Description of the Study Area

The Utcubamba River basin, situated in northern Peru, in the Department of Amazonas (Figure 1), is located between 5° 32' 36'' and 6° 50' 49'' south latitude, and 72° 22' 49'' and 78° 29' 59'' west longitude. The main channel is approximately 250 km long and occupies an area of 6 650.07 km². The source is located in the Punta de Arena hill (Chachapoyas province) at 4 130 masl, and flows into the Marañón River (Bagua province) at 360 masl.

Figure 1 Study area in the Utcubamba River basin, Amazonas, Peru. 

The middle and lower basins are located in ecosystems with seasonally dry tropical forests, which include the valleys of the Marañón, Chamaya, Chinchipe and Utcubamba rivers (^{Linares-Palomino, 2004}). The upper basin corresponds to the montane pluvial forests located in the high portion of the eastern slope and the strip between the Huancabamba and Chinchipe rivers (^{León, Pitman, & Roque, 2006}).

The Utcubamba basin contains a total of 220 048 inhabitants, and its main urban centers are the cities of Leymebamba, Chachapoyas, Pedro Ruíz Gallo, Bagua Grande, Cajaruro, El Milagro and Bagua (^{INEI, 2009}). The main pollutant loads in the basin come from domestic wastewater in these large population centers, in addition to livestock and agricultural runoff.

Selection of Sampling Stations

A total of 43 sampling stations were selected (Figure 2) following and adapting the methodology established by Freshwater Ecology and Management to evaluate the ecological status of the Mediterranean rivers (^{Prat, Rieradevall, & Fortuño, 2012}). The main tributaries included were: San Miguel (EM-2), El Chorro (EM-4), Pauca (EM-6), Singache (EM-9), Cedrocucho (EM-11), Yuyac (EM-13), Rumiyacu (EM-15), Tosán (EM-17), Sonche (EM-20), Cocahuayco (EM-22), El Chido (EM-24), Tingo (EM-27), Magunchal (EM-30), Naranjos (EM-32), Honda (EM-33), Goncha (EM-37), San Juan (EM-40), and La Peca (EM-41).

Figure 2 Sampling stations (EM) in the Utcubamba River basin, Amazonas, Peru. 

Sampling and Data Collection

Water samples were collected between 2014 and 2015, during three sampling campaigns that coincided with the early rainy season (throughout the month of October 2014), the rainy season (March 2015), and the dry season (August 2015).

For the collection of samples, the monitoring protocol of the sanitary quality of surface water resources was followed, established by the General Directorate of Environmental Health of Peru (^{Digesa, 2007}). The physicochemical and microbiological analyses were conducted according to APHA procedures (APHA, 2005).

The parameters studied were established according to the polluting activities in the basin, which were 19 in total: dissolved oxygen (DO), electrical conductivity (CE), temperature (T), pH, turbidity (TURB), nitrates (NITRA), nitrites (NITRI), phosphates (FOSFA), sulfates (SULFA), ammonium (AMO), chemical oxygen demand (COD), 5-day biochemical oxygen demand (BOD), alkalinity (ALCA), chlorides (CLO), hardness (DUR), total coliforms (CT), fecal coliforms (CF), Escherichia coli (EC), and fecal enterococci (EF).

Statistical analyses

The data obtained were entered into a Microsoft Excel spreadsheet and analyzed with Past 3.0 software (^{Hammer, Harper, & Ryan, 2001}).

Essentially, two statistical methodologies were used: principal component analysis (PCA) and cluster analysis (CA). The PCA is used to synthesize information among the established variables, and eliminate collinearity among them, creating new variables (principal components - PCs) that are not correlated. In this way, the first principal component (PC 1) accounts for the highest proportion of the variance (^{Terradez, 2002}). The CA is a technique that uses information from a set of variables related to an object, to divide the data into classes based on their similarity. Once the similarity is measured, they are grouped into homogeneous groups that are different from each other. For the CA, the Euclidean distance was used as the distance of measurement, and the grouping method used was the Ward method, through which the variables or the sampling stations could be grouped in a hierarchical way so as to minimize a certain function and form a new group with the two objects with the lowest distance (^{Hair, Anderson, Tatham, & Black, 1999}). The main objective of the Ward method is to unite, at each step of the clustering process, the two classes that result in lower intra-class inertia ( ^{Pardo & Del Campo, 2007}).

