Modeling and Monitoring of Drawdown Flushing and Dredging Toward Sustainable Sluicing in a Wide Philippine Reservoir

Abstract

Reservoir sedimentation, a global challenge causing an annual loss of 0.8–1% of reservoir storage capacity, disrupts fluvial sediment continuity and impacts ecology and societal needs. This study focuses on the Pulangi IV reservoir in the Philippines, a shallow and wide reservoir facing significant sedimentation issues. The research aims to investigate drawdown flushing and dredging of a flushing channel for future sustainable drawdown sluicing. A test flushing event was conducted and monitoring data, including discharge, suspended sediment concentration, bathymetry, and grain size distribution, were collected. Laboratory analyses, such as critical shear stress tests, were performed for model calibration. A 3D reservoir model and a 1D sediment transport model were applied incorporating cohesive sediment behavior. Scenarios were simulated to assess drawdown flushing, dredging and downstream impacts. Results highlight the importance of drawdown level, with cohesive sediment properties playing a critical role. Sedimentation downstream of the dam, resulting from dumped or flushed sediments, was low. However, downstream ecological and morphodynamic monitoring was found to be essential for all modeled strategies. This study demonstrates potential for establishing a flushing channel enabling future sustainable drawdown sluicing during floods by conducting repeated drawdown flushing in combination with dredging in the upper reservoir.

1. Introduction

Reservoir sedimentation totaling more than 100 billion metric tons is a critical global challenge interrupting sediment continuity from rivers to the ocean [1]. According to Annandale et al. [2], sedimentation causes an annual loss of 0.8–1% of reservoir storage capacity, as reported by the International Commission on Large Dams (ICOLD) [3], and this net loss persists despite the continued construction of new dams. Reservoir sedimentation disrupts ecological and societal systems both upstream and downstream, with upstream sedimentation degrading aquatic habitats [4], increased flood levels upstream of reservoirs [5,6] and downstream sediment starvation causing riverbed degradation [4,7], loss of habitat [4], as well as fine sediment deposits and infiltration into the riverbed [2,8,9,10,11]. Moreover, sediment accumulation in reservoirs reduces their storage capacity, compromising reservoir benefits and revenue [12]. These impacts develop incrementally until the end of the reservoir’s lifetime under the “design life approach” for reservoir planning, when deposits encroach on dam structures and intakes [2,13]. All reservoirs act as sediment traps [14] resulting in unsustainable storage capacity and sediment deficits downstream of the reservoir. Sustainable long-term sediment management is required to maintain long-term reservoir storage, instead of planning for reservoir lifetimes of e.g., 50 to 100 years [13,15]. This sustainable approach requires the appropriate design of new dams and, to a larger extent, extending the lifetime of existing reservoirs against the challenge of sedimentation [16]. The aim is to transform reservoir storage into sustainable capacity, satisfying inter-generational equity. This involves implementing sediment management techniques and modification of conventional economic evaluation of reservoir projects to favor sustainable outcomes [2]. Sedimentation problems exacerbated by additional factors including land use changes, changes in discharge, acceleration or deceleration of soil erosion, and climate change [17,18,19]. Impacts of sedimentation on benefits including hydropower generation, water supply, floods, aquatic ecology, infrastructure or dam safety, are summarized in [4,5,15].

Two categories of hydropower plants can be distinguished [20]. The first, storage hydropower, involves the retention of water in a reservoir, released as needed to shift wet season releases into periods of lower flows to sustain power production [20]. Active storage is large compared to dead storage, and may additionally be used for irrigation, water supply or flood retention [15]. The second, run-of-river hydropower plants, as described by [20,21], operate without storage capacity or with capacity only sufficient for daily power peaking. Dead storage may be larger than active storage [15]. Okumura & Sumi [12] found a negative impact on power generation (reduced water use efficiency) due to the loss of storage capacities in both storage and run-off river hydropower plants.

The Pulangi IV hydropower plant in the Philippines exemplifies the challenges posed by reservoir sedimentation, in this case for a shallow and wide reservoir. Commissioned in 1985, the reservoir operated as a run-of river hydropower plant with an initial storage capacity of 70 Mm³. However, the 2019 bathymetric study revealed a 76% loss of capacity, with only 17 Mm³ remaining [22]. Without intervention, projections suggest that reservoir capacity could drop to as low as 5 Mm³ by 2030 [23], which may be considered the end of the reservoir’s useful lifetime. The project’s feasibility study [24] estimated a reservoir lifetime of 50 years (reached in 2037), by which time the original trap efficiency, estimated at 40% (after [25]), would be reduced to zero. Data from the National Power Corporation (NPC) [22] estimated a sedimentation rate 1.70 Mm³y⁻¹ from 1987 to 2014, but capacity loss has more recently decreased to 0.79 Mm³y⁻¹ due to three phases of dredging operations with a total volume of around 1.25 Mm³y⁻¹ starting in 2007, plus reduced trap efficiency. This reservoir’s shallow and wide geometry exacerbates sedimentation, with sediment deposits forming wide deltaic lobes and near-dam deposits, a geometry ill-suited to sediment removal by the much narrower channels created by sluicing or flushing. These challenges highlight the need for effective and sustainable sediment management strategies tailored to the unique characteristics of shallow and wide reservoirs. Further challenges arise from the effect of Pulangi IV reservoir on the downstream reach, sediment starvation as sediments accumulated in the reservoir, and the potential for extreme sediment releases during management, both of which may impair ecological conditions below the dam.

Multiple sediment management options for reservoirs are available [26,27,28,29]. Morris [13] published a classification of management alternatives regarding reservoir sedimentation which included: (i) reducing sediment yield, (ii) routing sediments to minimize deposition, (iii) removing deposited sediment and (iv) applying adaptive strategies to minimize sediment impacts. With regard to management category (i) “reducing sediment yield” from Morris [13], sediment yield due to land conversion and climate change in the Pulangi IV catchment are already higher than a tolerable amount (11.2 t ha⁻¹y⁻¹) and simulations of the Pulangi IV catchment indicate that sediment yield may change between −2.4% (2050) and +12.4% (2080) for RCP 4.5, and between −0.6% (2050) and +20% (2080) for RCP 6.0 [30]. For category (ii), drawdown sluicing during floods and ‘compartmented reservoirs’ were mentioned as sediment pass-through strategies [13]. For category (iii) dredging as well as empty flushing were given as examples for sub-categories ‘mechanical removal’ and ‘hydraulic scour’, respectively [13].

Sluicing was described in detail by Morris [13] as the maximization of sediment passage through a reservoir by lowering the water level during a flood, which is most effective in hydrologically small reservoirs (such as in this case study). It was also noted that sluicing is inexpensive compared to other methods, such as dredging, and that it might become increasingly relevant later in the reservoir’s lifetime because of shrinking reservoir capacity. One major advantage cited is that it does not result in high suspended sediment loads downstream during flood events. This contrasts to emptying and flushing, which produce concentrations much higher than natural levels. This makes sluicing an environmentally friendly technique. An example of optimized sluicing operations for shallow and wide reservoirs reducing trapping of sediments is given in [31]. A “compartmented reservoir” refers to the hydraulic short-circuiting of sediment-laden inflows within the reservoir by using internal barriers or modified geometries created by dredging [13]. This strategy is most relevant for wide and shallow reservoirs, like Pulangi IV.

In contrast, Morris [13] mentioned that flushing releases high peaks in suspended sediment concentrations (SSC > 100 g/L), which can require complex environmental monitoring and mitigation. Morris and Fan [16] also describe flushing as the emptying of the reservoir to allow the river to erode deposits and discharge them through low-level outlets (LLO). The process was noted to typically consist of three stages: (i) the ‘drawdown’ stage, where upstream sediments are transported to the LLO with relatively smaller amounts of sediment being released; (ii) the ‘empty’ stage, during which the transported sediments are released through the LLO, producing peak sediment concentrations; and (iii) the ‘refill’ stage, a period when clear water may be released downstream to help flush the sediments through the river system. Flushing and sluicing operations are mentioned by CIGB, ICOLD [32] to have a major constraint, which is the requirement for excess water availability. Sediment pass-through is mentioned to be successful when the reservoir capacity is less than 5% of the mean annual runoff, which is the case for the Pulangi IV reservoir. Favorable conditions for drawdown flushing include: steep longitudinal slope and narrow valley (narrow reservoir) [16,33]. Lacking these conditions, flushing efficiency is low for shallow and wide reservoirs. Lai et al. [34] propose factors for drawdown flushing efficiency, including width, and depth, sediment composition, inflow of water and sediment, outlet configuration and reservoir operation. Kantoush et al. [35], for example, experimentally investigated the influence of free-flushing, pressure flushing and drawdown flushing regarding the width and shallowness of reservoirs and found a negative correlation of flushing efficiency with the width and depth of the reservoir. Drawdown flushing was by far the most efficient measure for sediment removal, but the sustainable reservoir capacity depends on the ratio between the width of the self-forming flushing channel and the width of the reservoir. A flushing channel much narrower than the overall width of a wide reservoir limits the ratio of sustainable capacity to original capacity [34] when dependent on flushing alone. A RESCON2 [36] assessment of Pulangi IV reservoir indicated that drawdown flushing, sluicing and dredging could contribute to sustainable management of the reservoir. For category (iii) use of dredging as management strategy [13], the disposal of dredged material from the reservoir and its downstream effects on morphology and ecology need consideration [4,8]. Permitting limitations and costs can also represent important constraints [5].

