Trend study and assessment of surface water quality in the Ebro River (Spain)

Depósito de investigación de la Universidad de Sevilla https://idus.us.es/ “This is an Accepted Manuscript of an article published by Elsevier in: JOURNAL OF HYDROLOGY on 2008, available at: https://doi.org/10.1016/j.jhydrol.2008.07.048” - 1 - TREND STUDY AND ASSESSMENT OF SURFACE WATER QUALITY IN THE EBRO RIVER (SPAIN) BOUZA-DEAÑO, R.; TERNERO-RODRÍGUEZ, M.*; FERNÁNDEZ-ESPINOSA, A.J. Department of Analytical Chemistry, Faculty of Chemistry, University of Seville. c/ Profesor García González, 1. Campus of Reina Mercedes. 41012 Seville (Spain). *terner[email protected]. Telephone: +34 954557170. Fax: +34 954557168 ABSTRACT 34 Physical-chemical and chemical variables were analysed in surface water samples collected every month over a period of twenty-four years. They were determined from thirteen sampling stations located along the Spanish Ebro River affected by anthropogenic and seasonal influences. The trend study was performed using the Mann-Kendall Seasonal Test and the Sen´s Slope estimator. Results revealed parameter variation over time due mainly to the reduction in phosphate concentration and increasing pH levels at the Ebro Basin during the 1981-2004 period. Exploratory analysis of data was also carried out by display methods (cluster analysis), and unsupervised pattern recognition (principal component analysis) in an attempt to differentiate between sources of variation in the water quality. PCA has allowed the identification of the following factors: geologic, climatic and anthropogenic. Spatial and seasonal sources of variation were identified that affect the quality and hydrochemistry of river water. Key words: Mann-Kendall, Sen´s slope, water quality, surface water, hydrochemistry, principal component analysis, cluster analysis. - 2 - 1. Introduction Sampling networks for monitoring the water quality of rivers constitute a great source of data for obtaining a local and temporal vision of the river state. In addition, a wide network gives an insight into the evolution of the ecosystem over time. This gives a better knowledge of its temporal state with regards to future evolution and forecasts if the number of variables analysed is extensive enough given that the study of seasonal or geographical evolution is quite complex (Simeonova, P. et al., 2003; Simeonov, V., 2002). Furthermore, challenging information is difficult for the public to digest and would require interpretation and/or simplification to be useful. The European Union has established a framework for water policies. The European Directive 2000/60/EC urges its members to conclude periodical trend studies of different water pollutants (Annexe V Section 2.4.4). This document considers the term “trend” as the monotonic variation of the pollutant concentration with regard to time. Observational and historical environmental time-series data are generally used for planning and designing environmental projects. There are a lot of statistical tools for trend and seasonal variations assessment (Wei, 2005; Shumway & Stoffer, 2006), but the procedure for calculating the trends has been widely discussed by the scientific community. There exist two groups of mathematical tools to calculate these trends (Zhang et al., 2006; Kundzewicz & Robson, 2004). The parametric method, based principally in linear and residual models, and the non-parametric method, such as the Mann-Kendall's Test (Mann (1945), Kendall (1975)) and the Sen’s slope estimator (Sen (1968)). The parametric methods, what are more powerful than non-parametric ones, but they require the data be independent and normally distributed, examine the linearity of the regressions (Montgomery (1982)), using the t test to calculate the significance of slope (Snedecor and Cochran (1980)). In the cases of data set with a seasonal component or with variables correlated, these techniques show false - 3 - positives. The majority of non-parametric methods are based