The Irrawaddy (Ayeyarwady) is the largest river basin in Myanmar, covering more than 50% of the land area. The Irrawaddy river is ∼2,100 km long, while the drainage area of 410,000 km² is primarily located in Myanmar (91%) and small parts in China (4%) and India (5%) (Figure 1A). The Irrawaddy River originates at the confluence of the Mali and N’Mai rivers and is fed by its main tributary (the Chindwin River) at Pakokku. It is the most important commercial waterway in Myanmar and ends in the Andaman Sea to form the second-largest delta system in Southeast Asia. The delta system begins ∼120 km downstream of the Pyay station and propagates ∼2.5 km/100 years (on average) into the Andaman Sea (Rodolfo, 1975). The basin’s topography varies from hilly mountain ranges upstream and low-lying delta downstream, passing through middle flood plains and plateaus. More than 65% of the basin area is covered by forest and agricultural lands. The most commonly found soil type of the IRB is clay-rich Acrisols. The basin receives a spatially varied annual rainfall of 500–4,000 mm, mainly during the monsoon season (May to October), with the average daily temperature ranging between 11 and 34°C within a year. The Irrawaddy river carries ∼380 Billion m³ of water and ∼325 Million tons of sediment annually at Pyay station.

The Kalu River Basin is the second largest river basin in Sri Lanka and receives high rainfalls leading to high river flows. The Kalu River (Kalu Ganga) originates from the Samanala mountain range in the South-central part of Sri Lanka and falls out to the sea at Kaluthara after traversing ∼129 km (Figure 1B). The drainage area of the Kalu River is ∼2,787 km². The coastal zone adjacent to the Kalu river outfall comprises a small tidal inlet system experiencing a .6 m oceanic tidal range. Much of the basin is utilized for rain-fed paddy cultivation, rubber, tea, and other commercial crops scattered throughout the basin. Clay-rich Acrisol is the most dominant soil type found in the basin. The average annual rainfall in the basin is ∼3,800 mm, mainly driven by the southwest monsoon (May–September). The average daily temperature in the basin is ∼25°C. The Kalu river carries ∼4,000 Million m³ of water and ∼0.7 Million tons of sediment load annually (on average) to the sea.

The Soil Water Assessment Tool (SWAT), developed by USDA Agricultural Research Service (Arnold et al., 1998; Neitsch et al., 2011), is a process-based continuous-time model for catchment simulations. In SWAT, a river basin is partitioned into sub-basins and further divided into hydrological response units (HRUs) based on land use, soil type, and slope classes. HRUs are the primary computational units of SWAT. Major catchment processes modelled in SWAT are hydrology, soil erosion, nutrients/pesticides (water quality), plant growth, and channel routing. SWAT estimates the surface runoff of each HRU using the Soil Conservation Services-Curve Number (SCS-CN)(Soil Conservation Service, 1971; Soil Conservation Service Engineering Division, 1986) or the Green and Ampt infiltration method (Mein and Larson, 1973). In this study, the SCS-CN method was used in all the simulations. The model estimates the soil erosion at each HRU caused by rainfall and runoff using the Modified universal Soil Loss Equation (MUSLE) (Williams and Berndt, 1977; Wischmeier and Smith, 1978) and assumes that all eroded sediments reach the channels. MUSLE (Eq. 1) is the modified version of the Universal Soil Loss Equation (USLE). In this version, the rainfall energy factor in USLE is replaced with the runoff factor, which improves sediment predictions and allows for the simulation of individual storm events (Neitsch et al., 2011). s e d = 11.8 ∙ Q s u r f ∙ q p e a k ∙ a r e a h r u 0.56 K U S L E ∙ C U S L E ∙ P U S L E ∙ L S U S L E ∙ C F R G (1) where, sed is the sediment yield (tons), Q s u r f is the surface runoff (mm), q p e a k is the peak runoff (m³/s), a r e a h r u is the area of HRU (ha), K U S L E ,   C U S L E , P U S L E ,   a n d   L S U S L E are USLE soil erodibility factor, cover, and management factor, support practice factor, and topographic factor, respectively, and C F R G is the coarse fragment factor.

The sediment transport model used in SWAT consists of two processes (i.e., deposition and degradation) which determine the magnitude of sediment generated within a river reach. The amount of sediment deposition or degradation depends on several factors, such as the maximum sediment concentration transported with river flow (Eq. 2), according to Bagnold’s Equation (Neitsch et al., 2011), flow velocity, flow rate, soil cover, and erodibility of the reach. c o n c s e d , c h , m x = c s p ∙ v c h , p k s p e x p (2) where, c o n c s e d , c h , m x is the maximum sediment concentration transported by water (ton/m³), c s p is the sediment transport coefficient; v c h , p k is the peak velocity in the river (m/s), and s p e x p is the exponent of the velocity.

When the maximum sediment concentration that can be carried by the water flow is less than the sediment concentration of the reach, sediment deposition (Eq. 3) occurs and vice versa for sediment degradation (Eq. 4). s e d d e p = c o n c s e d , c h , i − c o n c s e d , c h , m x . V c h (3) s e d deg =   c o n c s e d , c h , m x − c o n c s e d , c h , i . V c h ∙ K C H ∙ C C H (4) where, s e d d e p and s e d deg are the deposited and reentratined sediment volumes in the reach, respectively (tons), c o n c s e d , c h , i is the sediment concentration of the reach at the initial time (ton/m³), c o n c s e d , c h , m x is the maximum sediment concentration, which can be carried by water (ton/m³), V c h is the water volume in the reach (m³), K c h is the erodibility factor of the channel, and C C H is the cover factor of the channel.

