Changes in frequency and intensity of daily precipitation over the Iberian Peninsula

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2006jd007280, 2006 Changes in frequency and intensity of daily precipitation over the Iberian Peninsula M. C. Gallego, 1 J. A. García, 1 J. M. Vaquero, 2 and V. L. Mateos 1 Received 9 March 2006; revised 25 July 2006; accepted 11 August 2006; published 19 December [1] There is a general consensus within the climate community that any change in the frequency or severity of extreme climate events would profoundly impact nature and society. Such changes can be studied in terms of rainfall indices derived from daily data. The great variability of the rainfall over the Iberian Peninsula and the irregularity of its water regime make any study of the rainfall in this geographical region very interesting. In this work, we contribute some information about the characteristics and distribution of the daily precipitation series over the Iberian Peninsula, in particular the study of trends in series of several selected indices all related to the frequency or intensity of the daily rainfall. Data used are taken from a set of 35 series of daily rainfall in the aforementioned region during the period The selected observatories constitute a representative sample of the orography and geographical diversity present in the peninsula. From the daily rainfall data, series of frequency and intensity rainfall indices were constructed on the basis of a prior definition of four rainfall categories reflecting certain percentiles of the precipitation distribution in the peninsula: total rainfall (0.2 mm), light (0.2 mm and <2.5 mm), moderate (2.5 mm and <7.5 mm), intense (7.5 mm), and very intense (15 mm). The indices studied were number of rainy days in each rainfall category, medians and maxima of length of dry spells (both related to the frequency of the precipitation), accumulated precipitation in each rainfall category, proportion of rainfall in each category relative to the total accumulated rainfall, and mean precipitation per wet day in each category (related to the amounts). All the indices were evaluated seasonally. Significant trends in many of the selected series were found. The results for all the indices are coherent and point to an increase of the light rainfall events at the cost of a decrease of more intense events. Citation: Gallego, M. C., J. A. García, J. M. Vaquero, and V. L. Mateos (2006), Changes in frequency and intensity of daily precipitation over the Iberian Peninsula, J. Geophys. Res., 111,, doi: /2006jd Introduction [2] Changes in climate variability and extremes of weather and climate events have received increasing attention in the last few years. Nevertheless, understanding changes in climate variability and climate extremes is made difficult by interactions between the changes in the mean and the variability [Meehl et al., 2000]. Extremes are a key aspect of climatic change. Changes in the frequency of many extremes (increases or decreases) are often the most sensitive aspects of climate change for ecosystem and societal responses [Katz, 1999]. Results from the studies that have been conducted reveal that there has been an amplified response of the heaviest precipitation rates relative to the percentage change in total precipitation, and indicate that in many situations the overall impact is 1 Departamento de Física, Universidad de Extremadura, Badajoz, Spain. 2 Departamento de Física, Escuela Politécnica, Universidad de Extremadura, Cáceres, Spain. Copyright 2006 by the American Geophysical Union /06/2006JD determined more by changes in magnitude and frequency of extreme events than by variations of the mean value [IPCC Grupo de Trabajo II, 1992; Intergovernmental Panel on Climate Change (IPCC), 2001]. While the analysis of changes in precipitation extremes shows clear evidence of an increase in extreme and heavy precipitation events for some regions and times of the year, in others the differences in the way in which the data were analyzed make it difficult to interpret the results [Karl, 1999]. In order for these studies to be improved and fill one of the gaps highlighted by IPCC [2002], the ideal would be to access the longest records of daily data. An additional problem is that daily rainfall is a very difficult variable to deal with because of some of its statistical characteristics, such as the need for a decomposition into processes of occurrence of events and intensity of each event, the high degree of asymmetry of the distribution of intensities, the lack of spatial coherence, the marked local character and the strong dependence on orography, vegetation, existence of sources of humidity, etc. Hence it is insufficient to study the total monthly or seasonally accumulated quantities. Instead, it will also be necessary to analyze the frequency of occurrence of the 1of15

2 rainfall, intensity of rainfall, etc. In this context, Trenberth et al. [2003] observe that climate change is linked to local changes in intensity, frequency, duration, and quantity of rainfall. Following these ideas, the Expert Team on Climate Change Detection, Monitoring and Indices (ETCCDMI, recommend the study of series of rainfall indices as indicators of climate variability [Alexander et al., 2006]. [3] A summary of the state of the art in precipitation studies using rainfall indices in diverse locations around the world is given by Brunetti et al. [2001b, 2004], who includes an overview of the different trends found. Recently, Zhang et al. [2005] and Aguilar et al. [2005] studied the trends in the Middle East and central and northern South America, respectively, and Moberg et al. [2006] have just produced the most comprehensive study to date of changes in precipitation indices over Europe since 1901 through the EU EMULATE project. A focus on general Iberian rainfall, linking its behavior to the NAO, is given by Goodess and Jones [2002], Gallego et al. [2005], and García et al. [2006]. The present work has the idea of contributing some further information on the study of daily rainfall trends over the Iberian Peninsula. The Iberian Peninsula (IP) precipitation is very irregular due to its geographical situation between two moisture sources, the Atlantic Ocean and the Mediterranean Sea, and its great geographical diversity with mountainous chains, valleys, rivers, etc. Precipitation presents both large spatial and temporal variability. The great influence of precipitation on many aspects of the country s life (agriculture, water supply, tourism, etc.) motivates our study. Taking into account that changes in rainfall can be in its frequency, intensity, or a mixture of both, we used daily rainfall data series over this area to construct series of rainfall indices based on the frequency and intensity of the rainfall events by defining categories of rainfall