In conclusion, the overall reasons why these two statistical tests were used are those noted by ^{Vieira et al. (2012)}, and are summarized below:

(1) Group the sampling stations according to water quality in order to optimize the number of sampling stations and thereby reduce the number of samplings for future evaluations.

(2) Identify the correlated variables in order to reduce the number of variables to be analyzed per sample, thereby reducing the cost of future monitoring.

Finally, the significant differences between the concentrations of the variables analyzed were evaluated with the help of correlation matrices for the three different sampling campaigns. The correlation coefficient established as significant was below 0.01 (p < 0.01).

Correlation of Variables and Sources of Contamination

To establish the existing correlation between the variables selected for this study, a correlation matrix was generated (Table 1), in which the statistically significant correlation coefficients (with a significance level of 0.01) are marked in bold. There were high correlations between all the microbiological variables (CT and CF, CT and EC, CT and EF, CF and EC, CF and EF, and EC and EF). The Environmental Protection Agency recommends the use of Escherichia coli and members of the genus Enterococcus as indicators of fecal contamination in surface waters (^{USEPA, 2000}). The appearance of large concentrations of Escherichia coli, which inhabits the intestines of homeothermic animals (^{Sunantha & Vasudevan, 2016}), whose reproduction outside the intestine of these animals is favored by the existence of adequate organic matter and pH conditions (^{Prescott, Harley, & Klein, 1996}; ^{Arcos-Pulido, Ávila-de-Navia, Estupiñán-Torres, & Gómez-Prieto, 2005}) is found in much of the Utcubamba River basin.

In this sense, high correlations between physicochemical and microbiological parameters are associated with anthropogenic contamination (^{Vieira et al., 2012}), derived from both domestic and livestock wastewater. This was reflected in the COD and BOD, which are correlated with each other and also showed high correlations with CF, whose origin can be animal or human. There are no wastewater treatment plants in any locality in the region, and it is noteworthy that the capital and most populated city in this area, Chachapoyas, which is not located in the Utcubamba Valley, pours its sewage into the river, through the Santa Lucía Creek and the Sonche River, tributaries of the Utcubamba. This increases the bacteriological and physicochemical levels in the water of the receiving basin (^{Chávez, Leiva, & Corroto, 2016}).

Likewise, the high correlations that occurred between BOD, COD, nitrates, phosphates, nitrites and electric conductivity, verified contamination sources similar to the previous ones, that is, of fecal origin. However, when evaluating the activities carried out throughout the basin, some differences were evident. Specifically, in the high and middle sections, between Sampling Station 1 (2 834 masl) and Station 26 (1 088 masl), the source of pollution that led to these high correlations of the elements mentioned resulted from the dumping of wastewater from the main population centers (Leymebamba, Chachapoyas and Pedro Ruíz), and cattle scattered in the area. The oxidation of nitrogenous wastes of human origin and excreta from animals and industrial effluents contribute to this situation (^{Chidya, Sajidu, Mwatseteza, & Masamba, 2011}). It is worth mentioning the extraordinary presence of two rainbow trout fish farms (Oncorhynchus mykiss) in these two sections of the basin. Although the hatcheries are local and do not offer much production, the discharge of their effluents into the Utcubamba River could be reflected in the concentration of nitrates and phosphates, and according to ^{Schulz, Gelbrecht and Rennert (2003)}, the standard remnant in trout farm manure is composed of 7 to 32% of total nitrogen, and 30 to 84% of total phosphorus. Another unique aspect in the river basin was the presence of two pig farms in the middle section. Most of the wastewater from the pork industry is rich in nitrogen and phosphorus compounds (e.g., ^{Martínez-García et al., 2009}; ^{Xu, Zhao, Zhao, & Zhang, 2015}), a fact that also contributed to the mentioned correlations. Similarly, the existence of a chicken farm in this middle section also added to the presence in the river of the compounds mentioned, especially AMO, which increases with the increase of stock in poultry farms (^{Rossignol, Paerl, Fear, & Braddy, 2011}).

