Trends in daily rainfall in the Iberian Peninsula from 1951 to 2002

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 27: (27) Published online 9 October 26 in Wiley InterScience ( DOI: 1.12/joc.149 Trends in daily rainfall in the Iberian Peninsula from 191 to 22 F. S. Rodrigo a, * and Ricardo M. Trigo b,c a Department of Applied Physics, University of Almería, La Cañada de San Urbano, s/n, 412 Almería, Spain b Centro de Geofísica da Universidades de Lisboa, Departamento de Física, Faculdade de Ciências, Universidade de Lisboa, Campo Grande, Ed C, Piso 6, , Lisboa, Portugal c Universidade Lusófona, Departamento de Engenharias, Lisbon, Portugal Abstract: An analysis of 22 sites of daily precipitation records over the period for the Iberian Peninsula is presented. Annual and seasonal total precipitation (P ), number of wet days (N), precipitation intensity (I), the 9th percentile (P 9), and percentage of rain falling on days with rainfall above the 9th percentile (%) are investigated. The annual and seasonal trends for these variables and for all 22 rain gauges are analysed, using the Mann Kendall statistic, and a linear regression model. Moreover, a t-test is applied to the difference between the means of two subperiods, respectively and Principal results indicate a decreasing trend in P, I, andp 9 for several northern and southern stations in winter; P, I, andp 9 in some southern stations in spring, I and P 9 in some southern stations in summer, and I in some northern and southern stations in autumn. The general behaviour is a decrease in the daily intensity of rainfall, while the number of wet days does not reveal pronounced changes. This pattern is valid for both annual and seasonal values of the indices. The decreasing trend found for I in winter and annual series for some localities may be related to the predominance of the positive phase of the North Atlantic Oscillation (NAO), but it is necessary to find other mechanisms for those stations and seasons not linked directly to NAO. Copyright 26 Royal Meteorological Society KEY WORDS Iberian Peninsula; daily rainfall; Mann Kendall test; linear regression; NAO Received 2 October 2; Revised 29 June 26; Accepted 11 July INTRODUCTION Variations in total precipitation can be caused by a change in the frequency of precipitation events, or the intensity of precipitation per event, or a combination of both. In order to provide a better understanding of precipitation behaviour as an indicator of climate changes, daily precipitation series must be analysed (Jones et al., 1999; Karl et al., 1999; Brunetti et al., 21). It is essential to perform the analysis of the following variables: total amounts of precipitation, number of wet days, or intensity index, defined as total precipitation divided by number of days with precipitation. These indices are insufficient, however, to describe the nature of precipitation changes. It is important to know whether the change in precipitation frequency is due to a change in the number of days with heavy precipitation or with light precipitation. Among the indices recommended by the CLIVAR/GCOS/WMO workshop on indices and indicators for climate extremes (Karl et al., 1999), the variations in magnitude of 9th percentile and the percentage of annual/seasonal precipitation falling on * Correspondence to: F. S. Rodrigo, Department of Applied Physics, University of Almería, La Cañada de San Urbano, s/n, 412 Almería, Spain. frodrigo@ual.es days with rainfall above 9th percentile are proposed as high-priority indicators (Nicholls and Murray, 1999). The statistical analysis of daily data in the Iberian Peninsula is very interesting from a climatological point of view, particularly because monthly and annual values may conceal highly different precipitation regimes on the daily scale. The occurrence, or not, of one of this high daily amounts can change the character (dry or rainy) of any given month, season, or year, principally for the southern and eastern sectors of Iberia. This leads to considerable uncertainty in the average pluviometric contributions, which, in turn, has environmental and social repercussions (Martín-Vide, 24). However, the importance of daily precipitation has not been matched by sufficient scientific attention. Most papers recently published, dealing with the daily precipitation over the Iberian Peninsula, are limited to certain areas or regions within the country, for instance, the analyses of rainfall spatial variability by De Luis et al. (2) and that of daily rainfall trend in the region of Valencia (eastern Spain) by González-Hidalgo et al. (23); Lana et al. (23, 2) studied pluviometric indices at the Fabra Observatory (Barcelona, NE Spain) and Lana et al. (24) analysed the spatial and temporal variability of the daily rainfall regime in Catalonia (northeastern Copyright 26 Royal Meteorological Society