State-of-the-art methods for assessing sediment management measures span various spatial dimensions and approaches, including economic evaluation tools like RESCON2, numerical models combined with monitoring, and physical models. Sumi et al. [27] provide examples of 1D and 2D numerical studies, while highlighting the growing application of 3D numerical models, particularly in scenarios where velocity variations over the flow depth, such as in channel bends, are critical. In contrast, Annandale et al. [15] primarily discuss 1D and 2D models, with 1D models being used for long-term assessments of entire reservoirs and 2D models for detailed simulations near dams, where lateral water and sediment movement is significant, or for short-term events such as individual floods. Although Annandale et al. [15] explicitly mention 3D models only in the context of physical models, the increasing use of multidimensional numerical models is attributed to advancements in computational capacity. Lai et al. [34] provide a comprehensive review of hydraulic flushing in reservoirs, utilizing numerical models ranging from 1D to 3D, including fully Navier–Stokes-based models without the hydrostatic assumption, offering insights into their application and limitations. In this review, non-hydrostatic’ 3D model applications on drawdown flushing are described [37,38,39,40,41], reproducing flushing channels in various degrees of similarity. E.g., Ref. [40] simulated a relatively narrow reservoir using 3D hydrodynamics with a 2D numerical sediment transport model, successfully validated with observations, where a flushing channel developed during drawdown. The model was then found to be a successful tool to propose strategies to increase flushing efficiency. Lai [34] mentions that model calibration is not critical for 3D hydrodynamics, however limitations on 3D models are given especially due to scales in space and time. For large reservoirs, one must restrict time scale to a short period, e.g., one empty flushing event, as contrasted to 1D models which can simulate multiple events each year over a period of many decades. Concerning sediment transport, Lai [34] further mentions that the accuracy depends on the adopted empirical sediment transport equation and, due to the occurrence of cohesive material near the gates, proper measurement of cohesive properties. Modeling of cohesive sediments, including their erosion behavior, considers various influencing factors such as physical, electrochemical, and biological processes [42] and largely affects sediment transport and morphology for both rivers and reservoirs. Consolidation also affects cohesive sediment transport, particularly in reservoirs, as sediment layers exhibit varying bulk densities over consolidation time [42].

Moreover, successful modelling of flushing operations heavily relies on a detailed spatial-temporal monitoring of these operations, both within the reservoir and in the river downstream. This includes monitoring discharge, water levels, sediment concentrations, conducting bathymetric surveys, collecting sediment samples, performing laboratory analyses, and accounting for the cohesive properties of the samples [5,15]. Furthermore, ecological monitoring on habitat quality in both the reservoir and the river downstream is essential, especially due to the high amounts of SSC released during flushing. As an example, ref. [8] conducted an event-based monitoring on reach and local scale considering morphology with respect to fine sediments, turbidity, bed composition and fine sediment infiltration combined with monitoring macroinvertebrates and fish.

For the Pulangi IV reservoir, Tabios [43] summarizes a numerical study for pressure flushing (flushing with either zero or only partial water level drawdown), performed in 2005. It showed local effects near the intake, resulting in around 0.05 Mm³ removal for a release of 730 m³s⁻¹ for 12 h through the bottom sluice gates (BS). Tabios [43] introduced an additional numerical study from 2011 (2D model based on [44]) that proposed a sediment flushing protocol based on inflow and reservoir width. The study recommended that higher inflow should correspond to a greater gate opening, resulting in a larger drawdown. He found for 12 h flushing duration that flushed sediment volume increased with greater sluice gate opening and more water level drawdown, whereas reservoir inflow had less effect, with an optimal inflow range of 150 to 200 m³/s with no reduction in hydropower generation. For the investigated sluice gate openings of 0.2 to 2.0 m and the reservoir inflow between 100 and 300 m³, flushed sediment volume was between 10,000 and 50,000 m³. Tabios [43] also found consistency with flushing operations conducted in 2007, indicating 18,000 m³ flushed out sediments in 15 h. Comparisons with test flushing operations performed in 2007 indicated that sediment drawdown flushing operations were successful and should be continued. For [45,46], drawdown sluicing was modeled (numerically and physically) in combination with a dredged channel and a dike (compartmented reservoir created by side-casting dredged material), showing evidence for a highly effective sediment routing through the reservoir during a flood with a two-year return period. However, the dredging volume was substantial (5 Mm³), prompting consideration of a combined approach involving dredging in the upper reservoir and flushing in the lower reservoir before a continuous flushing channel is established. The more sustainable drawdown sluicing operations, such as in [31,45,46], can be considered in the future, as well. This combined sediment management approach also includes discharge of dredged material to the river downstream. Due to the shallow and wide reservoir geometry and the various bifurcated channels crossing the multi-lobe reservoir delta, a finer spatial scale and the use of more detailed, three-dimensional numerical methods were appropriate. Cohesive sediments, which dominate in the lower and middle sections of the reservoir, need to be accounted for during modeling.

The aim of this detailed case study on a shallow and wide reservoir was to evaluate and optimize sediment management strategies, with a particular focus on establishing a flushing channel to enable sustainable sediment passage through drawdown sluicing in the future. This included:

• Monitoring of drawdown flushing based on a single test flushing event.
• Preparation of a calibration dataset including laboratory analyses (e.g., grain size distribution, critical shear stress test).
• Calibration of a 3D numerical reservoir model and a 1D model for the river downstream, both incorporating cohesive sediment behavior.
• Modeling scenarios to develop a continuous flushing channel in the shallow and wide Pulangi IV reservoir enabling sustainable sediment routing through drawdown sluicing in the future.
• Modelling the downstream impacts of drawdown flushing on sediment transport and morphology.
• Analyzing the scouring and transport of dredged sediments from the reservoir, which were dumped in the river downstream of the dam.

2. Materials and Methods

2.1. Pulangi IV Case Study (Mindanao, Philippines)

The Pulangi IV case study (Mindanao, Philippines: coordinate location 7°47′12″ N Latitude and 124°1′28″ E Longitude) was conducted by ESMAP of the World Bank (funded by the Austrian Ministry of Finance). The hydropower plant has an installed capacity operating at 270 m³s⁻¹ and 110 m design head. The earth-filled dam has 7 spillways and 2 bottom sluice gates (BS, Figure 1) adjacent to the intake headworks (HW), delivering water to a headrace channel terminating in a surge pond and penstocks. Pulangi IV is operated as a run-of river project with daily peaking. Catchment size is 3114 km² with a mean discharge of 152 m³/s [24] and a coefficient of variation of 0.22 (data from 1953–1977), indicating a uniformly distributed discharge throughout the year. Pulangi IV was commissioned in 1985 with a volume of 70 Mm³ [47,48] and an average depth of only 6.3 m.

2.1.1. Scenarios

Three scenarios were considered in the reservoir with specific settings of inflow Q_in, initial reservoir water level W_init, drawdown level W_min and gate openings B (

Table 1. Summary of investigated scenarios for the reservoir model and the Pulangi river downstream of the dam.

Parameter	Parameter Description	Unit	Scenarios for Pulangi IV Reservoir Model
			C-75	1-250	2-250
Qin	reservoir inflow	m3s−1	75	250	250
Winit	initial reservoir water level	m	284.62	max OL = 285.5	max OL = 285.5
Wmin	minimum drawdown water level	m	281.0	280.0	280.0
Δh	water level drawdown height	m	3.62	4.5	4.5
B	gate opening for the two bottom sluice gates (BS) during drawdown flushing	m	range: 0 to 5.0 1 average: 1.36	3.0	3.0
QT	turbine discharge	m3s−1	60 ± 22 2	275	275
Qout	reservoir outflow through the two BS gates and the turbines (HW): Qout = f(W, B, QT)		355 ± 269	959 ± 21	959 ± 21
T	simulation time	h	5.83	4.40	4.80
ΔD	dredging volume	Mm3	-	-	2.2
			scenarios for Pulangi river downstream of the dam
			C-75		D
Qin	discharge	m3s−1	Qout of reservoir 2		200 to 800
T		h	5.83		72
ΔD	dumping volume	m3	-		6000 m3

Notes: ¹ Gate opening B was increased stepwise over T. Differences in B between both gates occurred. ² measured during monitoring of the test flushing event.

