Integrated modeling of water quantity and quality in the Araguari River basin, Brazil

. The Araguari River basin has a huge water resource potential. However, population and industrial growth have generated numerous private and collective conflicts of interest in the multiple uses of water, resulting in the need for integrated management of water quantity and quality at the basin scale. This study used the AQUATOOL Decision Support System. The water balance performed by the SIMGES module for the period of October 2006 to September 2011 provided a good representation of the reality of this basin. The parameters studied were dissolved oxygen, biochemical oxygen demand, organic nitrogen, ammonia, nitrate and total phosphorus. The coefficients of biochemical reactions, sedimentation rates and sediment dissolved oxygen release for this period were calibrated and validated in the quality modeling using the GESCAL module. A sensitivity analysis indicated that the coefficients of carbonaceous matter decomposition, nitrification, water temperature, and sediment oxygen demand interfered more significantly in the variables of state. To prevent eutrophication in the Nova Ponte reservoir and in the other cascade reservoirs, the local River Basin Committee should adopt restrictive actions against the use of agricultural fertilizers. On the other hand, in the sub basin of the Uberabinha River, new alternatives for public water supply to the city of Uberlândia and improvements in the treatment efficiency of the main wastewater treatment plant (WWTP) should be proposed, since the biochemical oxygen demand, ammonia and total phosphorus failed to meet the requirements of COPAM (2008) in the driest months.


INTRODUCTION
In developing countries, such as Brazil, which lack financial resources for basic sanitation and proper wastewater treatment, the problem of dissolved oxygen consumption in waterways after wastewater has been discharged into them is still significant, justifying the use of the assimilative capacity of waterways to complement the treatment process.Sustainable development and rational water use require the existence of a proper relationship between water quantity and quality.In this context, joint mathematical modeling allows for the diagnosis and prediction of impacts resulting from multiple water uses and the discharge of pollutant loads.
Numerous researchers have designed a variety of models and Decision Support Systems (DSS) that are useful for water resource planning and management at the basin scale.It is well known that the main focus of computational tools is quantitative water resource management and planning, considering the increasing demands and need to implement optimal rules for the operation of water resources.In this context, with different mathematical complexities, the main quantity models that stand out are: HEC-HMS (Klipsch & Hurst, 2007;Fan et al., 2009) and the MIKE SHE (McMichael et al., 2006) models, designed to simulate the precipitation-runoff processes of watershed systems which integrate all the important processes of the hydrologic cycle at catchment scale.HEC-ResSim and WRAP (Wurbs, 2005) models are used to model reservoir operations at one or more reservoirs and the interactions with rivers.MODFLOW (Rodriguez et al., 2008;Xu et al., 2012) and IRAS (Salewicz & Nakayama, 2004;Matrosov et al., 2011) models are used to simulate flow of groundwater through aquifers interactive river-aquifer simulation.However, environmental concerns regarding water quality at the basin scale, driven by the continuous discharge of domestic and industrial wastewater, have led to the design of increasingly complete water quality models (De Paula, 2011).These models have been in use since the development of Streeter & Phelps's classical model (Streeter & Phelps, 1925), which is a benchmark in the history of sanitary and environmental engineering.Several other models have been designed with increasing complexity and number of modeled variables.Those models can be used to simulate different water quality problems.For example, while the Qual2E model (Palmieri & De Carvalho, 2006;Chapra, 2008) and its updated version Qual2K model (Von Sperling, 2007;Chapra et al., 2008;De Paula, 2011) are used to model water quality in river and stream, WASP model (Lai et al., 2012;Zhang & Rao, 2012;Yenilmez & Aksoy, 2013) has been used to examine eutrophication in lakes or streams and heavy metal pollution in rivers.AQUATOX model (Mamaqani et al., 2011;McKnight et al., 2012) is a valuable tool in ecological risk assessment for aquatic ecosystems.
This brief review reveals the marked existence of river and reservoir water quality models that are not linked with any DSS in the quantitative management and planning of water resources.According to Paredes-Arquiola et al. (2010a), many scientific researches disregard the interactions between qualitative and quantitative aspects in water resource management at the basin scale.Due to this situation, many researchers around the world, e.g., Dai & Labadie (2001), Paredes & Lund (2006), Argent et al. (2009), Zhang et al. (2010), Paredes-Arquiola et al. (2010a, 2010b), Zhang et al. (2011), Sulis (2013) and Welsh et al. (2013), are focusing on relating water quality within a DSS in water management at a basin scale.
According to the State Environmental Foundation, the state of Minas Gerais has the highest water resource potential in Brazil and accounts for the generation of 18.5% of all the electricity produced in the country.Nevertheless, there is a lack of scientific research on the integrated management of water quantity and quality at the basin scale.Many water resource management proposals have been put forward by local river basin committees.However, these proposals are not underpinned by integrated studies of water quantity and quality in lentic and lotic environments, but instead focused only on the implementation of quantitative and qualitative telemetric information systems, on user registration and updating, on the creation of criteria for granting water rights, on charging for the use of water and on payment to the surrounding municipalities, watercourse guidelines, conflict prognosis between demands and capacities, and the creation of environmental protection units.
In this context, based on the AQUATOOL Decision Support System (DSS), this article presents an integrated modeling of water quantity (using the SIMGES module) and quality (using the GESCAL module) of the three main watercourses of the Araguari River basin (Araguari, Quebra-Anzol and Uberabinha rivers).Based on water flow and water quality data monitored by the National Water Agency (ANA), the Minas Gerais Water Management Institute (IGAM) and the Minas Gerais Electric Company (CEMIG), this article presents the results of the water balance and calibration of the water quality model for the period of October 2006 to September 2009, and its validation for the period of October 2009 -September 2011.The calibration and validation of the biochemical reaction coefficients, sedimentation rates and sediment oxygen demand will serve as a basis for future studies on quantitative or qualitative interventions in this basin.
The coefficients that are part of the natural selfpurification process of a watercourse, be it lentic or lotic, have distinct influences on the final water quality in the water system.Thus, using the factor model, this study performed a sensitivity analysis of the four main coefficients of biochemical reactions involved in the modeling (the re-aeration coefficient K a , decomposition coefficient of CBOD K d , coefficient of decomposition of organic nitrogen KN oa , and coefficient of ammonia nitrification KN ai ), of the water temperature (Temp) and the sediment oxygen demand (S OD ).