on the Mann-Kendall's Test (MK), which assumes the observations are independent. The application of Mann-Kendall's Test for an n independent series and data (x1, x2, ..., xn) being identically distributed suggests the previous acceptance of the null hypothesis H0 (non existence of trend). However this does not apply to information that presents a seasonal component, such as the information of water quality. In 1982 Hirsch et al. developed a modification of the MK test, the Mann-Kendall's Seasonal Test (SMK). It consists of the sum of the individual MK’s values for each cyclic period, thereby removing the seasonal component. Using the Mann-Kendall's Test it is possible to determine the existence of an increasing or decreasing trend, but it is hard to measure if the method, Sen’s Slope Estimator developed by Sen (1968), allows the calculation of the slope of the regression line for each parameter over any time without the influence of outliers (Bouza-Deaño et al. 1999). In order to apply and check the use of non-parametric statistical methods, the aim of the present study consists of the detection and quantification of the water quality trends in a Spanish River using both the Mann-Kendall Seasonal Test and the Sen’s Slope Estimator. The objective is to obtain a deeper understanding of the evolution in water quality through the years. 2. Methods 2.1. Monitoring area The Ebro River (Figure 1) Map showing the water quality monitoring sites on the Ebro River) is the largest River of the Iberian Peninsula. It originates from a natural reservoir in a forested area (elevation about 1200 m) near the town of Fontibre in Cantabria, Spain, and it flows into the Ebro Delta (outlet). The river flows from the North Eastern peninsula and traverses 800 Km before terminating in the Mediterranean Sea. The river drains a catchment area of about 85924 km2. Segre, - 4 - Cinca, Aragón, Gallego and Zadorra are the main tributaries of the Ebro River. Zaragoza (population about 650000), Vitoria (population about 225000), Pamplona (population about 190000), Logroño (population about 145000), Lérida (population about 125000) and Huesca (population about 48000) are the major urban settlements on the banks of the river. The river is usually used as a supply of water for agriculture, cattle breeding (89.3%), domestic (7.2%) and industrial activities (3.5%) from a total supply of 18217 Hm3/year in 2004. In addition, the river receives domestic and industrial wastewater from numerous minor settlements along the bed. Discharges into the Ebro River vary at different locations, showing an increase downstream, probably due to inputs from the tributaries or natural recharge of the stream. The drainage of the Ebro basin extends over the NW-SE direction with similar climatic conditions throughout. During the summer months, the maximum annual temperature is up to 34ºC and during the winter months the minimum temperature drops to -5ºC. In the present study, a total of 13 sampling sites were selected. They correspond to sampling stations from the National Water Quality Control Network (ICA Network) directed by Spanish Government for the Ebro River (Spanish Ministry of the Environment). The sampling sites were the following: Cereceda (Site-1), Miranda (Site-2), Conchas de Haro (Site-3), Mendavia (Site-4), Castrón (Site-5), Pignatelli (Site-6), Zaragoza (Site-7), Presa Pina (Site-8), Sastago (Site-9), Mequinenza (Site-10), Ribarroja (Site-11), Ascó (Site-12) and Tortosa (Site-13). The first three sites (1-5) are located in low pollution regions with a low population density. Another six sites (6-9) are located in high pollution zones and are affected due to domestic drains and emptying of polluted tributaries into the river at this stretch. The last four sites (10-13) are in a downstream region with moderate pollution. 