exceeding certain thresholds. For frequency indices we took the number of rainy days and dry spells in each category, and for intensity indices the values of accumulated rainfall, proportion of daily rainfall amounts compared with the corresponding total, and the mean amount of precipitation per wet day, all in each rainfall category. Because of the high degree of seasonality of rainfall in the Iberian Peninsula, each index was evaluated separately for each season defined as follows: spring (March-April-May), summer (June-July-August), autumn (September-October- November), and winter (December-January-February). In this seasonal study, results for summer must be interpreted with more caution. This is because rainfall regimes in the IP generally present scarce precipitation events in this season with which to evaluate the indices. Nevertheless, for coherence, the summer is included for all the indices. For each observatory and season, we evaluate the existence of trends in these series of rainfall indices by means of the Kendall test. The slope of the trends is calculated through the Theil- Sen estimator, a popular nonparametric point estimator of the slope parameter in the simple linear regression model that has many attractive properties: it is robust, easy to compute, and competitive in terms of mean squared error with alternative slope estimators [Dietz, 1989]. These nonparametric tests were selected because of the nonnormality of the daily rainfall events in Iberia. In order to identify spatially coherent regions with common trend behavior, the Theil-Sen estimator of the slope and the Kendall statistic significance level were also subjected to a spatial study. [4] Thus the main aims of the present work were (1) to analyze possible changes in precipitation over Iberia by means of the study of trends in daily rainfall frequency and intensity indices in a network of observatories in this area and (2) to identify spatially coherent regions with common trend behavior during a study period that spans the forty years from 1958 to The structure of the paper is as follows: Section 2 describes the data used for the study and the construction of the rainfall index series. The methods used for trend detection and slope estimation are presented in section 3. The trends found for each rainfall index series are discussed in section 4 (frequency indices) and section 5 (intensity indices), including a spatial study of these results. The pertinent conclusions are presented in section Data [5] The data used in the study were daily rainfall series at 35 observatories distributed over IP during the period This is a network of stations that, although relatively coarse, is broadly representative of the range of climate regimes experienced across the peninsula (see Figure 1). They were kindly supplied by three sources: Instituto Nacional de Meteorología de España (INM, European Climate Assessment and Dataset (ECA, and Real Instituto y Observatorio de la Armada Española en San Fernando (Cdiz) (ROA, The selection was made from a larger set of data series, taking into account their temporal extension with as few gaps as possible (in all cases less than 2%), and their spatial location in order for the sample to be representative of the orographic diversity of the Iberian Peninsula. Also it must be noted that the stations underwent no changes in location or instrumentation. The first verification of the goodness of the behavior of the data was their graphical representation. Further checks of the homogeneity of the series were made by means of Buishand tests [Buishand, 1982] and Pettitt test [Pettitt, 1979]. These tests did not detect any inhomogeneities. A monthly evaluation of the Alexandersson test [Alexandersson and Moberg, 1997] shows inhomogeneities for two months at five locations (5 from 35: Barracas, Cáceres, Embalse de Cierva, Embalse de Fuensanta and Grazalema). Rainfall series from neighboring sites to these observatories were requested from the Meteorological Services even where no relocations were documented. Then, a new analysis was performed with these series to determine whether differences existed or not between the suspect observatories and neighboring locations. The time evolution between each set of monthly rainfall series was very coherent. This fact, combined with the lack of inhomogeneities identified by performing new Alexandersson tests on the neighboring stations, led us to trust the appropriateness of the selected series and agreed with metadata reports for all the observatories. [6] As noted above, the daily rainfall series were broken down into frequency and intensity in order to extract information about the rainfall characteristics. It is therefore necessary to distinguish different types of rainfall by defining specific thresholds that are representative of each type. There does not seem to be an uniform criterion in the 2of15

3 Figure 1. Geographical situation of the selected observatories. literature regarding the selection of an appropriate threshold for appreciable precipitation, and one can find an assorted sample, sometimes according to the precipitation regime of a region or to the instrument resolution, for example, 0.1 mm according to Liu et al. [2005], Gallego et al. [2005], García et al. [2006], and Alpert et al. [2002], 0.2 mm according to Bodri et al. [2005], 0.3 mm according to Osborn et al. [2000], 0.5 mm according to Akinremi et al. [1999], and 1 mm according to Tarhule and Woo [1998], Brunetti et al. [2004], and Frich et al. [2002]. The pluviographs in the selected observatories over the IP are accurate to 0.1 mm and no instrumentation changes have been made in our period of study, according to the information obtained from Meteorological Services. Even so, following the criterion noted by the World Meteorological Organization [1996] and taking into account the IP precipitation regime we have taken an appreciable precipitation threshold of 0.2 mm. Thus we established four categories of rainfall in the IP: light (between 0.2 mm and 2.5 mm), moderate (between 2.5 mm and 7.5 mm), intense (greater than 7.5 mm), and very intense (greater than 15 mm). These, together with the total rainfall (greater than 0.2 mm), constitute the set of categories for which the rainfall indices will be evaluated. A detailed discussion of the choice of thresholds and the definition of these categories is given by Gallego et al. [2005]. [7] Seasonal series of rainfall indices were constructed from the daily rainfall series for each of the 35 observatories. One set of series, related to the frequency of rainfall, consisted of the number of rainy days in the categories of total, light, moderate, intense, and very intense rainfall, for each season of the year. Also related to the frequency, and complementary to the previous set, were the series of medians and maxima of dry spells counting the number of dry days between two total rain events. Series related to the intensity were constructed from the accumulated rainfall in each category during each season. To normalize the last index, the proportion is defined of daily rainfall amounts in each category relative to the corresponding total. Finally, as a ratio between accumulated rainfall and number of rainy days, the mean amount of precipitation per wet day is defined also in each rainfall category. The description of this set of indices is summarized in Tables 1 and 2 for frequency and intensity, respectively. Further information about the spatial distribution of the mean values of the frequency indices over Iberia is given by Gallego et al. [2004]. 