AQUATOOL DSS
There are few computational tools or models that simulate water quality linked to quantity at a basin scale.Andreu et al. (1996) developed a DSS called AQUATOOL, which is an interface for editing, simulating, reviewing and analyzing basin management simulation models, including a lentic and lotic water quality simulation module, that is widely used in Europe, Africa, Asia and Latin America (Paredes-Arquiola et al., 2010a, 2010b;Nakamura, 2010;Sulis & Sechi, 2013).The GESCAL and SIMGES modules are intercon-nected, sharing georeferenced quality and quantity data through a graphical interface (Paredes-Arquiola et al., 2010a).Thus, hypothetically considering a basin with multiple and transient uses, water quality can be simulated for any simulated outfall, recharge and environmental flow scenario.

SIMGES module
In this study, the quantitative water management module SIMGES was used in the water balance model in the Araguari River basin.In this water balance was considered the flow in rivers and reservoirs at the basin scale, based on the spatial and quantitative definition of outfalls (point wise outfall for irrigation, industries and human consumption).Simulations were performed by means of a network flow optimization algorithm, which controls the surface flow within the basin while aiming to minimize the deficits and maximize the liquid levels in reservoirs to meet irrigation, human consumption and hydropower demands.

GESCAL module
In order to simulate water quality linked to quantitative management in lentic and lotic environments previously defined in the SIMGES module, Paredes-Arquiola et al. (2009) developed the water quality module GESCAL.Although GESCAL allows modeling eutrophication, temperature, toxics and conventional contaminants, in our case, due to the lack of data and planning purpose of the study, the contaminants modeled were DO, CBOD, organic nitrogen, ammonia, nitrate and total phosphorus.In the modeling process adopted in this study, the relationship between nitrogen cycle and carbonaceous organic matter and the effect on dissolved oxygen, and total phosphorus as an arbitrary parameter was considered, according to the scheme illustrated in Fig. 1.

Study area: Araguari river basin
The Araguari River basin (Fig. 2) is located in the western region of the state of Minas Gerais, Brazil (18°20'-20°10'S, 46°00'-48°50'W).Headwaters are located in Serra da Canasta National Park, in the municipality of São Roque de Minas, covering 475 km to its mouth in the Parnaíba River (which is a tributary of the Grande River, that belongs to Transnational Paraná River basin).This basin covers an area of approximately 22,000 km 2 , with altitudes ranging from 465 m to 1,350 m and rainfall exceeding 1600 mm year -1 .The weather condition is warm, with the dry season between May and September and a wet season between October and April (Rosa et al., 2004).It has a resident population of approximately 1.2 million, distributed in 18 municipalities, 14 of which discharge their wastewater into the basin (Fig. 2).Only the municipalities of Araxá, Nova Ponte, Patrocínio and Uberlândia (which accounts for approximately 70% of the total population in the basin) have wastewater treatment plants (WWTPs), while the other 10 municipalities discharge their untreated wastewaters directly into the surface water bodies.According to the IGAM, surface and groundwater demands allocated in 2006 for human consumption, irrigation, industry, and livestock watering were 250.6 and 3.6 hm 3 year -1 , respectively.This basin has six hydroelectric power stations (HP), the four largest ones located on the Araguari River with cascade reservoirs (Fig. 2).The first one, situated on the upper Araguari River, is a regulation reservoir with a storage capacity of 12,792 hm 3 (Nova Ponte HP), while the other three reservoirs, located on the lower Araguari River, are trickle reservoirs (from up to downstream, Miranda HP, Capim Branco HP 1, and Capim Branco HP 2).There are also two small hydroelectric power stations (SHP) situated on the  Uberabinha River (Martins SHP and Malagone SHP).However, in the 2006-2011 period they had not yet entered into production that, for modeling purpose, make us to consider this region as a simple river segment.
In the 1980s, the joint effect of economic valuation of soybeans and the scientific discovery of suitability of the crop to the soil of the Araguari River, transformed the region through the practice of a modern agriculture, associated with the intensive use of phosphate fertilizers and agrochemicals.Also, the presence of phosphate rocks in the region contributes to the existence of that nutrient from their natural deposits (EPE, 2006;Rosolen et al., 2009;Flauzino et al., 2010;Danelon et al., 2012).Figure 2 shows that the basin may be divided into 18 sub basins, whose

Quantity modeling
The initial procedure in the quantity modeling was to outline the topology of the model using AQUATOOL, which basically corresponds to the situational diagram of the Araguari River basin, including the unscaled elements of the model, as illustrated in Figure 3.To improve visualization, the elements that represent the smaller tributaries and the diffuse distribution along the Quebra-Anzol, Araguari and Uberabinha rivers were removed from Figure 3.
In the quantity and quality modeling processes, the three main watercourses of this basin (Araguari, Quebra-Anzol and Uberabinha rivers) were divided into 20 segments, each of which was identified by a numbered node upstream and another numbered node downstream (Fig. 3).

Data input
Based on the water flow data monitored by the National Water Agency and the Minas Gerais Electric Company (Fig. 2), a text file was arranged containing the model's quantity input data for the calibration and validation periods.According to Figure 3, all the tributaries and point wise discharges of domestic wastewater with and without the wastewater treatment plant (WWTP) are identified as inputs.

Quebra-Anzol and Araguari rivers
The quantity data of the upper Araguari River and upper Quebra-Anzol River were used directly as input data in the simulation.However, the diffuse and point wise inputs from the other tributaries were obtained from the specific outfall in m 3 s -1 km -2 (Eq.1), taking into account the existing quantity data of the upper Araguari and Quebra-Anzol rivers and of the four cascading hydroelectric plants (data on turbine flow, downstream flow and volume variations in the reservoir, which enabled the flow upstream from each hydropower plant to be estimated).
( ) where: Q i = inflow i Q upstream = flow at any point upstream Q downstream =flow at any point downstream from the inflow Q i ; A n = total area between two monitoring stations, A i = area contribution of the inflow i, obtained by means of a GIS tool that enables the simultaneous acquisition of the area from the perimetral outline.