). Drawdown water level represents the elevation immediately upstream of the sluice gates. Two scenarios were modeled for the Pulangi river downstream of the dam. Scenario C-75 in the reservoir (Figure 2a) reproduces the test flushing event undertaken during monitoring in January 2020 (Section 2.2), where the reservoir inflow Q_in was around half of mean flow. According to the operator’s experience, gates were opened stepwise leading to a maximum reservoir outflow Q_out of 835 m³s⁻¹ before gates were closed. Due to the highly unsteady flow process, the model simulation ended 5.8 h after gates were opened, when reservoir level had lowered from 284.6 to 281.0 m. By comparison, the minimum operating level (OL) is 279.0 m. Scenario C-75 also monitored parameters in the river downstream of the dam (Table 1).

Scenario 1-250 (Figure 2a) and 2-250 (Figure 2b) aimed at the investigation of drawdown flushing effects under more favorable conditions (Table 1) such as elevated inflow (higher than mean flow), additional water level drawdown Δh (1m above min. OL), increased gate opening B at BS and maximum turbine discharge Q_T (set to the design discharge of the hydropower plant). The initial reservoir water level was set to the maximum operation water level (max OL), and drawdown was applied to a level where the drawdown cone was numerically stable (280.0 m). Scenario 1-250 was performed on the existing bathymetry (2019).

For scenario 2-250 (Figure 2b), a dredged flushing channel in the upper reservoir was considered, with the idea of extending the flushing channel in the lower reservoir through drawdown operation alone. The width of the channel was estimated by the empirical relation from Atkinson [49], correlating flushing channel width W_f (m) to flushing discharge Q_f (m³s⁻¹):

W_f = 12.8 Q_f^0.5

(1)

For a flushing inflow of 250 m³s⁻¹ (Q_f) the channel width (W_f) is 202 m (Equation (1)), and this was used as an initial estimate of the flushing channel design width at the midpoint along the channel length. Part of the dredged material was considered for deposition to form a dike along the right-hand side of the channel to help funnel flow along the dredge cut. The total dredged material from the flushing channel amounted to 2,220,445 m³ (Table 1), of which 374,502 m³ was designated for the dike. The other portion of the material was considered to be discharged below the dam (scenario D, Table 1). Side casting of dredged material to form this dike supports the sediment routing strategy ‘compartmented reservoir’ [13]. Figure 2c shows a potential final state of such a flushing channel, where a channel bed slope of 0.003 is predetermined by the connection of bed levels of the river upstream with the level at the bottom sluice gate. This flushed channel would represent 5 Mm³ of sediment if dredged completely and could then be used for future efficient sediment routing through drawdown sluicing.

For the Pulangi river downstream of the dam, an additional scenario (D) was considered to evaluate discharge of material from the dredged channel (Figure 2b) into the river gorge downstream of the dam (Table 1). A potential and accessible site was identified at G (Figure 1a). Dredged material of 6000 m³ was deposited, corresponding to a height of 2 m, and the scouring and transport in the river downstream was modeled for 72 h (Table 1).

2.1.2. Topography

The first available bathymetry and topographic survey dates to year 2000 [22], followed by measurement campaigns in 2010, 2012 (only lower reservoir), 2014 and 2019. Those campaigns, using GPS RTK and total station measurements, had different point densities with highest being in 2010. Part of the bathymetry in 2014 was surveyed using echolot and completed with data from GPS RTK and total stations. The modeling study of Pulangi IV reservoir was performed with the most recent bathymetry data from the 2019 measurement campaign, and gaps were filled up with data from earlier campaigns.

In earlier years the main flow in this wide and shallow reservoir followed its northern and western side. However, as the delta grew in this area, the main flow shifted progressively eastward after 2010, eventually creating various bifurcated channels and three main depositional lobes of in the delta. Bathymetry data were denser along those more-accessible bifurcated channels crossing the delta and in the lower part of the reservoir, representing areas of discharge during mean flow conditions. The various channels were represented by a Digital Elevation Model (DEM) with the Triangulated Irregular Network (TIN) grid established within ARCGIS. A DEM model for the Pulangi River old channel downstream of the dam was also created as a TIN surface. Bathymetric data for Pulangi River downstream of the dam, representative of 2011, were provided by NPC [22], which were combined with the DEM from [50].

2.2. Monitoring and Laboratory Analyses

2.2.1. Field Measurement Campaign 2020 (Test Flushing Event)

Monitoring was performed from 20 to 24 January 2020, focusing on documenting the test flushing event of 22 January 2020. At the reservoir inflow location (BB, Figure 1a) discharge and suspended sediment concentration (SSC) were measured. Reservoir inflow was measured continuously using a 2D electromagnetic current meter (Infinity FM with data logger, JFE Advantech) in combination with water level loggers (HOBO, Onset) using the velocity area method [51]. It consists of measurements of stream velocity (1 s interval), water depth and distance across the channel between observation verticals. Discharge of a vertical was derived from the product of mean normal velocities over the depth v_m, the water depth at the vertical profile and the distance between the verticals. Total discharge in the cross-section was calculated by summing up the discharges of the verticals. Eight verticals were measured at the 34.4 m wide trapezoidal cross section BB. Instantaneous cross-sectional profiling was performed at regular intervals during the test flushing event. SSC was measured continuously using turbidity sensors (Infinity-CLW, Nishinomiya, Japan; JFE Advantech Co., Ltd., Nishinomiya, Japan) in combination with water level loggers (HOBO, Onset, Bourne, MA, USA). Furthermore, bailing samples were taken to validate the turbidity sensors. Cross section sampling using a depth integrated sampler (SGLN, Bundesamt für Umwelt, Bern, Switzerland) and an improvised sampler (water bottle, holder frame and rope) were also performed at BB. Average cross-sectional SSC was calculated by using the integration method by dividing the flow area into several verticals. Depth-averaged velocity-weighted SSC is achieved when the inflow into the measuring device is isokinetic [52].

During test drawdown the reservoir level (provided by [22]) dropped from 284.62 m to 281.0 m. The duration of the water level drawdown was 5.83 h between opening and closing of sluice gates, with turbines operated continuously. Total discharge at the outlet of the reservoir is BS and HW (Figure 1a). Turbine discharge Q_T, the HW abstraction rate (Figure 1a) was provided by the owner [22]. Discharge by the bottom sluice gates (BS) was calculated using a sluice gate equation [43], as discharge could not be measured directly due to high flow velocities and the lack of secure and appropriate accessibility.

Turbidity sensors (Infinity-CLW Nishinomiya, Japan; JFE Advantech Co. Ltd., Nishinomiya, Japan) in combination with water level loggers (HOBO, Onset, Bourne, USA) were installed downstream of the sluice gates (BS, Figure 1a) for continuous measurements. However, data could not be retrieved due to cable breakage by the high discharge velocity. Instead, direct near-bank samples were taken at 5 min intervals. Sampling verticals in the water column was neither possible nor needed as high turbulence created completely mixed conditions.

In the power channel (CB in Figure 1a) discharge was measured using a 2D electromagnetic current meter (Infinity FM with data logger, JFE Advantech, Nishinomiya, Japan) and compared with ADCP (Teledyne Stream Pro, Manila, Philippines) measurements by applying the same method as adopted at the reservoir inflow (BB). Due to the uniform cross-section of the power channel, only 6 verticals were measured covering the total channel width of 42 m. Water level was monitored with level loggers (HOBO, Onset). Temporal variation of SSC was measured continuously using a turbidity sensor (Infinity-CLW, JFE Advantech Co., Ltd.). Cross-sectional SSC was measured by depth integrating samples using similar verticals as for discharge measurement.

A photogrammetric survey was conducted of the gorge of the old river at G. Monitoring was performed approximately 200 m downstream of the Pulangi reservoir, selecting the site based on its accessibility and suitability for water level monitoring. To derive the geometry of the cross-section, terrestrial photogrammetry was employed and analyzed with Agisoft Photoscan Professional (version 1.2.6.2834). Additionally, a water level logger (HOBO, Onset) was installed in the channel at G which, in combination with calculated discharge at BS and HW, served as a basis for calculating a rating curve (Q-W) during reservoir drawdown. This rating curve was needed for the hydrodynamic calibration of the 1D numerical model downstream of the dam.