2 14 F. S. RODRIGO AND R. M. TRIGO Spain); Vicente-Serrano and Beguería-Portugués (23) studied the extreme dry-spell in the middle Ebro Valley (northeastern Spain), and Trigo et al. (2) studied the influence of the North Atlantic Oscillation (NAO) on landslide episodes near Lisbon using daily data. The analysis of daily precipitation for the whole of Iberia has received scarce attention, probably due to the lack of high-quality daily data. Martín-Vide (24) analyses a concentration index that evaluates the contribution of the days of greatest rainfall to the total amount in 32 meteorological stations across peninsular Spain. Goodess and Jones (22) produced the most comprehensive work linking atmospheric circulation patterns and precipitation for the entire Iberian Peninsula on a daily scale. Paredes et al. (26) have used a relatively similar procedure to characterise the strong decrease of precipitation in March. In this paper, trends in mean seasonal rainfall, wet-day amount, and number of rainy days were calculated using least-squares linear regression, and trends in the contribution to total rainfall from 1% rainfall amount quantiles from 198 to 1998 for 18 Iberian stations (1 Spanish stations and 3 stations in Portugal) were explored. This work corresponds to the first step of a research program aiming at the analysis of daily climate variables for the whole Iberian Peninsula. The main objective is to analyse the behaviour of daily rainfall in Spain and Portugal using the methodology proposed in the CLIVAR/GCOS/WMO workshop on indices and indicators for climate extremes (Karl et al., 1999). Among the indices proposed to be calculated from daily data, in this work we analyse the total seasonal and annual amount of precipitation (P ), number of wet days per season (N), daily intensity index (I), variations in magnitude of 9th percentile (P 9), and percentage of seasonal precipitation falling on days with rainfall above P 9 (%). Unlike previous studies, in this work the 9th percentile is determined by fitting an appropriate gamma distribution to the data. Although thresholds can be estimated simply by counting the observed data, we have fitted a distribution for the following reasons: (1) estimates of extreme quantiles are more accurate, (2) distribution fitting approach copes better with missing values, (3) more sophisticated analyses are possible afterwards, and (4) there are techniques available for checking whether the distribution provides a good fit to the data (Nicholls and Murray, 1999). This analysis is made for 1 meteorological stations in peninsular Spain and 7 in Portugal from 191 to 22. Although the network of stations is coarse, it is, nevertheless, broadly representative of the range of climate regimes experienced across the Peninsula, including stations located on the north coast, the Mediterranean coast, and in the central, western, and southern areas of the country. The study of daily rainfall trends is performed using the non-parametric Mann Kendall test, least-squares linear regression, and comparisons between two sub-periods within the total time period between 191 and 22. Because of the high degree of seasonality of the Iberian Peninsula rainfall, these analyses are made for the entire year, and also on the seasonal scale. It should be stressed that winter values refer to December January February, spring to March April May, summer to June July August, and autumn to September October November. The paper is organised as follows: Section 2 describes the database and the methodology used in this study, Section 3 shows the results obtained with particular emphasis on those that are statistically significant, and, finally, in Section 4, the main results are discussed, and perspectives of future work are presented. 2. DATA AND METHODS The database used in this study comprises daily amounts of rainfall for 1 Spanish localities, covering the Iberian Peninsula area, and 7 Portuguese stations, for the period Figure 1 shows the locations of the rain gauges used here, while Table I provides the geographical data of the meteorological stations. It should be emphasised that both Portuguese and Spanish stations were selected from a larger set of rain gauges supplied by the Spanish Institute of Meteorology (INM) and the Portuguese Water Institute (INAG), on the basis of record length, completeness, and quality. The purpose is to provide a reasonable spatial coverage over much of the Iberian Peninsula ranging in longitude from 9 1 W(Lisbon) to 2 1 E (Barcelona), and in latitude from N (La Coruña) to N(Málaga). Most of the stations have not changed their position, but the metadata relative to methods and instruments is known only for a few. Table I. Daily rainfall data series in the Iberian Peninsula for the period Station (CODE) Altitude (m a.s.l.) Latitude Longitude Alicante (A) N 29 W Barcelona (B) N 2 1 E Bilbao (BI) N 2 W Burgos (BU) N 3 38 W Granada (GR) N 3 37 W La Coruña (LC) N 8 2 W León (LE) N 39 W Madrid (M) N 3 4 W Málaga (MA) N 4 28 W Murcia (MU) N 1 14 W Salamanca (SA) N 29 W Sevilla (SE) N 3 W Soria (SO) N 2 28 W Valencia (V) N 22 W Zaragoza (Z) N 1 W Lisboa (L) N 9 1 W Pinhel (PI) N 7 W Serpa (SER) N 7 37 W Gafanh (GA) N 8 42 W Monfor (MO) N 7 26 W Reliqu (RE) N 8 29 W Grando (G) N 8 34 W Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

3 DAILY RAINFALL IN THE IBERIAN PENINSULA 1 Figure 1. Map of the study area. Rain gauges are identified by the code indicated in Table I. Almarza et al. (1996) analysed the homogeneity of the Spanish series and concluded that these series are of high quality and do not present any heterogeneity problems. In the case of the Portuguese stations, the responsible institution (INAG) classifies the time series of its own rain gauges into three distinct classes (poor, fair, very good), and all the seven Portuguese stations used here fall into the third category. Twelve rain gauge stations (nine in Spain and three in Portugal) have no missing data, while for the remaining rain gauges missing data corresponds to less than 1.3% for each season and station. When data from neighbouring stations were not available to develop estimates using linear regressions, gaps in daily variables shorter than 3 days of duration were filled by simple linear interpolation using values available prior and after the missing data (otherwise the seasonal value was treated as missing). In a first step, for each station and season, the following variables were calculated: Total amount of rainfall (P ) Number of rainy days (N) with a minimum quantifiable rainfall of.1 mm (the resolution of the pluviometer). Daily intensity index (I) defined as total seasonal precipitation divided by number of days with precipitation. It is widely recognized that the distribution of daily precipitation totals can be approximated by the gammadistribution (Groisman et al., 1999) represented by Ɣ(x; α, β) = xα 1 e x/β β α Ɣ(α) (1) where x> and Ɣ(x; α, β) = when x. The α parameter defines the shape of the distribution, while the β parameter characterizes the scale of intensity of the daily precipitation. Ɣ(α) is the complete Gamma function. The preferred fitting method is maximum likelihood, because of its power and generality (Nicholls and Murray, 1999). In this method, we start by computing the sample statistic A A = ln x ln x (2) Then, the estimations of the shape and scale parameters can be estimated with ˆα = 3 A ˆβ = xˆα (3) 4A The distribution function of precipitation totals P(X x) is expressed as P(X x) = x Ɣ(x; α, β)dx (4) The distribution functions were compiled only for seasons composed of months with less than 1 days of missing data, according to common practice (Wettstein and Mearns, 22). This threshold was applied only on.8% occasions in winter and.1% in both spring and summer. The missing seasons were summer 197 in Barcelona, autumn 2 and winter 21 in Bilbao, winter 1997 in La Coruña, winter 196 in León, winter 1991 in Sevilla, and winter 1991 in Soria. The following step was taken to calculate the 9th percentile (P 9) for each season and station separately by fitting days with precipitation to a specific gamma distribution. For each station and season, the mean value of the 9th percentile for the reference period was calculated. Finally, the percentage (%) of seasonal precipitation falling on days with rainfall above this value was estimated. In this way, five variables were obtained to analyse the behaviour of daily rainfall: P,N,I,P9, and %. Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

