DSpace at VNU: Nutrient Dynamics During Flood Events in Tropical Catchments: A Case Study in Southern Vietnam

Nguyen Hong Quan1Günter Meon2 1 Institute for Environment and Resources, Vietnam National University, Ho Chi Minh City, Vietnam 2Leichtweiss Institute for Hydraulic Engineering and Water

Soil Air Water

CLEAN

www.clean-journal.com

Renewables Sustainability Environmental Monitoring

5 | 2015

Focus Issue:

Water Management for Agriculture and Energy Security in Asia

Nguyen Hong Quan1

Günter Meon2

1

Institute for Environment and

Resources, Vietnam National

University, Ho Chi Minh City, Vietnam

2Leichtweiss Institute for Hydraulic

Engineering and Water Resources,

University of Braunschweig,

Braunschweig, Germany

Research Article Nutrient Dynamics During Flood Events in Tropical Catchments: A Case Study in Southern Vietnam

Assessing surface water quality variation as well as chasing water pollution sources is essential for water quality management However, for existing conditions in developing countries, this assessment may not be done properly in many affected catchments due to limited data and lacking of tools In particular, pollutant transport from the catchment

to its river system during flood events needs quantification and is the aim of the study For this, a combined water quality monitoring and modeling approach is proposed The study was exemplarily performed for a typical ungauged medium scale catchment located in the southern area of Vietnam The available budget allowed at least a limited monitoring of nutrients and driven parameters (e.g., flow, sediment) These data were used to, in total, successfully calibrate the complex Hydrological Simulation Program-Fortran (HSPF) model The results lead to three main conclusions: (1) the contributions of point and diffuse sources to nutrient loadings could clearly be identi fied with the help of monitoring; (2) water quality sampling during flood events is critical to assess pollution sources, especially, diffuse ones However, just a monitoring of data alone is not adequate to interpret the observed concentrations; modeling is required (3) Despite of the limited amount of data, which could be recorded and processed during the study, a representative catchment modeling during floods could be performed It delivered essential information for linking pollution sources with water quality data Furthermore, the limits of an application of the complex HSPF model under given conditions were shown.

Keywords: HSPF model; Point and diffuse sources; Tapioca; Tropical regions; Water quality monitoring

Received: April 9, 2013; revised: August 27, 2013; accepted: September 18, 2013 DOI: 10.1002/clen.201300264

1 Introduction

Water pollution is one of the challenging problems in water

resources management At different impact levels, this problem still

exists in both developed and developing countries [1, 2] For example,

in Vietnam, surface water pollution at river basin scale stays at the

highest priority to be concerned, for example, said by Mr K N Pham,

Minister of the Ministry of Environment and Resources, that“the

Vietnam Environment Administration must concentrate well in

management and pollution control” (www.vea.gov.vn) National and

international agencies, scientists as well as other water resource

stakeholders have been looking for water pollution reduction

solutions Typically, for example, is the Clean Water Act 1972 for

the implementation of the total maximum daily load [3] in the

United States Other examples are the Drinking Water Directive, the Water Framework Directive in European countries [4], the Vietnam Law on Environment Protection [5] However, achievement to proposed objectives (i.e., good water quality status) is questioned due to, for instance, complexity of diffuse pollution [6]

Research in nutrient dynamics duringflood events is limited in literature, especially at medium-sized catchments ranging from 10

to 100 km2 The most concerned reason is the lack of monitoring data [7, 8] This problem is increasingly recognized in the scientific community, for example, during the Predictions in Ungauged Basins program [9] In the developing countries, for example, Vietnam, available observations from responsible agencies are often limited in temporal scale (e.g., from four to twelve times per year) Finer data resolution is sometimes available atfield scale (few hectares) or at a few small catchments (<5 km2), which are equipped mostly for research purposes, for example, in Mai [10] Furthermore, analysis on the nutrient dynamics during a flood event is often based on statistical analysis [11, 12] without quantification of complex anthropogenic impacts [8] Predictions based on such a few observations are uncertain Models can be used to compensate this gap, not only for prediction but also for assessing different scenarios including wastewater allocation, change of land use, variation of climate, etc [13–15]

Watershed water modeling has been developed extensively in the last decades Popular models with focus on pollutant transport are

Correspondence: Dr N H Quan, Department of Natural Resources

Management, Institute for Environment and Resources (IER), Vietnam

National University of Ho Chi Minh City, 142 To Hien Thanh, District 10,

Ho Chi Minh City, Vietnam

E-mail: hongquanmt@yahoo.com

Abbreviations: HSPF, Hydrological Simulation Program-Fortran; LSUR,

length of overlandflow plane; LZSN, lower zone nominal storage; NSE,

Nash–Sutcliffe efficiency; PBIAS, percent bias; RMSE, root mean square

error;SLSUR, slope of overland flow plane; TSS, total suspended solid; US

EPA, United State Environmental Protection Agency; UZSN, upper zone

nominal storage

SWAT [16, 17], HSPF [18], AGNPS [19, 20], SHETRAN [21], and

GLEAMS [22] The models have been used extensively not only for

research purposes but also for practical application such as water

balance analysis, nutrient variation, and best management practice

However, as pointed out by Borah and Bera [23], adopting a suitable

model for a particular objective is rather limited They selected

the DWSM model as the most suitable one among 11 models for

the simulation of the nutrient variation during flood events [24]