In contrast, the correlations between the physicochemical variables mentioned in the lower section, between Sampling Station 27 (880 masl) and Station 43 (386 masl), corresponded to agricultural development. The dry valley of Bagua has had agricultural activity over the last 25 years, mainly for the seeding of rice in pools, and consequently a great percentage of the valley has been deforested for this purpose. In the provinces of Bagua and Utcubamba, this crop has an annual production of around 250 000 metric tons, making it the most important agricultural crop in the region in terms of volume and value of production (^{Peru-IIAP, 2008}). The seasonal distribution of NITRA, FOSFA, CLO and AMO in the lower section of the Utcubamba River is related to the application of chemical fertilizers in the agricultural areas of the drainage area, with diammonium phosphate (DAP), urea, and ammonium chloride being the most used in the region. Since nitrates are not well retained by soils because they are not used quickly by plants, they are lost through drainage (^{Martínez-García et al., 2009}; ^{Arias, Pérez, & Rueda, 2004}). DAP, a source of nitrogen (18%) and phosphate (20%) (^{Tanaka, Diagne, & Saito, 2015}), is universally used in rice cultivation (^{Wopereis-Pura, Watanabe, Moreira, & Wopereis, 2002}; ^{Solaraj, Dhanakumar, Murthy, & Mohanraj, 2010}; ^{Zhang, Gao, Fan, Lan, & Zhao, 2017}) and is considered the most widely used phosphate fertilizer (^{Azizullah, Nasir, Richter, Lebert, & Häder, 2011}). Urea contains 46% nitrogen (^{Zou, Lu, Li, & Li 2011}) and is usually used in a complementary way with DAP. Its drainage in soils and its direct drainage caused the high concentrations of these three chemical parameters in the water in Utcubamba. In addition to the large bacterial load that reaches this section of the basin, there is also the discharge of untreated domestic waters from the cities of Bagua Grande and Bagua, the most populated in the study section.

Finally, the high correlation found between DUR and ALCA was also common in several studies (e.g., ^{Segnini & Chacón, 2005}; ^{Chidya et al., 2011}) and is caused by the geology of the study area, which is composed mainly of calcites and dolomites, giving the pH of the soil a basic character.

PCA of the variables studied

The selection of the number of principal components (PCs) and the classification of the influence of the variables in each PC were based on the research line reported by ^{Vieira et al. (2012)} on the Lis River (Portugal), which was adapted. Thus, regarding the selection of the number of PCs, the eigenvalues ​​greater than 1 were chosen (^{Kaiser, 1958}). The first five PCs were selected accordingly, which represented 80.50% with respect to the original variance. The values ​​corresponding to each of these five PCs were classified based on the influence of the variables in each one, resulting in values ​​≥ 0.3 being those that contributed the most (Table 2). Thus, there were 13 variables that contributed the most, out of the initial 19. These data are similar to those obtained by ^{Singh, Malik, Mohan and Sinha (2004)} in his study of the Gomti River using multivariate statistical techniques, where as a result of the PCA their variables were reduced from 24 to 14, explaining 71% of the total variance of the data set.

In PC 2, the variables with the most influence (CF, EC, and OD) were associated with fecal contamination that occurs throughout the basin. The first two, CF and EC, were directly related to fecal waste, both human and animal (^{Gemmell & Schmidt, 2012}), and the DO can be linked to changing physicochemical conditions caused by suspended bacteria that are indicators of fecal contamination in water (^{Tshibanda et al., 2014}). Meanwhile, in PC 3, the presence of pH, DUR and ALCA, variables with the greatest contribution, was associated with the chemical characteristics of the water, resulting from the chemical properties of the calcareous soil existing in the basin. PC 5 was moderately influenced by pH, NITRO, COD and CLO. This group of associated variables was similar to that determined by ^{Arpajón et al. (2011)}, in their study conducted in the Almendrares River (Cuba), where the presence of large concentrations of nitrites in the water reflected organic contamination of fecal origin from populated areas.

In PC 4, the contribution of the TURB had its origin in the seasonality of the region, with the rainy season having a direct affect on this physical characteristic of water. In addition, variables such as pH and CLO reappeared, which could be a result of fecal contamination, however their presence together with the TURB prevents finding a common origin.

CA of the variables studied

The CA made it possible to create groups of variables according to their similarity (Figure 3). The groups formed (GF) were broadly the following:

Figure 3 Dendrogram for the variables of water quality. 

- GF 1: TURB, T, FOSFA, NITRA, BOD and COD. These variables, taken as a whole, are associated with pollution of agricultural origin, where FOSFA and NITRA are from common fertilizer used by farmers in the region. COD, and to a lesser extent BOD, are indirectly related to the effluents coming from agriculture, however, they are more closely associated with domestic wastewater (^{Shrestha & Kazama, 2007}). ).

Finally, it is necessary to highlight the presence of the TURB in this group, related to seasonal variations in the study area, with rainy summers that cause changes in the turbidity of the river water. In terms of the TURB, this GF is similar to PC 4 resulting from the PCA, while the variables COD and BOD correspond to PC 2.