A single-beam bathymetric survey was performed in the area of the lower reservoir (Figure 1a) before and after the test sluice event. Reservoir samples (Figure 1a) were taken in the lower and middle reservoir by inserting plastic pipes (5 cm dia.) 1.0 m into the deposited sediments. Samples were capped hermetically for transport to the laboratory for analyses.

2.2.2. Laboratory Analyses

SSC analysis for 15 samples was performed in the Laboratory at CMU (Central Mindanao University, Philippines) by 45 µm pore-size filtration assisted by a vacuum pump in accordance with DIN 38 409, 2 [53]. At BOKU University, plastic pipes from the reservoir sampling were cut into various samples representing layers, each around 10 cm depth. Grain density was analyzed for 3 samples using the gas pycnometer method after EN ISO 17892-3 [54] using a Helium pycnometer (AccuPyc II 1340, Micrometrics, Norcross, GA, USA). Dry bulk density and porosity for 10 samples was determined by the ratio of mass of the material of a selected depth layer to the volume defined by the plastic pipe. After drying at 105 °C and weighting, bulk density and porosity were calculated. Critical shear stress tests [55,56] for 16 samples were performed in the laboratory at BOKU University using the test rig illustrated in Figure 3.

This allowed the determination of shear stress limits for erosion as described in [42]. Critical bed shear stress was determined on samples which were aligned with the bed in the measurement section, using hydraulic loss and local measurements based on disturbance-free laser-optical techniques (Particle Image Velocimetry and Laser Doppler Velocimetry). The test reach extends for a length of 2.7 m with a rectangular cross section (0.188 × 0.094 m). By constantly increasing discharge shear stress, threshold values for surface and mass erosion of the sample were detected. To evaluate possible bias caused by storing and possible drying out of samples, 6 samples were also submerged in water for 1 week before performing the tests.

2.3. Numerical Methods

2.3.1. Three-Dimensional Hydrodynamic Model (RSim-3D), 2D Sediment Transport Model (iSed)

Pulangi IV reservoir was simulated using the sediment transport model iSed version 2.16 [57,58], which was coupled with the 3D hydrodynamic model RSim-3D version 2.27 [59]. The model RSim-3D solves Reynolds-averaged Navier–Stokes equations on an unstructured grid using the Finite Volume method. Polyhedral cells with vertical walls are generated from cell centroids by applying Voronoi decomposition [60] and dividing the water column into n cells in the vertical direction. The location and density of cell centroids is defined by the user, which enables full control of the stored variables in the domain. Moreover, orthogonality of cell faces is guaranteed enabling non-orthogonal terms to be neglected in the discretized equations, reducing numerical diffusion. Turbulence closure is provided by two-equation models, where the k-ε model by Launder and Spalding [61] is the standard method. The model works with steady state as well as unsteady conditions. Details on the model are given in Tritthart [59] and Tritthart & Gutknecht [62]. The model was successfully applied in various studies on rivers [63,64], reservoirs [65,66] and laboratory experiments [67,68].

The sediment transport model iSed [57,58], coupled (two-way) with a hydrodynamic model (e.g., RSim-3D), is capable of simulating both bedload and suspended sediment transport. For bedload transport a variety of formulas that account for non-uniform sediment transport can be selected. Suspended sediment transport is implemented by a 2D or 3D equation of convection-diffusion type including additional exchange terms for the interaction with the riverbed [69,70]. The evolution of the riverbed over time follows the Exner equation. Grain sorting is implemented by an exchange layer concept. Cohesive sediments are implemented in iSed after [42]. The integration in the model is shown in [65]. Concerning aggregation of particles due to cohesion, settling velocity can be adapted by the modeler with respect to the sediment concentration to account for hindered settling [71]. On the other hand, a computation using the empirical formulation of settling velocity after Krone [71], derived in laboratory studies, can be selected:

w_c = Ks^4/3,

(2)

where K represents an empirical constant and s is the suspended sediment concentration in g/L. Deposition occurs when the bed shear stress falls below the critical shear stress (τ_d,part), with the deposition rate determined by its relation to the critical shear stress thresholds (τ_d,full and τ_d,part), the settling velocity (w) and a probability factor P_d:

$$P_d=1-{\displaystyle \frac{\tau }{{\tau }_{d,full}}}$$

(3)

The partial deposition rate Q_d (for τ_d,full < τ < τ_d,part) additionally considers an equilibrium sediment concentration c_eq:

Q_d = P_dw⋅(c − c_eq)

(4)

P_d is set to 0 when τ < τ_d,part, turning off deposition. Erosion is implemented in two stages after [42]. Surface erosion applies when bed shear stress τ is larger than a critical shear stress for surface erosion (τ_c,s), shifting to mass erosion when the critical shear stress value for mass erosion (τ_c,m) is exceeded. Erosion rate is integrated using surface erosion rate M_se and mass erosion rate M_me, respectively. Erosion flux after [65] for surface and mass erosion (s_ero,s, s_ero,m) is implemented according to the following equations:

$$s_{ero,s}={\displaystyle \frac{M_{se}}{{\rho }_b}}{\displaystyle \frac{\tau -{\tau }_{c,s}}{{\tau }_{c,m}-{\tau }_{c,s}}} \mathrm{for}{\tau }_{c,s}\le \tau \le {\tau }_{c,m}$$

(5)

$$s_{ero,m}={\displaystyle \frac{M_{me}}{{\rho }_b}}{\displaystyle \frac{\tau -{\tau }_{c,m}}{{\tau }_{c,m}}}+{\displaystyle \frac{M_{se}}{{\rho }_b}} \mathrm{for}\tau \le {\tau }_{c,m}$$

(6)

where ρ_b represents the bulk density. Further details on the cohesive model, implemented in iSed, are given in [42,65].

2.3.2. One-Dimensional Hydrodynamic and Sediment Transport Modeling by HEC-RAS

HEC-RAS is a widely used 1D numerical model including hydrodynamic and sediment transport modeling. Similar to the sediment transport model of the reservoir, the model for the river downstream of the dam is also capable of incorporating cohesive sediments [72]. Details are given in [73].

2.4. Numerical Model Implementation

2.4.1. Reservoir Model (RSim-3D, iSed)

Steady state 3D hydrodynamic model runs were performed by matching model outflow and inflow, and these were used as the initial condition for modeling unsteady water level drawdown. These unsteady 3D hydrodynamic model simulations were conducted using a fixed bed. This model run was refined by unsteady modeling using the two-way coupled approach of the models RSim-3D and iSed. Such a model approach was needed to avoid instabilities regarding the outflow condition at BS and the establishment of a stable drawdown cone. Time step Δt in the fixed bed model was chosen between 360 and 2880 s. The number of iterations per time step, as well as the relaxation parameters essential for 3D hydrodynamic modeling [62], were optimized to reach sufficiently small residuals as well as a reliable mass conservation.

The following unsteady morphodynamic model runs, featuring two-way coupling between iSed and RSim-3D, were applied using the unsteady fixed bed runs as initial condition including flow fields and boundary conditions. The morphodynamic time step was set between 90 s at the beginning of the drawdown, but requiring refinement to a value between 15 and 5 s during the later stages of simulation. The level of refinement strongly depended on the model’s ability to reach the target water level at the sluice gate, estimated by the previous fixed-bed 3D hydrodynamic model. The bed at the model outlet needed to be fixed to reach numerically stable solutions in the drawdown cone near the BS.

The unstructured computational mesh for the 3D numerical model used 284,466 hexahedral cells distributed over 6 vertical layers. Cell size varied over the DEM domain to represent the 2019 surface. Distance between cell centroids was generally 35 m. Smaller average cell sizes (5 to 15 m between cell centroids) were used at the inflow channel to the reservoir, the bifurcated channels in the upper reservoir, and the central part of the middle and lower reservoir, where scouring was expected during drawdown. The smallest cell sizes were applied near the dam at the BS using a rectangular configuration (2 × 1.5 m). A mesh adaptation was needed along the flushing channel (scenarios 1-250 and 2-250) using a rectangular arrangement instead of a hexagonal one for improved representation of the linear geometrical configuration.

The DEM was interpolated onto the mesh, and the location of the upper- and lower boundary Q_in, and Q_out were defined. Discharge through the power intake (HW) was added to the outflow boundary of the BS gates, as these two structures are adjacent to one another and this simplification was not expected to influence results. The sluice gate geometry was approximated by blocking the upper three cell rows in the 3D mesh.