Given extensive agricultural activities, a main contribution to

diffuse pollution, are dominant in the Vietnam, responsible

authorities, farmers, and concerned public should be aware of the

diffuse pollution and its adverse impact on water quality best

possibly [25–28] For example, overland runoff during storms can

bring enormous loads of pollutants to water body systems from the

agriculture areas [29] Especially, in tropical region, for example, in

Vietnam, extreme flood events occur very often during the rainy

season However, this aspect does not get enough attention even in

the most recent years and is ignored in most legal documents

with regard to both water quality monitoring programs and

modeling efforts [25, 30] In addition, the control of point sources

is still difficult because of limited management capacity For

example, illegal wastewater disposal duringflood events is often

reported by local residents, no official evidence is available

because of limited observations (see also VEDAN company, www

nea.gov.vn/Sukien_Noibat/Tinkhac/Thang%201-2009/sggp_3-1.htm)

These circumstancesfinally result in an inadequate water quality

management

Combining water quality monitoring and water quality modeling

is not a common practice yet [8], in this study, the combined

approach is implemented in order to investigate its implementation

in water quality management Effects of both point and diffuse

sources during flood events at a medium-sized Vietnamese catchment are examined and compositions of inorganic nutrients (i.e., phosphorus phosphate [P-PO4], nitrogen ammonium [N-NH4], and nitrogen nitrate [N-NO3]) are addressed based on three flood events

2 Materials and methods

2.1 Introduction to study areas The study catchment, namely Tra Phi (Fig 1), is a tributary to the Tay Ninh canal (river) catchment which is one of the most polluted spots within the Dong Nai River basin, the biggest national river basin in Southern Vietnam [25] The study is linked to an joint Vietnamese/ German research project named“Tapioca” funded by the German Ministry of Education and Research and the Vietnamese Ministry of Science and Technology which had been completed in 2012 [31] An increase of water hyacinths in the river during rainy season has been observed Therefore, this study focuses on the highflow period while the model-based management system of the Tapioca project is based

on long-term simulations on a daily time step

The catchment is affected by the tropical monsoon climate with two distinguished seasons (rainy season from May to November, dry season from December to April) Extreme rainfall events often occur

in the area whose maximum daily rainfall observed during the last decade was in the range of 180 mm [32] The catchment is characterized by highly topography variation (ranging from 2 to

30 m above sea level (m a.s.l.) in low land areas, to 1000 m a.s.l at the water head The catchment covering about 21 km2is identified by its corner coordinates of 11°1905800N, 106°503000E and 11°2303000N, 106°

1001500E There are three main land-use units including agriculture

Figure 1 Study area in 3D view by draping with LandsatTM(Source: Landsat.org, Global Observatory for Ecosystem Services, Michigan State University (http://landsat.org)) composite bands 1, 2, 3 (up right) in relation to Dong Nai–Sai Gon river basin [25] (down right, where sub-catchments are in color, provincial borders are in black line) and Vietnam (left)

(the agriculture is classified according to their specific plants

including rice (20%);“tree” like rice, sugar cane, cassavas (15%) as

short-term plants;“high trees” are mainly rubber (31%) as long-term

plants) (66%), forest areas (11%), wetland (“wetland” is river and

existing canals, ponds) (3%), semi-urban areas (semi-urban area is

defined here as mainly pervious areas except main roads and houses

are classified as urban areas) (13%), and urban areas (7%) (Fig 2) Gray

soil (acrisols) covers uniformly the catchment [32] The only one

identified point source comes from a factory which produces tapioca

starch from cassava (tapioca) roots The company produces averagely

30 tons of tapioca starch per day which generates about 360–600 m3

wastewater per day [33] Normally, from April to July each year, the

company often stops working due to a limited supply of tapioca roots

Wastewater from the company is kept inside linked ponds;

nevertheless, a part of the wastewater which occurred during the

washing of tapioca, is obviously discharged directly and continuously

into river since the pond of washing wastewater is always full during

the operating period (estimated as 3 m3/h:3 m3/h was estimated by

random observations; regular monitoring was not possible due to

private sector)

There were two samples of wastewater collected in 2007 and

2008 [34] In addition, it was observed from the field that the

wastewater can be easily increased duringflood events Company

restrictions did not allow taking measurements of released

wastewater during rainfall

2.2 Field measurements

Field measurements were mainly done within July and August 2009

The measurements included water level and water quality

param-eters at the outlet of the Tra Phi River The water-level records were

transformed into discharge with the help of a stage-discharge curve

The stage-discharge (rating) curve was developed during the 2007 and

2008 rainy seasons on the basis of 25flow measurements using an

acoustic Doppler current profiler Water samples were collected

before, during and afterflood events every 1, 2, or 3 h Only one

sample was taken for each time in the middle of the stream, 40 cm

below the water surface Several parameters were analyzed

immediately after sampling including temperature (T0), pH,

dis-solved oxygen, total disdis-solved solid using the handheld optical

Instrument (WTW 197) [35] with accuracy of1 digit P-PO4, N-NH4,

and N-NO3were analyzed within 24 h using a photometer, here the

Spectroquant1NOVA 60 is used [36] The equipment implements specific methods with accuracy were as follows:

(1) N-NH4: indophenol blue method (reference to: analogous EPA 350.1, APHA 4500-NH3 D, ISO 7150/1, DIN 38406 E5) with accuracy of0.2;

(2) N-NO3: 2,6-dimethylphenol method (reference to: analogous ISO 7890/1, DIN 38405 D9) with accuracy of0.5; and (3) P-PO4: phosphormolybdenum blue method (reference to: analogous EPA 365.2þ 3, APHA 4500-P E, DIN EN ISO 6878) with accuracy of0.02%

The total suspended solid (TSS) was analyzed at the laboratory of the Institute for Environment and Resources, Vietnam National University,

Ho Chi Minh City following the Standard Methods [37] Synthesized information on these three events is presented in Table 1

Water sampling was done based on observation offlow discharge as well as at a certain time interval, for example, 1 or 2 h interval when water level rises rapidly; 3 or 4 h interval when water level recedes slowly) It was also consistent with those found literature as“for a good estimation of load at least eight time-integrated samples are needed per runoff event to reach the level of accuracy comparable to a single flow-composite sample and consequently we can lose any advantage over grab sampling at such high sampling frequency” [38] As shown in Table 1, 16, 10, and 12 samples were taken for each event, respectively Thus, the water sampling frequency in this study was sufficient for an event sampling Only one sample was taken for each time in the middle

of the stream, at a depth of 40 cm from water surface

2.3 Modeling approach Research on nutrient dynamic modeling at small catchment scale during flood event in tropical regions is limited Modeling applications in tropical regions mostly focus on stream flow and sediment dynamics For example, Campling et al [39] apply the TOPMODEL model to simulate rainfall—runoff relationship; Marsik and Waylen [40] use the CASC2D model to assess the changes of land cover on hydrological cycles; Millward and Mersey [41] use the RULSE model with some modifications to adapt tropical conditions; Diaz-Ramirez et al [42] provide an example of utilization of the Hydrological Simulation Program-Fortran (HSPF) model to study the hydrology, soil erosion, and sediment transport for tropical island watersheds at monthly time steps Polyakov et al [43] apply the AnnAGNPS model to simulate runoff and sediment in a tropical catchment for daily and monthly time steps Other works on nutrient dynamics are based on analyzing sampled data [44, 45] in a statistical manner or applying model at farm scale, for example, nitrogen leaching [10] Consequently, simulation of nutrient dynamics at small catchment scale duringflood events, especially at an hourly time step, is not common in tropical regions

Figure 2 Land cover map of Tra Phi Catchment

Table 1 Summarized information of three observed events

Event 1 Event 2 Event 3 Total duration (h) 36 36 24 Total rainfall (mm) 58.6 18.6 33.6 Stream water level variation (cm) 68 9 80 Hourly maximum rainfall

intensity (mm)

40.5 7.3 27.6 Total samples 16 10 12

HSPF [18] was selected among a range of available models codes

including eight often-cited models These are two empirical models

(ANGPS [19] and CNS [46]), four conceptual models (SWAT [16],

HSPF [18], HBV [47], and ANSWER-2000 [48]), and two physically based

models (SHE and SHETRAN [49] and DWSM [50]) HSPF is a

comprehensive catchment model, simulating land surface and

subsurface hydrological and water quality processes Furthermore,

streamflow routing and reservoir operation is integrated into HSPF

The HSPF model suits to all requirements needed for the study, in

particular Nguyen [34]:

(1) simulation at hourly step to capture the high dynamics offlow

and nutrient load duringflood events;

(2) consideration of point and non-point sources;

(3) consideration of all relevant processes in nutrient

transforma-tion, transport in upland areas as well as in the river;

(4) availability of a detailed user manual as well as user friendly

interface

Above characteristics contributed to the widespread utilization of

HSPF For example, Bicknell et al [18] stated that“the HSPF is the

primary catchment model included in the EPA BASINS modeling

system”

The following data were acquired for model implementation

including: (1) meteorological data which were obtained from the Tay

Ninh meteorological national station which is 1 km away from the

outlet; (2) land use and topographic maps obtained from a local

agency; soil parameters; and (3) point sources loadings

Since soil data are not available for a long-term simulation (e.g.,

soiltemperature), nutrient transformation processes in the soil were

not considered Nevertheless, the model was used to assess the

contaminant transport during flood events During the event,

overlandflows may be dominant to groundwater flows and thus

bring most pollutants to the receiving water bodies It was assumed

that the negligence of these processes may not significantly impact

on the results

2.3.1 Model discretization

The Tra Phi catchment is discretized in the BASINS system intofive

sub-catchments accompanied withfive reaches Physical information

of the catchment and land-use distribution are summarized in

Table 2

2.3.2 Model parameterization

Model calibration follows the hierarchy of catchment model

calibration (BASINS 4 lectures, datasets, and exercises at www.epa

gov/waterscience/basins/training.htm) Firstly, the hydrological

com-ponent is calibrated Next, the simulated sediment loadings are compared to observation ones Finally, nutrients variables like P-PO4, N-NO3, and N-NH4 are considered Sensitive parameters were recommended in BASIN’s lecture notes as well as from Radcliffe and Lin [51] Model parameters were manually calibrated using given values adopted from technical notes [52, 53] The parameterization of the HSPF model refers to the following groups: (1) hydraulic, (2) hydrology, (3) sediment, and (4) nutrients (ammonium, nitrate, and phosphate)