2.2. Monitored pollutants and analytical methods The data used in this study is part of the database maintained by the Spanish Ministry of the Environment. This monitoring network and sampling strategy was - 5 - designed to cover a wide range of pollutants. The monitoring program collects samples each month at the thirteen sampling stations. With this frequency we can monitor changes caused by the seasonal hydrological cycle over a 24 year period (1981-2004). The network provides enough data (3620 samples, due to 124 samples could not be sampling by different circumstances) to determine the quality of the river system, including tributaries and pollutants that have an impact on downstream. Preservation and transportation of samples to the laboratory were in accordance with standard method (APHA, 1992). Water temperature, Air temperature, Flow rate and Dissolved oxygen were measured in situ. The other parameters were analysed in the Central Laboratory of the Ebro Basin Authority (Confederación Hidrográfica del Ebro, Zaragoza, Spain) following the standard protocols (APHA, 1992). The quality of analytical data during the last years was ensured by using the UNE-EN-ISO 17025 Norm and Accreditation of the Central Laboratory. During the previous years it followed Good Laboratory Practices, such as standard protocols, blank measurements, spiked and duplicate samples. The laboratory regularly participates in international programs on quality control. The samples were analysed for 34 parameters listed in Table 1. - 6 - Table 1. Physico-chemical parameters determined and analytical techniques used in the study. Variable Abbreviation Analytical technique Units Total alkalinity T-ALK Titration mgl-1 Arsenic AS ICP - OES µgl -1 Calcium CA ICP-OES mgl-1 Water flow FLOW Area - velocity m 3 day -1 Cadmium CD ICP-OES mgl-1 Chloride CL Titration mgl -1 Cyanide CN Spectrophotometry µgl-1 Total coliform T - COLI MPN method MPN/100 ml Electrical conductivity EC Conductometry µScm -1 at 20ºC Total chromium T-CR ICP-OES µgl-1 Copper CU ICP - OES µgl -1 5-days biochemical oxygen demand BOD Potentiometry/DO probe mgl-1 Tensioactives TEN Spectrophotometry µgl -1 Chemical oxygen demand COD Redox titrometry (K2CrO4) mgl-1 Iron FE ICP - OES µgl -1 Phenols PHE Spectrophotometry µgl-1 Fluor F Ion selective electrode µgl -1 Mercury HG ICP - OES µgl -1 Dissolved hydrocarbons D-HID Spectrophotometry mgl-1 Potassium K ICP - OES mgl -1 Total suspended solids TSS Filtration/drying at 105ºC/weighing mgl-1 Magnesium MG ICP-OES µgl-1 Manganese MN ICP - OES mgl -1 Sodium NA ICP-OES mgl-1 Ammonium NH4 Spectrophotometry mgl -1 Nitrate NO3 Spectrophotometry mgl -1 Dissolved oxygen DO Potencipometry/DO probe mgl-1 Lead PB ICP - OES µgl -1 pH PH Potenciometry/pH probe pH units Phosphate PO4 Spectrophotometry mgl -1 Sulphate SO4 Spectrophotometry mgl-1 Water temperature W - TEMP Temperature probe ºC Air temperature A-TEMP Temperature probe ºC Zinc ZN ICP - OES µgl -1 - 7 - 1 The Water quality index (WQI) was also calculated. It was adopted by the Spanish 2 Government in 1983 (Ministerio de Obras Públicas y Urbanismo, 1983) and based on the 3 development by Provencher and Lamontagne from the Water Quality Service on the 4 Canadian Ministry of Natural Resources (Natural Resources Canada, NRCan) 5 (Provencher and Lamontagne, 1977). The WQI index uses 23 physical, chemical and 6 physico-chemical parameters to obtain a numerical indicator from 0 (poor water quality) 7 to 100 (high water quality). 8 9 2.3. Statistical data treatment and trend study 10 2.3.1. Parametric statistical tools 11 The Pearson product-moment coefficient (PPMC) was the first parametric statistical 12 technique applied to the data set. This univariate tool assumes that the two variables are 13 measured on at least interval scales, and determines the extent to which values of the two 14 variables are "proportional" to each other. To interpret the coefficients correlation, a 15 common source must be assigned to the correlated variables with regard to the main 16 sources of this River. Linearity was also verified by graphical examination. This procedure 17 consisted of verifying the linearity and observing the linear profile of the points (the pairs of 18 data x-y). Sometimes a false linearity is due to an accumulation of points plus an outlier. 