al., 2000(Romero et al., , 1997Tudurí and Ramis, 1997) and may not be well-captured in the back trajectory analyses, which are computed for the previous 120 hours 405 (see Methods). Therefore, convection processes, that may be associated with easterly trajectories, are under-represented in this methodology (see 24 hours analyses in Figure S1 where more trajectories with different origin appear more frequently). Therefore, it is important to note here that this method provides information on the air mass origin (source effect) but not in the moisture uptake regions. In that way, it is clear the dominant WNW trajectory for the three studied stations.
Despite the three sites sharing a common dominant WNW trajectory, they behave quite differently in terms of the associated 410 amount of rainfall and  18 O p values. Oviedo (with a temperate oceanic climate -Cfb, Table 1) presents a narrower range of rainfall amounts and  18 O p values than at the other two sites, as shown in Figure 5A by the negligible frequency of rainfall amounts above 32 mm (orange) or below 2 mm (purple), while "extreme" events are much more common in Borrastre or northwesterly trajectory associated with more negative  18 O p values and southwesterly trajectory associated with less negative values ( Figure 5B). Borrastre station is chosen to further evaluate back trajectories for all rainfall events over one whole year (2014, n=126 rainfall events) since the presence of rainfall events where moisture comes from the SW, with usually less negative  18 O p values, is significant compared to, for example, Oviedo station. Thus, one example from every 425 trajectory is presented in Figure 6.
Above 80% of winter trajectories recorded in Borrastre rainfall events originate in the North Atlantic, Artic or inland USA or Canada. They cross the Atlantic Ocean north of Madeira Island and usually enter the IP by the west, next to the Galicia and Portugal border. Those trajectories arriving from the N-NW reach Borrastre site at the Pyrenees almost without crossing the IP, thus providing the more negative  18 O p values (e.g., 7 th February, Figure 6A, with  18 O p =-6.5‰). On the contrary, those 430 arriving from the W-SW enter via Lisbon and cross central IP providing less negative  18 O p values (e.g., 16 th January, Figure   6B, with  18 O p =-1.2‰). If the trajectory of the air mass travels larger distances over the continent, the contribution of reevaporated land moisture to the water vapour travelling inland may be significant and thus  18 O p values may appear higher, as it has been shown to occur in other regions (Krklec and Domínguez-Villar, 2014). In that case, the progressive rainout effect may be compensated by the moisture uptake of evaporated (high  18 O) surface water. 435 During spring, the typical situation of air masses entering from the W alternates with those arriving from the SW, entering at the latitude of the Cape San Vicente and crossing the IP from south to north (e.g., 20 th April, Figure 6C; with  18 O p =-2.1‰).
Some spring trajectories are subject to Mediterranean influence (eg. 20 th May; Figure 6D) and are characterized by higher  18 O p values ( 18 O p =-1.3‰). In general, the penetration of subtropical Atlantic air masses, which becomes a very common situation in summer, results in higher  18 O p values (e.g., 6 th July, Figure 6E, with  18 O p =-2.2‰). Therefore, the less negative 440  18 O p values usually associated with SW trajectories in Borrastre can be explained by (1) the origin in the subtropical Atlantic Ocean with higher  18 O p values (1.5 ‰) compared to North Atlantic (0.5 ‰) (LeGrande and Schmidt, 2006) and, (2) the recycling of surface moisture over land incorporating enriched  18 O p values from surface waters that have been subject to evaporation over time (Krklec and Domínguez-Villar, 2014).