Uberabinha River
Existing data for the upper Uberabinha River were used directly as input data in the simulation of the model.The absence of water flow data from the mouth of this sub basin and from the two small hydroelectric plants precluded the use of the specific discharge method to estimate the diffuse and point wise flow rates.Thereby a specific rainfall-runoff model is needed for the water balance in this sub basin.
The curve number method (CN) for urban sub basins was used in our study (SCS, 1986).This is a distributed model widely accepted worldwide due to the reduced number of parameters and their relationship with the physical characteristics of the basin (Tucci, 2005;Rezende, 2012).
The HBV model developed by Bergström (1995) was used for the rural sub basins.This is a semidistributed model that is part of a range of models which use the most important surface runoff processes by means of a simple structure and with a reduced number of parameters.The model functions on a daily or monthly time scale and uses precipitation, groundlevel air temperature and average monthly evapotranspiration as input data (Hundecha & Bárdossy, 2004;Das et al., 2006).Detailed descriptions of the equations used in the HBV model are given by Bergström (1995) and Paredes-Arquiola et al. (2011).
The parameters of the HBV model were calibrated using the evolutionary algorithm for calibration, SCE-UA (Shuffled Complex Evolution method, University of Arizona) (Duan et al., 1992).To this end, the results of the time series of surface flow obtained from the HBV model were compared with the existing time series of surface flow in the upper Uberabinha River.Self-calibration was performed adapting the original code of the SCE-UA algorithm from Duan et al. (1992) and reprogrammed in a Visual Basic platform.Each assessment of the objective function implies the execution of the HBV model.This algorithm has been used successfully to solve nonlinear problems in various applications of hydrological models at the basin scale (Paredes-Arquiola et al., 2011).
In our study, the model was applied to the sub basin corresponding to the single water flow monitoring station existing in the upper Uberabinha River (Fig. 2), whose area of contribution is 801.6 km 2 .Due to the similarity of climate, geology, land use and occupation throughout the Uberabinha River sub basin, the initially calibrated parameters for this sub basin were used as input data to estimate the surface flow into the other rural sub basins.As it can be seen in Fig. 2, the Bom Jardim River sub basin (394.6 km 2 ) and the Das Pedras River sub basin (389.4 km 2 ) are the main rural sub basins.

WWTP
The WWTP's inflows were calculated using the drinking water flow distribution equation multiplied by the coefficient of return, which, according to the Brazilian standards ABNT: NBR 9649 (1986) and ABNT: NBR 14486 (2000), is set to 0.80 for these situations in which there are no observed data available.

Point wise demand with and without consumption
The data on granted and georeferenced surface water demands for human consumption, irrigation, industry, and livestock watering were obtained from the IGAM, based on 2006 data.Data relating to variable requirements for hydroelectric purposes were obtained from CEMIG.

Water balance
The water balance was determined using the SIMGES module after completion of the topographic map, along with inputs of quantity data required for each element of the model, which include the point wise surface consumption demands, point wise requirements for hydroelectric purposes without consumption, point wise entries of tributaries, point wise effluents with and without WWTP, and the diffuse inputs from the main rivers (Quebra-Anzol, Araguari and Uberabinha).Various input data on storage reservoirs and hydroelectric plants are also essential in modeling, such as the dead volume of each reservoir (hm 3 ), volume set aside in each reservoir at the beginning of the simulation (hm 3 ), maximum storage capacity in each reservoir (hm 3 ), base depth (m), minimum turbine depth (m), energy coefficient (GW hm 3 m -1 ), maximum turbine requirement (m 3 s -1 ), evapotranspiration for each month, and bathymetric data of the reservoirs.

Quality modeling
In AQUATOOL, quality modeling with the GESCAL module is performed after quantity modeling.Another text file was created containing data on the water quality of tributaries and point wise discharges of WWTP treated and untreated domestic wastewater.The text file was introduced into the GESCAL module to start the simulations.
The data on water quality of the tributaries and the WWTPs were obtained from IGAM and CEMIG.
With respect to the 10 municipalities that discharge their untreated wastewaters directly into the water courses (approximately 30% of the total population of this basin), the water quality was estimated based on the characteristics of raw wastewater.The per capita gross load of BOD of 54 g day -1 was adopted based on the recommendation of the Brazilian standard ABNT: NBR 12209 (2011), in the absence of available measured data.Likewise, the per capita gross pollutant loads of organic nitrogen, ammonia, nitrate, and inorganic and organic phosphorus were estimated, to be 5.0, 7.0, 0.5, 1.0 and 1.5 g day -1 , respectively.These estimates are based on the numerous experimental results reported by several authors, such as Tchobanoglous et al. (2003) and Von Sperling (2007).The number of inhabitants per municipality was obtained from census of the Brazilian Institute of Geography and Statistics (IBGE, 2013).
The simulated water quality parameters are: dissolved oxygen, biochemical oxygen demand (BOD 5 ), organic nitrogen, ammonia, nitrate and total phosphorus.Due to the absence of eutrophication in the reservoirs for the time series under study, the modeling of water quality assumed thoroughly mixed reservoirs, for which the simulations were performed adopting only the upper region of the epilimnion.Although we thought that the behavior of the water quality in the reservoirs are enough defined with the model, overall, based on the available data, new information regarding temperature profiles and dynamics of nutrients could improve the model of the reservoir.Generally, the model is related to phosphorous and the internal sediment source of phosphorous.In this case, the developed CSTR model could be incremented to two layer model and could include the effect of the sediment, improving the knowledge of the system and the robustness of the model.Fig. 4 shows the line diagram of the integrated modeling of water quantity and quality in the Araguari River basin.This plot shows the longitudinal distance between all the elements of the model, the longitudinal distance of the 20 river segments, and the location of the water quality monitoring stations used in the calibration model and its validation process.To calibrate the model in each segment of the river, existing water quality data was used in the node downstream from the segment (Figs. 3, 4).The GESCAL module allows the re-aeration coefficient in each segment of the river to be obtained by the Covar method (Von Sperling, 2007;Paredes-Arquiola et al., 2009) or through the direct introduction of its value in the calibration process.The Covar method (empirical equations that depend on the mean flow velocity and the net depth) showed a good fit between observed and simulated dissolved oxygen data only in the headwater segments of the rivers involved.Table 1 identifies the 20 segments, the longitudinal length of each segment, and the hydraulic relationships used in the headwater segments.