Hydraulic parameters include mean width, depth of the streams, rivers, side of floodplain, manning numbers of channel, and floodplain which were mostly taken from the reference manual [52] andfield observation Representative parameters for the hydrologi-cal, sediment, and nutrients components are explained and parameterized (calibrated) in Table 3 and are explained in detail

as follows

Following the detailed instructions, that is, US EPA [52], model parameterization of the hydrological processes was straightforward Some parameters (lower zone nominal storage [LZSN], upper zone nominal storage [UZSN], length of overlandflow plane [LSUR], and slope of overlandflow plane [SLSUR]) were calculated using available formula or GIS processing, while other parameters were calibrated based on look-up table or using default values The initial value LZSN was calculated based on a formula given in Al-Abed and Whiteley [54] as:

LZSN¼ 100þ 0:25 P ðSeasonal precipitationÞ

100þ 0:125 P ðPrecipitation distributed throughout the yearÞ (

where P the mean annual precipitation (mm)

UZSN is equal to LZSN multiplied by 0.06 at steep slope areas (i.e., forest on rock) and multiplied with 0.08 at moderate slope areas The geometry parameters (LSUR and SLSUR) were calculated by means of GIS processing and kept without calibration The values of the calibrated LZSN, INTFW, UZSN, LSUR, SLSUR, and NSUR parameters stay within the ranges The infiltration parameter (INFILT) was calibrated according to the instructions [52] where the soil is most likely within groups A and B Thus, the INFILT was rather high after calibration This is suitable with the soil characteristics (acrisols) as well as given observations in thefield (except for the wetland and rice due to clay deposition on the land surface) The AGWRC, IRC was calibrated after using default values in US EPA [52] until the recession curve can capture the baseflow

Initial values for sediment parameters were chosen from the BASIN’s lecture notes and US EPA [53] Calibration was done based on thefirst event observation The parameters were then kept for the successive events Since the catchment is covered by the unique acrisols soil, the sediment generation parameters were kept similar

Table 2 Physical characteristics of sub-catchments and reaches

Parameter

Sub-catchments SWS1

(reach 2)

SWS2 (reach 3)

SWS3 (reach 4)

SWS4 (reach 5)

SWS5 (reach 1)

Drop in water elevation from the upstream

to the downstream (m)

Average channel slope (%) 0.21 0.005 0.006 0.006 0.002

for each land-use type except those affected by land covers The gully

erosion, deposition from atmosphere as well as in-stream sediment

transport processes was ignored here, since gully erosion was not

identified during the fieldwork campaign Thus, default values

were applied for model parameters of these processes Similarly to

the hydrological part, most of the sediment parameters were within

the ranges provided in look-up tables or by default values [53] The

coefficient in the detached sediment wash-off equation was out of

the range (0.01–0.5) as it was observed also in Diaz-Ramirez et al [55]

The calibration was straightforward and soon delivered to reasonable

conditions

The calibration of hydrological and sediment parameters followed

standard procedures as given by the US EPA [52, 53] This is not the

case for nutrient parameters as also mentioned by Radcliffe and

Lin [51] which can be considered as the most challenging part in

parameterization processes In addition, given the site-specific

problems such as simulation at hourly time steps, limited data do

not allow to simulate nutrient transformations in soil or using

default values for in-stream process parameters Since the focus of

this simulation was nutrient dynamics duringflood event only, an

assumption of the interactions and transformation processes in soil

and river was made Moreover, it was also observed during model

calibration (e.g., the changes of in-stream parameters affect model

results to a minor extent only) In order to cope with the problem of

nutrient transformation during normal condition (e.g., continuous

simulation during dry days) as well as management practice such as

fertilizing the soil, the monthly nutrient accumulation rate and

nutrient storage (MON-SQOLIM and MON-ACCUM, respectively) in the

model has been utilized Model parameters comprising of SQO, POTFW, POTFS, ACQOP, SQOLIM, WSQOP, ACCUM, and MON-SQOLIM mostly relate to storage and transport processes In addition, because of lacking data for identifying contributions from different land uses, each of the model parameters was kept unchanged An example of model parameters for phosphate are explained and parameterized in Fig 3, more information on data and parameter is referred elsewhere [34]

Some parameters are only roughly described in the BASINS’ lectures, for example, POTFW, ACQOP, SQOLIM, and WSQOP Other model parameters are calibrated by trial and error The later ones are site specific Some other studies using HSPF do not provide parameter sets for comparison For example, the work given by Radcliffe and Lin [51] implementing HSPF for simulating phosphorus dynamics at catchment scale provided only parameters related to nutrient transformation in soil Therefore, comparative studies that is applying the HSPF model in other areas in the regions or an up-scaling of the model are strongly recommended