3. Method [8] The trends in the rainfall index series were studied through the nonparametric Mann-Kendall (M-K) test [Kendall, 1976]. The null hypothesis H 0 is that the data (X 1, X 2,..., X n ) are identical and independently distributed random variables, and the alternative hypothesis H 1 is that thy are distributed according to an increasing or decreasing trend. The statistic of the test is the Kendall t, whose expression is the following t ¼ Xn 1 i¼1 X n j¼iþ1 sgn X j X i ; where sgn(x) is the sign function, with values equal to 1, 0, or 1 depending on whether the argument is negative, zero, or positive, respectively. The variance of t under the null hypothesis is varðtþ ¼ nn ð 1Þð2n þ 5Þ : 18 3of15

4 Table 1. Description of the Set of Rainfall Frequency Indices a Rainfall Category Description Number of Rainy Days Total seasonal count of days when PRCP 0.2 mm Light seasonal count of days when 0.2 mm PRCP < 2.5 mm Moderate seasonal count of days when 2.5 mm PRCP < 7.5 mm Intense seasonal count of days when PRCP 7.5 mm Very intense seasonal count of days when PRCP 15 mm Medians Maxima a PRCP is precipitation per day. Dry Spells Seasonal median of number of consecutive days with PRCP < 0.2 mm Seasonal maximum of number of consecutive days with PRCP < 0.2 mm The exact distribution of t can be evaluated. For n > 10, the distribution approaches a normal, especially if the correction t 0 ¼ t sgnðtþ is made. The normalized variable t 0 Z ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi varðtþ is evaluated from t 0. Obviously, Z is normally distributed with mean 0 and variance 1. It will give the values of the probability of acceptance of the null hypothesis. The power of the M-K test, the probability of rejecting the null hypothesis (detection of a trend), is as high as that of the parametric tests (for example, the t test of the slope of the regression). Its advantage lies in its independence of the form of the distribution [Press et al., 1990]. Nevertheless, this method can be inappropriate when data show substantial serial correlation. Generally, when lag-1 autocorrelation coefficients exceed 0.3 the M-K test yields results that overestimate trend magnitude and statistical significance [von Storch, 1995], but for autocorrelations 0.1 the test runs correctly. According to this assumption, we have computed the autocorrelation function for our indices series. For all the cases, the lag-1 autocorrelation coefficients are minor to 0.1. Thus the results are appropriate. [9] The M-K test described above detects the existence of a trend, but does not provide an estimate of its magnitude. For this purpose, we use the algorithm proposed by Hirsch and Smith [1982], an extension of that suggested by Theil [1950a, 1950b, 1950c] and Sen [1968]. The statistic of this Theil-Sen test is related to the slope of the trend found by the M-K test, and is defined in the following form: B k ¼ med D ijk ; where D ijk =(x ik x jk )/(i j) for all of pairs (x ik, x jk ) with k each one of the seasons and 1 i < j n k, n k being the number of observations in season k during the period considered. The slope estimator B is related to the Kendall statistic t 0 as follows: If t 0 > 0, then B 0. Moreover, B is a measure of the slope that is insensitive to the effect of the existence of extreme values in the data, since it is evaluated across the median [Lettenmaier et al., 1994]. [10] The M-K test was evaluated two-sidedly for two significance levels: at 5% and at 10%. Since one does not know a priori the type of positive or negative trend that Table 2. Description of the Set of Rainfall Intensity Indices a Rainfall Category Description Accumulated Rainfall Total seasonal rainfall amount when PRCP 0.2 mm Light seasonal rainfall amount when 0.2 mm PRCP < 2.5 mm Moderate seasonal rainfall amount when 2.5 mm PRCP < 7.5 mm Intense seasonal rainfall amount when PRCP 7.5 mm Very intense seasonal rainfall amount when PRCP 15 mm Light Moderate Intense Very intense Total Light Moderate Intense Very intense a PRCP is precipitation/day. Proportion proportion of rainfall amount when 0.2 mm PRCP < 2.5 mm respect to the total proportion of rainfall amount when 2.5 mm PRCP < 7.5 mm respect to the total proportion of rainfall amount when PRCP 7.5 mm respect to the total proportion of rainfall amount when PRCP 15 mm respect to the total Mean Rainfall per Wet Day mean rainfall amount per wet day when PRCP 0.2 mm mean rainfall amount per wet day when 0.2 mm PRCP < 2.5 mm mean rainfall amount per wet day when 2.5 mm PRCP < 7.5 mm mean rainfall amount per wet day when PRCP 7.5 mm mean rainfall amount per wet day when PRCP 15 mm 4of15

5 the data series may present, the null hypothesis is that no trend exists, and the alternative hypothesis is that a trend exists: H 0 : t ¼ 0 y H 1 : t 6¼ 0: This requires to a two-sided application of the test. [11] The calculation of the Theil-Sen statistic provides an absolute value of the slope in each of the cases. In order to be able to compare the results among observatories as homogeneously as possible, one has to normalize the aforementioned statistic with respect to the standard deviation of each series. One thereby obtains values of relative slopes for the study period in each of the observatories. [12] After calculating the values of the normalized trends and their significance level, a spatial interpolation was performed by kriging [Nychka et al., 1998] using the FIELDS R software package (Fields Development Team, Tools for spatial data, to explore the spatial coherence of the trends calculated over each IP index series. The results were mapped in figures that contain the information relative to the trend of every index at each observatory independently, as well as the spatial distribution of the slopes obtained in the procedure. [13] From the results obtained for each location, it would be desirable to be able to obtain regional conclusions that take into account the behavior of all the observatories as a whole. Because of the high number of indices and the variety of cases, we have not performed any test of field significance, but n independent repetitions, with