19 Additionally, cluster Analysis (CA) was the unsupervised multivariate technique 20 used to classify the objects into categories or clusters based on their nearness or similarity 21 (Vega et al., 1998). It is the most common approach in which clusters are formed 22 sequentially by starting with the most similar pair of objects and forming higher clusters 23 step by step (Otto, 1998). Cluster analysis was applied to the data set with the aim of 24 grouping similar sampling sites (spatial variability) and spreading them over the river 25 (dendrogram). Hierarchical agglomerative clustering was performed on the normalised 26 data by means of the Ward’s method and using Euclidean distances (Dlink/Dmax) 27 (Wunderlin et al., 2001; Simeonov et al., 2003). 28 - 8 - Principal Components Analysis (PCA) was performed on the rearranged data 1 matrix, so that it explained the structure of the underlying data set (Liu et al., 2003). PCA 2 is a widely used tool in science and engineering. We performed the PCA with a Varimax 3 rotation, extracting the factors or PCs, which explain more than 4 – 6 % of the total 4 variance of the data set, and also more than 40 % of accumulated variance. Kaiser 5 criterion was also used (eigenvalue >1). Experimental data was standardised in order to 6 avoid misclassification due to wide differences in data dimensionality (Liu et al., 2003). 7 Standardisation tends to increase the influence of small variance and reduce the influence 8 of large ones. Furthermore, the standardisation procedure eliminates the influence of 9 different units of measurement and renders the data dimensionless. 10 Consequently, multivariate analysis of the data was performed through CA and 11 PCA (Wunderlin et al., 2001; Simeonov et al., 2003). All mathematical and statistical 12 computations were made using Excel 2003 (Microsoft, 2003) and STATISTICA 6.0 13 (StatSoft, Inc, USA, 2001) software. 14 15 2.3.2. Non-parametric statistical tools 16 Spearman´s R test (ρ) is a non-parametric measure of correlation, which assesses how 17 well an arbitrary monotonic function can describe the relationship between two variables 18 without making any assumptions about the frequency distribution of the variables. ρ is 19 simply a special case of the Pearson product-moment coefficient in which the data is 20 converted to ranks before calculating the coefficient. The raw scores are converted to 21 ranks, and the differences D between the ranks of each observation on the two variables 22 are calculated. ρ is then given by: 23 24 Where R(x) and R(y) are ranks of a pair of variables (x and y) each containing n 25 observations. 26 - 15 - 3.2. Multivariate statistical analysis 1 3.2.1. Cluster analysis 2 3 The water-quality data (3620 observations) was further subjected to different multivariate 4 statistical techniques to explore their temporal and spatial trends 5 Cluster analysis rendered a dendrogram (Figure 3. Dendrogram showing clustering 6 of sampling on the Ebro River) where all thirteen sampling sites of the river were grouped 7 into three statistically significant clusters at (Dlink/Dmax)x100 < 70%. The clustering 8 procedure highlighted two groups of sites in a very convincing way because the sites in 9 these groups have similar characteristics and natural source types. Cluster I (Sites 1-5), 10 cluster II (Sites 6-9) and cluster III (Sites 10-13) correspond to highest water quality sites 11 (clusters I and III) and lowest water quality sites (cluster II). This implies that for rapid 12 assessment of water quality, only one site in each cluster may be an accurate assessment 13 of the water quality for the whole network. 14 This classification is coincidental with the WQI medium values in each sampling site 15 as shown in Table 2, where sampling sites 6, 7, 8 and 9 have the lowest values. 