ABSTRACT
Hydrological models are vital for water management to determine in-stream flow, irrigational water, domestic water supply, and biodiversity conservation. This study formulates a hydrological model with a novel approach for streamflow and sediment load in the QGIS-supported Soil and Water Assessment Tool for the Halda River catchment, a unique ecological habitat for natural carp spawning and freshwater sources. The daily simulation uses an innovative stage–discharge relationship technique from available 15-day interval flow data. The model evaluation parameters R2 values 0.80 and 0.62, and NS values 0.81 and 0.61 for calibration and validation of streamflow suggested excellent agreement in the seasonal cycle and most of the monsoon peak flow. The streamflow/precipitation ratio indicates a significant influence of groundwater through infiltration. The baseflow shows a decreasing trend. The sediment load based on suspended sediment concentration at a downstream location is 1,625 tons/day. On the contrary, the model prediction is 30 times lower. The scattered sediment load data support the model estimate by considering relatively lower intervention or land use change in its upstream. This model provides a baseline for daily flow and sediment load for scenario modeling (e.g., climate change, land use change) for environmental flow estimation of the fish habitat, freshwater supply, irrigation, and salinity intrusion.
HIGHLIGHTS
This study reports flow and sediment flux simulation of a unique carp-breeding habitat.
The research uses a novel approach of rating curves to estimate discharge from water level data to use for daily flow simulation in SWAT.
The performance matrix of the flow and sediment flux simulation is satisfactory and comparable.
While the flow monitoring data are not reliable, this approach is highly pertinent.
INTRODUCTION
Halda River is the only tidal river among a few river basins that lies inside the political boundary of Bangladesh and serves as a natural source of fertilized carp eggs (Saimon et al. 2016). The river is fed by several hilly streams starting from its origin. It has 20 sub-canals, 34 small hilly streams (Kibria 2012), and 12 tributaries (Badiuzzaman 1978; Podder et al. 2017) from which Dhurung, about 56 km long, is very turbulent and joined at Sundarpur, Dhurung Union (Tsai et al. 1981). The downstream of this river is well recognized as the only pure Indian carp natural spawning habitat in South Asia (Tsai et al. 1981; Azadi & Alam 2011). The flood plain of this river is relatively stable and has reported somewhat less riverbank erosion phenomenon, though it has highly braided channel characteristics throughout its course. The overall length of the river is approximately 98 km, and the total catchment area is about 1,671 km2 (Badiuzzaman 1978). The average depth of the Halda River is 6.4 m, and the maximum depth is 9.1 m. Raihan et al. (2020) reported a significant increase in the number of rain days at the northern sites during the monsoon season, with an increase per decade of 3 days in Sitakunda and 7 days in Rangamati. They also suggest a change in the flow and sediment load of the Halda Basin, which the anthropogenic land use change may trigger. A substantial increase in agricultural land use in its riverbank and catchment has also been reported in recent years (Akter & Ali 2012), which may change the river's flow and sediment load.
However, historical remotely sensed data with lower resolution cannot measure discrete small land use changes. As a result, a total change in the catchment did not spell out the effect on river sediment load and siltation. Since fish habitat sustainability and usability of the water, as a domestic source, heavily depend on water quality, many studies reported water quality and sediment load of this river (Azadi & Alam 2013; Patra & Azadi 1985; Tsai et al. 1981; Karmakar et al. 2020). Bhuyan & Bakar (2017) reported the heavy metal contaminants in surface water and sediment of the Halda River of industrial, municipal, and agricultural origin. Podder et al. (2017) recorded 34 watersheds of the Halda River, from which 7 major watershed areas were identified as significant ecological habitats. However, the ecological significance within the river varies with the season, which heavily depends on flow and watershed runoff. This river supplies 1.8 × 105 m3/day (MLD) freshwater to the nearby Chattogram city through two water treatment plants installed downstream of this river. Additionally, a few more proposals for the water treatment plant for fresh water supply to the nearby area are under consideration. In contrast, salinity was commonly reported to water treatment plants operating during the dry season.