The sediment transport model iSed was built on the same mesh configuration defined in RSim-3D using 4 layers with a total depth of 3.5 m defined based on reservoir sampling (Section 2.2.1). The sampled grain size distribution was interpolated over the computational domain for each layer based on the samples (Section 2.2.2). Sediment properties including grain density, pore content and cohesive parameters were defined according to the values from sample analysis to compute suspended sediment transport for cohesive behavior implemented after [65]. Bedload transport was omitted due to lack of information on the sediment properties in the upper reservoir including the various bifurcated channels (Figure 2a). Additionally, the bed was assumed to be fixed in these areas to avoid instabilities arising from high flow forces in combination with missing information on larger grain size fractions (e.g., sand).

Modelling reservoir drawdown was computationally expensive due to the large domain and the small time scales. One of the 790 nodes at the Austrian VSC4 supercomputer [74], comprising two Intel Skylake Platinum 8174 Processors with 24 cores each, was used. Depending on the number of iterations and the relaxation coefficients the simulation time was up to 25 times real time. Thus, the sensitivity analysis of the reservoir model considering cohesive parameters was applied only for the last 20 min of drawdown, where changes in bed levels due to erosion were expected to be highest. Sensitivity of other sediment transport parameters is evaluated in Tritthart et al. 2011 [57,58].

2.4.2. One-Dimensional Sediment Transport Model of Pulangi River Below the Dam (HEC-RAS)

The HEC-RAS model was built using the QGIS Plugin RiverGIS [75] River geometries, such as the stream centerline, cross-sections, flow paths dividing the main channel and floodplains, as well as data for estimating roughness and ineffective flow areas, were derived in conjunction with the DEM prepared for this reach.

Calibration of the 1D hydraulic model HEC-RAS was conducted at site G, where measurements were performed to derive a rating curve. The calibration process involved adjusting the Manning coefficient and comparing simulated water levels to observed values derived from the rating curve. Sensitivity analyses were conducted to evaluate the influence of geometric variations, roughness, number of discharge classes (6, 13 and 26 classes between Q = 200 m³s⁻¹ and 800 m³s⁻¹), and boundary conditions on model performance. Statistical metrics, including the Nash-Sutcliffe Efficiency (NSE) and Sum of Square Errors (SSE), were used to assess the model’s accuracy and reliability.

The test flushing event C-75 in the Pulangi river downstream of the dam was modeled with the 1D numerical model using sediment concentration SSC derived from the measurement campaign as the upper boundary. A quasi-steady sediment transport modeling approach was used, as the unsteady approach based on the Saint-Venant-equations was not numerically stable. However, this provided sufficient accuracy as the gates were opened stepwise. A plausibility check of the 1D sediment transport model was applied after [76] since data for model calibration and validation were unavailable. Thereby, compared to hydrodynamic models, it is recommended to change only one parameter at once. Initial model boundary parameters comprised a fixed bed (river incised in rock, Figure 1c), sediment load input time series from the reservoir model c(t), grain size distributions of the transported sediment d_i (averaged over all reservoir samples), cohesive sediment parameters (τ_crit,M, τ_crit,S) and the active layer thickness (d^ex). The next steps included:

• Selection of sediment transport formulas based on model assumptions and numerical instabilities (sensitivity analysis),
• Selection of the sorting method (plausibility check and dismissal of unsuitable methods), and
• Selection of the fall velocity formula (sensitivity analysis).

Based on this selection process a further sensitivity analysis was performed including parameters for roughness n, cohesion [72] and active layer thickness d^ex.

For the additional scenario D investigating the scouring of dredged material from the reservoir, which was discharged to the gorge downstream of the dam, a potential site was identified at G (Figure 1a). Dredged material of around 6000 m³ was deposited, corresponding to a thickness of 2 m along 85 m of channel length. Scouring and transport of this sediment along the river downstream was modeled for 72 h.

3. Results and Discussion

3.1. Monitoring

3.1.1. Sediment Analysis

Grain size distribution analysis (Figure 4) was performed for 15 samples. The dominant sizes were silt (37.2 to 56.8%) and clay (40.3 to 62.7%). Only three samples contained a sand fraction (>0.0625 mm) greater than 15%, with two of these taken from the northernmost sample point P5 near the river inflow.

The top layer sample (0 to 0.20 m) of P5 (Figure 1) contained 83% sand and the following layer (0.20 to 0.30 m) contained 15.9% sand. The average of all samples from 2020 results in a mean diameter d_m of 0.035 mm and a d₉₀ of 0.091 mm. Pre-dam analysis from 1976 showed a d_m of 0.158 mm and a d₉₀ of 0.356 mm [24] for riverbed sediments. The difference between pre-dam analysis and the 2020 monitoring demonstrates the accumulation of fine sediments (mostly clay and silt) in the lower reservoir, supporting the need to consider cohesion during modeling of this reservoir.

Grain density was analyzed for three samples giving an average of 2656 kg/m³ + 0.095. Dry bulk density for 10 samples at different locations in the reservoir, distributed over various depths, ranged from 590 kg/m³ to 790 kg/m³ with an average value of 680 kg/m³. Porosity ranged from 70.38% to 77.93% with an average value of 74.43%. Comparing initial weight of the analyzed samples with values from [77] for clay and silt showed good agreement.

Critical shear stress values measured at 16 locations and depth layers demonstrated a high variability in terms of location, depth layer as well as in the visual process evaluation during the critical shear stress tests. Mean and standard error values for of τ_c,se and τ_c,me are listed in

Table 2. Critical shear stress test results.

Description	τc,se (Nm−2)	τc,me (Nm−2)
16 samples	2.84 ± 3.11	5.91 ± 3.23
6 samples, stored in water for 1 week	0.81 ± 0.40	4.81 ± 1.46
P3 (6 to 16 cm), stored in water for 1 week	0.819	3.29
P3 (42 to 48 cm), stored in water for 1 week	0.398 to 0.496	2.321 to 4.033
P3 (48 to 52 cm), stored in water for 1 week	0.340 to 0.513	no value 1
P11 (0 to 15 cm), stored in water for 1 week	1.373	5.815
P11 (30 to 37 cm), stored in water for 1 week	0.913	4.033
16 samples, minimum value	0.340	2.321
16 samples, maximum value	12.205	12.141

Notes: ¹ Transition from surface to mass erosion was smooth or not detectable (no value). Sample contained sand.

.

The six samples stored in water for a week to evaluate the potential effects of drying during storage, resulted in lower critical values. These were found to be more realistic, as potential effects of drying during storage were reduced. Sample P3 is shown as a representative example of the area between the upper and lower reservoirs, whereas sample P11 represents the area near the BS. Cohesion was apparent for all layers except for the deepest analyzed layer of P3 highlighting the high variability of deposited sediments. Moreover, neighboring samples could also vary significantly. Finally, minimum and maximum values from all 16 samples were evaluated as a basis for estimating the model’s sensitivity on cohesion.

3.1.2. Discharge, Water Level and SSC

Water level at the reservoir inflow (BB) was clearly influenced by the backwater from reservoir operation, prohibiting calculation of a rating curve. A steady state inflow of 73.5 m³s⁻¹ defined reservoir inflow Q_in (BB) during the test flushing event, as derived from measurements at 9 cross sections (Q_in = 73.5 ± 2.17 m³s⁻¹, Figure 5).

Turbine discharge Q_T (60 ± 22 m³s⁻¹) provided by the operator [22] was in agreement with ADCP measurements. Cross sectional averaged SSC at BB was determined as 207 ± 63 mg/L (Figure 5). Results from the dam construction feasibility study in 1976 (pre-dam condition, [24]) showed lower concentration values at BB for a comparable discharge. The increase may be attributed to enhanced erosion in the basin due to land use changes and climate change [30]. The individual cross sectional SSC measurements and the calculated SSC (Figure 5), using derived rating curves (RC) to combine those measurements with continuous near-bank turbidity sensor measurements, were in agreement for BB (reservoir inflow) and CB (power channel). Inflow was relatively constant for around 24 h during monitoring after opening the gates. Discharge Q_out (BS) (Figure 5) represents the sluice gate equation [43] based on gate opening B and reservoir water level. Maximum and minimum SSC measurements at BS using near-bank bailing samples were delineated (Figure 5), although some individual measurements deviate.