16 It is evident that the CA technique is useful in offering reliable classification of surface 17 waters in the whole region and will make it possible to design a future spatial sampling 18 strategy in an optimal manner. Thus, the number of sampling sites in the monitored 19 network was reduced; and therefore cost, without loosing any value to the outcome. There 20 are other reports (Wunderling et al.. 2001; Simeonov et al. 2003) where this approach 21 was successfully applied in water-programs. 22 23 3.2.2. Principal Components Analysis 24 The Screen plot was used to identify the number of PCs to be retained in order to 25 comprehend the underlying data structure (Jackson, 1991; Vega et al. 1998). In the 26 present study, the Screen plot showed a pronounced change of slope after the four 27 eigenvalues. Rotation of the axis defined by PCA produced a new set of factors, or PCs. 28 - 16 - Each PC involves a subset of the original variables divided into groups. Thus, PCA result 1 from the present data set minimised the contribution of less significant variables. The 2 Varimax rotation reduces all of the variables to four different PCs, which explains more 3 than 80% of the total variance (Table 3). Variables with communalities lower than 0,5 were 4 removed due to it low significance. 5 6 - 17 - Table 3. Loadings of experimental significative variables (18) on the first four 1 rotated PCs for complete data set. 2 PARAMETER PC1 PC2 PC3 PC4 T-ALK 0,720 0,160 -0,448 0,010 CA 0,966 0,004 -0,058 -0,064 CL 0,946 0,081 0,190 0,032 EC 0,957 0,062 0,122 -0,011 BOD 0,016 0,300 0,051 0,761 COD -0,225 0,529 -0,121 0,620 FE 0,023 0,878 -0,059 0,060 F 0,575 -0,060 0,205 -0,217 K 0,824 0,041 0,178 0,098 TSS 0,094 0,922 -0,050 0,024 MG 0,951 -0,005 0,187 -0,103 MN 0,131 0,874 0,133 0,105 NA 0,923 0,076 0,190 0,075 NH4 0,067 -0,157 -0,077 0,743 NO3 0,760 0,169 -0,453 -0,030 SO4 0,940 -0,044 0,218 -0,043 W-TEM 0,239 0,008 0,902 0,013 A-TEM 0,262 0,026 0,856 -0,087 Eigenvalue 7,888 3,147 2,061 1,324 %Total Variance 43,82 17,48 11,45 7,35 %Variance 43,82 61,30 72,75 80,10 3 Tendrías que decir en algún lado qué significan las negritas y los subrayados o ponerlas todas en negrita ya que son 4 todas mayores de 0.75 como dices aquí abajo 5 6 Liu et al. (2003) classified the factor loadings as ‘strong’, ‘moderate’, and ‘weak’ 7 corresponding to absolute loading values of > 0.75 (bold values in Table 3), 0.75-0.50 8 (cursive values in Table 3) and 0.50-0.30. respectively. 9 PC1 explains 43,8% of total variance and has strong positive loadings for total 10 calcium, conductivity, magnesium, chloride, sulphate, sodium, potassium and nitrate. This 11 can be attributed to a mineral component of the river water and could be conditionally 12 named the “geogenic” factor. PC2 (17,5% of total variance), with high correlation of 13 suspended solids, iron and manganese, is the “turbidity” factor. PC3 (11.4%) has a strong 14 positive loading for water and air temperature, so it could be conditionally named the 15 “climatic” factor. PC4 (7,4%) explain only biological oxygen demand, ammonium and 16 chemical oxygen demand and could be conditionally named an “anthropogenic” factor. 17 - 18 - 3.3. Trend analysis 1 The Mann-Kendall Seasonal Test was applied in order to study trends of water quality 2 parameters for 24 years (1981-2004). In the cases in which a trend was detected (positive 3 or negative) quantified levels of the above mentioned trend were found by means of Sen's 4 slope. The results of the trends study are shown in Table 4. In 16% of the cases the 5 counterfoil information did not contain sufficient information to help the analysis of non - 6 parametric trends. 7 8 - 19 - Table 4. Results of the Sen’s Slope Test (n.t. = “not trend”, - = “not data enough”). 