2.1 Hydrological and physical conditions

The Kelani River Basin is bounded by latitudes 6° 47′ and 7° 05′ and eastern longitudes 79° 52′ and 80° 13′ and is completely located in the wet zone of the country, as illustrated in Figure 2. The catchment covers an area of 2230 km² (Dissanayaka & Rajapakse, 2019). The upstream area is steep, whereas the downstream area is flat below Hanwella (De Silva et al., 2012). The catchment receives an average annual rainfall of 3718 mm. Southwest monsoon rains contribute mostly to the annual rainfall; however, the contribution of the northeast monsoon and inter-monsoon rains is also considerably high (Hettiarachchi, 2018). Streamflow fluctuated in the range of 600 and 1800 m³/s during the monsoonal season (Dissanayaka & Rajapakse, 2019).

Figure 3 illustrates the annual maximum discharges recorded from 1995 to 2018 at Hanwella (downstream) and Glencourse (upstream). The annual maximum discharge at Hanwella ranged from 500 to 1600 m³/s, and at Glencourse ranged from 300 to 1800 m³/s. Furthermore, the average annual maximum discharge at Hanwella and Glencourse was approximately 950 and 800 m³/s, respectively.

Figure 4 illustrates the cumulative frequency of the daily streamflow at Hanwella from 1995 to 2018, which provides the percentage of time that a particular discharge was equal to or exceeded. The flow duration curve is useful for identifying river flow thresholds. Discharges at 95%–90% and 5%–10% exceedance were classified as low and high flows, respectively. Accordingly, the low and high flows were approximately 30–35 m³/s and 400–255 m³/s, respectively.

The lower basin experiences recurring flooding, with the most recent severe instance occurring in 2016. Figure 5 illustrates the historical flood levels at Nagalagam Street recorded from 1837 to 2018, along with the corresponding flood discharges at Hanwella. According to Figure 5, the magnitude of floods has significantly decreased after the year 2000 when compared to the past. This may be attributed to the current riverbed lowering and changes in the climate. Although the magnitude of floods has decreased in the present, the cost of flood damage has increased due to the prolonged duration of flooding and urbanization in the floodplain.

Figure 6 shows the longitudinal profile of bed elevation measured in 2016. As shown herein, the riverbed profile varies randomly, which may be caused by sand mining, channel meanders, sand bars, and bedrock hindrance. Therefore, the channel bed elevation varies about −2 and −15 m from the mean sea level (MSL) in the downstream reach and changes to a steep stream in the upstream section starting from 40 km. The bed slope of the reach is an important factor that controls sediment transportation. It is estimated to be 5.5 × 10⁻⁵ in the downstream and 5 × 10⁻⁴ in the upstream. The riverbed elevation also varies approximately in the range of −3 to 7 m MSL from 40 to 60 km upstream of the river mouth, which is nine times steeper than the downstream riverbed.

Figure 7 illustrates the channel widths. Although it widens at the proximity of the river mouth by about 160–200 m, it can be estimated on average to be 80 m for the downstream section and 60 m for the upstream section. At some upstream locations, the channel width narrows unexpectedly by 40–20 m due to hindrance by bedrock and meanders.

2.2 Impacts of sand mining on the physical environment

River morphological characteristics reflect the flow and sediment transport processes in the river (Church, 2006; Joshi et al., 2019). Figure 8 shows photographs taken in January and February 2023 to introduce the present morphological characteristics in the downstream and upstream reaches. Photos (a) and (b) show that the downstream reach with an erodible bank is filled deeply by water due to bed lowering and the backwater effect. Photos (c) and (d), which were taken in the upstream reach, suggest that the upstream reach is characterized by mountain streams, gravel beds, and rocky banks and is very steep in comparison to the bed slope in the downstream reach.

Figure 9a,b shows the features of the downstream reach, which was located 4 km from the river mouth in 1825 and 1906, respectively. The two photographs suggest that the flow depth was shallower than half the height of a mature person, and the channel bed was covered thickly by sediment, considering recent degradation. In addition, the impact of sand mining on the physical channel environment can be recognized.

Cross-sectional shapes of the river channels were measured in 2003 and 2016. Figure 10 shows the longitudinal profiles of their channel beds together with the channel bed elevation measured in 1961. The observed riverbed lowering and uneven changes in longitudinal riverbed profiles in the 2003 and 2016 profiles, compared to the approximate natural riverbed in 1961, are primarily attributed to riverbed erosion resulting from excessive sand mining. Using these data, the lowered volume was estimated to be 1,618,000 m³ for 42 years from 1961 to 2003, and 2,834,000 m³ for 13 years from 2003 to 2016, respectively, in the channel reach from 0.7 to 20.4 km. The mean annual riverbed lowering volumes were 1.95 and 11.1 m³/m for each period, implying that the latest annual sediment loss was 5.7 times larger than the annual average loss for 42 years until 2016.