The influence of rainfall type on isotopes.
The influence of rainfall type on the  18 O p is well documented globally, with different  18 O p observed depending the type of precipitation (convective showers, frontal, continuous stratiform precipitation, etc.) (Aggarwal et al., 2012). This relationship is observed in previous studies both at daily or monthly timescales (Aggarwal et al., 2016), with few examples in the 450 Equatorial Indian Ocean (Gat, 1996) and California, USA (Coplen et al., 2015), both indicating that  18 O p values were lower when precipitation was dominantly stratiform and higher when it was mostly convective. The main reason to explain this difference lies on the processes of condensation and riming associated with boundary layer moisture which produced higher https://doi.org/10.5194/acp-2020-861 Preprint. Discussion started: 8 September 2020 c Author(s) 2020. CC BY 4.0 License.
isotope ratios in convective rain (Aggarwal et al., 2016). Some studies in the Mediterranean region (Celle-Jeanton et al., 2001) also directly link the isotopic signature of the precipitation to the prevailing weather conditions during the rainfall 455 event.
Here we explore how the specific synoptic situation, i.e., rainfall types or rainfall components, influence  18 O p values across the studied transect. Table 4 shows the percentage of rain events associated with each type of precipitation, that were previously defined following (Millán et al., 2005) and represented in Figure 1B: (i) Atlantic frontal systems (westerly winds), (ii) convective-orographic storms, and (iii) backdoor cold fronts from the Mediterranean Sea (easterly winds). Backdoor 460 cold fronts from the Mediterranean Sea are sporadic events occurring in autumn (secondarily in winter-spring), but they cause heavy precipitation and flooding (Llasat et al., 2007).
The prominence of rainfall associated with Atlantic fronts is evident (above 40% in the seven studied stations North Atlantic, and their movement is associated with westerly frontal systems (Smith et al., 2016). This situation appears to be true along the studied transect, however for the Mediterranean and Iberian Range sites, Atlantic and Mediterranean sources are balanced (including backdoor cold fronts as Mediterranean) (Table 4). Distance to the Mediterranean and elevation are important factors in determining the frequency of rainfall associated with backdoor cold fronts. Thus, backdoor cold fronts are associated with 38.78% of Mallorca rain events and are still frequent situations at the two sites from the 470 Iberian Range (20.6% in Ortigosa de Cameros and 23.9% in Molinos). The frequency of convective precipitation is higher at the three mountain sites (20.6% in Ortigosa de Cameros, 24.3% in Molinos and 23% in Borrastre), compared to those sites at lower elevation (17% in Oviedo;11.9% in El Pindal,17% in Barcelona and 20.4% in Mallorca).
The Kruskal-Wallis test was applied to investigate if there were significant differences in the  18 O p values of the three rainfall types analysed (Atlantic, backdoor frontal precipitation, and convective) in the studied stations at the daily scale. 475 Test values shown in Table 5 (Aggarwal et al., 2016). The highest  18 O p values associated with convective precipitation may relate to the critical role played by the re-evaporation of droplets, a circumstance that usually takes place during convective rainfall (Bony et al., 2008). In any case, what is relevant here, is the similarity among  18 O p values of the two types of frontal rains (Atlantic fronts and Mediterranean backdoor cold fronts) 485 while there is a difference considering the type of precipitation.
Besides  18 O p values associated with the three rainfall types, variations of air temperature and precipitation have an effect in separating the three rainfall types (Figure 7). Regarding air temperature, backdoor cold front events are the ones occurring with colder temperatures while convective rains are more associated with the warm season. Thus, air temperature (and its variation along a vertical profile) is another variable clearly associated with the type of rainfall, with higher temperature 490 during convective rains and lower for the Atlantic and backdoor types. This is a clear reflection of the seasonal pattern of convective rains, which are more abundant in summer months (Table S1) thus preventing an isolation of the effect of the type of rainfall which appears mixed with the temperature effect. In contrast, the high number of outliers in the box plots of the amount of precipitation when organized by rainfall type (Figure 7) indicates that this parameter is determined more by local factors (e.g.. topography) than by the specific synoptic situation. 495

Conclusion
The major findings in this study are summarized as follows:  The analysis of  18 O p and  2 H p at seven stations along a west-to-east transect in northern Iberia and Balearic Islands 500 yields similar LMWLs but all with lower slope and intercept values than the GMWL. In conclusion, the northern Iberian region, is under the influence of two climatic regimes (Atlantic and Mediterranean) and 520 affected by different moisture sources. Therefore, synoptic-scale atmospheric circulation is playing a key role in determining the ranges, values and seasonal distribution of  18 O p variability. Future detailed studies focusing on particular events that can be traced along the whole west-to east transect will be conducted to further understand the air masses trajectories over northern Spain and their influence on  18 O p variability.

Data availability
All data are included in the Supplementary Tables S1 and S2. and the European Centre for Medium-Range Weather Forecasts for the ERA-Interim dataset. IC also thanks the ICREA Academia program from the Generalitat de Catalunya. We dedicate this study to our colleague Carlos Sancho who intensively worked to produce this large  18 O rainfall dataset for northern Iberia.

800
As supplementary Table S1. Event  18 O p and  2 H p data for the stations considered in this study. Meteorological data from nearby stations (Table 1) are also included.