probability p, of a certain attribute A, will produce a natural form of the binomial distribution, B(n;p). Applying the binomial distribution to our problem, each experiment contains 35 statistically independent events (the series corresponding to each of the observatories). In our case, the hypothesis of independence of rainfall events between observatories is assumed since we propose that they are located far enough apart from each other that local effects will dominate for some of the indices. This fact must be taken into account when interpreting the results. As noted above, n = 35, p = 0.05 (for the more restrictive significance level, 5%) and A is the existence of positive or negative trend. From this binomial distribution, in order that a regional result is significant at the 5% level in the IP from the evaluation of M-K test at 35 observatories distributed over its geography, it is necessary to find at least 5 trends of the same sign in the whole peninsula. A minor number of trends might be randomly found. 4. Trend Analysis of the Frequency Indices 4.1. Number of Rainy Days [14] Figure 2 shows the trends found when one applies the M-K test to the series of number of rainy days for each category in every season and observatory for the period The meaning of the symbols used is as follows: Upward triangles represent increasing trends, and downward triangles decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation of each seasonal index series at each observatory), according to the Figure 2 legend. Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. The trends with absolute value < 0.5 are represented by circles. Isolines correspond to interpolated values of the normalized trend (in percentages). In order to provide a high enough resolution for the figures, the legends are scaled respective to each individual figure and this must be considered when comparing results. Nevertheless, the color palette is always the same: blue implies decrease and red increase. [15] Figure 2a shows the trends for the case of the total number of rainy days. In spring, very many negative trends are observed over the entire peninsula (see Table 3), with only the Mediterranean region, except Catalonia, presenting positive trends. In summer, on the contrary, we found over most of the peninsula an increase in rainy days. Only one station, Torre de Juan Abad, shows a negative significant trend. In autumn, there are generally weaker trends and a more mixed pattern of behavior: the south of Atlantic seaboard, the peninsula center and Barcelona had small negative trends, while the northwest and the southeast had positive trends, the last being that with the most marked trends. In winter, there was again a more generalized decreasing behavior. Positive trends appeared only in the peninsula southeast. This winter behavior is consistent with the association with the NAO index that Gallego et al. [2005] noted for the total number of rainy days in the peninsula. The negative trend at San Fernando in spring and autumn confirms reports of other workers [Lana and Burgueño, 2000]. Lana et al. [2003] have also described a negative trend at the observatory of Barcelona in spring for the total number of rainy days, that is very slightly here. Globally, other studies have indicated a decrease of the total number of rainy days in different regions of the world [Brunetti et al., 2001a, 2001b; Manton et al., 2001; Haylock and Nicholls, 2000]. It is particularly worthy of note that an increase of the total number of rainy days is observed for every season in the IP southeast. The maps reflect a certain persistence in the structures from winter to spring. [16] The results for the light rain category are shown in Figure 2b. This is the index for which there exist most significant trends in the series, and most of them are positive (see Table 3). One can therefore say that, for the entire Iberian Peninsula, and at a significance level of 5%, there exists an overall behavior pattern: light rain has been increasing over the study period ( ), since in every season we found fairly high numbers of significant positive trends (9 in spring, 14 in summer, 10 in autumn, and 9 in winter). This possible increase seems to be consistent with the results of Alpert et al. [2002] that show an increase in the category of light rain. The only zones with a decreasing trend for the number of days of light rain are: the Atlantic seaboard in autumn and winter, the interior of the IP and the Pyrenees region in spring, and the Barcelona observatory in all seasons. [17] For the moderate rain, winter is the season that presents the greatest number of significant trends and of negative sign (31.43% of the series), as is reflected in Table 3. One observes in Figure 2c that the magnitude of the trends is less than for the previous indices. We found 11 significant negative trends in the peninsula overall in winter. Because of this number of significant trends, and with the 5of15

6 Figure 2. Mann-Kendall test for number of rainy days in spring, summer, autumn, and winter (from left to right): (a) total, (b) light, (c) moderate, (d) intense, and (e) very intense. Upward triangles represent increasing trends, and downward triangles represent decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation). Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. Isolines correspond to interpolated values of the normalized trend (in percentage). Circles show trends with absolute value < 0.5. caution noted above, one could state that there was a decrease in moderate rainfall in this season during this period over a great part of the IP, except in the southeast where we found an increase strongly influenced by the behavior at the Embalse de Fuensanta. The case is somewhat similar in spring, but to a lesser degree. [18] The categories of intense and very intense rain also show generalized overall patterns of behavior for spring and winter in the common period of work. For both categories (see Table 3), there were 9 significant negative trends in spring and in winter 9 significant negative trends for intense rainfall and 8 for the very intense. Because of this high number of trends, one could thus state that there was a decrease of the precipitation in these categories in these two seasons, except in Mediterranean area. Four observatories were common to all four situations (intense and very intense 6 of 15