The Soil and Water Assessment Tool (SWAT) model (Arnold et al. 1998), a physically based semi-distributed, computationally efficient open-source code (Yasin & Clemente 2014), is better suited for scarce data regions (Ndomba et al. 2008) compared to other commonly used models, such as MIKESHE (Refsgaard & Storm 1995), TOPMODEL (Beven & Kirkby 1979), or WASIM (Schulla & Jasper 2007). The semi-distributed model in SWAT subdivides a watershed into smaller sub-basins and hydrological response units (HRUs). SWAT has demonstrated strength in simulating catchment hydrology, considering climate and land management practices on water, sediment, and agriculture chemicals in large, complex watersheds (Neitsch et al. 2002). Badiuzzaman (1978) studied the hydrology of this river using unit hydrographs at Narayanhat, South Sunderpur, and Panchpukuria stations of this river. Habiba (2016) estimated the sediment load at Panchpukuria point in the river system using remote sensing data corresponding to suspended sediment, which is 288,850 tons/year. Moreover, freshwater flow and environmental consequences during the dry season have been analyzed by Akter & Ali (2012). The most recent study considers that the rubber dam in this river's upper stream negatively impacts the river's downstream flow and reduces the water-holding capacity of the river (Raihan et al. 2020). Parvez et al. (2019) also reported the sediment load and salinity intrusion downstream.
To date, flow simulation for this river remains a significant challenge due to the lack of continuous data and the effect of flow regulation on the observation data supplied by the Bangladesh Water Development Board (BWDB). This phenomenon is common to observe in many regional river basins in many regions of the world. Although daily water level data are available for this river at several stable cross-sections, these data were never used to validate discharge data of only measuring stations nor used for hydrological modeling. The flow model on this river carries the limitations of lack of daily flow data, inadequate data, a significant amount of missing data, and a few outliers supplied by the only monitoring agency, BWDB. Moreover, varied values for discharge with the same water level and extremely low or high discharge made the discharge data challenging to corroborate with water level and other ecological data to use in many instances. Several environmental flow modeling and discharge estimations reveal lower confidence in using them in decision support, such as integrated river management schemes for the carp-breeding habitat. Developing a watershed management tool for the Halda Basin is vital for identifying key hydrological processes influencing water availability and evaluating how these may change in the future. Hence, the goals of this study are (i) to reproduce hydrological conditions for the Halda Basin using the SWAT modeling framework to provide a basis for future water resource management and monitoring, (ii) to suggest a novel approach to integrating water level data using the rating curve method to estimate discharge and compare an earlier model prepared by Raihan et al. (2020), (iii) and finally, to estimate sediment load in the river basin, which may significantly affect the flow and fish habitat.
MATERIALS AND METHODS
Halda River catchment area
The SWAT model setup
The SWAT is a semi-distributed, continuous, and semi-physical based deterministic simulation model widely used in assessing hydrology and water quality (Arnold et al. 1998, 2012; Querner & Zanen 2013). Here, the catchment is divided into sub-catchments and further into HRUs based on unique land use or land management, soil attributes, and slope definition (El-Nasr et al. 2005; Gassman et al. 2007; Golmohammadi et al. 2014). Hence, the input information includes climate, soil properties, topography, vegetation, and land management practices (Gassman et al. 2007; Golmohammadi et al. 2014). The SWAT uses the water balance equation for the hydrological phase, divided into land and water or routing phases (Neitsch et al. 2002; Moriasi et al. 2012; Sarwar 2013). SWAT follows the water balance equation for all processes in a catchment (Arnold et al. 2012) and routing of water through the channel network of the basin, carrying the sediment, nutrients, and pesticides to the outlet. SWAT also models the transformation of chemicals in the stream and streambed (Neitsch et al. 2009). Surface runoff is computed by modifying the soil conservation service curve number (Singh et al. 2019).