Calibration, validation and sensitivity analysis
In this study, the coefficients of biochemical reactions, sedimentation rates and sediment oxygen release in the 20 segments identified in Figures 3 and 4 were calibrated through a process of trial and error.The coefficients of reactions and sedimentation rates include: re-aeration, decomposition of carbonaceous organic matter, sedimentation rate of carbonaceous organic matter, hydrolysis of organic nitrogen, sedimentation rate of organic nitrogen, ammonia nitrification and denitrification, phytoplankton growth, phytoplankton death/respiration, phytoplankton sedimentation rate, organic phosphorus decay rate and organic phosphorus sedimentation rate.
A sensitivity analysis was performed of all the segments defined in Figures 3 and 4 in view of the changes in the input values of the four main previously calibrated coefficients of reactions (reaeration coefficient K a , coefficient of carbonaceous organic matter decomposition K d , decomposition coefficient of organic nitrogen KN oa , and coefficient of ammonia nitrification of KN ai ), sediment oxygen demand S OD and water temperature Temp.
Unlike what was done in the calibration process, in which each segment was calibrated separately, using the data observed in the node downstream from the segment as the base for calibration, the sensitivity analysis joined two or more sequential segments in some cases in which the simulated and calibrated values of the node downstream from the last sequential segment were used as the standard in the analyses.The analyses of sequential segments were organized as follows: Araguari segments (1), ( 2), (3-4-5-6), (7-8) and (9-10-11) correspond, respectively, to the nodes 2, 3, 7, 9 and 12; Quebra-Anzol segments (1) and (2-3-4) correspond, respectively, to the nodes 15 and 3; finally, Uberabinha segments (1), (2-3) and (4) correspond, respectively, to the nodes 19, 21 and 13.
The factor method used in the sensitivity analysis enabled the assessment of changes in the concentrations of quality parameters based on the simulta- Table 1.Identification of the 20 segments, longitudinal length (L) of each segment, and hydraulic relationships used in the headwater segments.Q: average flow (m 3 s -1 ); u: average velocity (m s -1 ); h: average depth (m); b: width of the transverse section (m); α1, β1, α2, β2, α3 and β3 are coefficients of the potential relationships of u = f(Q), h = f(Q) and b = f(Q), adjusted by optimizing the Nash-Sutcliffe efficiency coefficient (Nash & Sutcliffe, 1970).

Segment
Between nodes  neous variation of K a , K d , KN oa , KN ai, and S OD by ± 10% from their calibrated value, which is called a dual-level analysis.According to Loucks et al. (2005) and Nakamura (2010), in dual-level analysis, 2 n different simulations are performed, where n is the number of coefficients.However, for each river segment, 1 x 2 x 2 n simulations were made, in which the number 1 corresponds to the number of + andpairs, the first number 2 corresponds to the two simulations +10% and -10%, and n corresponds to the number of coefficients (n is equal to 4 in the segments that have no sediment oxygen demand S OD ).With respect to temperature, a relative method was used to assess changes in the concentrations of quality parameters on the isolated variation by +10% and -10% from water temperature.

Quantity modeling
Figure 5a illustrates the variation of simulated flow during the period of calibration and validation of the main sections in the basin.The flow at the mouth of Uberabinha River varies from 54.28 to 310.81 hm 3 month -1 .In the upper Araguari River (node 2) and upper Quebra-Anzol River (node 15) vary, respectively, from 80.06 to 603.56 hm 3 month -1 and from 83.89 to 761.98 hm 3 month -1 , while at the mouth of the Araguari River basin (node 13) the flow varied from 799.23 to 2654.52 hm 3 month -1 .
Figures 5b, 5c and 5d, respectively, illustrate the longitudinal profiles of the simulated flows of the Araguari, Quebra-Anzol and Uberabinha rivers in the driest and rainiest months, along with the maximum flow observed, minimum flow observed, average flow observed and 25-75% percentile observed.Downstream to the Nova Ponte reservoir, the box-plot graph (Fig. 5) shows that the extreme model scenarios-driest and rainiest months-are between 25-75% percentiles.In the upper Quebra-Anzol and Uberabinha Rivers, it is observed that the extreme model scenario in the driest month is between minimum observed and 25% percentile observed and the extreme model scenario in rainiest month is between maximum observed and 75% percentile observed.