7 Table 3. Number of Significant Trends Found in Series of Seasonal Number of Rainy Days for Each Rainfall Category a Index (Number of Rainy Days) Negative Positive WT Spring Total rainfall 6(17.14%) 2(5.71%) 27(77.14%) Light 3(8.57%) 9(25.71%) 23(65.71%) Moderate 5(14.29%) 1(2.86%) 29(82.86%) Intense 9(25.71%) 1(2.86%) 25(71.43%) Very intense 9(25.71%) 0(0.00%) 26(74.29%) Summer Total rainfall 1(2.86%) 4(11.43%) 30(85.71%) Light 0(0.00%) 14(40.00%) 21(60.00%) Moderate 1(2.86%) 1(2.86%) 33(94.29%) Intense 3(8.57%) 0(0.00%) 32(91.43%) Very intense 2(5.71%) 0(0.00%) 33(94.29%) Autumn Total rainfall 0(0.00%) 3(8.57%) 32(91.43%) Light 2(5.71%) 10(28.57%) 23(65.71%) Moderate 1(2.86%) 4(11.43%) 30(85.71%) Intense 1(2.86%) 0(0.00%) 34(97.14%) Very intense 3(8.57%) 2(5.71%) 30(85.71%) Winter Total rainfall 3(8.57%) 2(5.71%) 30(85.71%) Light 5(14.29%) 9(25.71%) 21(60.00%) Moderate 11(31.43%) 1(2.86%) 23(65.71%) Intense 9(25.71%) 0(0.00%) 26(74.29%) Very intense 8(22.86%) 0(0.00%) 27(77.14%) a WT indicates without any significant trend. rain in spring and winter): Grazalema, Pozo Alcón, Torre de Juan Abad, and Villameca (see Figures 2d and 2e). At these stations, therefore, this behavior extended over the rainfall categories and over time, such that a marked decrease is observed in these categories over half the year (winter and spring) for the study period Length of Dry Spells [19] Table 4 presents the number of trends found for the length of dry spells, in values of both the medians and the Table 4. Number of Significant Trends Found in Series of Dry Spells a Index (Dry Spells) Negative Positive WT Spring Medians 3(8.57%) 6(17.14%) 26(74.29%) Maxima 2(5.71%) 5(14.29%) 28(80.00%) Summer Medians 0(0.00%) 1(2.86%) 34(97.14%) Maxima 6(17.14%) 0(0.00%) 29(82.86%) Autumn Medians 2(5.71%) 2(5.71%) 31(88.57%) Maxima 4(11.43%) 2(5.71%) 29(82.86%) Winter Medians 2(5.71%) 2(5.71%) 31(88.57%) Maxima 2(5.71%) 3(8.57%) 30(85.71%) a WT indicates without any significant trend. maxima, for the period The number of significant trends and their magnitude for the case of the medians was less than for the last indices, with the local behavior patterns having more influence. Even so, there seems to be a certain persistence of the decrease shown by the IP northwest in spring and summer, and southeast in autumn and winter. There was a significant increase in spring in the interior zone of the IP, influenced by the Navacerrada and Torre de Juan Abad observatories (see Figure 3a). [20] In the case of the maxima, there were a greater number of trends at some stations (see Table 4). Locally, there are many observatories that show significant trends opposite to that of the total number of rainy days. This was to be expected, since the two indices can be regarded as complementary. Four examples are: the growth of the dry spell maxima in spring and winter at San Fernando and Torre de Juan Abad, according to the decrease of the total number of rainy days at those same stations and seasons, and in summer at Málaga and Embalse de Figure 3. Mann-Kendall test for dry spells in spring, summer, autumn, and winter (from left to right): (a) medians and (b) maxima. Upward triangles represent increasing trends, and downward triangles represent decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation). Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. Isolines correspond to interpolated values of the normalized trend (in percentage). Circles show trends with absolute value < of15

8 Figure 4. Mann-Kendall test for accumulated rainfall in spring, summer, autumn, and winter (from left to right): (a) total, (b) light, (c) moderate, (d) intense, and (e) very intense. Upward triangles represent increasing trends, and downward triangles represent decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation). Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. Isolines correspond to interpolated values of the normalized trend (in percentage). Circles show trends with absolute value < 0.5. Fuensanta (Figure 3b), complementary to the increase that shows the total number of rainy days at the same stations (Figure 2a). For the peninsula as a whole, this complementarity is also observed with the spatial patterns appearing as the negatives of those found for the total number of rainy days in spring, summer, and autumn. In spring, a decrease in the maxima of dry spells in the east and northwest is observed, while the intermediate zone shows a positive trend with a most noticeable increase in the IP southwest. In summer, the highest values of the normalized trends are basically grouped in the south, with a decreasing length of maxima of dry spells over much of the IP, excepting Catalonia. In autumn, the northern IP and Portugal show increases, while there is a marked decrease in the middle toward the southeast. In winter, the Portuguese and southern coasts present a more noticeable 8 of 15

9 Table 5. Number of Significant Trends Found in Series of Seasonal Accumulated Rainfall for Each Rainfall Category a Index (Accumulated Rainfall) Negative Positive WT Spring Total rainfall 13(37.14%) 0(0.00%) 22(62.86%) Light 5(14.29%) 3(8.57%) 27(77.14%) Moderate 5(14.29%) 1(2.86%) 29(82.86%) Intense 13(37.14%) 1(2.86%) 21(60.00%) Very intense 9(25.71%) 0(0.00%) 26(74.29%) Summer Total rainfall 1(2.86%) 0(0.00%) 34(97.14%) Light 1(2.86%) 6(17.14%) 28(80.00%) Moderate 0(0.00%) 1(2.86%) 34(97.14%) Intense 4(11.43%) 0(0.00%) 31(88.57%) Very intense 4(11.43%) 1(2.86%) 30(85.71%) Autumn Total rainfall 3(8.57%) 1(2.86%) 31(88.57%) Light 4(11.43%) 8(22.86%) 23(65.71%) Moderate 2(5.71%) 4(11.43%) 29(82.86%) Intense 3(8.57%) 0(0.00%) 32(91.43%) Very intense 3(8.57%) 1(2.86%) 31(88.57%) Winter Total rainfall 6(17.14%) 0(0.00%) 29(82.86%) Light 5(14.29%) 5(14.29%) 25(71.43%) Moderate 11(31.43%) 1(2.86%) 23(65.71%) Intense 6(17.14%) 0(0.00%) 29(82.86%) Very intense 7(20.00%) 0(0.00%) 28(80.00%) a WT indicates without any significant trend. increase, while in the rest of the IP a decrease in the maxima of dry spells is observed. 5. Trend Analysis of the Intensity Indices 5.1. Accumulated Rainfal [21] Figure 4 shows the trends found by applying the M-K test to the series of seasonal accumulated rainfall for each rainfall category. [22] Figure 4a represents the case of the seasonal total accumulated rain. As summarized in Table 5, in spring and winter there were 13(37.14%) and 6(17.14%) significant negative trends, respectively, and no positive trends. There existed strong decreasing trends in spring in most of the IP, except in the Pyrenees zone. This finding is consistent with the negative trend reported by Serrano et al. [1999] for the IP accumulated rainfall in the month of March. Similar patterns to those of spring are observed in summer, but they are weaker and less significant. In autumn, a northwestsoutheast division of the IP is observed, with some positive trends in the south and more negative in the northern half. The latter is stronger and more significant, especially at the Barracas observatory. Winter trends are decreasing almost everywhere on the IP except for a small region in the east. This decrease agrees with the results of Klein Tank et al. [2002] for the winter accumulated rainfall in the south of Europe. Some observatories, i.e., Villameca, Grazalema, Pozo Alcón, and Torre