Digital elevation model, slope, land use, and soil data as model input
DEM: The multi-error-removed improved-terrain (MERIT) DEM version 1.0.3, published on 15 October 2018, is used for this study. It was developed at 3 arc-s resolution (about 90 m) by removing multiple error components from the DEMs like SRTM-90, v. 2.1 and AW3D30, v.1, with a varied temporal extent and 2 m global vertical accuracy (Yamazaki et al. 2017). It covers land areas between 90N and 60S and is projected to WGS84 and EGM96. From DEM during the HRU creation process, five quantile slope classes were derived, viz. flat to nearly level, very gentle, gentle slope, sloping, and moderately steep, covering 23.61, 11.11, 13.79, 14.43, and 37.06% of the total area, respectively.
Land use: This study used the GlobCover, a European Space Agency (ESA) initiative to serve a 300-m global land cover map, produced from an automated classification based on the ENVISAT's (ESA Environmental Satellite) Medium Resolution Imaging Spectrometer (ESA & UCLouvain 2011). The GlobCover image of December–January 2009 with 13 land use classes reclassified to 11 categories, projected to WGS84 ellipsoid, according to the SWAT land cover/plant growth and urban growth database (Figure 2).
Soil data: The harmonized world soil database is used in this study, which combines existing regional and national updates of soil information worldwide with the information of FAO-UNESCO World Soil (1995) Map (FAO et al. 2009). Three major soil classes – ferric acrisols (2.07%), dystric cambisols (62.05%, weakly to moderately developed soils in steep hilly slopes and Eutric Gleysols (35.87%, permanent or temporary wetness near the surface or the stream networks) – are present in the Halda Basin (Figure 2).
Weather generator (WGEN) and observed precipitation data
Climate data are considered one of the primary inputs for hydrological process simulation in SWAT. Climate data includes precipitation, maximum and minimum temperature, solar radiation, wind speed, relative humidity, and the weather generator file. We have used climate data from the National Centers for Environmental Prediction Climate Forecast System Reanalysis to prepare the WGEN user database using two applications, ‘dewpoint’ and ‘pcpSTAT’ as described by Dile & Srinivasan (2014) as weather generator data. Moreover, daily rainfall data from 1967 to 2017 for Narayanhat, Fatikchari, Nazirhat, and Hathazari stations of this river catchment were collected from the BWDB (Figure 2).
Hydrological model setup in QSWAT
The open-source Quantum GIS (QGIS) 3.16 and QSWAT 3.1, SWAT 2012 editor, were used to develop the hydrological model. For automatic watershed delineation, the DEM raster MERIT 90 m was used. Stream networks may differ (Reddy & Reddy 2015) from the observation corrected for the drainage network by digitizing and georeferencing from the Google Earth and outlet points of ‘khals’ from field survey verification, respectively. A 5 km2 area threshold value was used for stream network creation in this river catchment, as suggested by Datta et al. (2022). During the HRUs creation process, land use, soil, and slope bands data were added using only the dominant proportion, assuming a negligible influence from these parameters to streamflow to reduce processing time (Her et al. 2015). Then WGEN and rainfall data were added to the SWAT editor. Here, we defined the rainfall distribution as skewed normal. The model was run daily from January 1990 to June 2015, with the first 3 years of simulation as a warm-up period to stabilize the model. For model calibration, 1993–1997 and validation, 1998–2005 period was used. The simulation was extended for an additional 10 years as an extended validation and prediction period to understand the performance of the model.
Water level and discharge data
The time series of water levels from 1967 to 2017 (2.12% missing data) and discharge data from 1983 to 2017 of the Halda River were collected from the BWDB for this study. Monthly flow data from the Panchpukaria Hydrometric Station (Station ID: 119.1) were used for model calibration and validation. These data were incomplete and inconsistent. As daily discharge data are not continuous, measured discharge and water level (twice per month) data are used to construct a stage–discharge relationship to generate daily discharge data after filtering outliers (Figure 3).