4 16 F. S. RODRIGO AND R. M. TRIGO Linear trends in indices for climate extremes have been previously investigated for various regions or globally. Time series of these indices may have properties (nonnormal distribution, autocorrelation, presence of outliers) that make them unsuitable for ordinary least-squares trend fitting (Moberg and Jones, 2). Although there is no universally preferred technique, non-parametric methods may be more appropriate in this circumstance. Therefore, indices time-series were subjected to the non-parametric Mann Kendall test to detect any significant trends. In the Mann Kendall test, for each element x i (i = 1,...,n)of the series, the number n i of lower elements x j (x j <x i ) preceding it (j <i) is calculated (Sneyers, 199), and the test statistic t is given by t = i n i () In the absence of any trend (null hypothesis), t is asymptotically normal, and independent from the distribution function of the data. The statistic u(t) is defined as u(t) = t t (6) var(t) with t and var(t) given by t = n(n 1) 4 var(t) = n(n 1)(2n + 7) 72 (7) In the absence of any trend, u(t) has standard normal distribution. The null hypothesis, therefore, is rejected for high values of u(t), being the probability α 1 of rejecting the null hypothesis when it is derived by a standard normal distribution table, α 1 = P( u > u(t) ) (8) The sign of the statistic u(t) indicates if the trend is positive (+) or negative( ). The linear model was used as a first approach to analyse the changing rate in the cases where the Mann Kendall test detected a significant trend. The slopes of the trends were calculated by least-squares linear fitting. The adjustment of a model to a series of observations by least squares generally involves the underlying hypothesis that the residual given by the difference between the observed and estimated values corresponds to a simple random variable with a normal (or Gaussian) distribution (Sneyers, 1992). Therefore, the residual was tested by means of two non-parametric tests of randomness, runs up and down, and runs above and below the median. In addition, the normal character of the residual was tested using the Kolmogorov Smirnov test, and the residual was tested for the null hypothesis of null mean. Only when the correlation coefficient was significant at the 9% confidence level and the null hypothesis of the residual (randomness, normality, null mean) was verified, the linear model was accepted. Finally, the total period was divided into two halves, and , and the indices corresponding to each sub-period were compared by means of a t-test (9% confidence level). The first half coincides with an episode with little temperature change over Europe, while the second one corresponds to an episode of pronounced warming in Europe on average (Klein-Tank et al., 2). The purpose is to explore if these changes detected in the temperature series are also found in the rainfall series in the Iberian Peninsula. For annual values, a similar methodology was adopted. 3. ANALYSIS AND RESULTS 3.1. Winter Table II shows the average value of the total amount of winter rainfall in the 22 rain gauges for the reference period Maximum values correspond to the north-western coastal stations (LC, BI, GA, L), with rainfall above 3 mm, meanwhile minimum values correspond to east and south-eastern stations (A, MU, Z), with values below 1 mm. In relation to the average number of wet days, the northern stations have the highest values (above 4 days), while the Mediterranean stations (B, V, A, MU) show the lowest number of wet days (below 2 days). The average daily intensity shows maximum values in south-west sites (SE, MA, G, RE), above 8 mm/day, with the northern inland stations Table II. Mean values of: total amount rainfall P (mm); number of rainy days N (days); daily rainfall intensity I (mm/day); 9th percentile P 9 (mm); and percentage of rainfall falling in days with rainfall above P 9 (%), corresponding to winter, reference period Station P (mm) N I(mm/day) P 9(mm) % A B BI BU LC GR LE M MA MU SA SE SO V Z GA G L MO PI RE SER Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