de Juan Abad, showed a persistence of this significant negative trend from winter through to spring. There is also a region in the southeast with a slight positive trend that persisted throughout the year. [23] In the case of the accumulated rainfall in the light rain category (see Figure 4b), similar patterns are observed for spring and winter with fairly high values of the normalized trend. Both seasons present zones of positive trends in the northwest and southeast, and negative trends in the rest of the IP. In summer, most of the peninsula geography shows an increase in the values of summer accumulated light rain; only in the Catalonia region and at the Cáceres observatory was there no significant decrease of this index. In autumn, the increase is again fairly general as in the last season, but now the Atlantic coast observatories and Barcelona are added to the significant negative zone. The slight increase that was observed in the total accumulated rainfall in the southeast is now seen to be explained by the significant increase of the accumulated light rainfall in this zone in every season. [24] The trends of the moderate rain are shown in Figure 4c. Winter shows the most generalized behavior: of the 12 significant trends found, 11 are negative (distributed over the entire IP, with a major grouping in the peninsula center that also exists in spring) and 1 positive (Embalse de Fuensanta, in the southeast, which also showed significant increases in spring and autumn in the moderate category, and in winter in the light category). These high number of significant negative trends would allow one to state that there was a certain decreasing trend in the accumulated moderate rainfall in winter over the peninsula. In spring, there were 5 negative trends and 1 positive, indicating a decrease of the accumulated moderate rainfall in this season. In summer, the northern and western zones of the IP present increases, while the rest show decreases. In autumn, only four observatories, Pozoblanco, Embalse de Fuensanta, Barracas and Barcelona, present significant increases in this category. The rest of the IP present decreases. [25] In the case of the intense rainfall (Figure 4d), spring is the season in which most trends appear, with the negative sign predominating over positive (13 and 1, respectively, as observed in Table 5). Also here, and with the caution one must be present, one could state that there is a certain generalized pattern of decreasing accumulated rainfall of this category in spring. This result is consistent with the total rainfall findings. Indeed, the patterns obtained for each of the seasons are very similar to those of the total accumulated rainfall (Figure 4a). The stations of Villameca, with negative trends throughout the year, and Grazalema, with significant negative trends in spring, summer, and winter attract attention. Their results are consistent with previous findings for the maxima block seasonal precipitation at these locations [García et al., 2006]. [26] The case is somewhat similar for the very intense rainfall (see Figure 4e and Table 5): a predominance of the negative trends in spring (9 negative versus 0 positive) and winter (7 negative versus 0 positive). This is indicative of a certain generalized behavior analogous to the previous case of intense rainfall. The positive trend in southeastern Spain that was persistent for the rest of the accumulated categories is, however, no longer present. Villameca and Grazalema again present significant negative trends for the same seasons as in the previous case. This behavior extends to the total accumulated rainfall and to the maxima block seasonal precipitation [García et al., 2006] Proportion of Each Rainfall Category Relative to Total Rainfall [27] Figure 5 shows the trends found by applying the M-K test to the series of proportion of rainfall in each 9of15

10 Figure 5. Mann-Kendall test for proportion of each rainfall category respect to the total in spring, summer, autumn, and winter (from left to right): (a) light, (b) moderate, (c) intense, and (d) very intense. Upward triangles represent increasing trends, and downward triangles represent decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation). Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. Isolines correspond to interpolated values of the normalized trend (in percentage). Circles show trends with absolute value < 0.5. category relative to the amount of total rainfall for each station and season for the period [28] The case of the proportion of light rain is shown in Figure 5a. The positive trends prevail over the negative for every season (see Table 6). There were 9(25.71%) significant positive trends in spring, 12(34.29%) in summer, 11(31.43%) in autumn, and 9(25.71%) in winter. In spring, the northeastern zone of the IP and the station of Salamanca were the only places showing a decrease of the proportion of events of light rainfall. The rest presented a fairly generalized increase. In summer, the differences in the values were smoothed out: the negative zone was restricted to the Catalonian region, the Pyrenees, and, but now not significantly, Salamanca. In autumn, the Atlantic seaboard and Catalonia show negative trends, while the rest of the IP shows positive trends. In winter, only southern Portugal shows a decrease in light rain events relative to the total. For the rest of the IP, there were significant increases at many of the observatories. A case worthy of note is that of Villameca which shows significant positive trends in every season. This pattern of behavior was also observed in the number of days of light rain (see Figure 2b). In spite of these cases of more local behavior, one can state that there was a certain generalized increasing trend in the proportion of light rain relative to the total in all seasons over the study period. This implies that there were increasing numbers of light rain events, confirming the result described in the previous section. One could think that this result runs contrary to the suggestion of Karl and Knight [1998] that there is an intensification of the hydrological cycle. However, changes in rainfall are very regionally dependent and although there has generally been an increase globally in precipitation means and extremes [e.g., Groisman et al., 2005; Alexander et al., 2006], studies do show large areas with decreasing trends. Of particular relevance, Alexander et al. [2006] show decreases in heavy precipitation days, contribution 10 of 15