Stage–discharge relationship formulation for stream discharge estimation
From the rating curve relationship of observed discharge data with stage, the value for C is 0.0034, a is 0, b is 4.007696, and the model R² value is 0.69.
Sediment load estimation
SY is the sediment yield on a given day (metric tons), is the surface runoff height (mm H2O/ha), is the peak runoff (m3/s), is the Area of HRU, is the USLE soil erodibility factor (0.013 tons-m2 h), is the USLE cover and management factor, is the USLE support practice factors, is the USLE topographic factor and F is the course fragment factor (Wischmeier & Smith 1978).
Calibration and validation
Calibration and validation play important roles in decreasing uncertainty and increasing the predictive abilities to make the model effective. For the calibration period (1993–1997), values were adjusted by inputting the fitted values (Figure 4). Once the model is calibrated with soil and vegetation parameters, the model is validated with modified parameters. The model has been validated from 1998 to 2005.
Global sensitivity analysis and statistical evaluation of uncertainty
The sensitivity of the flow simulation in SWAT is usually defined by using p-factor, r-factor, and t-statistic as global estimates. The p-factor denotes the proportion of observed data within the 95% prediction uncertainty (95PPU), and the r-factor defines the average width of the 95PPU band divided by the standard deviation of the corresponding measured variable. The 95PPU is estimated at 2.5 and 97.5% levels of the cumulative distribution of an output variable obtained through Latin hypercube sampling (multiple regression system) by omitting 5% of the worst simulations against the selected objective function values (Van Griensven et al. 2006, Abbaspour 2007). The t-stat is the coefficient of a parameter divided by its standard error, used to identify the relative significance of each parameter. In this analysis, the larger the absolute value, the value of the t-stat, and the smaller the p-value, the more sensitive the parameter. Though the r-factor ranges from 0 to infinity, p > 0.70 and r < 1.5 were recommended for discharge (Abbaspour 2007).
Model performance evaluation
Calibration and validation performance was further assessed using performance ratings (Moriasi et al. 2007, 2015) of the coefficient of determination (R2), the Nash–Sutcliffe efficiency (NS), percent bias (PBIAS), and the ratio of the root mean square error (RMSE) to the standard deviation of measured data (RSR) (Table 1).
Model name . | Performance evaluator equation . | Description of the variables . | Qualitative range . | References . |
---|---|---|---|---|
Ratio of RMSE to observation standard deviation | q is a variable (discharge). m and s are measured and simulated data. Q is the arithmetic mean for respective observations | 0.0 ≤ RSR ≤ 0.50 | Moriasi et al. (2007) | |
Coefficient of determination | 0.75 ≤ R2 ≤ 1.0 | Moriasi et al. (2007, 2015) | ||
Percent bias | PBIAS ≤ ±10 | Gupta et al. (1999) | ||
Nash–Sutcliffe efficiency | 0.75 ≤ NS ≤1.0 | Nash & Sutcliffe (1970) |
Model name . | Performance evaluator equation . | Description of the variables . | Qualitative range . | References . |
---|---|---|---|---|
Ratio of RMSE to observation standard deviation | q is a variable (discharge). m and s are measured and simulated data. Q is the arithmetic mean for respective observations | 0.0 ≤ RSR ≤ 0.50 | Moriasi et al. (2007) | |
Coefficient of determination | 0.75 ≤ R2 ≤ 1.0 | Moriasi et al. (2007, 2015) | ||
Percent bias | PBIAS ≤ ±10 | Gupta et al. (1999) | ||
Nash–Sutcliffe efficiency | 0.75 ≤ NS ≤1.0 | Nash & Sutcliffe (1970) |
The coefficient of determination, denoted R2, is the proportion of the variance of the dependent variable to the predicated independent variable(s). The NS efficiency is a normalized statistic that determines the relative magnitude of the residual variance compared to the measured data variance (Nash & Sutcliffe 1970). The NS efficiency indicates how well the plot of observed versus simulated data fits the 1:1 line. The NS = 1 means a perfect match, and the NS = 0 corresponds to the model data being as accurate as the observed mean. The Inf < NS < 0 represents that the observed mean is a better predictor than the model data. PBIAS measures the average tendency of the simulated data to be larger or smaller than the observations. The optimum value is zero; low magnitude values indicate a better simulation. Positive values indicate model underestimation and negative values indicate overestimation (Gupta et al. 1999). RMSE is the standard deviation of the residuals (prediction errors), which measures the distance from the regression line data points. The ratio of the RMSE to the standard deviation of estimated data gives a standardized RMSE value as RSR that varies from zero to positive values. The lower the RSR, the better the model fit (Moriasi et al. 2007).