5 DAILY RAINFALL IN THE IBERIAN PENINSULA 17 Table III. Mann Kendall statistic for total precipitation (P ), number of wet days (N), daily intensity (I), 9th percentile (P 9), and percentage of rainfall falling in days with rainfall above P 9 (%) for winter from 191 to 22. s = slope of the linear regression ± standard error; d = difference between the averages for and (s and d only are indicated when Mann Kendall test is significant). Bold numbers: when confidence level is higher than 9%. Station P N I P9 % A B BI s= 4 ± 1 s=.8 ±.2 s=.14 ±. d= 79 d= 1.9 d= 3.6 BU GR s=.3 ±.1 d= 1. LC LE M s =.3 ±.2 d = 6.8 MA MU SA SE SO V Z s=.3 ±.1 d =. GA G L MO PI RE s = 2. ± 1. s=.9 ±.2 s=.24 ±.7 s=.3 ±.1 d = 6. d= 3. d= 7. d= 9.8 SER (Z, SO, SA, BU) presenting the lowest values (below 4 mm/day). The average of the 9th percentiles shows minimum values in the central and western stations, around 1 mm, and values above 2 mm in the northern, south-western and north-eastern stations. The percentage of rainfall falling in days with rain above P 9 reaches values in the range of 2 3% for the vast majority of the stations. These statistics are in agreement with the known behaviour of the Iberian winter rainfall, with three areas well defined: (1) northern coast, (2) central and western area, and (3) the Mediterranean coast (Muñoz- Díaz and Rodrigo, 24a). During winter, the large-scale circulation is mainly driven by the position and intensity of the Icelandic low, and most of the Iberian Peninsula is affected by westerly winds that carry moist air and produce rainfall events (Trigo and Dacamara, 2). This precipitation is intensified by the passage of cold fronts associated with families of transient depressions. In the northern coast, rainfall is mostly due to meridional fluxes (Goodess and Jones, 22) that, associated with the local orography, force an ascent of the air mass and, consequently, produce precipitation. Northern fluxes provoke the high values of P,N,I,andP 9. On the other hand, central and western sectors of Iberia are mainly affected by western fluxes (Trigo and Dacamara, 2), with a clear influence of the NAO (Muñoz- Díaz and Rodrigo, 24b; Trigo et al., 24). In the Mediterranean coast, precipitation is mainly produced by easterly air flows (Romero et al., 1999). This region is sheltered from the intense Atlantic disturbances by the central Spanish plateau and the Pyrenees, and also by higher land flanking the Mediterranean coast (Sumner et al., 21). These conditions explain the low values of P and N in this area. The convective nature of rainfall in this zone explains that P 9 and % have values very similar (or even higher) to those attained by other stations. Table III shows the Mann Kendall statistic, calculated to detect any possible trends in the time series of the five variables analysed. There are significant trends in BI, with decreasing P,I and P 9, in RE with decreasing trends in P,I,P9, and %, in the intensity of GR and Z, and in % Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

6 18 F. S. RODRIGO AND R. M. TRIGO 1 BI, winter 2 RE, winter 8 P = 46-4t r =,-43 2 I = 194 -,9t r = -,46 P (mm) 6 4 I (mm/day) I (mm/day) BI, winter I = 9,93 -,8t r =, P9 (mm) RE, winter y = 31-,24t r = -, Figure 2. Least-squares linear regression for P and I for winter in Bilbao (a) and for I and P 9 in Reliq (b) from 191 to 22. Regression equation and correlation coefficient are included. for M. The decrease in total rainfall of BI and RE seems to be provoked by the decrease in the corresponding daily intensity, with the consequent decrease in the 9th percentile. Despite the lack of significance for the remaining statistics, a slightly different behaviour between the Mediterranean coastal stations and the others is worth mentioning. As a rule, V, A, and MU show a positive trend signal for all the variables, whereas the other stations show a negative sign in general. In relation to extreme events (measured by means of %), only significant negative trends are there in the southern time-series of M and RE. Figure 2 shows the least-squares linear regression for the BI variables P,I. In all the cases, the correlation coefficient is significant at the 9% confidence level and the residuals behave as simple random and normally distributed. The slopes indicate a decreasing trend of 4 mm/1 years for P,.8 (mm/day)/1 years for I, and 1.4 mm/1 years in the value of P 9. The percentage of variance explained by the linear model is 18., 36, and 14.4%, respectively. In order to assess the influence of the large initial value (191) in the previous results, the procedure was repeated without this year. The analysis of the linear trend from 192 to 22 yielded significant decreasing trends in P,I, and P 9 again, with slightly less pronounced values: 3 mm/1 years for P,.7 (mm/day)/1 years for I, and.9 mm/1 years for P 9. Figure 2 also shows the linear regression for the same variables in RE. In this case, the result corresponding to P was not significant, perhaps as a consequence of the slight positive behaviour of N. The decreasing trend in I and P 9 corresponds to a change rate of.9 (mm/day)/1 years, and 2.4 mm/1 years, with percentages of explained variance of 21 and 17%, respectively. In GR and Z, the slopes indicate a significant decreasing trend in I of.3 (mm/day)/1 years (graphs not shown) with a % of variance explained by the linear model of 1. and 9.9%, respectively. In M, the correlation of % with time was non-significant. The comparison of the two sub-periods and for P in BI shows a significant (9% confidence level) difference between the means (397 mm and 318 mm, respectively). The decrease of 79 mm corresponds to 21% of the reference period mean. Also, in BI there is a significant difference in the I and P 9 means, from 8.8 to 6.8 mm/day, and from 27.2 to 23. mm, respectively. In GR, the difference between the means for I of the two sub-periods is also significant, with a decrease from.4 to 4.4 mm/day, and, in Z, where the mean of the total period for I is 3.1 mm/day, the decrease from 3.3 to 2.8 mm/day is not significant, perhaps due to the low value of I in this station. The difference of % in M is not significant, therefore the trend detected in this case must be considered cautiously. In RE, the comparison of the two sub-periods does not yield a significant difference for P, coinciding with the result of linear regression. On the contrary, other variables show significant differences, 3. mm/day for I, 7. mm for P 9 and 9.8 for % Spring Table IV shows the mean value of the five variables for the 22 stations over the reference period The spatial distribution of the P values is very similar to that in winter, with high values in the north-western stations (above 2 mm), and the minimum values (below 1 mm) in the eastern stations. The strong seasonal character of the Iberian precipitation regime implies that spring corresponds to a transition period from winter to summer (when minimum values are obtained for all the stations). Therefore, it is expectable that, in general, Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