Quality modeling
Figures 6 and 7 show longitudinal profile of simulated quality parameters in the driest and rainiest months, and average values, maximum and minimum flow rates observed and 25-75% percentiles observed in the period of calibration and validation in the Araguari, Quebra-Anzol and Uberabinha rivers.In the three major rivers, the longitudinal profile of simulate quality parameters always remained within the minimum and maximum values observed in all the nodes studied.
Table 2 presents the calibrated values of the main coefficients of biochemical reactions (K a , K d , KN oa , KN ai and K phosph ), the sedimentation rates (V Sd , V SNo and V Sphosph ) and sediment oxygen demand (S OD ) in each river segment.The values in this table are within limits recommended in the literature (Chapra, 2003;Von Sperling, 2007;Paredes-Arquiola et al., 2009).Also in Table 2, the values set at -1 for K a in some segments indicate that this coefficient was estimated by the Covar method.Note that there was sediment oxygen demand in much of the basin, ranging from nodes 2 (upper course of Araguari River) and 15 (upper course of Quebra-Anzol River) to node 9 (Capim Branco HP 1).
According to Figure 8, a comparison was made for the main coefficients found in this paper with values from the literature.k a values upstream to the Nova Ponte reservoir are similar to the found by Paredes-Arquiola et al. (2010a, 2010b), Nakamura (2010) and Salla et al. (2013), which varied between 0.5 and 6.4 day -1 , k d values presented two bands, a range between 0.001 to 0.1 day -1 (similar to Paredes-Arquiola et al., 2010a) and another range from 0.1 to 0.6 day -1 (similar to Paredes-Arquiola et al., 2010b;Nakamura, 2010;and Salla et al., 2013).With respect to S OD , the same range of values found in this study was found in Paredes-Arquiola et al. (2010a), which varied between 0.10 and 0.23 day -1 .In all references consulted KN oa coefficient ranged from 0.002 to 0.6 day -1 .The range of values found in this study to KN ai (0.007 to 0.2 day - 1 ) is within the limits found by Paredes-Arquiola et al. (2010a, 2010b) and Salla et al. (2013).
The low value of the constants of biochemical reactions (Table 2) are associated with the high pollutant dilution capacity due to high surface water flows in all the river segments under study and to the low pollutant loads discharged point wise by the 13 aforementioned municipalities (Figs. 3, 4).The models that have been calibrated are intended for basin planning so that the aim is not to obtain the same adjustment to specific models or detail of water masses, as general data have been used.This approach allows to consider a reasonable fit between the time series of the simulated and observed values of water quality parameters studied here, with the best results achieved in the upper course of Araguari River, the upper course of Quebra-Anzol River and in Uberabinha River, according to the results indicated by the most representative nodes of this basin (Fig. 9).some segments, also the sediment oxygen demand S OD .
Figure 10 illustrates the percentage of variation of the parameters DO, BOD 5 , organic nitrogen, ammonia and nitrate as a function of the segments.In general, it was found that variations in the coefficients and in sediment oxygen demand display a low sensitivity with respect the previously calibrated results, while water temperature generated the largest one.With regard to the parameter DO, the highest sensitivities occurred as a result of changes in S OD and Temp in the segments of the Nova Ponte reservoir.With respect to S OD , the parameter DO ranged from -2.1 to +10% S OD and +1.9 to -10% S OD in Araguari segment 2 and from -3.8 to +10% S OD and +3.3 to -10% S OD in Quebra-Anzol segments 2, 3 and 4. With respect to Temp, the parameter DO has reached -6.4 to -10% Temp in Araguari segment 2 and +6.7 to +10% Temp in Quebra-Anzol segments 2, 3 and 4. The variation of K a generated little sensitivity in the calibrated results of DO (≤ 1.2% in all the segments).
The highest variations in the organic nitrogen occurred due to variations in the coefficient KN oa and water temperature Temp.The higher sensitivities observed where of ± 1.7% in Quebra-Anzol segments 2 to 4 and of ± 2.1% in Uberabinha segments 2 and 3 due to variations in the coefficient KN oa .With respect to Temp, organic nitrogen has reached -5.4 to -10% Temp in Araguari segment 2 and -4.4 to -10% Temp in Uberabinha river segments 2 and 3 (Fig. 10).
Ammonia showed low sensitivity (≤1.1%) due to variations in the coefficients KN oa and KN ai .With respect to water temperature Temp, the ammonia has reached -2.1 to +10% Temp and -3.2 to -10% Temp in Quebra-Anzol segment 1.
And nitrate showed the highest sensitivity due to variations in the coefficient KN ai and water temperature Temp, showed the highest sensitivity of ± 6.4% in Quebra-Anzol segments 2 to 4 and of ±3.4%

DISCUSSION
In the quantity simulations performed in the SIMGES module, from October 2006 to September 2011, the adjustments were satisfactory for scale work used in this paper, in which we tried to represent the mean behavior of the system.In Figure 5a, the greater amplitude of oscillation of the flow in the Nova Ponte HP (node 3) compared to the Miranda HP (node 7), Capim Branco HP 1 (node 9) and Capim Branco HP 2 (node 12) indicates the regulatory behavior of the Ponte Nova reservoir vis-à-vis the other three cascade reservoirs.The regulatory behavior of the Nova Ponte reservoir (node 3) is also shown in Figure 5b.An analysis of node 3 reveals that there is storage of liquid volume in the rainy season and release during the dry months, which causes a considerable decrease in the difference in flow between the rainy and dry seasons (note the segments upstream and downstream from node 3).
However, a general analysis of the longitudinal profiles of the quality parameters simulated for the rainiest and driest months (Figs.6, 7) reveals discrepancies with regard to the parameters BOD 5 in the Uberabinha River (Fig. 7) and total phosphorus in the Uberabinha, Araguari and Quebra-Anzol rivers (Figs. 6, 7).In Uberabinha River, downstream from the site where the municipality of Uberlândia discharges its treated wastewater, to the mouth of Uberlandia River (called Uberabinha segments 3 and 4), the BOD 5 and total phosphorus show Class 3 behavior.The BOD 5 ranged from 5.1 to 6.8 mg O 2 L -1 in the rainiest month and from 6.2 to 9.1 mg O 2 L -1 in the driest month.The total phosphorus parameter for lotic environments ranged from 0.10 to 0.14 mg P L -1 in the rainiest month and from 0.17 to 0.28 mg P L -1 in the driest month.The higher concentrations of BOD 5 and total phosphorus in the driest month are associated with the lower capacity for natural self-purification and dilution of pollutants due to reduced flows.This problem will increase due to the increasing population of this municipality.
In the Araguari and Quebra-Anzol rivers, the simulated profiles of the parameter total phosphorus show non-compliance with the COPAM (2008) in the Araguari 2, Quebra-Anzol 3 and Quebra-Anzol 4 segments.These segments, which correspond to the flooded areas of the Nova Ponte reservoir, behave like lentic environments, in which phosphorus ranged from 0.04 to 0.06 mg P L -1 in the rainiest month and from 0.02 to 0.04 mg P L -1 in the driest month in the Araguari segment 2, and from 0.03 to 0.09 mg P L -1 in the rainiest month in the Quebra-Anzol segments 1, 2 and 3.In this region of the Araguari River basin, the higher concentrations of total phosphorus in the rainiest month are associated with land use in terms of the excessive application of this nutrient in annual and perennial crops.In the period of this study, land use for pasture, and annual and perennial crops represented approximately 53% of the area of contribution to the sub basin of the Quebra-Anzol River, according to the Committee of Araguari river basin.
An overall analysis of the time series of observed values of quality parameters (Figs. 6, 7) reveals a behavior that does not comply with the recommendations of the COPAM (2008) on certain dates within the period studied.In Uberabinha River, the parameter DO showed values of less than 5.0 mg O 2 L -1 on only four occasions in the dry months, downstream from Uberlandia's municipal wastewater treatment plant (nodes 20 and 21).Point wise DO values of less than 5.0 mg O 2 L -1 were found in Araguari River segments 3, 4 and 5, indicating the influence of the bottom discharge of Nova Ponte reservoir (lower concentrations of dissolved oxygen).BOD 5 values far exceeding the maximum of 5.0 mg O 2 L -1 were found only in Uberabinha River downstream from the municipality's WWTP, which reached up to 32.0 mg O 2 L -1 to observed data in node 20, However, the box-plot graph (Fig. 7) shows that the extreme model scenarios -driest and rainiest month-are between 25 and 75% percentile observed in node 20.Ammonia values exceeded the maximum of 3.7 mg NH 4 + L -1 only in Uberabinha River, also downstream from the municipality's WWTP, which reached up to 11.0 mg NH 4 + L -1 in a single month without rain.The nitrate parameter was in compliance with the COPAM (2008) in the analyzed time series.However, in most of the nodes, the total phosphorus parameter presented values not in compliance with the maximum of 0.03 mg P L -1 , except for nodes 4, 5, 6, 10, 14, and 18.
The calibrations in the upper courses of Araguari and Quebra-Anzol rivers (Figs.9a, 9b) showed satisfactory fits to the DO, nitrate and total phosphorus parameters.As for organic nitrogen, the reduced number of observed data precluded a good assessment of the fit.The time series of observed data of the BOD 5 and ammonia parameters showed practically constant values, which also made it difficult to assess the fit between observed and simulated data, indicating the possibility that the laboratory measurements of these parameters have methodological limitations.
Figure 9c shows the time series of simulations and observed data at node 21, located downstream from Uberlandia's municipal WWTP at the lower course of Uberabinha River.The calibrations achieved satisfactory results for the DO, BOD 5 , ammonia, nitrate and total phosphorus parameters, despite a few observed data scattered of the ammonia, nitrate and phosphorus parameters.The quality of the observed data for the nitrogen parameter hindered their fit to the simulations, as indicated by a comparison of the oscillatory behavior of the data observed for ammonia and its fixed behavior and with the value of 0.2 mg N L -1 for 60% of the data observed for nitrogen.Calibrations in the middle and lower course of Araguari River (Figs. 9d, 9e) showed different behaviors in relation to the nodes located in the upper course of Araguari River and in Uberabinha River.In general, the time series of observed data are highly scattered in these regions of the basin, which hindered the satisfactory fit of the simulations, except for the of dissolved oxygen and phosphorus parameters.