1 PARAMETER SITE 1 2 3 4 5 6 7 8 9 10 11 12 13 T-ALK n.t. 1.06 n.t. n.t. n.t. n.t. 0.31 n.t. n.t. n.t. n.t. 0.50 n.t. AS - <1E-5 <1E-5 - - <1E-5 <1E-5 <1E-5 - - - <1E-5 <1E-5 CA n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. FLOW n.t. n.t. n.t. -0.89 n.t. n.t. n.t. - n.t. n.t. n.t. n.t. n.t. CD - <1E-5 - - - 0.000 7 0.001 0 0.0030 - - - 0.012 5 0.006 8 CL -0.4 n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. CN - <1E-5 0.1796 1 <1E-5 <1E-5 <1E-5 <1E-5 0.1298 1 - <1E-5 <1E-5 <1E-5 <1E-5 T-COLI -100 -134 -424 n.t. n.t. n.t. n.t. n.t. -416 n.t. n.t. -234 -1320 EC 3.5 4.3 n.t. n.t. 6.0 3.8 2.4 n.t. n.t. 10.5 5.9 9.2 6.8 T-CR - <1E-5 <1E-5 - - <1E-5 <1E-5 <1E-5 - - - <1E-5 <1E-5 CU - <1E-5 0.080 - - <1E-5 <1E-5 0.025 - - - <1E-5 n.t. BOD n.t. 0.11 n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. TEN n.t. -1.7 -3.5 -1.7 -2.5 n.t. n.t. -3.6 -2.2 n.t. n.t. n.t. n.t. COD n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. FE - 5.0 - - - 4.4 5.3 8.7 - - - 1.7 n.t. PHE - n.t. -0.32 n.t. - -0.06 -0.10 -0.25 - - - -0.10 n.t. F 2.0 2.5 1.6 n.t. 1.7 2.5 4.1 n.t. 0.6 6.0 4.4 n.t. n.t. HG - <1E-5 - - - <1E-5 <1E-5 <1E-5 - - - <1E-5 <1E-5 D-HID - - 0.006 - - - n.t. -0.004 - - - 0.019 - -0.007 -0.003 K n.t. -0.03 n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. TSS 0.17 0.11 0.24 0.90 0.50 1.00 0.52 1.04 n.t. 0.13 n.t. -0.20 n.t. MG -0.23 n.t. -0.10 n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. MN 2.5 0.7 - - - 1.7 1.5 2.4 - - - 0.5 n.t. NA n.t. 0.27 n.t. n.t. n.t. 0.21 -0.43 n.t. n.t. n.t. n.t. n.t. n.t. NH4 - n.t. -0.008 n.t. n.t. n.t. n.t. n.t. n.t. - 0.005 - 0.007 -0.004 n.t. NO3 n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. n.t. DO n.t. - 0.050 0.078 n.t. 0.051 0.020 0.050 0.141 0.100 n.t. 0.073 0.038 n.t. PB - <1E-5 - - - <1E-5 <1E-5 <1E-5 - - - <1E-5 <1E-5 pH 0.015 0.017 0.019 0.020 0.017 0.017 0.014 0.021 0.023 0.022 0.022 0.021 0.019 PO4 -0.005 - 0.008 n.t. - 0.025 - 0.031 -0.029 -0.028 -0.047 - 0.050 - 0.010 - 0.019 -0.024 -0.026 SO4 n.t. 0.72 n.t. n.t. n.t. 1.08 -0.44 n.t. n.t. n.t. n.t. 1.40 n.t. W-TEMP 0.060 0.057 0.050 n.t. 0.042 0.103 0.181 0.089 0.156 n.t. 0.030 0.078 0.063 A-TEMP n.t. - 0.100 -0.057 n.t. 0.089 n.t. 0.272 n.t. 0.167 0.200 0.180 0.101 n.t. ZINC - 1.8 - - - 1.5 1.9 n.t. - - - 1.0 1.7 WQI -0.44 -0.44 n.t. n.t. -0.50 -0.24 -0.28 0.37 -0.29 -0.45 -0.20 n.t. -0.55 2 3 Of the studied trends, 48 % of the cases were not detected. A positive trend was detected 4 in 37 % of the stations, and negative in 15 %. Amongst the parameters that present a 5 major number of detected trends, the pH stands out. A concentration increase was 6 - 20 - detected on many of the studied stations: Cyanides (85 % of the stations), the temperature 1 of the water (85 %), electrical conductivity (69 %), fine (69 %), solid in suspension (69 %), 2 dissolved oxygen (62 %), total chromium (54 %) and arsenic (54 %). As for detected 3 negative trends, the most significant are phosphates (92% of the stations) and the WQI 4 (69 %). For parameters such as Ca wealth, Cl, BOD, COD, K, Mg, NO3, a trend was 5 detected on less than 80 % of the stations. 6 Phosphate is the parameter that shows the greatest degree of variation throughout 7 the studied period. The variations were superior to 3 % per year in all the cases, especially 8 in station 9, where there was an annual decrease of 11.6 %. Figure 5 (Yearly temporal 9 variation of phosphate (1981-2004)) details the annual evolution of this parameter for each 10 of the stations. A decrease in concentration can be clearly observed over time in all points. 11 The reason for the reduction in concentration of phosphates in the whole basin could be 12 due to the reduction of the quantity of phosphate base fertilizers in the agriculture. This is 13 mainly caused by modernisation in agriculture techniques by means of drip irrigation 14 system. 15 The light increase of the pH (0.5 % per year) in the whole basin must also be 16 highlighted. This is a very small variation in relation with the totality of the studied stations. 17 The reason for this variation has not been determined. 