7 DAILY RAINFALL IN THE IBERIAN PENINSULA 19 Table IV. As Table II for spring, reference period Station P (mm) N(days) I(mm/day) P 9(mm) % A B BI BU GR LC LE M MA MU SA SE SO V Z GA G L MO PI RE SER spring P values are lower than those obtained for winter. However, the average number of rainy days N is similar in winter and spring. Therefore, we can expect that, in general, the average intensity I and the 9th percentiles will be lower in spring than in winter. Finally, % values are very similar to winter values, with percentages also falling within the 2 3% range. Table V shows the Mann Kendall statistic calculated for the five variables and the 22 sites. It is worth mentioning that there are significant decreasing trends in the intensity I for several northern (BI, BU, LE) and southern (MA, MU, SE, RE) stations. This behaviour reflects the decreasing trends found for P 9 in BI, MA, SE, and RE and for % in MA, MU, and RE. Similar to previous results obtained for winter, the number of rainy days does not show significant trends, except for two Portuguese stations, MO (negative), and RE (positive). The positive sign of the statistic corresponding to N in many stations (although rarely significant) seems to compensate for the decreasing trends of I, resulting in trends that are not significant in P for many stations: only southern stations (MA, SE, MO) show a significant decreasing trend in P. Therefore, the conclusion is that there is a trend to lower amounts of rainfall during the rainy days. Finally, two stations (G and PI) show significant negative trends in extreme values (P 9 and %), but without an impact on the remaining statistics. Figure 3 shows the least-squares linear regression of the variables P and I for SE and P and N for MO. In all the cases, the correlation coefficient was significant at the 9% confidence level and the residuals behave as simple random and normally distributed. In the case of SE, the slopes indicate a decreasing trend of 19 mm/1 years and 1.1 (mm/day)/1 years for P and I respectively. The large values at the start of the time series correspond to a very wet period in the beginning of the 196s decade. The NAO index corresponding to the maximum in 1962 was 3.1, one of the highest values of this index in the study period (NAO index available in In the next section, the relationship between NAO index and daily rainfall in the Iberian Peninsula is discussed in more detail. The percentage of variance explained by the linear model is, respectively, of 16 and 26%. In MO, the decrease in P is 18 mm/1 years, basically provoked by the descent in the number of wet days, N, of 1.8 days/1 years. In this station, the percentage of explained variance by the linear model for both variables is, respectively, 19 and 9%. For MA (figures not shown), the slopes indicate a decreasing trend in 13 mm/1 years, 1.2 (mm/day)/1 years, 3 mm/1 years, and %/1 years for P,I,P9, and % respectively. In RE, the decrease in I (1.2 mm/1 years), along with the increase in N (2.3 days/1 years) yields total spring rainfall P without a significant trend. However, a negative trend in I yields significant decreasing trends for P 9 (2.4 mm/1 years) and % (3%/1 years). With respect to northern stations, the slopes corresponding to BI indicate a decreasing trend in.4 (mm/day)/1 years for I, and.9 mm/1 years for P 9, and for the intensity of BU and LE.3 (mm/day)/1 years. The comparison of mean values of I corresponding to the two halves of the complete period show significant differences between the two sub-period means of BI (from 6.9 to 6. mm/day), BU (from 4.9 to 4.3 mm/day), LE (from.2 to 4.4 mm/day), MA (from 9.4 to. mm/day), MU (from 6.3 to 4.6 mm/day), SE (from 8.9 to.6 mm/day), and RE (from 9. to 6.2 mm/day). These differences suppose percentages with respect to the reference period average of 14% for BI, 13% for BU, 1% for GR, 17% for LE, 2% for MA, 31% for MU, 4% for SE, and 44% for RE. These results, along with the analysis of linear trends, seem to indicate that the decreasing trend of daily intensity is more prominent in southern stations. As a result, the decreasing trend in the total amount of rainfall P is only detected in MA and SE. If we compare the mean values of the two subperiods, we find significant differences, with a fall of 3 mm in MA and 64 mm in SE (43% of the complete period average in both cases). In the case of MO, the descent in P seems to be related to the decrease in the number of wet days. Recent works have highlighted the important issue of intense descents of precipitation in early spring, using both monthly (Trigo and DaCamara, 2; Serrano et al., 1999) and daily (Paredes et al., 26) data. The paper by Paredes et al. (26) clearly shows that this decrease is most significant over western, central, and southern sectors of Iberia, being almost coincident with the month of March. Moreover, these authors show that this decrease is mostly due to a northward shift on the average latitudinal location of the storm-tracks with Atlantic origin, Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