INTRODUCTION
Myctophids are typically pelagic fish of the open ocean (Hartel & Craddock, 2002) and, together with members of Sternoptychidae, Gonostomatidae, Chauliodontidae and the suborder Stomiatoidei, represent the characteristic families in mesopelagic depths (Haedrich, 1997).Among these, Myctophidae is the dominant family (Nafpaktitis et al., 1977) and the most speciose, including almost 250 species referred to as lanternfish due to a variety of luminous organs, among which photophores are the most characteristic (Nelson, 2006).Lanternfish range from arctic to antarctic waters, and from the surface at night to depths exceeding 2000 m (Nafpaktitis et al., 1977).
The family also includes species known as pseudoceanic, associated with continental shelf and slope regions and in the neighbourhood of oceanic islands (Hulley, 1981).Continental slopes are particularly important due to the topographic and hydrographic gradients, and are considered areas of dynamic tension (Merrett & Haedrich, 1997).
Continental slopes also encompass a wider set of physical niches, and provide an environment for the development of a recognizable and trophicallydependent community of benthic and benthopelagic fish (Haedrich et al., 1980).Down-slope zonation of lanternfish may result from the combined effects of depth and water column structure (Hulley, 1992).Much of the current knowledge on Atlantic myctophids resulted from the study of the collections of the Woods Hole Oceanographic Institution (WHOI) (Nafpaktitis et al., 1977) and Institut für Seefischerei (Hulley, 1981).In the southwestern Atlantic (0º-60ºS), 79 species (22 genera) were collected during the 11th cruise of the R/V Akademik Kurchatov (Parin & Andriyashev, 1972;Parin et al., 1974).The distribution of 40 of these species, with respect to the water masses between 40º30'-47º00'S, was further examined (Konstantinova et al., 1994;Figueroa et al., 1998).Off the coasts of Suriname and French Guiana, 15 species from 7 genera were reported (Uyeno et al., 1983).In the Eastern Central Atlantic, Wienerroither et al. (2009) reported 52 species for the Canarian archipelago.
The present study provides knowledge about southwestern Atlantic lanternfish, including samples from Vitória-Trindade chain, an area understudied (Clark et al., 2010), adjacent to a transition zone between tropical Atlantic and temperate South America biota.We report the occurrence and distribution of lanternfish in relation to oceanographic conditions and attempt to examine whether species associations are spatially different.

Study area
The eastern coast of Brazil is a typical oligotrophic system (Gaeta et al., 1999), and the most important oceanic surface feature is the southward flowing Brazil Current (BC: 22-27 o C, 36.5-37.0psu), the warm western boundary current of the subtropical gyre (Silveira et al., 2001).The continental shelf of the study area (11-22ºS) extends for only 8 km off Salvador (França, 1979) and widens to the south to form the Royal Charlotte Bank (RCB, 16 o S) and the Abrolhos Bank (AB, 18 o S).The Vitória-Trindade chain that extends along 20.5ºS comprises seamounts that have shallow summits at depths of 34-76 m (Miloslavich et al., 2011).These topographic barriers produce a complex hydrographic structure including vortices, upwellings and vertical mixing processes, which alter the oligotrophic condition mainly south of AB (Ciotti et al., 2007;Valentin et al., 2007).The subsurface layer beneath the BC is occupied by the cold and nutrient-rich South Atlantic Central Water (SACW: 6.0-18.5 o C, 34.5-36.0psu) flowing north, between 400-700 m (Schmid et al., 1995).Periodic upwelling of SACW beyond the deep thermocline (80-120 m) enhances primary production (Nonaka et al., 2000).In the subthermocline region there are three major water masses, Antarctic Intermediate Water (AAIW) near 800-900 m, North Atlantic Deep Water (NADW) centered at about 2500 m, and Antarctic Bottom Water (AABW) below about 3500 m (Hogg & Owens, 1999).