18 WQI decreased in nine sites, increased in Site-8 and showed no trend in Site-3, Site-19 4 and Site-12 (Fig. 6) during the study period (1981-2004). This means a decline of the 20 general water quality. However, in the last five years there has been an increase in WQI 21 parameter, which means a general improvement of the water quality. 22 4. Conclusions 23 River water quality networks generate multidimensional data that needs to be 24 statistically studied to analyse and interpret the underlying information. PCA has found a 25 reduced number of ‘latent’ variables (principal components) that explain most of the 26 variance of the experimental data set. A varimax rotation of these PCs led to a reduced 27 number of PCs, each of them related to a small group of experimental variables with a 28 - 21 - hydrochemical meaning: natural factor PC1 (geogenic), PC2 (turbidity), PC3 (climatic) and 1 anthropogenic factor PC4. 2 Cluster analysis has found spatial similarities in station variations across the river, 3 reducing the number of sites to three (one per group). The first group (sites 1-5) is located 4 at the top of the river (clean waters). In the middle of the river we get the second group 5 (sites 6-8, highly contaminated) and the last group is located at the end of the river, similar 6 to the first group (sites 9-13). Highly polluted waters located in the middle river are 7 surrounded by higher population density across the basin. 8 Trend analysis with the Mann-Kendall Test and Sen’s slope estimator, 9 demonstrated higher reduction in phosphate concentration across the whole river 10 (reduction >3% per year in all cases), due to a reduction of the quantity of this kind of 11 fertilizers. 12 WQI decreased in nine sampling sites during the study period (Site-1, Site-2, Site-13 5, Site.6, Site-7, Site-9, Site-10, Site-11 and Site-13) (Figure 6. Yearly temporal variation 14 of WQI (1981-2004)). It shows no trend in three sites (Site-3, Site-4 and Site-12) and an 15 upward trend in one (Site-8). 16 The combination of WQI and trend detection gives a unique insight into general 17 water quality evolution of each sampling site with time so that it is easily understood by 18 the general public. 19 Acknowledgements 20 The authors wish to thank the Ebro River Basin Authority: Confederación Hidrográfica del 21 Ebro (Zaragoza. Spain) for providing water quality data of Ebro River. 22 References 23 APHA. 1992. Standard Methods for the Examination of Water and Wastewater. eighteenth 24 ed. American Public Health Association. Washington. DC. 25 Berzas. J.J.;García. L.F.; Rodríguez. R.C.; Martín-Alvarez. P.J. 2000. Evolution of the 26 - 22 - water quality of a managed natural wetland: Tablas de Daimiel National Park. Water 1 Research 34. 3161-3170. 2 Bordalo. A.A.; Nilsumranchit. W.; Chalermwat. K. 2001. Water quality and uses of the 3 Bangpakong River (Eastern Thailand). Water Research 15. 3635-3642. 4 Bouza-Deaño, R.; Ternero-Rodríguez, M.; Fernández-Espinosa, A.J. 2004. Métodos no 5 paramétricos para la detección de tendencias de calidad de aguas. Aplicación a datos 6 históricos del río Ebro (España) 1981-2000. IV Congreso Ibérico de Gestión y 7 Planificación del Agua. Tortosa (España). ISBN:84-689-0052-4. 8 Briggs. J.C. 1978. Nationwide surface water quality monitoring networks of the U.S. 9 Geological Survey. In Establishment of Water Quality Monitoring Programs. American 10 Water Resources Association. San Francisco. California. 49-57. 11 Chang. N.; Chen. H.W.; Ning. S.K. 2001. Identification of river water quality using the 12 Fuzzy Synthetic Evaluation approach. Journal of Environmental Management 63. 293-13 305. 14 Confederación Hidrográfica del Ebro. http://www.chebro.es. June, 2006 15 Fore. S.A.; Guttman. S.I.; Bailer. A.J.; Altfater. D.J.; Counts. B.V. 1995. Exploratory 16 analysis of population genetic assessment as a water quality indicator I. Ecotoxicology 17 and Environmental Safety 30. 24-35. 18 Fore. S.A.; Guttman. S.I.; Bailer. A.J.; Altfater. D.J.; Counts. B.V. 1995. Exploratory 19 analysis of population genetic assessment as a water quality indicator II. Ecotoxicology 20 and Environmental Safety 30. 