Sediment load at BS during the drawdown flushing event (Q75) was calculated using delineated minimum and maximum SSC, which ranged between 30,050 m³ and 51,796 m³. By considering the average gate opening B of 1.36 m and a maximum of B = 5 m (Figure 5), the observed load is approximately comparable with numerical results from Tabios [43], where a 12 h hour drawdown flushing event at 100 m³/s inflow and 2 m gate opening resulted in loads between 46,000 m³ and 52,000 m³. However, Tabios [43] does not mention the level of drawdown reached by the 2D hydrodynamic model. Sediment load at the power channel during C-75 was calculated with 4348 ± 459 m³ between opening and closing of the gates, which was slightly lower than the cumulative sediment load measured at the reservoir inflow (5754 m³). Considering the relatively constant inflow of water and sediment into the reservoir, it is clearly evident that SSC (up to 19.9 g/L) and calculated sediment loads (up to at least 51,796 m³) peaks during drawdown sluicing at BS while the gates were open (Figure 5). Elevated SSC concentrations were mentioned by [13] as a disadvantage of drawdown flushing compared to other management operations (e.g., drawdown sluicing). However, concentrations >100 g/L were not observed due to the very limited drawdown. Nevertheless, a monitoring program targeting macroinvertebrates and fish [8] is required to monitor environmental impact, and the maximum concentration released can be controlled by limiting the drawdown rate. SSC in the power channel (Figure 5), measured at CB 2350 m downstream of HW, also peaked during the gate operations (SSC up to 8.5 g/L), and reached a plateau of around 1.2 g/L (value from cross-sectional SSC measurement t = 21.8 h after gate opening). This plateau lasted until around t = 30 h, indicated by the SSC calculated from rating curves, before reaching SSC concentrations close to the inflowing concentrations. Risks of turbine abrasion (depending on the sand fraction in the sediment mixture) and siltation in the power channel and SP should be carefully considered and monitored. Turbine abrasion will become increasingly important as the delta with higher sand fraction approaches the power intake (HW).

Calculated discharge from the reservoir Q_tot, including bottom sluice gates and turbines, was approximated with a polynomial function over the flushing time t (Q_tot = 2 · 10⁻¹⁹ t⁵ − 4 · 10⁻¹⁴ t⁴ + 6 · 10⁻¹⁰ t³ + 2 · 10⁻⁶t² + 0.002 t + 87.142, R² = 0.95) and served as the boundary condition for the reservoir numerical model. Measurements at BB, BS, and CB were used as sediment input to the reservoir model and the 1D model in the old river downstream of the dam.

Water level measurements in the gorge downstream of the gates (G), using the cross sections derived with the photogrammetric survey, were clearly correlated to gate opening B during the drawdown event. Therefore, a rating curve at G was derived under steady-state conditions (W = 3.0 · 10⁻⁷ Q² + 7.8 · 10⁻³ Q + 261.51, R² = 0.98) as the flushing gates were opened incrementally, allowing for stable flow measurements. In contrast, the closing of the gates resulted in unsteady flow conditions, making the rating curve unreliable during this phase. To ensure accuracy, only the portion of the rating curve corresponding to increasing discharge was used for calibration of the 1D model downstream of the dam.

3.2. Simulation of Pulangi IV Reservoir

3.2.1. Model Calibration and Scenarios

Morphodynamic calibration of the sediment transport model iSed was performed with the help of measured bathymetry changes in the lower reservoir before and after the test flushing event in 2020. Parameters for the cohesive model were set to values presented in [42] (

Table 3. Cohesive parameters determined during calibration and serving as a basis for estimating the sensitivity of the sediment transport model based on [42].

Parameter Name	Parameter	Value
Critical shear stress value for full deposition	τd,full	0.000010
Critical shear stress value for partial deposition	τd,part	0.60
equilibrium concentration	ceq	823 mg/L = cin
Surface erosion rate constant	Mse	2.5 · 10−5 ms−1
Mass erosion rate constant	Mme	2.1 · 10−4 ms−1
Critical shear stress value for surface erosion	τc,se	1 Nm−2
Critical shear stress value for mass erosion	τc,me	3.5 Nm−2
Settling velocity	w	Ref. [71]

), whereas critical shear stress values were set in accordance with test runs considering critical shear stress test results using wetted samples (Table 2). Equilibrium concentration c_eq was estimated to be equal to the concentration of reservoir inflow c_in at Q = 250 m³/s. It fell in the range of values proved in literature [42,78]. Settling velocity was chosen using equation 2 proposed by [71]. Surface and mass erosion rates (M_se, M_me) were adjusted using a power law [79]. Calibration performance was assessed by the difference between the modelled and observed net volume change considering erosion and sedimentation along the low flow channel in the observation area in the lower reservoir. Figure 6 shows measured and modeled erosion and deposition in the observation area in the lower reservoir.

Bathymetry from monitoring for the test flushing event in 2020 was different from the modeled bathymetry based on 2019 measurements, as is obvious in cross section CS1 (Figure 6a) and the longitudinal profile LS (Figure 6c). As bed levels near the BS needed to be fixed to achieve numerical stability, the model was not able to reproduce the observed erosion in the proximity of BS (around 16,000 m³), as seen in Figure 6. The need for the fixed bed should be seen as local effect having in mind the large model domain of the shallow and wide reservoir modeled in detail with 3D hydrodynamics, which is larger than other 3D reservoir models [37,39,40,41]. Moreover, gates are not included in the model domains of the 3D reservoir models, mentioned above. By using the parameters summarized in Table 3, net volume difference ΔS in the observation area including the 16,000 m³ associated with the fixed bed near the BS, the modeled release ΔS was slightly higher (+ 6563 m³) than observed ΔS, representing a volumetric error of 10%. However, modeled erosion at the upstream end of the observation area was larger and may be underestimated by the bathymetric survey as profiles were not taken at the exact same locations before and after the test drawdown event (Figure 6b). Those unquantifiable measurement errors could reduce the 10% volumetric error. In general, measured and modeled bathymetry changes do match well, especially regarding the evolving flushing channel. This is in accordance with other 3D modeling studies, where flushing channels could also be reproduced [37,39,40,41].

Water level drawdown for the calibration scenario (C-75) is shown in Figure 7a. Modeled drawdown from 284.6 m to 281.0 m is comparable with observations during the simulation period, which ended when gate closure caused the water level to rise. Calibration of the hydrodynamic model RSim-3D against measured water levels at the model inflow (BB) for the test flushing event (C-75) showed relatively small deviations (<0.15 m) between measured and modeled water levels. The agreement indicates proof of a successful 3D hydrodynamic model calibration. The slightly elevated modeled water level can also be attributed to insufficient resolution or outdated bathymetry measurements in the river entering the reservoir.

Modeled suspended sediment volume change ΔS for C-75 was compared with suspended sediment load calculated from sediment transport field measurements during the test flushing event (Figure 7b). The upper boundary of measured suspended sediment concentrations (C-75 obs, max), transferred to sediment load (L), showed agreement with the modeled calibration scenario C-75. By comparing the cumulative reservoir inflow of 5754 m³ against the released sediments at BS and HW, the maximum difference between observation (C-75 obs, max) and the model is 2486 m³, representing an error of only 4.8% to 5.4%, corresponding to the maximum and minimum sediment loads at BS of 46,000 m³ to 52,000 m³.

Modeled sediment volume change ΔS showed a steeper gradient at the end of drawdown flushing due to the abrupt start of mass erosion, whereas the observed sediment load indicated an earlier onset of mass erosion. This effect can be attributed to the fixed bed at the model outlet, where local erosion of deposited sediments is expected during the initial stage of drawdown [34]. Three-dimensional numerical results of bed shear stress at the drawdown cone support this, as critical shear stress values are exceeded much earlier. The need to fix bed levels in parts of the upper reservoir for numerical stability may also contribute to the shallower modeled gradient in ΔS at the beginning of the simulation. Another reason for the lag between simulation and observation may be because bank erosion—as implemented in [38]—is not considered. This could reduce numerical instabilities in the upper part (bifurcated channels in the delta), but on the other hand it increases simulation time and adds further instabilities [38]. Additionally, that only one parameter set for modeling cohesion across the entire reservoir was possible within the model, may also contribute to this issue, as measured critical shear stress values were not uniform in the lower reservoir and may vary even more significantly in the sandier, deltaic region of the upper reservoir. Deposit consolidation, expressed by varying bulk densities, may also contribute to differences in timing of modeled and observed erosion. Nevertheless, the final sediment volume change at the end of the flushing period corresponded well with measured loads.