2.3 Temporal changes in water quality, and impacts on the freshwater ecosystem

Excessive sand mining activities tend to cause sediment supply deficiency, enlargement of channel cross-sections, and increase river channel flood water transport capacity. Besides, such physical changes cause declining water levels, resulting in shallow water levels during the dry season and also affect badly the water supply, water quality, and aquatic habitats (Lai et al., 2014).

Figure 11 illustrates the monthly conductivity measurements from 2015 to 2019 taken at Victoria Bridge, which is located at a distance of 4 km from the river mouth, along with the daily observed discharge at Hanwella on the conductivity measurement day and the maximum marginal standard value of 700 µS/cm set by the Central Environmental Authority (Central Environmental Authority CEA, 1992). As shown in Figure 11, the majority of the observed conductivity values surpassed the marginal maximum value and reached unacceptably high conductivity ranges that exceeded 10,000 µS/cm, indicating the occurrence of salt intrusion.

The conductivity in water routinely exceeds the acceptable marginal maximum value when the flow rate is less than 50 m³/s, which corresponds to 66% flow exceedance in Figure 4.

Physical changes in habitat, hydro morphology, groundwater level, and water quality caused by sand mining affect freshwater fish species (Koehnken et al., 2020). The Kelani River Basin exhibits a vital ecosystem, as it comprises 63% of the freshwater fish species in Sri Lanka and is home to a total of 60 fish species, including 30 endemic and 22 threatened species (Surasinghe et al., 2019). Fish species such as Garra ceylonesis, Systomus pleurotaenia, Schistura notostigma, and Puntius titteya, which are specific to certain environmental conditions, are mostly affected by sand mining and are rare in the Lower Basin because of habitat loss, changes in the physical environment, and water quality (Kotalawala, 1994; Sudasinghe et al., 2014; Surasinghe et al., 2019). Such degradation of fish species is correlated to the decrease in sediment mobility and variability of bedforms. A stable riverbed provides an optimal environment that ensures the existence of many species (Rentier & Cammeraat, 2022).

3.1 Sediment size distribution

As mentioned in the previous section, riverbed materials were investigated to evaluate sediment transport processes in the basin. Three samples of riverbed material in the mainstream at the middle of the river, and closer to riverbanks from each cross-section were collected at seven downstream locations as well as four upstream locations, as illustrated in Figure 2, between January 15, 2023, and February 13, 2023, using a scraper bucket sampler. Wet and dry sieve tests were conducted to determine sediment grain size distribution curves. Figure 12 illustrates the accumulated size distribution curves of the bed material samples at locations 1–11 in Figure 2. In addition, a reference curve of sediment particle size, which is averaged, is illustrated in the downstream and upstream regions. Table 1 lists the sediment size fractions and finer percentages of each sediment size class of the accumulated upstream and downstream sediment distribution curves that were employed to estimate the downstream and upstream sediment transport rates. As shown in Table 1, the average diameter of the sediments in the upstream and downstream regions was approximately 1 mm. Additionally, according to ASTMD-2487, the soil type of the riverbed material at all locations except for location 7 was sand. At locations 4, 6, 8, 9, and 10, the riverbed material is classified as clean sand. The mean grain size of sediments at locations 3, 4, 6, 8, 9, and 10 was in the range of 0.5 and 2 mm, which is the best for fine aggregates (Amalan et al., 2014).

3.2 Estimations of bedload and suspended load

To manage sediment transport processes properly in a drainage basin, including sand mining in river channels, we need to estimate the amount of sediment supplied from upstream to downstream reach and the amount of sediment discharged from the downstream reach to the sea during floods. Herein, the target reach is downstream of the Kelani River. The amount of sediment supplied to the target reach and the amount of sediment discharged to the sea during floods were estimated using bedload formulas and suspended sediment concentration formula, where we employed Egashira et al.'s formula (1997), Egashira & Itoh's formula (2006) and Ashida & Michiue's formula (1972) for evaluating bedload rates, and Harada et al.'s method (2022) for evaluating equilibrium concentration of the suspended sediment.

The validity of the derived bedload formula was evaluated using experimental data (Egashira et al., 1997) and through its application to analyses of channel changes with active sediment transportation in river basins (Nagumo et al., 2024; Rahman et al., 2024; Subedi et al., 2024). Furthermore, this bedload formula is consistent with the functional form of the well-known Brown's empirical formula (1950), and provides almost the same results that are computed by empirical formulas proposed by Shinohara and Tsubaki (1959), and Graf and Suszka (1987).