11 Table 6. Number of Significant Trends Found in the Series of Seasonal Proportion of Each Rainfall Category Relative to the Total Rainfall a Index (Proportion) Negative Positive WT Spring Light 2(5.71%) 9(25.71%) 24(68.57%) Moderate 2(5.71%) 6(17.14%) 27(77.14%) Intense 6(17.14%) 2(5.71%) 27(77.14%) Very intense 4(11.43%) 2(5.71%) 29(82.86%) Summer Light 0(0.00%) 12(34.29%) 23(65.71%) Moderate 2(5.71%) 1(2.86%) 32(91.43%) Intense 8(22.86%) 0(0.00%) 27(77.14%) Very intense 4(11.43%) 1(2.86%) 30(85.71%) Autumn Light 4(11.43%) 11(31.43%) 20(57.14%) Moderate 2(5.71%) 3(8.57%) 30(85.71%) Intense 3(8.57%) 2(5.71%) 30(85.71%) Very intense 4(11.43%) 2(5.71%) 29(82.86%) Winter Light 0(0.00%) 9(25.71%) 26(74.29%) Moderate 1(2.86%) 2(5.71%) 32(91.43%) Intense 5(14.29%) 2(5.71%) 28(80.00%) Very intense 5(14.29%) 0(0.00%) 30(85.71%) a WT indicates without any significant trend. from very wet days and intensity and increasing trends in the maximum number of consecutive dry days over the IP. These findings agree with our results. [29] The trends for the proportion of moderate rainfall events relative to the total are shown in Figure 5b. The number of significant trends is notably less than for the light rain category (see Table 6). In spring, there appears a decrease for the moderate events in an isolated zone of the north central IP. In the rest of the peninsula, there are many positive trends, which are particularly noticeable in the southeast and northwest. In summer, the west and the station of Barracas show an increase for this rainfall category, while the rest of the IP shows decreases. A division crossing the peninsula appears in autumn, with the northeastern half being positive in character, especially in Barracas and Barcelona, and the southwestern half having a far less marked negative character. In winter, there were negative trends in the northeastern half of the IP, part of the east, and the station of Beja. The rest of the IP presents an increase in the moderate rainfall category. One notices that there is again in the southeastern IP (involving several stations) a persistence of the positive trend from winter to spring. At Villameca, there are significant positive trends in spring, summer, and winter, as was observed for the case of the proportion of light rain relative to the total. There thus seems to be an increase in low rainfall events, which will have to be at the cost of the high rainfall events. [30] Figure 5c shows the results for the intense rain category. The maps for spring, autumn, and winter seem to be the inverses of those for the proportion of moderate rain, but with a generally greater number of significant trends (see Table 6). In summer, the only positive zone corresponds to the increasing trend for the Haro observatory and Pyrenees zone. The case of Villameca again stands out: it presents negative trends throughout the year, coherent with its increasing trends in every season in the light and moderate categories (see Figures 5a and 5b). [31] The proportions of very intense rain relative to the total are summarized in Figure 5d. The behavior is very similar to the previous case of intense rain. Again, as was the case for the intense category, Villameca stands out in its decrease of very intense rain events. The decreases in these two categories, despite the increase in the light and moderate categories, leads to a significant decrease in the value of the total accumulated rainfall in spring, autumn, and winter (see Figure 4a). The case is somewhat similar for the observatory of Grazalema in spring and winter. These results are consistent with the previously observed decrease in maxima block seasonal precipitation at these two observatories [García et al., 2006] Mean Precipitation per Rainy Event [32] Figure 6 shows the trends found by applying the M-K test to the series of mean precipitation per rainy event for each category, season, and observatory for the period [33] The trends found in the total rainfall series are shown in Figure 6a. As is observed from Table 7, the number of trends is considerable, especially in spring, summer, and autumn, when up to 25% of the series present an increasing or decreasing pattern of behavior. The negative trends predominate over the positive in every season: 9(25.71%) in spring, summer and autumn, and 7(20%) in winter. As one observes in the maps, in spring, summer, and winter there are significant decreases almost everywhere in mean precipitation per rainy event. In spring, there is an increase in this index in the northeastern zone and the Pyrenees, with a significant increase corresponding to Barcelona and San Esteban de Gozmar. This region has a decreasing trend in summer, and is practically limited to the Barcelona observatory in winter but without significance. In autumn, the Atlantic seaboard and the observatory of Cáceres present positive trends in the value of the mean precipitation per rainy event; the rest of the IP shows a decrease. This decrease in the value of mean precipitation per rainy day at many observatories is consistent with the decrease in intense rainfall events noted in the previous section. The following are some specific cases of the observed trends: Villameca again appears with significant negative trends throughout the year, especially marked in autumn; Grazalema, Málaga, Pozo Alcón, Embalse de Fuensanta and Pozoblanco, all in the south to east of the IP, also have negative trends in spring and winter; Barcelona presents a significant increase in spring, and Cáceres in autumn. [34] The light rain category shows again many negative trends in every season, especially in spring, autumn, and winter (see Table 7). Indeed, in spring and winter, there were the fewest series without any significant trend, 19(54.29%)in both cases, and also the most outstanding values of the relative trends. In Figure 6b, one can see that the positive trends at Embalse de Fuensanta and Torre de Juan Abad, and the negative trend at Haro persist all year round. Most of the rest of the trends are negative. There are also many observatories that show persistence of negative trends from winter to spring. Some of them also 11 of 15

12 Figure 6. Mann-Kendall test for mean precipitation per rainy day in spring, summer, autumn, and winter (from left to right): (a) total, (b) light, (c) moderate, (d) intense, and (e) very intense. Upward triangles represent increasing trends, and downward triangles represent decreasing trends. The size of each triangle is proportional to the magnitude of the normalized trend (normalized by the standard deviation). Black triangles represent trends significant at the 5% level, and gray triangles indicate trends significant at the 10% level. Isolines correspond to interpolated values of the normalized trend (in percentage). Circles show trends with absolute value < 0.5. present decreases of the mean value of light rain per rainy day in autumn. The 11 significant negative trends in spring, 8 in autumn, and 13 in winter would allow one to state that there is a certain decrease of the mean value of precipitation per light rainfall event, except at the Torre de Juan Abad and Embalse de Fuensanta observatories in the southwest of IP where an increase is observed. In summer, also in the Atlantic seaboard an increase is shown. For this light rain category, we had obtained increasing trends for both the accumulated rainfall and the number of days cases series. For those results to make sense in light of the decreasing behavior obtained now for the mean value per rainy event, it would seem that the growth in the number of days was relatively stronger than the growth in accumulated rainfall. [35] Figure 6c shows the case of the moderate rainfall. It shows that there are now fewer significant trends (as confirmed by the data given in Table 7), and that they 12 of 15