3 Results and discussions

3.1 Monitoring data Three different magnitudes of rainfall and consequentlyflow for each event were observed The maximum hourly rainfall amounts of the three recorded events were 7, 40, and 26 mm, while maximum discharge was 1.3, 4.1, and 4.7 m3/s, respectively (Fig 3) The correlation coefficient between the maximum rainfall intensity

Table 3 Process and physical parameters used after calibration of the HSPF model for hydrology

Process parameter Description (unit)

Parametric value

Forest Urban and road

Cropland and pasture Rice Tree Wetland Hydrological parameter

Sediment parameter

COVER The fraction of land surface which is shielded from erosion by rainfall 0.8 0.9 0.7 0.5 0.6 0.2 AFFIX Fraction by which detached sediment storage decreases each day as

a result of soil compaction

Nutrient parameter

ACQOP Rate of accumulation of QUALOF (kg ha1day1) 0.045 0.002 0.054 0.045 0.045 0.045 SQOLIM SQOLIM is the maximum storage of QUALOF, (kg/ha) (recommended value) 0.027 0.027 0.027 0.027 0.027 0.027 WSQOP Rate of surface runoff which will remove 90% of stored QUALOF/h (recommended value) 0.5 0.5 0.5 0.5 0.5 0.5 MON-ACCUM Monthly values of accumulation rate of QUALOF at start of each month (kg ha1day1) 0.05 0.05 0.05 0.05 0.05 0.05 MON-SQOLIM Monthly values limiting storage of QUALOF at start of each month

(from July to September) (kg/ha)

and maximum river discharge is very high (0.96) The time lag

between maximum rainfall and maximum discharge is about 2–4 h

and depends on event durations One should note that there were

also preceding rainfall-runoff events which influenced the soil

moisture of the unsaturated zone and thus contributed to overland

flow

Behaviors of TSS, P-PO4, N-NH4, and N-NO3were highly dynamic

during these events (Fig 4) TSS was the most sensitive parameter to

event durations and as well as event magnitudes However, P-PO4,

N-NH4, and N-NO3exhibited partially different behavior during the

first and third events, the concentrations of P-PO4and N-NH4varied

in correspondence toflow discharge (i.e., 0.4–1 and 0.1–1.4 mg/L for

P-PO4and N-NH4, respectively) In contrast, during the second event,

these variables increased in a much higher magnitude (i.e., 0.25–3.6

and 0.1–2.3 mg/L for P-PO4and N-NH4, respectively), although this was

the smallest event The concentration of N-NO3reached the highest

value of 2.4 mg/L during thefirst event During the second event, a

strong smell of tapioca starch was noted in the sampled water It was

assumed that the different behaviors between the second on one

hand and thefirst and third events on the other hand were highly

influenced by more wastewater disposal

Table 4 summarizes the derived coefficient of determination (R2

) based on an exponentialfitting curve between constituents and flow

as well as between constituents and TSS, and those are presented as R2 (Q), R2(TSS), respectively Between TSS andflow discharge (Q), the highest R2was 0.78 for event 3 and R2was 0.41 and 0.19 for events 1 and 2, respectively It should be noted that during event 3 it was not affected by point sources disposal since the factory did not produce tapioca starch at that time Therefore, the sediment (TSS) was in very high correlation with the river discharge The highest correlation between P-PO4 and Q was obtained in event 1 (R2¼ 0.48) and the coefficient remained low in events 2 and 3; while the highest correlation between P-PO4 and TSS was 0.47 (event 3) and the coefficient was 0.38 and 0.1 for events 2 and 1 The correlation between N-NH4and Q, as well as between N-NH4and TSS was slightly different compared to the correlation between P-PO4 and Q, TSS, especially during thefirst event There was no significantly clear relation between N-NO3and Q as well as between N-NO3and TSS The correlation was 0.33, 0.3, and 0.38 (events 1, 2, and 3, respectively) and with TSS was 0.29, 0.24, and 0.26 for events 1, 2, and 3, respectively In addition, another similar test was given to the observed nutrient parameters and results are shown in Table 5 Among the three

Figure 3 Three rainfall-runoff events measured at the catchment outlet

Figure 4 TSS, N-NO3, N-NH4, and P-PO4measured at catchment outlet for three recorded events, respectively

Table 4 Correlation between TSS, discharge, and P-PO4, N-NH4, N-NO3during the three events

R2(Q) R2(TSS) R2(Q) R2(TSS) R2(Q) R2(TSS) R2(Q)

R2

(Q): correlation of determination based on exponentialfitting curve with flow discharge; R2