The method based on a convection–diffusion equation and the equilibrium profile of sediment concentration has been widely employed. However, there remain unresolved issues regarding the reference level and the specification of reference sediment concentration. Consequently, we utilize a novel approach proposed by Harada et al. (2022) to evaluate the entrainment rate from the bedload layer.

As described in Equation (6), an entrainment coefficient, which is defined as the entrainment velocity divided by the mean velocity in the upper layer, depends on the inverse of the Richardson number (the square of the internal Froude number). Recently, we found that the same parameters were used by (Guo et al., 2008) in evaluating the suspended sediment concentration in the lower reaches of the Yellow River. According to their study, observed data on sediment concentration illustrated in Figure 2 of their article from 0.0566 × 10⁻³ to 0.943 × 10⁻³ at ($v/w_o){\mathit{Fr}}^2=1$, and range from 0.151 × 10⁻³ to 7.55 × 10⁻³ at ($v/w_o){\mathit{Fr}}^2=1$0, where these values are recomputed by volume concentration. While Equation (9) of our manuscript gives 0.91 × 10⁻³ to 9.1 × 10⁻³, respectively, assuming that the sediment particle size of the bed material is the same as that of the suspended sediment one. Given that Equation (9) represents the equilibrium sediment concentration and there exists a discrepancy between the bed sediment and suspended sediment computed concentrations, the computed sediment concentration may be deemed acceptable.

Both of the bedload formula and the formula of suspended sediment concentration employed in our research applied to evaluating inundation processes with active sediment transportation on the flood plain, and their applicability is confirmed (Rahman et al., 2024; Subedi et al., 2024).

The computation of bedload rate and suspended load rates was determined at the reach scale. A reach, also referred to as a unit channel, is a section of a stream or river extending from the point where a tributary joins a river to the subsequent confluence, characterized by two inflow points and one outflow point, exhibiting consistent hydrologic parameters, including discharge, depth, area, and slope (Egashira & Matsuki, 2000; Harada & Egashira, 2024). Figure 13 presents a schematic representation of a unit channel, illustrating its planform and longitudinal bed profile.

Equation (2) is applied to Equation (10) to evaluate the bed load transport rates, where Equation (2) was reformed to compute bedload transport rates for nonuniform sediment, and Equation (9) is applied to Equation (11) to evaluate the suspended sediment transport rates.

The sediment hydrographs for bedload and suspended load corresponding to different typical floods that occurred between 2012 and 2018 were computed using Equations (2), (3), (9), (10), and (11). In these computations, reference cross-sections are specified as follows: The bed slope and the channel width are $i_b$ = 5 × 10⁻⁴ and $B$ = 60 m in the upstream reach (unit channel) and $i_b$ = 5.5 × 10⁻⁵ and $B$ = 80 m in the downstream reach (unit channel). Besides, size fractions of bed material don't change in these estimations to obtain stable, prospective results on sediment hydrographs. Figure 14 shows the results together with the flood hydrographs. Both bedload and suspended load were larger in the upstream region than in the downstream region, where the sediment hydrographs changed in accordance with the flood hydrograph. Suspended load dominates both the upstream and downstream reaches, and the bedload transport rates are negligibly small.

3.3 Sediment budgets at the downstream reach

To compute the sediment discharge simply and stably, we assumed that the bed slopes and bed material did not change in the upstream and downstream reaches, allowing parallel bed aggradation and degradation. Figure 15 illustrates the sediment budget for the downstream reach, where $Q_{\mathit{ts},\mathrm{in}}$ is the influx to the downstream reach and $Q_{\mathit{ts},\mathrm{out}}$ is the outflux from the reach to the sea.

Figure 16 shows the sediment fluxes, such as the annual amounts of bedload and suspended load transported from upstream and towards the sea, together with the approved mining amount. Annual bedload was estimated using the formula proposed by Egashira et al. According to Figure 16, it is apparent that the annual bedload and suspended load, except in 2012, which entered the downstream reach, were higher than the outflux from the reach. Outflow-suspended loads are dominated mostly by sediments finer than 0.075 mm (clay and silt), which can be transported toward the sea. Therefore, there are no silt or clay deposition issues in the downstream reaches. The approved mining amount is 14 times larger than the estimated upstream bedload for low- or non-flooded years, although the approved mining amount is 0.6 times the bedload supply during large floods and 0.9 times the bedload supply for years with medium floods. However, the actual amount of mining cannot be estimated (Global Water Partnership, 2018).