Due to higher discharges at BS, water level drawdown for 1-250 and 2-250, shown in Figure 7a, indicated a shorter duration in comparison to C-75, although the initial water level was higher. However, the duration for 2-250 was longer than for 1-250 due to the higher volume in the dredged channel (Figure 2b). In the observation area, ΔS was highest for 2-250, although the difference to 1-250 was small with + 2.5% (Figure 7b). When considering the total reservoir, 1-250 was higher than 2-250 (+12%), a result of bed shear stress far below critical shear stress values in the dredged channel (2-250). Considering a drawdown level of 181.0 m, differences in ΔS between the scenarios reached +7% and +33% for 2-250 and 1-250, respectively, when compared to C-75. These relatively small changes may be related to the shorter duration of the scenarios with higher discharges when critical shear stress values are exceeded and cause erosion of the cohesive sediments. This provides indications that the gradual increase in gate openings, as applied for C-75, is more effective than the theoretically constant opening over the whole period.

Similarity with 2D numerical results from Tabios [43], as described in analogy for the monitoring of scenario C-75 (Section 3.1.2), cannot be concluded with certainty. For 250 m³s⁻¹ reservoir inflow and less sluiced discharge (~−30%) distributed over a longer period, the 2D model estimates sediment loads of 47,000 m³ to 53,000 m³. Tabios [43] does not mention the level of drawdown reached by the 2D hydrodynamic model, which—as already described in Section 3.1.2—is an important innovation reached by the application of the 3D hydrodynamic model approach. However, Tabios seems to consider a drawdown flushing event including drawdown over all three stages according to Morris [16]: drawdown, free-flow period while the reservoir is empty (or close to empty), and the refill. Ongoing modeling of the free-flow period (nearly empty reservoir) using the 3D reservoir model would thus indicate a higher erosion potential, which is limited by the flow retreating into the developed flushing channel.

Figure 8 shows erosion differences in the lower reservoir for the higher discharge scenarios (1-250 and 2-250) and lower drawdown level. Figure 8a,d can be compared with Figure 6c given the same drawdown level of 281.0 m. 1-250 predicts larger erosion areas than scenario 2-250 and C-75 (Figure 6c), although magnitudes are smaller.

Comparable findings were achieved for the lower drawdown level (280 m), simulated for 1-250 and 2-250. The deeper channel in Figure 8d is clearly distinguishable when compared to Figure 8b. For both scenarios in Figure 8, lower drawdown level increased erosion near the BS. However, an upstream to downstream erosion progress was detectable for 2-250. The upstream area with the beginning erosion was located immediately downstream of the end of the dredging channel. This increases the chance that the dredging channel is possibly connecting the upper reservoir to the lower reservoir when multiple drawdown events are performed. By simulating a single drawdown flushing event, it cannot be stated with certainty that the initial dredging channel can be transformed to a fully established sluicing channel through the entire reservoir, as depicted in Figure 2c. However, scenarios with higher reservoir inflow might increase the chance due to progressive erosion (upstream to downstream direction) [34]. On the other hand, drawdown flushing is less effective for 2-250 than for 1-250, when the width of the flushed channel in the lower reservoir is considered (Figure 2a,b). Lai [34] defines the ratio between this width and the width of the reservoir as a criterion for reaching a sustainable reservoir capacity. In the case of the shallow and wide Pulangi IV reservoir, this ratio is unfavorable in any case, suggesting that drawdown flushing—combined with dredging—is just one component of a broader approach toward sustainable sediment management. However, for all considered scenarios drawdown was considered effective even though the water level was not lowered below the min OL, thus sustaining continuous power generation during drawdown flushing. Hauer et al. [4] highlight the necessity of a significant (or complete) drawdown, which conflicts with continuous hydropower operation. Drawdown below min OL was not tested in the scenarios considered in this case study, to test the effectiveness of sediment release without interrupting power production.

3.2.2. Sensitivity Analysis

A sensitivity analysis for cohesive model parameters on ΔS was performed for the last 20 min of drawdown flushing, where changes in ΔS (%) were highest due to the highest water level drawdown gradient. ΔS was measured in the observation area in the lower reservoir (Figure 6). The reduction of settling velocity w on average by 80%, using values for hindered settling [42] instead of Equation (2), increased ΔS by 13% for C-75, and decreased ΔS for 1-250 by −5%. Th86e increased ΔS (less net erosion) seems counterintuitive for C-75. However, while erosion volume increased in the flushed channel, sedimentation along the banks of the channel increased to an even greater extent. In Figure 6c, this sedimentation, also mentioned in [13], is clearly discernible for C-75. In contrast, the sedimentation outside the channel was lower for 1-250, as shown in Figure 8a,b, which might be attributed to the higher reservoir inflow. A decrease of c_eq by 75% was not sensitive regarding ΔS. As a result, critical shear stress values for surface erosion (τ_c,se) and mass erosion (τ_c,me), were identified as the most sensitive adjustment parameters for the cohesive sediment transport model iSed (

Table 4. Sensitivity analysis for cohesive parameters of the sediment transport model iSed.

Parameter	Parameter Value/Change (%)	Change in ΔS (%)
		C-75	1-250	2-250
τc,min	τc,se = 0.398 Nm−2/−60%	−103%	−72%	−56%
	τc,me = 2.321 Nm−2/−34%
τc,max	τc,se = 12.205 Nm−2/+1121%	+72%	+73%	+51%
	τc,me = 12.205 Nm−2/+249%
w1	after [42]/−80%	+13	-5%	-
ceq2	207 mg/L/−75%	−1%	+3%	-

Notes: ¹ Decrease in % was calculated on average over all concentration classes. ² Value of 207 mg/L represents concentration c_in at reservoir inflow for Q = 75 m³s⁻¹.

).

By setting critical shear stress values to the minimum and maximum values determined during monitoring (Table 2), modeled ΔS changed between −103% and +73%. Sensitivity of critical shear stress values to changes in ΔS is also plotted in Figure 7b, indicating the range of erosion volume achievable by drawdown flushing. As the model allows just one general cohesive parameter set for layers at all depths over the whole reservoir, this method was found suitable to cope with model uncertainties regarding the high variability of critical shear stress values determined by sampling various locations and depth layers. Future improvements to the iSed model code could include the consideration of varying cohesive parameters based on the location and layer depth. Additionally, consolidation could be addressed by accounting for varying bulk densities and porosities.

3.3. Pulangi River Downstream of Reservoir

Results of the 1D sediment transport model HEC-RAS for the old river downstream of the dam between BS and C (Figure 1a) were worked out in Tasser [80].

3.3.1. Calibration and Sensitivity Analysis of the Hydrodynamic Model

Calibration of the hydraulic model downstream of Pulangi reservoir (G) using cross section measurement taken during the test flushing event achieved best results with a Manning coefficient of n = 0.037. The model demonstrated a high degree of accuracy, as indicated by a Nash-Sutcliffe Efficiency (NSE) coefficient of 0.984 and a Sum of Square Errors (SSE) value of 0.41 m². The mean absolute error between modeled and measured water surface elevation was 16 cm, with an error range between -29 cm and +18 cm. Sensitivity analysis revealed that the model is highly sensitive to geometric variations, particularly in the narrowing of the riverbed, which significantly impacts flow behavior due to the Venturi effect. Due to limited survey data, the riverbed profile was adjusted by increasing its width based on airborne imagery, which improved the model’s performance. Conversely, the model showed low sensitivity to changes in roughness and no influence from the number of discharge classes (13 classes between Q = 200 m³s⁻¹ and 800 m³s⁻¹) or the upstream boundary condition, respectively.

The agreement between the simulated and observed water levels, particularly in the mid-range discharge scenarios, validates the model’s reliability for hydraulic predictions of existing conditions. It was also found to be reliable to use a single Manning coefficient (n) for the entire model, as applied in e.g., [81,82]. This approach was justified by the low sensitivity of the model to variations in roughness, similar channel characteristics along the length of the channel, and the lack of sufficiently detailed data to support spatially variable roughness coefficients.