(TSS): correlation of determination based on exponentialfitting curve with TSS

nutrients, P-PO4and N-NH4are best correlated, especially during the

second event (R2was 0.98) where the contribution of point sources

was expected This correlation was also observed in the wastewater

sample Correlation between N-NO3and P-PO4as well as between

N-NO3and N-NH4was not clear: with P-PO4, R2was 0.29, 0.24, and

0.26 (events 1, 2, and 3, respectively); with N-NH4, R2was 0.33, 0.3,

and 0.38 (events 1, 2, and 3, respectively) These data indicate a

poor correlation between flow (as well as suspended sediment)

and nutrient constituents (i.e.,<0.5) However, the suspended solid

(sediment) is considerably correlated withflow discharge because of

little point source effects The correlation coefficient can go up to

0.78 A relation between nutrient constituents and discharge or

sediment is hardly visible However, it was observed that the

correlation between P-PO4and N-NH4is quite high as compared to

N-NO3, especially for the second event where there was also a good

agreement between these two parameters in the wastewater samples

Although data availability for statistical analysis is quite limited

(only three events because of the project budget), it is obvious that

monitoring alone is not enough to explain the variation of water

quality owing to system complexity and various anthropogenic

impacts Therefore, other tools (i.e., modeling) are needed for

reasoning the nutrient dynamics

3.2 HSPF results

Model results are compared with measured data for the three events

The model results are assessed by both qualitative and quantitative

means Qualitatively, model results versus observed variables are

presented in Figs 5–7 Quantitatively, model results are assessed by a

number of criteria including agreement index (d), coefficient of

determination (R2), root mean square error (RMSE), Nash–Sutcliffe

efficiency (NSE), and percent bias (PBIAS)

The agreement index (d) can be given as:

d¼ 1:0 

Pn

i¼1ðOi PiÞ2

Pn

i¼1ðjPi Ojþ Oj i OjÞ2

where Oi, observed value; Pi, predicted value; O, average of the observed value; P, average of the predicted value

The NSE was estimated by:

NSE¼ 1:0 

Pn i¼1ðOi PiÞ2

Pn i¼1ðOi OÞ2

where Oi, observed value; Pi, predicted value; O, average of the observed value

The PBIAS was calculated with:

PBIAS¼

Pn i¼1ðOPi PiÞ  100

n i¼1ðOiÞ where Oi, observed value; Pi, predicted value; O, average of the observed value

The coefficient of determination R2was predicted:

R2¼

Pn i¼1ðOi PiÞ2

Pn i¼1ðOi OÞ2

 0:5Pn

i¼1ðPi PÞ2

 0:5

)2

(

where Oi, observed value; Pi, predicted value; O, average of the observed value; P, average of the predicted value

The RMSE can be estimated by:

RMSE¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Xn i¼1ðOi PiÞ2

s

where Oi, observed value; Pi, predicted value

Table 5 Correlation of determinations among P-PO4, N-NH4, and

N-NO3

P-PO4

and N-NH4

P-PO4

and N-NO3

N-NO3

and N-NH4

Event 1 0.49 0.27 0.23

Figure 5 Observed and simulated hydrographs using HSPF model for

three recorded events, respectively

Figure 6 Observed and simulated TSS using HSPF model for three recorded events, respectively

Figure 7 Observed and simulated P-PO4using HSPF model for three recorded events, respectively (simulated results on N-NO3, N-NH4can be found in MERCK [36])

A summary of the evaluation parameters is shown in Table 6 The

observed values in the third events are not the same to the model in

term of recorded time (measured and predicted values are shifted a

30 min) Therefore, in the evaluation table, the contaminants like

TSS, nutrients are only applied for thefirst and second events, and

evaluation for event 3 is limited to graphical aspect

Figure 5 shows thatflow discharge variations are captured in the

model simulation although the variations are very large (e.g., nearly

four times between the extreme condition and the normal ones) In

Table 6, it can be seen that the simulated values agree well with

observed ones (d, approach to 1) However, the model performance is

rather limited with regard to the R2, RMSE and NSE values, except the

first event The PBIAS values show that the model overestimates for

thefirst and second events and underestimates for the last events

(event 3) Nevertheless, the peaks of flow discharge were well

represented This aspect is very critical since it is assumed that diffuse

contaminants are transported during these times One reason for

above over- and under-estimations is the impact of the of rainfall data

error Data from only one meteorological station were used in this

study Although the catchment is small, its highly topological

differences can induce variations of rainfall in time and space, for

example, by orthographic lifting [56] Event 2 is a clear example of

errors in rainfall data Given observation in the catchment outlet

(nearby the rainfall station) that the rain was heavy (18.6 mm in 2 h),

rainfall observed in a daily rainfall station (Nui Ba station, upstream

of the catchment) was only 5 mm This leads to a simulatedflood

hydrograph, which is higher than the hydrograph induced by rainfall

in reality

The simulation results of sediment transport (TSS, Fig 6) seem

similar to those observed in the flow discharge Considering the

uncertainties involved in the input data, in the model algorithms,

and in model parameters; the predicted and observed values have a

moderate to good agreement especially during the high flow

However, the model cannot reproduce well the observed

contami-nant during lowflow conditions (event 2) In addition, errors caused

by sampling may contribute to these differences

For nutrients simulation, in thefirst simulation runs, the model

could not yet represent the values observed in thefield Based on own

observations in thefield (at the tapioca company), it could be possible

that the pollutant loadings during theflood events were increased

because of thefilled-up settling pond or illegal wastewater disposal

Consequently, the input data were modified by increasing

wastewa-ter loadings from the company Afwastewa-ter several times of“trial”, the model catches the observed nutrient patterns at different magni-tudes higher than the normal wastewater loadings For thefirst and second events, the expected loadings were three and twelve times higher than the normal one In the last event, no point sources existed since the company was closed for renovation It is observed that the behavior of P-PO4(Fig 7), N-NH4is quite similar, while N-NO3

is not The N-NO3is quite sensitive toflood events, whereas P-PO4and N-NH4were strongly related to point sources and extreme events only With the introduction of“illegal” wastewater disposal (denoted

as simulation 2 in Fig 7), for thefirst events, only the simulation of

P-PO4is improved; the simulation of N-NO3does not change much, and the simulation of N-NH4is even worse For the second event, the simulation of N-NO3has a similar behavior, while the simulation of