Figure 17 shows the annual sediment stored volume in the downstream reach and the approved mining amount by the Geological Survey and Mines Bureau from 2012 to 2018, where it was evaluated assuming that there was no sand mining.

Referring to the flood discharge data illustrated in Figure 3, the bar graph suggests that the annual sediment volume stored in the downstream reach was significantly influenced by the magnitude and occurrence frequency of the flood. For example, in 2012, fine sediment was only transported to the sea without significant upstream sediment supply because no flood occurred. In 2013, 2014, and 2015, the stored sediment volumes ranged from 80,000 to 308,000 m³/year. Floods ranging from middle to large sizes took place in 2016, 2017, and 2018; thus, these floods resulted in the accumulation of substantial amounts of sediment in the downstream reach, which amounted to 4,679,000, 1,126,000, and 1,924,000 m³, respectively.

4.1 Restoring channel condition

Equation (13) shows that the sediment supply rate from the upstream balances the elimination rates due to sediment mining and the flushing rate to the sea. This condition illustrated by Equation (14) shows that the sediment stored volume in the downstream reach should be larger than the critical sediment volume causing saline water issues and less than the critical volume causing floodwater inundation.

Sediment mobility is a critical aspect of river dynamics, as it is responsible for the erosion and formation of bedforms, as well as lateral and vertical changes in the channel that are essential for the diversity of riparian landscapes and the conservation of pioneer species. To fulfill condition (3), the river restoration channel must be capable of creating at least alternate bars, and sediment mobility must increase to a level that ensures significant sediment transportation.

4.2 Conditions of the 1961 channel

4.2.1 Equilibrium bed slope of the downstream

It is possible to identify whether the restoring channel can be reproduced by investigating the equilibrium bed slope of the downstream reach to the sediment supply rates from the upstream. In practice, the equilibrium bed slope of the downstream reach can be evaluated using the continuity principle for sediment transportation, composed of bedload and suspended load. The continuity equation for bedload is as follows:

$$\sum q_{b*,u}{B}_u=\sum q_{b*,d}{B}_d,$$()

where $q_{b*,u}{,B}_u$, and $q_{b*,d},B_d$ are the nondimensional bedload rates and channel widths in the upstream and downstream reaches, respectively. The nondimensional bedload transport rate has been previously defined. The continuity equation is given for the suspended load, taking Harada et al.'s formula described by Equation (9) into consideration, as follows:

$$\sum v_u{{\mathit{Fr}}_u^{2}Q}_u=\sum v_d{{\mathit{Fr}}_d^{2}Q}_d,$$()

where $v_u,{{\mathit{Fr}}_u,\mathit{Q}}_u,$ and $v_d,{{\mathit{Fr}}_d,\mathit{Q}}_d$are the mean velocity, Froude number, and flow discharge in the upstream and downstream reaches, respectively. The mean velocity, Froude number correspond to $Q_d={(A}_d/A_u)Q_u$ where, $A_d$ is upstream drainage area at Hanwella and $A_u$ is the upstream drainage area at Glencourse for a uniform rainfall intensity throughout the basin.

Substituting Egashira et al.'s bedload formula and Manning's flow resistance into Equation (15) and using sediment size fractions, the equilibrium bed slope of the downstream reach is obtained as follows:

$$i_{e,d}/i_u=0.928,$$()

where $i_{e,d}$ is the equilibrium bed slope of the downstream reach, and $i_u$ is the bed slope of the upstream reach, where $i_u=0.0005.$ Substituting Manning's flow resistance formula into Equation (16) and using the sediment size fractions, the equilibrium bed slope of the downstream reach for suspended sediment is obtained as follows:

$$i_{e,d}/i_u=0.906.$$()

Equations (17) and (18) show that the equilibrium bed slopes for the bedload and suspended load were 0.00046 and 0.00045, respectively. As shown in Figure 16, suspended sediment transportation significantly dominated in both the upstream and downstream reaches. The equilibrium bed slope described by Equation (18) is shown in Figure 18. The equilibrium bed slope was steeper than the bed slope formed in 1961, which means that we were able to restore the 1961 river reach while performing sand mining in the downstream reach.

4.2.2 Transport capacity of floodwater

The transport capacity of the floodwater was investigated for the channel formed in 1961, the restoring channel, by comparing the water surface profiles and levee heights along the reach. The water surface profiles along the reach were computed using the continuity and energy conservation equations for a one-dimensional steady flow. Figure 19 shows the water surface profiles for the restoring channel, where Manning's roughness of 0.03 was employed, and the discharges corresponding to the peak flow discharges for several flood magnitudes.