8 2 F. S. RODRIGO AND R. M. TRIGO Table V. As Table III for spring from 191 to 22. Station P N I P9 % A B BI s=.4 ±.1 s=.9 ±.4 d=.9 d = 2. BU s=.3 ±.1 d=.6 GR LC LE s=.3 ±.1 d=.8 M MA s= 1.3 ±.6 s=.12 ±.3 s=.3 ±.1 s=. ±.2 d= 3.4 d= 3.9 d= 1.2 d= 16.1 MU s=. ±.2 s =. ±.2 d= 1.7 d= 1.1 SA SE s= 1.9 ±.6 s=.11 ±.3 s=.24 ±.8 d= 64.1 d= 2.9 d= 7. SO V Z GA G s=.14 ±.7 s =.4 ±.2 d = 2.3 d = 8.4 L MO s= 1.8 ±. s=.18 ±.8 d=.8 d= 6.7 PI s=.1 ±.4 s=.4 ±.1 d= 2.7 d= 9.9 RE s= +.23 ±.7 s=.12 ±.2 s=.24 ±.6 s=.3 ±.2 d= +.7 d= 3.3 d= 7.3 d= 11.1 SER associated with a positive trend over the last four decades of the NAO index for the month of March Summer During summer, the large-scale atmospheric circulation is dominated by the Azores anticyclone, which is displaced towards its north-westerly position, producing northerly or north-easterly winds that bring warm and dry air into Iberia (García-Herrera et al., 2a). This circulation is usually reinforced at the regional scale by the development of a thermal low, centred over the Iberian Peninsula, and the few precipitation episodes are usually related to convective mesoscale systems (Trigo and Dacamara, 2; Garcia-Herrera et al., 2b). In this framework, the Iberian Peninsula registers scarce summer rainfall, making it difficult to calculate trends and to assess their significance (Serrano et al., 1999; Mosmann et al., 24). Most papers dealing with Iberian rainfall (e.g. Goodess and Jones, 22) do not present analysis for the summer season. However, in this work, we have attempted to make an analysis of summer rainfall similar to the one developed for other seasons. Results must be interpreted cautiously, as in many situations the Gamma distribution may not be the most appropriate distribution to model the amount of rainfall in this season (with reduced number of rainy days). Table VI shows the average values of the five variables studied for the reference period A Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

9 DAILY RAINFALL IN THE IBERIAN PENINSULA 21 P (mm) SE, spring P = 2 1,9t r =, P (mm) MO, spring 3 3 P = 191 1,8t r =, P (mm) SE, spring I = 1,11t r =, P (mm) MO, spring N = 26,18t r =, Figure 3. Least-squares linear regression for P and I of spring in Sevilla (a) and P and N in Monfor (b) from 191 to 22. Regression equation and correlation coefficient are included. Table VI. As Table II for summer, reference period Station P (mm) N(days) I(mm/day) P 9(mm) % A B BI BU GR LC LE M MA MU SA SE SO V Z GA G L MO PI RE SER clear difference between the north and south areas of the country is seen, with rainfall above or close to 1 mm in a few northern stations, and rainfall below 4 mm in the southern area. Similar differences are seen in N, with N above (below) 1 days for stations located in the northern (southern) area. However, the average values of I and P 9 are very similar in all the stations, mostly in the range of 4 mm/day and 1 2 mm, respectively. It is worth mentioning that the highest value of P 9 corresponds to B (29.4 mm), to the north-east of the country. Rainfall behaviour in this season is related to local and convective phenomena, especially important in the Mediterranean coast. This fact may explain the high values (above 3%) found for % in the Mediterranean stations (B, V, A). Table VII shows the results of the application of the Mann Kendall test to detect trends. Decreasing trends Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

10 22 F. S. RODRIGO AND R. M. TRIGO Table VII. As Table III for summer from 191 to 22. Station P N I P9 % A s=.9 ±.4 s=.2 ±.1 s=.6 ±.3 d= 3. d= 8.7 d = 16.3 B BI s =.4 ±.2 s =.8 ±.9 d =.6 d =.8 BU GR LC LE s =.4 ±.2 d =.8 M MA MU SA SE s =.8 ±.7 s =.3 ±.2 d = 2. d = 7.2 SO V Z GA G s =. ±.3 s =.13 ±.9 d= 2. d=.2 L MO s =. ±.3 d= 2. PI RE s=.9 ±.3 s=.27 ±.8 s =.4 ±.3 d= 2.8 d= 7. d= 12.8 SER for I have been detected in some southern (A, SE, G, RE) and northern (BI, LE) stations. These trends are reproduced in the behaviour of P 9 for A, BI, SE, G, and RE and % for A and RE, with decreasing trends. A recent work by Mosmann et al. (24) has shown that there are significant trends of precipitation over southern Spain, with June and September (July and August) being dominated by negative (positive) tendencies. The use of summer (JJA) time series in our work merges these conflicting monthly trends, therefore hampering the appearance of any significant trend in P (Table VII). Nevertheless, we stress again that the number of days N with precipitation in the summer can be extremely low, particularly in southern Iberia (Table VII). The least-squares linear regression only yields significant results in A and RE for I and P 9 (Figure 4). In A, the slope indicates a decreasing trend of.9 (mm/day)/1 years and 2 mm/1 years, respectively. The maximum at the beginning of the series (1962) may affect the results, therefore the analysis was repeated without this value (considering 1962 as missing data). The result still indicates a significant decreasing trend with a slope of 1.7 mm/1 years. In RE, significant decreasing trends were found for I and P 9 with respectively.9 (mm/day)/1 years and 2.7 mm/1 years. The comparison of the two sub-periods indicates a significant change in A from 6.2 to 3.2 mm/day for I, and from 2. to 11.3 mm for P 9. In RE, there was a descent of 2.8 mm/day and 7 mm in the value of I and P Autumn Autumn is a transition season from summer to winter conditions. Autumn pattern reflects the influence of convective and local storms in early autumn, and the influence of westerly circulation types from October onwards (Sumner et al., 21; Garcia-Herrera et al., 2b). As a result, the total amount of rainfall in this season P (Table VIII) shows a behaviour very similar to winter, with values around 3 mm in northern stations Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