3.3.2. Plausibility Check and Sensitivity Analysis of the Sediment Transport Model

A plausibility check after Dahl et al. [76], applied for the test flushing event (2020) with a discharge of Q = 75 m³s⁻¹, resulted in the selection of two sediment transport formulas—Toffaleti [83] and Laursen [84]—which exhibited numerical stability and gave plausible results. Other formulas, such as Meyer-Peter & Müller [85], Yang [86], Wilcock-Crowe [87] and Engelund & Hansen [88] were disregarded due to their limited applicability for fine sediments flushed from the reservoir. The active layer concept was selected as sorting method because of its stability and the fine sediments. Following the suggestion by Tritthart et al. [89], the initial value of the active layer depth d^ex, typically ranging from 1 to 4 × d₉₀, was selected as 4 × d₉₀, resulting in 0.364 m. Based on the plausibility check, a sensitivity analysis was performed for the test flushing event including two sediment transport formulas and five fall-velocity methods. The 28,000 t of cumulative sediments flushed during the event, derived as a mean value from the monitoring campaign, were mostly transported through the model reach. Deposition varied depending on the transport formula used, ranging between 13% (Toffaleti [83]) and 5% (Laursen [84]) of the sediments discharged from the reservoir. The Toffaleti formula led to a more continuous deposition over the reach whereas the Laursen equation indicated deposition located in pools. The selection of fall velocity method was not sensitive with regard to the fraction of deposition in the reach. A further sensitivity analysis resulted in a high sensitivity to cohesion assumptions. The difference in total depositions along the reach between enabled and disabled cohesion was 6.3% and 0.7% for the Toffaleti and Laursen equations, respectively. On the other hand, the critical shear stress values τ_c for cohesion, the variation of the roughness parameter n and the active layer depth d^ex were not found to be sensitive when modeling the test flushing event in the Pulangi River downstream of the reservoir. The model resulted in distributed deposits in the upper reach (up to 0.03 m for Toffaleti) or locally in pools (up to 0.22 m for Laursen). Deposits consisted of fine and medium sand, whereas finer sediments were transported through the reach completely. The generally low sedimentation rates do not suggest that flushing with clear water after filling is necessary for the studied, as indicated to be a possibility in Morris [13], though it could be required for the release of greater sediment volumes and when considering the entire river below the dam, which has a declining bed slope. A long-term monitoring of downstream impacts by sedimentation, as well as the resulting ecological impacts is highly recommended [4,8]. Additionally, since drawdown flushing is not considered as environmentally friendly as sluicing [13], it can be considered as a transitional phase toward the development of a sustainable sluicing channel (cf. Figure 2c).

Detailed monitoring as a basis for model calibration should be performed in the future. A more differentiated variation of roughness, given the high sensitivity of roughness parameters, would lead to more reliable hydrodynamic results such as in Dahl et al. [76]. Concerning the 1D sediment transport model, the lack of calibration and validation data is of utmost importance. The provided results, based on the plausibility check and sensitivity analysis [76], can only offer an initial assessment, which should be supported by calibration data from field monitoring. Other studies on sediment transport [82,90,91] suggest that the applied 1D model can deliver highly accurate results regarding sediment transport when monitoring data are available.

3.3.3. Scouring of Dredged Material from a Containment Structure (Scenario D)

At the position (G) selected for a potential dredged sediment containment area, 130 m downstream of the dam, elevated shear stress between 10 and 20 Nm⁻² was modeled due to the narrower channel with a width of 52.0 m at the upstream end to 16.5 m at the downstream end of the containment area. A discharge of Q = 200 m³s⁻¹ was capable of completely eroding the contained material and transporting it downstream (Figure 9) within 72 h. 95% (Laursen) and 89% (Toffaleti) of the eroded material of the containment area could be transported through the modeled reach, resulting in low volumes of fine and coarse sand deposits. Large areas of deposits did not occur along the study reach.

Sensitivity was evaluated for parameters Q, τ_c,se τ_c,me, n, and d^ex with respect to the duration of the erosion in the containment area. In general, parameters were sensitive to shorter durations (

Table 5. Sensitivity analysis for relative erosion (%) based on completely deposited sediments at the Pulangi River downstream of the reservoir. Relative erosions are evaluated separately over time.

Parameter 1	Δ Parameter 1 (%)	Parameter 2	Δ Parameter 2 (%)	Relative Erosion (%)
Q (m3s−1)				t = 12 h	t = 24 h	t = 48 h	t = 72 h
200	-	-	-	55	88	97	100
300	+50	-	-	62	90	98	99
400	+100	-	-	68	90	98	99
500	+150	-	-	72	90	98	98
600	+200	-	-	73	90	99	99
700	+250	-	-	74	91	99	99
range (%) *:	+250			19	2.8	1.6	1.9
τc,se (Nm−2)		τc,me (Nm−2)		t = 12 h	t = 24 h	t = 48 h	t = 72 h
1	-	4	-	55	89	98	100
12.2	+1120	12.2	+205	51	85	95	98.5
0.340	−66	2.31	−42	54	88	97	100
range (%) *:	1186		247	3.2	3.0	1.9	1.5
n		-	-	t = 12 h	t = 24 h	t = 48 h	t = 72 h
0.037	-	-	-	54	88	97	100
0.033	−11	-	-	47	82	96	99
0.041	+11	-	-	65	91	98	99
range (%) *:	22	-	-	17.9	8.6	2.1	1.3
dex	-	-	-	t = 12 h	t = 24 h	t = 48 h	t = 72 h
4 d90	-	-	-	55	89	98	100
1 d90	−75	-	-	31	61	93	97
8 d90	+100	-	-	63	91	98	101
12 d90	+200	-	-	64	91	98	100
range (%) *:	275			33.0	30.2	5.8	4.2

Note: * Range in (%) is defined by maximum minus minimum value. The use of bold represents sub-headers.

).

After 12 h, 55% of the material was eroded for a discharge of Q = 200 m³s⁻¹. By increasing discharge Q, erosion volume in the containment area is enhanced, following a quadratic relationship (ΔS = 0.0054Q² − 7.1139Q − 2052.1; R² = 0.9981). This yields greater changes in erosion (ΔS) when lower discharge levels are adjusted. Higher discharges led to similar erosion rates ranging from 72% at Q = 500 m³s⁻¹ to 74% at Q = 700 m³s⁻¹. After 24 h, erosion rates were almost independent of Q, ranging between 88% and 91%. Thus, sensitivity to Q over the tested discharges ranged between 19% and 1.9% for 12- and 72-h simulation periods, respectively. The sensitivity of bed roughness ranged between 17.9% and 1.3%. Highest sensitivity on the erosion of dredged sediments was found for the active layer depth d^ex, ranging between 33% and 4.9%. Larger d^ex led to acceleration of erosion within the first phase (<48 h). In contrast to the previous parameters, the variation of critical shear stress values ranging between measured values, evaluated from reservoir samples during monitoring (Table 2), indicated a maximum range of only 3.2% after 12 h regarding the eroded fraction.

For reservoir flushing and scouring of dredged material from a containment structure (scenario D), extensive monitoring of downstream impacts regarding sedimentation, water quality and ecological impacts is highly recommended in the future, such as in [8]. Additionally, using dredged material from the channel in scenario 2-250 (Figure 2b), complete disposal of the dredged channel volume into the river downstream would represent about 360 times more sediment than simulated here. Dredging in combination with drawdown flushing could contribute to a sustainable pass-through strategy [13]. It is possible that repetitive drawdown events could reduce the amount of dredging. However, this is heavily dependent on the sand content in the upper reservoir.

4. Conclusions

Effective sediment management in shallow and wide reservoirs such as Pulangi IV requires a combination of measures for sustainable long-term management. Long-term sediment and ecological monitoring combined with scenario-based numerical modeling can support decision-making to preserve hydropower generation while minimizing environmental impacts.

Suspended sediment concentration during drawdown flushing is high and requires ecological and morphological monitoring, and drawdown rates and durations should be adjusted considering downstream limits in sediment concentration and river morphological impacts. In contrast, suspended sediment concentration during drawdown sluicing would typically not exceed the natural spectrum of floods, when conducted during flood inflow events.

The 3D reservoir model—a model larger than those used for most comparable studies—for the shallow and wide reservoir (surface area of 11 km²) provided detailed results for drawdown flushing in the lower reservoir area, with the modeling of cohesive sediment properties found to be crucial. Drawdown was combined with a dredged flushing channel and a dike in the upper reservoir to better channel flows during drawdown sluicing of floods. Sensitivity analysis allowed the assessment of variability, based on extensive sampling and critical shear stress tests. The model was found to align well with bathymetric measurements. Performing sufficient water level drawdown was crucial for effective erosion, but in the tested limited drawdown scenarios, hydropower generation continued uninterrupted for this detailed case study. A gradual opening of the gates was more effective than a theoretically time-invariant (constant) opening. Downstream sedimentation from flushed and dredged sediment discharge was low, though calibration and validation data could be improved by long term monitoring.

Combining flushing with dredging in the upper reservoir suggests the potential for establishing a flushing channel which, with repeated operations, especially during elevated discharge, could enable future sustainable drawdown sluicing during floods without exceeding natural SSC levels. On a larger scale, downstream sediment and aquatic ecological impacts must be addressed to reestablish the sediment balance from rivers to oceans. This effort should also include measures to reduce sediment yield caused by land conversion and climate change. These findings can inform sediment management practices in other reservoirs with similar challenges.