P-PO4, N-NH4is much improved The model simulates well the peaks of nutrient variations However, during the recedingflow, the model generally underestimates the concentration This can be explained by retention effects in the agriculturalfields, especially rice fields being covered by earth-dykes

Flow discharge and TSS were rather well simulated during high flood events, especially the peaks Reproduction failed for the low flow period One common reason is because of rainfall data errors The different rainfall distribution in time and space can lead to over-and under-estimation of the realflow by the model The nutrient dynamics duringflood events are well captured by the model In spite

of the abnormal observations, the model was adapted by introducing higher wastewater loadings given the fact that strong smell of tapioca starch wastewater was observed during monitoring, as well as during investigating the sewage discharge system of the tapioca company The abnormal nutrient variation during low flow (event 2) is explainable by considering illegal wastewater disposal (12 times more than under normal conditions) The improvement of model simulation is clearly seen for phosphate phosphorus and ammonium nitrogen This is not the case for the performance of nitrate nitrogen

It is concluded that the point sources contribute significantly to phosphate phosphorus and ammonium nitrogen but the diffuse sources control the nitrate nitrogen Nutrient dynamics were well simulated during the risingflow; it is not the case during receding flow This could be an effect of water retention in rice fields where water was released from the after event by farmers in order to keep water level stay at a certain level

Regarding the uncertainty aspects in implementing the HSPF model, several facts should be kept in mind: (1) the model is complex Some parameters can interfere with each other during the calibration processes Therefore, the problem of equi-finality cannot

be avoided; (2) despite of the monitoring campaign, the study catchment can be regarded as a“nearly ungauged” catchment since data are still limited Many model parameters were estimated from literature (e.g., soil data), not from the real catchment In addition, input data such as rainfall, pollutant sources (e.g., point sources, nutrient storages in soil– most related to management practice) were also to some extent uncertain Therefore, for further study (e.g., up-scaling), the estimated model parameters have to be checked with care; (3) data are limited for model calibration; no data are available for model validation Thus, model results for long-term prediction have to be re-assessed by collecting more data In addition, soil data were not available for a long-term simulation (e.g., several years) Furthermore, the contribution of contaminants from groundwater was ignored in this model application Thus, the model applied here

is limited for short-term simulation of single events

Table 6 Model evaluation based on different criteria

PBIAS d R2(1:1) RMSE NSE

Flow

Event 1 34.52 0.90 0.64 0.84 0.48

Event 2 12.05 0.61 0.12 0.29 1.70

Event 3 50.71 0.76 1.00 0.78 0.18

TSS

Event 1 6.62 0.83 0.02 223.78 0.52

Event 2 217.50 0.23 0.30 254.01 85.30

P-PO4

Event 1 29.66 0.64 0.17 0.38 2.95

Event 2 2.00 0.77 0.86 0.45 0.24

N-NH4

Event 1 120.88 0.57 1.47 0.41 3.82

Event 2 5.49 0.84 0.91 0.26 0.48

N-NO3

Event 1 7.92 0.57 0.32 0.71 0.59

Event 2 16.87 0.58 0.83 0.16 0.01

The comprehensive HSPF model has been implemented The model

can be used for (extreme) event-based simulations of tropical

catchments being exposed to various anthropogenic impacts (point

and diffuse sources) The model parameterization is difficult and

requires high-level expertise In addition, given a number of

uncertainty sources, especially from model input and model

parameters, collecting of more data as well as implementing

comparative studies are recommended It became evident, that a

simpler and more robust model is– in total – more reasonable for the

simulation of nutrient dynamics during flood events for given

tropical conditions and a poor database Details about such a robust

model, which had been developed by the first author, and its

application to the same catchment is given in elsewhere [34]