11 DAILY RAINFALL IN THE IBERIAN PENINSULA 23 2 A, summer 2 RE, summer I (mm/day) I =7,9t r =,33 I (mm/day) I =6,9t r =,43 P9 (mm) A, summer P9 = 22,2t r =, P9 (mm) RE, summer P9 = 19,27t r =, Figure 4. Least-squares linear regression for I and P 9 of summer in Alicante (a) and Reliq (b) from 191 to 22. Regression equation and correlation coefficient are included. (LC, BI, GA) and the lowest values (on the order of 1 mm) corresponding to eastern stations (Z, MU). However, P values in these stations are higher than winter values, possibly as a consequence of the important role of convective processes in this area. The number of rainy days show values similar to those obtained for winter in all stations, while I and P 9 present higher values than in winter for the vast majority of the stations. Finally, % values are very similar or slightly higher than winter values. Table IX shows the results of the application of the Mann Kendall test to detect trends. Decreasingtrends for I have been detected in northern (BI, LE) and southern (MA, RE) stations. These trends are reproduced in P 9 and the % variables for BI and RE. In the cases of MA and RE, to the south of the country, a positive trend in N has been found. For the remaining stations (except in A, B, and SER), the sign of the statistic for N is positive, although non-significant. As a result, in these stations with significant negative trend in I, the total amount P does not show a significant trend (except in BI). In most cases where the Mann Kendall test detects a trend, the application of least-squares linear regression yielded a significant correlation coefficient. So, the slopes for BI were.9 (mm/day)/1 years for I, 2.7 mm/1 years for P 9, and 4%/1 years for %. The slope for I in LE was.4 (mm/day)/1 years, in MA it was 1.1 (mm/day)/1 years, and in RE it was.9 mm/1 years. The slope for N in MA was +1.4 days/1 years and +2 days/1 years in RE. Figure shows the regressions corresponding to N and I for MA, with opposite trends. The percentage of variance explained by the linear model is 11% for N and 8% for I in MA. Table VIII. As Table II for autumn, reference period Station P (mm) N (days) I (mm/day) P 9 (mm) % A B BI BU GR LC LE M MA MU SA SE SO V Z GA G L MO PI RE SER The comparison of the two sub-periods shows a significant difference in N for MA (from 13 to 17 days), while a significant difference is not there in I for this station (from 12.9 to 1.9 mm/day). In BI, the decreasing trend in I and P 9 is detected when comparing both sub-periods, with a change from 1. to 7.6 mm/day and from 33. to 26.1 mm respectively. There is a significant difference in P of 89.8 mm (27% of the reference period Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

12 24 F. S. RODRIGO AND R. M. TRIGO Table IX. As Table III for autumn from 191 to 22. Station P N I P9 % A B BI s= 3. ± 1. s=.9 ±.2 s=.27 ±.9 s=.4 ±.2 d= 89.8 d= 2.4 d= 6.9 d =.8 BU GR LC LE s=.4 ±.2 d =. M MA s= +.14 ±. s=.11 ±. d= +4. d = 2. MU SA SE SO V Z GA G L MO PI RE s =+.2 ±.1 s=.9 ±.3 s=.21 ±.9 s=.4 ±.2 d = 4.9 d = 1.1 d = 1. d =.6 SER average). The comparison of I for LE, % for BI, and the variables corresponding to RE did not yield a significant difference. 3.. Annual values A similar methodology was applied to the entire series to investigate the behaviour of the annual values of the five variables in the 22 rain gauges. Table X shows the Mann Kendall statistic obtained for each variable and the rain gauge. In general terms, most rain gauges show a decrease of the total amount of annual rainfall, with the exception of the two north-western stations (LC and GA) that reveal positive (non significant) trends for P. However, the vast majority of these negative trends is non-significant, with the exception of BI (north) and MO and RE (south). For BI, the slope of the linear trend (significant at the 99% confidence level) is 71.3 mm/1 years, and the difference of mean values is significant, with a descent from 1297 to 1124 mm (that is, 173 mm, 14% of the total annual rainfall). In relation to N, most stations reveal a positive trend, although nonsignificant, except in LE and MA (positive) and MO (negative). In the case of MA, this result is similar to autumn, when the total amount of rainfall is the 32% of the total annual rainfall. Probably the most outstanding result from Table X is the negative trend detected in I in most of localities, along with positive (although non significant) trends in the number of wet days. Some of these results may be due to the influence of the negative trends detected in just a few seasonal values, for instance, in summer (A), in spring (B), or in winter (GR). These trends have a slight influence on the other two variables, with significant decreasing trends in P 9 (BI, BU, MA, and RE), and in % (BI, M, RE). An interesting case is SER, with positive trends in the indices of extreme values (P 9, %), reflecting the positive value of the seasonal statistics, although in each season the values obtained were not significant. The application of linear models (Figure 6) and the comparison between the two halves of the complete period yield similar results for the annual values. 4. DISCUSSION The main results obtained in this work may be summarised in the following way: In winter, the most important trends were detected in BI (north) and RE (south), both stations presenting a decrease of the total amount of rainfall P, the daily Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