- GF 2: pH, ALCA, DUR, SULFA and CE. These physical and chemical parameters are clearly related to the basic geology of the site, and its similarity in terms of the PCA would be PC 3.

- GF 3: NITRO, OD, CF, CT, EC, EF, AMO and CLO. This group is related to anthropogenic contamination, where coliforms are used by routine monitoring as an indicator of faecal contamination in water (^{Percival, Yates, Williams, Chalmers, & Gray, 2014}). In addition, the properties of EF as indicators of faecal contamination, and the indirect relationship that AMO and CLO have with them, associates these variables with anthropogenic contamination. Meanwhile, AMO and CLO also have a close link with the agricultural fertilizers used throughout the basin and its corresponding runoff into the riverbed (^{Sundaray, Panda, Nayak, & Bhatta, 2006}).

The GF 3 was found to be significantly associated with PC 2 in terms of the faecal origin of water contamination.

These groups are similar to those formed as a product of the cluster analysis in the Cértima River basin (Portugal) (^{Figueiredo et al., 2012}), which are primarily divided according to two groups: pollution derived from agricultural effluents and contamination derived from animal and human effluents.

PCA of the Sampling Stations

The same methodology previously used for the analysis of the variables was followed, but in this case the objective was to evaluate the spatial variation through the examination of the 43 sampling stations. In this case, the number of PCs from the PCA (with eigenvalues ​​greater than 1) was eight, which accounted for 80.18% of the variance (Table 3). Thus, the number of sampling stations was reduced from 43 to 10.

The results showed that PCs 1, 2, 3 and 5 grouped sampling stations (EM-32, EM-33, EM-36, EM-38 and EM-40) located in the lower section of the river, and were affected by agricultural surpluses from the cultivation of rice, and by domestic discharges primarily from the localities of Naranjos, Bagua Grande, Cajaruro and Bagua. In PC 4, only EM-24 obtained a value above 0.3, which is located in the El Chido Creek, a tributary on the right bank of the Utcubamba River, and has the particularity of being located downstream of a pig farm and a chicken farm, whose effluents go directly to this station. PC 8 (EM-15, EM-20 and EM-33) is composed of three tributaries that are affected by pollution from the discharge of solid waste upstream of them. The Sonche River (EM-20) is the main receiver of these wastes, which are generated in the city of Chachapoyas. Finally, PC 6 (EM-5 and EM-27) and PC 7 (EM-16) showed a relationship with the hydrogeomorphological transformation that have occurred at their sampling stations from the establishment of quarries, which have a negative impact, especially on the physical quality of the water. Despite this, these three stations were characterized by moderate pollution.

CA of the Sampling Stations

The groups from the CA were somewhat different than those of the PCA, resulting in three GFs in the cluster of sampling stations (Figure 4). The similarity with the PCA was mainly found in the lower section of the river, where sampling stations EM-32, EM-17, EM-40, EM-39, EM-38, EM-35, EM-34, EM-43, EM-42, EM-41, EM-37, EM-33, EM-36 and EM-24 were grouped into GF 3 (PCs 1, 2, 3, 4 and 5 of the PCA). Although stations 17 and 24 did not belong to this section of the basin, these presented common characteristics in terms of fecal contamination, either of human or animal origin. The dendrogram extracted from the CA reflected two other GFs, which were categorized according to their level of contamination: medium or low. This last division is the one reported by ^{Singh, Malik and Sinha (2005)}, who divided the Gomti River basin (India) into three groups according to the pollution they presented: high, moderate and medium. Thus, in the present study, GF 1 was mainly composed of tributaries with small volumes and without large anthropogenic effects. The contamination was accentuated in GF 2, consisting of sampling stations located in the main channel and in large tributaries. The stations located in the main basin were downstream of the first locality in the study (Leymebamba), EM-5, which discharges its domestic waste into the river. It is noteworthy that the EMs located in the middle section, such as 25, 26, 28, 29 and 31, which presumably should have a contamination of fecal origin similar to those grouped in GF 3, appeared in GF 2. This can be explained by this section of the basin having a greater slope than the others, with steep drop-offs, which ^{Domenech, Ollero and Sánchez-Fabre (2008)} describe as a "fluvial course with a closed canon valley." The waterfalls in this section oxygenate the water (^{Gonzalo & Camargo, 2013}) and help to decompose the accumulated organic matter coming from the farms and populations located upstream.

Figure 4 Dendrogram for the sampling stations.