4 Concluding remarks

In this study, thefield measurement of river discharge and water

quality (focusing on nutrients parameters) being relevant to the

research project offers an example of how data can be collected in a

small catchment in Vietnam being exposed to anthropogenic

impacts Data quality as well as limitations of data monitoring

was also emphasized The data showed that there are significant

correlations between N-NH4and P-PO4in connection to point sources

while N-NO3is typically related to diffuse sources in the catchment

However, monitoring alone is not enough to explain the variation of

water quality owing to system complexity and various anthropogenic

impacts Therefore, the HSPF model was used to investigate the

nutrient dynamics besides monitoring activities However, the

uncertain and limited data are typical issues when implementing

the model For example, when point sources are hardly controlled,

consequently, wastewater discharge is highly uncertain and it can be

increased significantly during flood event by either overloading the

wastewater pond or discharging illegally For this aspect, water

quality modeling becomes an important tool for a sound planning of

wastewater allocation Furthermore, it is an important tool for the

virtual reconstruction of disastrous historical wastewater discharge

scenarios as applied in several cases in Vietnam where monitoring

data are not available for inspection [57, 58] The results show that

although the model was implemented, there are still a number of

open issues that makes the model difficult, but not impossible, to be

used as an operational tool for the region Regarding the group of

available highly complex models like HSPF, comparative studies

using several tropical catchments are recommended A data bank of

parameter sets for the HSPF model under tropical conditions is

needed if it is used to support water quality management in the

developing countries Based on the monitoring data and the HSPF

model, further research can be developed, for example, long-term

studies about nutrient transport mechanism (e.g., at rice field),

groundwater and surface water interaction, water quality

manage-ment schemes (best managemanage-ment practice, wastewater allocation,

and reduction) Furthermore, for the given conditions, simpler

models as, for example, the developed by Nguyen [34], may replace

the HSPF model

Acknowledgments

The study was performed within the IPSWAT PhD Scholarship

Program of the German Ministry of Education and Research (BMBF)

Furthermore, the study was linked to the Vietnamese-German

research project “TAPIOCA” funded by the German BMBF and

the Vietnamese Ministry of Science and Technology MOST and coordinated by the Leichtweiss Institute for Hydraulic Engineer-ing and Water Resources, University of Braunschweig, Germany (www.tu-braunschweig.de/lwi/hywa/forschung-projekte/vietnam)

The authors have declared no conflict of interest

References

[1] United Nations Conference on Environment Development (UNCED), Agenda 21, Chapter 18: Protection of the Quality Supply of Freshwater Resources, United Nations Division for Sustainable Development, Rio

de Janerio, Brazil1992

[2] UN, World Water Development Report: Water for People, Water for Life, United Nations Educational Scientific Cultural Organization (UNESCO) and Berghahn Books, Barcelona2003, p 576

[3] NRC, Assessing the TMDL Approach to Water Quality Management: Committee to Assess the Scientific Basis of the Total Maximum Daily Load Approach to Water Pollution Reduction, Water Science Technology Board, National Research Council (NRC), National Academy of Sciences, Washington, DC2001

[4] H Blöch, EU Policy on Nutrients Emissions: Legislation Implemen-tation, Water Sci Technol.2001, 44 (1), 1–6

[5] NA SRV, Law on Environmental Protection (in Vietnamese and English) National Assembly of the Socialist Republic of Vietnam (NA SRV), Hanoi2005

[6] V Novotny, The Next Step-Incorporating Diffuse Pollution Abate-ment into Watershed ManageAbate-ment, Water Sci Technol.2005, 51 (3/4),

1–9

[7] M B Beck, Transient Pollution Events: Acute Risks to the Aquatic Environment, Water Sci Technol.1996, 33 (2), 1–15

[8] A L Højberg, J C Refsgaard, F van Geer, L F Jørgensen, I Zsuffa, Use

of Models to Support the Monitoring Requirements in the Water Framework Directive, Water Resour Manage.2007, 21 (10), 1649–1672 [9] M Sivapalan, K Takeuchi, S W Franks, H V Gupta, H Karambiri,

V Lakshmi, X Liang, et al., IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012: Shaping an Exciting Future for the Hydrological Sciences, Hydrol Sci J.2003, 48 (6), 857–880 [10] T V Mai, PhD Thesis, Wageningen University, Wageningen, The Netherlands2007, p 182

[11] M I Stutter, S J Langan, R J Cooper, Spatial Contributions of Diffuse Inputs Within-Channel Processes to the Form of Stream Water Phosphorus Over Storm Events, J Hydrol.2008, 350, 203–214 [12] W A House, M S Warwick, Hysteresis of the Solute Concentration/ Discharge Relationship in Rivers during Storms, Water Res.1998, 32 (8), 2279–2290

[13] NRC, Preparing for the Third Decade (Cycle 3) of the National Water-Quality Assessment (NAWQA) Program, Water Science and Technology Board, Division on Earth Life Studies, National Research Council (NRC), National Academy of Sciences, Washington, DC2012

[14] B T Neilson, S C Chapra, Integration of Water Quality Monitoring Modeling for TMDL Development, Water Resour.2003, 5 (1), 9–11 [15] L F Jørgensen, J C Refsgaard, A L Højberg, The Inadequacy of Monitoring without Modelling Support, J Environ Monit.2007, 9,

931–942

[16] S L Neitsch, G Arnold, J R Kiniry, J R Williams, Soil and Water Assessment Tool Theoretical Documentation-Version 2005, Grassland Soil & Water Research Laboratory, Temple, Texas2005, p 494

[17] J G Arnold, N Fohrer, SWAT2000: Current Capabilities and Research Opportunities in Applied Watershed Modelling, Hydrol Process.2005,

19, 563–572

[18] B R Bicknell, J C Imhoff, J L Kittle, T H Jobes, A S Donigian, Hydrological Simulation Program-Fortran (HSPF) Version 12 User’s Manual, National Exposure Research Laboratory, Office of Research Development, U.S Environmental Protection Agency, Athens2001,

p 873