RESULTS AND DISCUSSIONS
Discharge estimation using an innovative mixed rating curve method
A mixed rating curve combining power function (lean flow phase) and Microsoft Excel solver (peak flow phase) data was formulated for daily discharge estimation to calibrate and validate flow simulation. The rate of discharge change for a given portion of the stage–discharge curves differs in a hydrograph's rising and falling limbs. Low flow discharge occupies a relatively small part of channels, and overflows occur through the vast flood plains during flood flow in this river. As a result, a highly variable discharge with distance from the main channel was common throughout this river. Again, after the flood peak, the recession stage, water reenters the stream and causes an unsteady flow, producing a stream slope more diminutive than that for a constant discharge (Figure 5).
Flow simulation using SWAT
The model was run daily from January 1990 to December 2015, with the first 3 years of simulation as a warm-up period to stabilize the model. For model calibration, 1993–1997, and validation, the 1998–2005 period was used. This period included one major flood year (1991) and one drought year (1994), which allowed us to evaluate the model's ability to simulate extreme flow events associated with the monsoon-driven climate.
For discharge, the p-factor and r-factor recommended a value of 0.7 or 0.75 and a value of 1.5 to be adequate (Abbaspour et al. 2015). Hence, this study acquired acceptable values for both factors (Tables 2 and 3). The simulation performance matrix comparing the monthly flow model developed by Raihan et al. (2020) performed better, except for the PBIAS value for the calibration period. However, the relative improvement in the daily flow simulation compared to the monthly flow simulation done by Raihan et al. (2020) was insignificant. The lag time from the rainfall peak to the discharge point (time of concentration) in the Halda River catchment is less than a day. Since the rainfall–runoff relationship cannot be evaluated for a half-month or monthly simulation for this catchment, the runoff generated by a single storm may fall in an early or late phase of the observation, which limits the application of flow simulation for river management. Additionally, variation applies to DEM uses, land use change, and discharge data of BWDB (observation error). The BWDB data reports every 15 days, and a reliable interpolation method was required to comprehend the daily discharge. However, the lack of regular flow monitoring data did not affect the quality of flow simulation using water level data from a nearby station using the adapted rating curve method used in this study.