13 Table 7. Number of Significant Trends Found in Series of Seasonal Mean Amount of Precipitation per Wet Day in Each Category a Index (Mean Rainfall per Wet Day) Negative Positive WT Spring Total rainfall 9(25.71%) 2(5.71%) 24(68.57%) Light 11(31.43%) 5(14.29%) 19(54.29%) Moderate 4(11.43%) 0(0.00%) 31(88.57%) Intense 5(14.29%) 0(0.00%) 30(85.71%) Very intense 4(11.43%) 1(2.86%) 30(85.71%) Summer Total rainfall 9(25.71%) 0(0.00%) 26(74.29%) Light 2(5.71%) 2(5.71%) 31(88.57%) Moderate 6(17.14%) 2(5.71%) 27(77.14%) Intense 4(11.43%) 1(2.86%) 30(85.71%) Very intense 4(11.43%) 2(5.71%) 29(82.86%) Autumn Total rainfall 9(25.71%) 3(8.57%) 23(65.71%) Light 8(22.86%) 4(11.43%) 23(65.71%) Moderate 5(14.29%) 2(5.71%) 28(80.00%) Intense 3(8.57%) 2(5.71%) 30(85.71%) Very intense 2(5.71%) 3(8.57%) 30(85.71%) Winter Total rainfall 7(20.00%) 0(0.00%) 28(80.00%) Light 13(37.14%) 3(8.57%) 19(54.29%) Moderate 3(8.57%) 0(0.00%) 32(91.43%) Intense 4(11.43%) 0(0.00%) 31(88.57%) Very intense 3(8.57%) 0(0.00%) 32(91.43%) a WT indicates without any significant trend. now appear more in isolation. For the case of southeast Spain, one now observes a decrease in the mean values of rain in this category, consistent with the decrease that was observed in this zone for the case of the mean precipitation for total rainy events (see Figure 6a). [36] The trends for the category of intense rainfall are shown in Figure 6d. As one observes in Table 7, the number of negative trends relative to positive is fairly high for spring (5 negative trends, 0 positive) and winter (4 negative trends, 0 positive). The major trends, in particular the negative trends observed in Villameca all year round, and in Grazalema in spring and winter, appear again. The spatial patterns are very similar to those obtained for the intense accumulated rainfall (Figure 4d). [37] For the category of very intense rain, the significant trend patterns are more local (Figure 6e), although still apparently similar to the previous case. It is notable that the trends in Villameca and Grazalema appear again, in confirmation of the results obtained by García et al. [2006] for the rainfall maxima. Examples of significant local positive values are Cáceres Oporto in autumn. 6. Summary and Conclusions [38] Using the definition of frequency and intensity indices to study trends in daily precipitation is a key aspect of climate change assessment. In the last few decades, studies of this kind have been conducted for different regions around the world. We here examined the trends in six indices designed to highlight changes in precipitation at 35 stations covering the Iberian Peninsula for the period These indices were evaluated seasonally in five rainfall categories, chosen specifically to take the rainfall regime of this region into account. Hence the work involved the study of a great number of series (2940) in order to determine the evolution of the daily precipitation characteristics, and the spatial distribution of the trends that were found over the cited period. The results showed statistically significant and spatially coherent trends in some of the indices. Our findings are internally coherent, and correspond well with what has been reported by other workers in more global or regional studies (to the best of our knowledge, this is the first study of trends in frequency and rainfall intensity indices at a daily scale which covers the whole Iberian Peninsula). The patterns of behavior can be summarized as follows. [39] At many observatories over the IP, there were increasing trends in the total number of rainy days in summer and autumn. In spring and winter, there was a predominantly decreasing behavior, except in the southeast where the increase was evident all year round. Considering the evolution of this index for each rainfall category, one saw that there was an increasing trend in the number of days with light rainfall in every season over much of the IP, while number of days with moderate, intense, and very intense rainfall was diminishing, especially in spring and winter. [40] With respect to the length of dry spells, the pattern was not so generalized. However, one can observe a slight winter decrease of the medians in the south and of the maxima in the north. In spring, the maxima of length of dry spells presented a slight increasing trend in the southwest, and in summer somewhat more generalized decrease. As reflected in both the medians and the maxima, the southeast seemed to show negative trends in the length of dry spells, consistent with the positive trends in the number of rainy events due to the complementarity of these two indices. [41] The more generalized patterns of behavior found in the number of events were internally consistent. The increasing trends in the number of days of light rainfall in summer and autumn were sufficient to explain the aforementioned increase in the total number of rainy days. The spring and winter increase observed in the number of days of light rainfall was at the cost of the decrease in the rest of the rainfall categories (moderate, intense, and very intense) in these seasons. This implies that the real distribution of the daily peninsula precipitation is becoming ever more asymmetric. The decreasing trend observed in the maxima of length of dry spells in summer is coherent with the increasing trend in the total number of rainy days in that season. [42] For the accumulated rainfall, decreasing trends were observed in many observatories in spring and winter. The same was the case in summer, but to a lesser degree. In autumn, this index was decreasing in the northern half of the IP. The accumulated light rainfall was increasing in summer and autumn over much of the peninsula, and in the southeast all year round. On the contrary, the accumulated moderate, intense, and very intense precipitation was declining in spring and winter over much of the IP, and in the northern half of the IP in autumn. The behavior of these three rainfall categories had a major influence on that of the total accumulated rainfall. [43] The proportion of light rainfall relative to the total represented an increase at the peninsula level in every season. There were decreasing trends in spring and summer, 13 of 15