Biological sampling
The studied material was obtained with midwater and demersal trawls, both performed only during the day, on the continental shelf, continental slope and near oceanic banks and seamounts off eastern Brazil (Fig. 1).During the midwater cruise, the collections were obtained using a large midwater net (292 m circumference and 191 m long).During operations, maximum horizontal and vertical opening was 56 and 25 m, respectively.Mesh sizes were 8000 mm in the wings and 20 mm in the cod-end.A total of 62 pelagic midwater trawls were towed at depths from 19 to 910 m, among which 50 had myctophid catches (n = 28,645).Hauls were undertaken on acoustically detected fish aggregations.
During the demersal cruise, the individuals were obtained using a bottom-trawl net with a 26.8 m head rope and 47.2-m foot rope, equipped with 40 rubber bobbins (rockhoppers) attached to the foot-rope.Mesh sizes were 110 mm for the wings and 20 mm for the cod-end.During the fishing operations, the horizontal and vertical mouth openings ranged between 28.0 and 45.5 m and from 3.0 to 10.6 m, respectively.
Demersal trawls were decided based on the availability of trawlable bottoms.A total of 58 stations yielded more than 45,000 specimens, from which 2,720 specimens of myctophids were recorded in 47 stations, ranging in depth from 100 to 2,271 m.On both cruises, trawl depth was acoustically controled using SCANMAR system, which was also used to access trawl geometry, including the horizontal opening and distance from net to bottom.The vertical opening and its distance in relation to the bottom was controled by the Ossian Trawl-Eye 49 KHz transducer fixed in the headrope.These systems allowed maintaining the net operating at specific depths during fishing and further classify the stations into different depth strata.

Water mass distribution in the study area
Temperature and salinity profiles (n = 116) recorded during the midwater cruise were used to analyse the horizontal distribution of temperature contours at the 200 m isobath (e.g., beginning of the mesopelagic zone), throughout the studied area (11-22 o S).Water masses distribution was inferred from a T-S diagram using data compiled from the National Oceanographic Data Centre (Brazilian Navy), including CTD profiles down to 5,000 m.These data was sorted from the same geographic area and period (May-July) and processed using Ocean Data View (ODV) software.

Distribution
The distribution of myctophid assemblages was analysed with non-metric multidimensional scaling (Clarke & Warwick, 2001) using the Sorensen similarity index calculated with species presenceabsence in the samples.The final matrix used in the ordination was composed by 29 species and 53 samples (9 midwater and 44 demersal trawls).Stations with only one species (n = 11) were excluded from the matrix.A non-quantitative index was chosen due to differences in net sizes and sampling strategies between midwater and demersal fishing.

Water mass distribution in the study area
The thermal structure of the water column during the two cruises, both during winter, was similar.The water temperature ranged from 24-28ºC at surface, 20-24ºC at 100 m depth, 15.7-16.1ºCat 200 m, 8-9.5ºC at 500 m, and was always <3ºC beyond 1,000 m depth.Tropical Water (TW) from BC (T > 20 o C; S > 36.2 psu) was present at surface (29-68 m), and the subtropical SACW with lower temperatures (6-18.5 o C) and salinities (34.6-36.0psu) occuppied the subsurface layer (118-624 m) (Fig. 2).
The horizontal distribution of the water temperature at 200 m (Fig. 3) showed that this depth was occuppied by SACW throughout the studied area, with an east-west gradient of decreasing temperatures.The lowest temperatures (14-15ºC) were identified between 13-15 o S off Salvador, and near RCB, AB and the Vitória Channel, reflecting the upwelling of SACW as a result of BC meandering along the shelf edge and seamounts.
The number of species/trawl ranged from 1-13 (Fig. 5), but captures of more than 5 species/trawl were more frequent in demersal trawls (21%) than in midwater trawls (17%).The maximum number of species per trawl (13) was similar in demersal and midwater trawls.Midwater trawls near seamounts south of Abrolhos Bank yielded the highest number of species (Vitória: 13 spp.;Davis: 7 spp.).Demersal trawls that yielded the highest number of species occurred at the southernmost part of the area at 624.5 m (13 spp.) and at 2,126 m off Salvador (10 spp.).Nine trawls yielded 6-9 species trawl -1 .Among these, six were performed between 13-14ºS at depths from 522 to 1,929 m, and three were performed between 19-20ºS at depths from 895 to 1,649 m.

Species data
From a total of 31,365 myctophids examined, all but 278 (0.9%) were identifiable to species.Table 2 lists the species grouped according to abundance, along with their totals and frequency of occurrence in midwater and demersal trawls, depth of occurrence and size range.The identified material comprised 11 genera and 29 species.The top five most abundant species comprised approximately 95% of the total number of individuals.Only one species was extremely abundant, 4 were abundant, 5 were common, 8 were uncommon and 11 were rare.Diaphus dumerilii was the most frequent species, both in demersal (62%) and midwater (66%) trawls.The majority of species (79%) had broadly tropical and tropical affinities (as indicated by Hulley, 1981), while species with subtropical and temperate affinities were poorly represented and occurred in very low numbers (1-17 ind).
Although NAB and SAB did not significantly differ (P = 0.255), a change in dominance was evident when mean densities (ind h -1 ) of the nine most abundant and frequent species in midwater trawls were compared (Fig. 7).Diaphus garmani and D. dumerilii were caught throughout the area, though mostly abundant at RCB and Minerva seamount.M. obtusirostre occurred associated to the four seamounts sampled (Minerva, Hotspur, Vitória, and Davis).At Vitória seamount, except for D. garmani, the eight remaining species occurred together in catches, with yields that ranged from 4.2 to 110.2 ind h -1 .A monospecific school of D. brachycephalus caught at 15ºS yielded 1,115 ind h -1 .