Model performance . | Calibration . | Validation . | Preferred value range . | |
---|---|---|---|---|
Before . | After . | |||
Nash-Sutcliffe efficiency (NSE) | 0.45 | 0.81 | 0.61 | 0.75 ≤ NSE ≤ 1.0 |
R2 | 0.73 | 0.80 | 0.62 | 0.75 ≤ R2 ≤ 1.0 |
RSR | 0.74 | 0.45 | 0.62 | 0.0 ≤ RSR ≤ 0.50 |
PBIAS | +0.5 | −6.9 | −0.8 | PBIAS ≤ ±10 |
Model performance . | Calibration . | Validation . | Preferred value range . | |
---|---|---|---|---|
Before . | After . | |||
Nash-Sutcliffe efficiency (NSE) | 0.45 | 0.81 | 0.61 | 0.75 ≤ NSE ≤ 1.0 |
R2 | 0.73 | 0.80 | 0.62 | 0.75 ≤ R2 ≤ 1.0 |
RSR | 0.74 | 0.45 | 0.62 | 0.0 ≤ RSR ≤ 0.50 |
PBIAS | +0.5 | −6.9 | −0.8 | PBIAS ≤ ±10 |
Parameter . | Fitted value . | t-stat . | p-factor value . |
---|---|---|---|
R__CN2.mgt | −0.058 | 4.729 | 0.00 |
V__GWQMN.gw | 5 | −1.609 | 0.108 |
V__GW_DELAY.gw | 150 | 0.396 | 0.691 |
V__ALPHA_BF.gw | 0.55 | 4.287 | 0.00 |
V__REVAPMN.gw | 34.7 | −1.204 | 0.228 |
R__SOL_AWC(..).sol | 0.021 | −0.685 | 0.493 |
R__SOL_K(..).sol | 0.32 | −0.896 | 0.370 |
V__ESCO.hru | 0.81 | 1.563 | 0.118 |
R__SLSUBBSN.hru | 0.02 | 2.179 | 0.0297 |
V__ALPHA_BNK.rte | 0.0583 | 11.064 | 0.00 |
V__CH_N2.rte | 0.1 | −3.368 | 0.00 |
V__CH_K2.rte | 33.97 | −4.811 | 0.00 |
V__GW_REVAP.gw | 0.0447 | −0.734 | 0.462 |
Parameter . | Fitted value . | t-stat . | p-factor value . |
---|---|---|---|
R__CN2.mgt | −0.058 | 4.729 | 0.00 |
V__GWQMN.gw | 5 | −1.609 | 0.108 |
V__GW_DELAY.gw | 150 | 0.396 | 0.691 |
V__ALPHA_BF.gw | 0.55 | 4.287 | 0.00 |
V__REVAPMN.gw | 34.7 | −1.204 | 0.228 |
R__SOL_AWC(..).sol | 0.021 | −0.685 | 0.493 |
R__SOL_K(..).sol | 0.32 | −0.896 | 0.370 |
V__ESCO.hru | 0.81 | 1.563 | 0.118 |
R__SLSUBBSN.hru | 0.02 | 2.179 | 0.0297 |
V__ALPHA_BNK.rte | 0.0583 | 11.064 | 0.00 |
V__CH_N2.rte | 0.1 | −3.368 | 0.00 |
V__CH_K2.rte | 33.97 | −4.811 | 0.00 |
V__GW_REVAP.gw | 0.0447 | −0.734 | 0.462 |
Multiple simulation iterations were executed with 500 simulations each run to achieve the best model efficiency between the observed and simulated flows.
Baseflow flow characteristics
Following the rubber dam project operation just 5 km upstream of the monitoring station, Panchpukuria, since 2011, the stream's low flow water was significantly affected. Raihan et al. (2020), Saha et al. (2019), and Akter & Ali (2012) and field observation reported a lower base flow to the downstream and a reservoir due to the rubber dam being operational during dry months (January to May). However, low flow simulation is still a significant challenge in SWAT (Leisenring & Moradkhani 2012) due to higher uncertainty in groundwater flow in SWAT (Rostamian et al. 2008), which is affected mainly by not only the surface process but also subsurface processes and characteristics such as hydraulic conductivity, porosity, and geological formation. Considering the catchment characteristics of the Halda River following several field visits to the area, we find that the groundwater system would represent a complex one to correctly represent the recharge process in SWAT (Shao et al. 2019) (Supplementary information Figure S1). This could be attributed to the prolonged baseflow recession in its wide floodplain's thick soil layer on both riverbanks at some sub-catchments, which could not be measured by the gauge station. This limitation can be overcome by coupling groundwater flow (e.g., MODFLOW) modules or similar. The rainfall streamflow ratio is significantly higher, indicating a limited role of groundwater storage in the river flow. The influence of the soil layer on the groundwater flow has been articulated adequately in Raihan et al.’s (2020) study of this basin. We have observed a similar fitting parameter value for groundwater flow (Table 2).