36-46. 21 Jackson. J.E.. 1991. A User´s Guide to Principal Components. Wiley. New York. 22 Gilbert. R.O. 1987. Statistical methods for environmental pollution monitoring. Van 23 Nostrand Reinhold Company. New York. 24 Grapentine. L. Marvin. C. Painter. S. 2000. Initial development and evaluation of a 25 sediment quality index for the Great Lakes region. Human and Ecological Risk 26 - 23 - Assessment 8. 1549-1567. 1 Hirsch. R.M.; Slack. J.R.; Smith. R.A. 1982. Nonparametric tests for trend in water quality. 2 Water Resources Research 18: 107-121. 3 Jarvie. H.P.; Lycett. E.; Neal. C.; Love. A. 2002. Patterns in nutrient concentrations and 4 biological quality indices across the upper Thames river basin. UK. The Science of the 5 Total Environment 282-283. 263-294. 6 Kendall. M.G. 1975. Rank correlation methods. 4th Ed.. Charles Griffin. London. 7 Kundzewicz, Z.W., Robson, A.J., 2004. Change detection in hydrological records – a 8 review of the methodology, Hydrol. Sci. J. 49 (1), 7-19. 9 Kwiatkowski. R.E. 1987. Water Quality Branch monitoring. an historical review. In Water 10 Quality Monitoring Design. Proceedings of the Technology Transfer Workshop. Technical 11 workshop series no. 6. Environment Canada. Ottawa. 5-17. 12 Kondratyev. S.; Gronskaya. T.; Ignatieva. N.; Blinova. I.; Telesh. I.; Yefremova. L. 2002. 13 Assessment of present state of water resources of Lake Ladoga and its drainage basin 14 using Sustainable Development indicators. Ecological Indicators 2. 79-92. 15 Libiseller. C.; Grimvall. A. 2002. Performance of Partial Mann-Kendall test for Trend 16 Detection in the Presence of Covariantes. Environmetrics 13. 71-84. 17 Liu. C.W.. Lin. K.H.. Kuo. Y.M.. 2003. Application of factor analysis in the assessment of 18 groundwater quality in a blackfoot disease area in Taiwan. Sci. Tot. Environ. 313. 77-89. 19 Liou. M.L.; Lo. S.L.; Hu. C.Y. 2003. Application of two-stage fuzzy set theory to river quality 20 evaluation in Taiwan. Water Research 37. 1406-1416. 21 Mann. H.B. 1945. Non-parametric test against trend. Econometrica 13: 245-249. 22 Ministerio de Obras Públicas y Urbanismo. La vigilancia de la contaminación fluvial. 23 Madrid 1983. 24 Montgomery. D.C.; Peck. E.A. 1982. Introduction to linear regression analysis. John 25 - 24 - Wiley. New York. 1 Otto. M.. 1998. Multivariate methods. In: Kellner. R.. Mermet. J.M.. Otto. M.. Widmer. H.M. 2 (Eds.). Analytical Chemistry. Wiley-VCH. Weinheim. Germany 916 pp. 3 Provencher, M. and Lamontagne, M.P. 1977. Méthode de determination d’un indice 4 d’appréciation de la qualité des eaux selon différents utilisations (IQE). Ministère des 5 Richesses naturelles. 6 Sen. P.K. 1968. Estimates of the regression coefficient based on Kendall´s tau. Journal 7 of the American Statistical Association 63: 1379-1389. 8 Shumway, Robert H.; Dtoffer, David S.. Time Series Analysis And Its Applications. 9 Springer Verlag ed., 2006. 10 Simeonov, P.; Einax, J.; Stanimirova, I.; Kraft, J., 2002. Environmetric modelling and 11 interpretation of river water monitoring data. Analytical and Bioanalytical Chemistry, 12 374(5), 898-905. 13 Simeonov. V.. Stratis. J.A.. Samara. C.. Zachariadis. G.. Voutsa. B.. Anthemidis. A.. 14 Sofoniou. M.. Kouimtzis. Th.. 2003. Assessment of the surface water quality in Northen 15 Greece. Water Res. 37. 4119-4124. 16 Simeonova, P.; Simeonov, V.; Andreev, G., 2003. Water quality study of the Struma river 17 basin, Bulgaria. Central European Jounal of Chemistry, 2, 121-136. 18 Snedecor. G.W.; Cochran. W.G. 1980. Statistical methods. 7th ed. Iowa State University 19 Press. Ames. Iowa. 20 Theil, H. 1950. A rank-invariant method of linear and polynomial regression analysis, Part 21 3, in Proceedings of Koninalijke Nederlandse Akademie wan Wetenschatpen A 53: 1397-22 1412. 23 European Union. Directiva del Parlamento Europeo y del Consejo 2000/60/CE de 23 24 Octubre de 2000 por la que se establece el marco comunitario de acción en política de 25 aguas. Diario oficial L017. 19/01/2001- p0039-0039 IT. 26