13 DAILY RAINFALL IN THE IBERIAN PENINSULA 2 N (days) MA, autumn 3 3 N = 11 +,14t r = I (mm/days) RE, autumn 3 3 I = 12,9t r =.39t I (mm/days) MA, autumn 3 I = 1,11t 3 r =, P RE, autumn P9 = 33,21t r = Figure. Least-squares linear regression for N and I of autumn in Málaga (a) and I and P 9 of Reliq (b) from 191 to 22. Regression equation and correlation coefficient are included. I (mm/days) BI, year I = 9,6t r =,73 I (mm/days) SE, year I = 1,t r =, MA, year I = 13,1t r =,3 2 1 RE, year I = 11,1t r =,6 I (mm/days) 1 I (mm/days) Figure 6. Least-squares linear regression for I of annual values in Bilbao (BI), Sevilla (SE), Málaga (MA) and Reliq (RE). Regression equation and correlation coefficient are included. intensity I, and the 9th percentile. The number of rainy days did not change in a significant way. In spring, the decreasing trend detected in I in some stations is more prominent in MA, SE, and RE, to the south of the Peninsula, reflecting a decrease in the total amount of rainfall P in MA and SE. The number of rainy days only changed in a significant way in MO (decreasing) and RE (increasing). In summer, a decreasing trend was found for I,P9, and % in A and RE, both located to the south of Iberia. Other stations also recorded negative trends for the variables I and P 9. In autumn, an increasing trend in the number of rainy days N for MA and RE and a decreasing trend in the intensity I for these stations to the south of the Iberian Peninsula is worth mentioning. The general behaviour in the cases where a trend has been detected is a decrease in the daily intensity rainfall I, while the number of rainy days N does not change in a significant way. This result is also valid for annual Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

14 26 F. S. RODRIGO AND R. M. TRIGO Table X. As Table III for yearly values from 191 to 22. Station P N I P9 % A s =.2 ±.1 d =.3 B BI s= 7.1 ± 1.7 s=.6±.1 s=.14±.3 s=.19±.8 d= d= 1. d= 3.3 d = 3.7 BU s=.3 ±.1 s=. ±.2 d=.6 d = 1.2 GR s=.2 ±.1 d=.7 LC LE s =+.3 ±.1 s=.2 ±.1 d =+7.7 d =.4 M s=.22 ±.8 d= 6.1 MA s=.3 ±.1 s=.1 ±.2 s=.21 ±.8 d= +7.3 d= 3.1 d= 6.4 MU s=.3 ±.1 d= 1.2 SA SE s=. ±.2 d= 1.3 SO s =.1 ±.1 d=.3 V Z s=.2 ±.1 d=.6 GA G L MO s= 4. ± 2. s =.4 ±.2 d = 9.7 d= 14.1 PI RE s= 4. ± 2. s=.1±.2 s=.24±. s=.26±.9 d = 83.8 d= 2.6 d= 6.1 d= 7.8 SER s= +.18 ±.6 s= +.3 ±.1 d= +4.7 d= +9.1 values of the indices. Therefore, the conclusion is that there is a trend to lower amount of rainfall during the rainy days. Extreme rainfalls (measured by means of P 9 and %) do not present significant trends, except in a few stations, where a decrease in the value of the 9th percentile and the percentage of rainfall falling in days with rainfall above the threshold chosen has been detected, coherent with the descent of the daily intensity index I. The main purpose of this work is to present an analysis on extreme precipitation changes across the Iberian Peninsular under a unique methodology (based Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc

15 DAILY RAINFALL IN THE IBERIAN PENINSULA 27 Figure 7. Correlation coefficients between NAO index and winter rainfall indices (a): P ;(b):n;(c):i; (d)p 9; (e) %. Bold numbers: coefficients significant at the 9% confidence level. The continuous line divides areas with significant correlations from areas where the correlation is not significant. on recommendations by CLIVAR), using a seasonal and annual aggregation (including the analysis of summer rainfall). In general terms, results do not yield a clear spatial pattern, but they show the main changes that occurred in the intensity of northern and southern stations, with a decreasing trend. Comparisons of these results with those obtained for other places in Iberia, Europe, and other regions of the globe are difficult to establish because the period, the adopted variables, and the temporal scale (annual, seasonal, monthly or daily) of the studies do not coincide in most cases. Nevertheless, the trend to lower number of days with high amounts of rainfall has been found in other works (e.g. Goodess and Jones (22) in their analysis of 18 stations across the Iberian Peninsula, or De Luis et al. (2) and González-Hidalgo et al. (23), in their studies on the Valencia region, where they conclude that torrential events may have diminished in magnitude). The analysis of 1 series across the Iberian Peninsula for the period (Moreno-Rodríguez, 2) did not yield significant trends in the annual amount of rainfall and the number of rainy days. The analysis on the behaviour of rainfall in the eastern part of the country (where the rain gauges of B, V, A, MU are located) for (Chazarra and Almarza, 22), in the Ebro Valley (rain gauge Z) for (Abaurrea et al., 22), or in the Spanish plateau (M, SA), toward the central area of the country (Galán et al., 1999) did not detect significant trends, similar to our results corresponding to the stations located in these areas. Further work is necessary to obtain a complete view of the spatiotemporal variability of daily rainfall in the Iberian Peninsula. Copyright 26 Royal Meteorological Society Int. J. Climatol. 27: (27) DOI: 1.12/joc