DISCUSSION
In this study, myctophids were more frequent in demersal than in midwater trawls, possibly as a result of our exclusive daytime sampling, since under the normal diel vertical migration pattern these fishes hide from visual predators at depth during the day and forage on abundant plankton in upper waters at night (Pearre, 2003).Also, a number of specimens could have been caught during retrieval and/or deployment of the bottom trawl (the nets used were devoid of open/close mechanisms) and, for this reason, fish density estimates were not compared between midwater and demersal trawls.Moreover, the presence of myctophids in demersal trawls could represent the adoption of an adult benthopelagic life strategy, as indicated by Vinnichenko (1997).Gartner et al. (2008) suggested that persistent high-density near-bottom aggregations (NBAs) are a normal part of the life history of several species traditionally considered to be mesopelagic.These NBAs would enhance the probability of feeding success, as fishes could explore  food supplies in a two dimensional search area (i.e., near bottom).
Among the 29 myctophid species captured in this study, tropical and broad tropical distribution patterns dominated.Species with temperate and subtropical affinities were restricted to hauls that sampled depths below 700 m, which is the upper limit of AAIW in the area (Hogg & Owens, 1999).Although this number is  genera) is comparable to that reported for Hawaii (47 species, 18 genera; Clarke, 1973), eastern Gulf of Mexico, GOM (49 species, 17 genera; Gartner et al., 1987) and north-central GOM (38 species, 17 genera;Ross et al., 2010).Collectively, Brazilian waters have a high diversity of myctophids (79 species, 23 genera: Menezes et al., 2003) comparable to that registered in the North Atlantic (82 species, 20 genera: Nafpaktitis et al., 1977).These numbers include the 30 species reported by Hulley (1981) for waters beyond 3,000 m during the research cruises of FRV "Walther Herwig" to South America (1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975)(1976).Among these, many are typically associated with the STC, the frontal zone where subantarctic and subtropical waters meet, which is a circumglobal feature of the Southern Ocean (Williams et al., 2001).The STC is a major biogeographic boundary, as well as a region of enhanced productivity (Pakhomov et al., 2000), and much of the plankton and fish fauna in this region have a circumglobal distribution (McGinnis, 1982;Pakhomov et al., 2000).
A tendency of increasing species number with depth was observed, and since temperature correlates to lanternfish distribution (Brandt, 1983), this result could be associated with the marked thermal structure of the water column.During hauls that sampled depths higher than 1,500 m between 11º-22ºS, the net was towed through four waters masses (BC, SACW, AAIW, NADW), possibly increasing the probability of catching a higher number of species.Hulley (1992) also observed an increase in lanternfish species richness and in the complexity of the distributions across the slope, possibly as the result of a higher structuring of the water column.
The presence of a variety of reliefs between 11-22ºS adds topographic complexity, causing islandinduced disturbance, in which upwelled nutrients promote primary and secondary production in the island wake (Bonecker et al., 1992(Bonecker et al., , 1993) ) and probably act to affect the distribution of the mesopelagic fauna.Our hydrographic results showed the occurrence of SACW at 200 m between 13-15ºS, and near RCB, AB and the Vitória Channel; this possibly reflects the permanent cyclonic eddy that Schmid et al. (1995) detected to be formed south of the AB from the meandering movement of the BC after passing through the Vitória Channel.The spatial distribution of the SACW in the area studied seems to explain the distribution of the more speciose trawls and, for some species, the highest densities associated with seamounts and banks.
While D. dumerilii was the most frequent and second most abundant species in our study, it was dominant between 22-34ºS (47%; Figueiredo et al., 2002) and seemed to be important in the trophic relation on the slope, once it was found in the stomach contents of several demersal bony fish of STC ecosystem (Haimovici et al., 1994).Near RCB this species was most abundant in rather shallow depths (25-34 m), indicating a certain degree of land association, as it was observed by Wienerroither et al. (2009).Diaphus dumerilii dominated both the water column (deep scattering layer) and the bottom (NBAs) on deep coral banks off Cape Lookout middle slope, North Carolina (Gartner et al., 2008).
The occurrence of mesopelagic species in shallow waters is ascribed to the abrupt depth changeover around islands of volcanic origin (Uiblein & Bordes, 1999), and the direct interaction between pelagic and demersal organisms at the interfaces between submerged bottom features establishes and important link between epipelagic waters and the deep benthos (Marshall & Merrett, 1977).Lanternfish may be an important prey item for large pelagic species, abundant in longline catches from Vitória-Trindade seamounts (Olavo et al., 2005).Future research in the area should address the study of oceanic food webs.crew of the R/V Thalassa for their collaboration.Special thanks are given to the Chief Scientists and coordinators of the Brazilian cruises aboard R/V Thalassa, for the facilitation of logistics and successful integration of the multidisciplinary teams during cruises Bahia-1 and Bahia-2.Two anonymous reviewers made numerous comments to improve the manuscript.

Figure 1 .
Figure 1.Relationship among the modeled quality parameters.
are agriculture, aquaculture, farming, mining, power generation, manufacturing, agribusiness and tourism.

Figure 3 .
Figure 3. Model topology applied to the Araguari River basin.

Figure 4 .
Figure 4. Single line diagram of the model.

Figure 5 .Figure 6 .
Figure 5. a) Variation of the simulated flow over the period of calibration and validation at the main points in the basin.Longitudinal profile of the simulated flows in the driest and rainiest months, with values of average, maximum and minimum flow rates observed and 25-75% percentiles observed in b) Araguari River, c) Quebra-Anzol River, and d) Uberabinha River.

Figure 7 .
Figure 7. Longitudinal profile of the simulated quality parameters for the rainiest and driest months, and values of average, maximum and minimum flow rates observed and 25-75% percentiles observed in the period of calibration and validation in Uberabinha River.

Figure 8 .
Figure 8.Comparison of the main coefficients found in this paper with literature values.

Figure 10 .
Figure 10.Sensitivity Analysis -Percentages of variation of the DO, BOD 5 , organic nitrogen, ammonia and nitrate parameters as a function of the segments of river.

Figure 1 .
Figure 1.Study area and location of the sampling stations.

Figure 2 .
Figure 2. Water masses recorded in the present study, according to depth.TW: Tropical Water, SACW: South Atlantic Central Water, AAIW: Antartic Intermediate Water, NADW: North Atlantic Deep Water, AABW: Antartic Bottom Water.

Figure 3 .
Figure 3. Distribution of temperature contours at 200 m isobath in the study area.

Table 1 .
Number of stations, depth range, effort and characteristics of nets used in midwater and demersal sampling off eastern Brazil and details of myctophids catch.