According to the comparisons of the levee/ground level heights and water surface elevations for each flood, overtopping of floodwater may occur in the regions listed in Table 2.

Table 2. Locations where overtopping occurs for the flood with a 100-year return period.

Flood discharge (m3/s)	Affected channel cross sections from the river mouth where overtopping occurs for the flood with a 100-year return period (km)
1964	6–8
1583	13.4–20
1324	20–26
1583	26–28.8
1324	28.8–31.5
1200	31.5–33.2
1324	33.2–34

Countermeasures must be implemented to reduce flood damage. However, it should be noted that computations of the water surface profile were performed to demonstrate the process for setting the allowable sediment mining volume and not for designing the channel for flood control purposes.

4.2.3 Saltwater intrusion

It is important to note that the saltwater intrusion issues were not present 25 years ago at the Ambatale water intake. However, it now occurs when the flow is less than 90 m³/s (EIA for Alterations to the Salinity Barrier at Ambatale in Kelani River, 2022). To investigate the effect of the restoring channel on water quality, the intrusion lengths of the salt wedge were calculated using the frictional two-layer hydraulic model of the salt wedge (Krvavica & Ružić, 2020) based on the method of (Schijf & Schönfled, 1953), considering an estuary width of 120 m, a quadratic interfacial drag coefficient ($C_i$) of 6 × 10⁻⁴, and an ocean basin salinity difference of 35 psu. Figure 20 illustrates the extent of salt wedges for the 1961 and 2016 channels at a flow discharge rate of 35 m³/s, which corresponds to a 90% probability of occurrence in Figure 4. The findings indicate that the salt wedge intrusion lengths for the 1961 and 2016 channels were approximately 11 and 32 km, respectively, under the flow rate of 35 m³/s. This suggests that the 1961 channel was not subject to salt intrusion up to the Ambatale water treatment plant for 90% of the flow discharges. Therefore, the sediment stored volume corresponding to the 1961 river slope is larger than the critical volume, causing saline water issues $S_{\mathrm{cl}}$.

4.2.4 Diversity in the physical environment

Bed-lowering often influences bedforms, resulting in morphological changes that may affect the ecological system. Figure 21 shows regions of mesoscale bedforms that were investigated using (Kuroki & Kishi, 1984) for flow discharges ranging from 30 to 400 m³/s, which correspond to 5% to 95% exceedance for the cumulative frequency of the daily streamflow illustrated in Figure 4, where the channel bed conditions observed in 1961 and 2016 are employed. Data points for the 1961 and 2016 riverbed correspond to flows of 30 to 100 m³/s in 10 m³/s intervals and 100 to 400 m³/s in 100 m³/s intervals. Here, ${\tau }_{*}$ is the nondimensional bed shear stress (see Equation (1) in Section 3), $B$ is the channel width, $I$ is the channel bed slope, and $H$ is the flow depth.

Sediment mobility is usually evaluated using nondimensional bed shear stress, where sediment transportation is initiated when ${\tau }_{*}$ increases above a critical value (e.g., 0.05). Consequently, for the 2016 riverbed, points below τ∗ of 0.05 pertaining to flows of 30–100 m³/s indicate that sediment transport does not occur. The results suggest that alternate bars can be formed in the 1961 riverbed for a wide range of flow discharges and cannot be produced in the 2016 riverbed, and that the sediment mobility is much higher in the 1961 channel bed than in the 2016 channel bed. These results indicate that the channel bed morphology in 1961 had diversities in the physical environment, which is expected to be suitable for ecological systems.

4.3 Allowable mining amount

We propose a river channel formed in 1961 as the restoring channel, focusing on flood management, water quality (water use), and diversity of the physical environment. The bed slope of the restoring channel was approximately 2.31 × 10⁻⁴. Therefore, the sediment volume, which is the volume of the sandwiched region between the 2016-channel bed and the 1961-channel bed, was estimated to be 10.8 million cubic meters. To fill this volume with sediment, we can employ several methods such as prohibiting excavation until the filling up of sediment or allowing a small amount of excavation until developing the restoring channel.

The 1961 river channel was set as the restoring target in consideration of the river's environmental aspects; however, if the channel is not acceptable as a flood control measure, the requirements for the river channel to be reclaimed should be re-established. To specify the appropriate amount of sand mining, flood control, water utilization, and river environment issues, including sand mining related to the newly established reclaimed river channel, were evaluated, and the restoring river channel was defined based on these evaluations. The appropriate amount of sand mining should be determined during this process.