Effect of land use on streamflow and sediment load
Numerous studies suggested that hydrological responses to deforestation are very inconstant and often hard to comprehend. The evapotranspiration changes are regarded as the main effect of deforestation since they influence soil water-holding capacity and the interception of precipitation (Zhang et al. 2014). Deforestation generally increases, and reforestation decreases the annual flow (Ma et al. 2009; Hlásny et al. 2015). Usually, vegetation removal intensifies shallower root distribution and reduces soil porosity and soil moisture capacity (Zhang et al. 2014). Hlásny et al. (2015) specified that deforestation alters the pathways and the timing of runoff, thereby changing the volume and timing of flood peaks. Afforestation has a greater infiltration rate, which may reduce runoff peak and total runoff volume (Zhang et al. 2014), and a new approach in rainfall–runoff simulation may enhance the simulation performance (Esmaeili-Gisavandani et al. 2021). However, we have considered a minor effect on streamflow in this watershed due to the large proportion of its lowland covered with agricultural land or shrub vegetation, which would primarily affect the sediment load and nutrient flow rather than the fluid.
In addition to the total sediment estimate, the siltation survey data present river siltation-erosion dynamics and upstream–downstream balance. Figure 8 shows that the upstream locations, RHMLD10 to RHMLD12, experience a small volume of siltation between the siltation survey 2003–2006 (Supplementary Information Table 1); however, the river course remains stable until it meets a tributary before RHMLD05. The sediment load estimate at the Panchpukuria location using SWAT lies between siltation survey station RMHLD08 and RMHLD09. Here, the daily estimate of 50 tons/day in the SWAT simulation supports the siltation survey estimate in this location during the 2003–2014 period. However, erosion continued at the RMHLD05 survey section throughout the siltation survey period, 2003–2014, possibly due to higher flow from the tributary (Saha et al. 2019) but lower sediment load in river water causing higher riverbank erosion. The downstream cross-section data shows an accretion trend. In contrast, the middle and mid-downstream showed the most soil erosion during the study period 2006–2009. This river hypsometry depicts this river catchment mainly as plainland; consequently, the regulation or recession of flow affects most downstream compared to the upstream area. However, this river's accretion and bank erosion were not very intense (Figure 8) compared to other regional and transboundary streams in Bangladesh (Karmakar et al. 2009). From this river dynamics, it is also noticeable that flow regulation in the upstream region poses a more significant impact downstream than a very intimate part of the regulation point. From the hydrological data, we cannot univocally articulate the impact of flow regulation on the siltation or erosion, as well as the effects of any other land use measures that may since some changes, such as channel modification and sand quarry in the riverbed concurrent on this river (Saha et al. 2019). The flow regulations for irrigation and diversion canals have also become a regular phenomenon in the last two decades to boost agricultural production in the eastern part of this catchment. Since sediment data are unavailable at the discharge monitoring station, sediment estimation using SWAT provides a management implication and the importance of monitoring the data. Since this river provides many ecosystem services and freshwater supply to the city, conserving the downstream habitat and riverbank would best serve for flow and siltation dynamics and vice versa.
CONCLUSIONS
Despite medium confidence with the error matrix of the discharge estimation for calibration and validation periods, the daily flow simulation results agree very well with previous simulation results. Moreover, this study shows that the rating curve of flow-water level with an adaptive approach for low and peak flow regimes can be used in data-poor or low-temporal resolution data regions for watershed management. The hydrological water balance analysis revealed that groundwater flow is an essential element of the total discharge within the study area. Sediment load downstream was attributed to the steep slope area of the catchment, land use change, higher precipitation, sediment movement along the channel, and tidal resuspension of the river sediment. The effect of climate, land use, and watershed management such as afforestation and soil conservation practices can be integrated with this flow model to understand river hydrology in this region. Eventually, that would lead to water quality modeling research for nutrient transport in a data-poor catchment of Bangladesh to ensure sustainable and integrated river water management in critical water habitats.
ACKNOWLEDGEMENTS
This research is funded by the Research and Development cell funding of the University of Chittagong.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.