The amount of sediment (in tons) flowing out from a river reach (Eq. 5) is calculated based on the sediment amount (in tons–based on the initial amount of suspended sediment in the reach, deposited and degraded sediments), and volume of outflows. s e d o u t = s e d c h ∙ V o u t V c h (5) where, s e d o u t is the amount of sediment outflow from the reach (tons), s e d c h is the suspended sediment load in the reach (tons), and V o u t is the outflow volume (m³).

As model inputs, SWAT requires surface elevation data (digital elevation model (DEM)), land use, soil characteristics, land management, and daily climatic data (i.e., precipitation, maximum and minimum temperatures). In this study, all the spatial data (i.e., DEM, landuse, and soil) were obtained from freely available global products. Hydro-meteorological data (i.e., precipitation, temperature, streamflow, and sediment load) were obtained from local authorities in the two countries (i.e., Myanmar and Sri Lanka) and previous studies and reports. The SWAT model setup for each basin was calibrated and validated for streamflow and sediment loads. At first, streamflow was calibrated and validated with the available data at eight stations in IRB and 3 stations in KRB. These models were subsequently calibrated and validated for sediment load at 3 stations in IRB and at the basin outlet of KRB. The detailed information on input data, calibration, and validation of the SWAT models for streamflow and sediment loads at the Irrawaddy and Kalu River Basins are presented in Sirisena et al., 2018, Sirisena et al., 2021a, and Sirisena et al., 2021b.

In SWAT simulations for future periods, precipitation and temperature data were obtained from three General Circulation Models (GCMs) for the Irrawaddy Basin and three Regional Climate Models (RCMs) for the Kalu Basin under RCP 2.6 and RCP 8.5. Here, simulations were performed for the two future periods considered (i.e., 2046–2065 (mid-century) and 2081–2100 (end-century)). The detailed descriptions of GCMs and RCMs are provided in Supplementary Table S1 and the selection of the respective GCMs and RCMs are summarized in Sirisena et al. (2021a), Sirisena et al. (2021b), respectively. It is assumed that prevailing land-use conditions will remain invariant throughout the modelling period, and the inclusion of planned reservoirs is the only future human activity considered. For the future periods, out of the several planned reservoirs in the Irrawaddy basin, six large reservoirs having capacities of 17.7, 2.2, 2.6, 11.2, 13.2, and 8.6 billion m³ were considered to analyze their impacts on streamflow and sediment transport using SWAT (more details can be found in Sirisena, 2020; Sirisena et al., 2021a). Each reservoir and its trapping efficiency are individually represented in the SWAT model.

Many coastal studies have used BQART presented by Syvitski and Milliman (2007) to project annual fluvial sediment supply to the coast (e.g. Balthazar et al., 2013; Bamunawala et al., 2018a; Bamunawala et al., 2020a; Bamunawala et al., 2020b; Bamunawala et al., 2021). BQART was developed using data from 488 global river basins, and it estimates the long-term average annual suspended sediment load using the following equations. Q s = ω   B   Q 0.31   A 0.5   R   T f o r   T ≥ 2 0 C (6) where Q _s is the sediment load in kg/s or MT/yr with ω = 0.02 or 0.0006, respectively, B is the lithology and human impact index, Q is the annual streamflow from the basin (km³/yr), A is the basin area (km²), R is the relief of the basin (km), T is mean annual temperature of the basin (^oC).

The term ‘B’ accounts for geology and human activities, which is defined as; B = I L 1 − T E E h (7) where I is the glacial erosion factor (I ≥ 1), L is the basin-wide lithology factor, T E is reservoir trapping efficiency, and E h is the human-induced erosion factor.

The term I is defined as I = 1 + 0.09 A g (8) where, A g is the percentage of ice coverage area in the basin.

The main rivers in the two study areas are mostly in pristine condition. However, as mentioned earlier, six large planned reservoirs were considered in the hydrological simulations of the Irrawaddy basin. Since BQART is a lumped model, the trapping efficiency is defined for the entire basin. Therefore, for BQART applications, the basin-wide Trapping Efficiency (TE) from the six reservoirs was estimated using the method proposed by Vörösmarty et al. (2003) (from Eqs 6–11). Vörösmarty et al. (2003) developed the basin-wide trapping efficiency model for large reservoirs (>500 MCM) by considering 633 reservoir data across the world. This basin-wide trapping ( T E b a s i n ) is derived by considering the geographical locations of the reservoirs and the sediment residence time from 236 regulated river basins. Figure 2 shows the schematic diagram of a representative basin as considered by Vörösmarty et al. (2003) in this derivation. ∆ τ r e g , j = ∑ 1 n V i Q j e g . V 2 + V 3 + V 4 + V 5 Q B (9) T E r e g , j = 1 − 0.05 ∆ τ r e g , j (10) T E b a s i n = ∑ 1 m T E r e g , j   Q j Q m (11) where, ∆ τ r e g , j is the residence time of reservoir areas of j ^th regulated sub-basin of the basin (j is A, B, and C here), V i is the operational volume of i ^th reservoir in j, Q j is the discharge at regulated sub-basin j, T E r e g , j is approximated trapping efficiencies of j ^th sub-basin, T E b a s i n is the discharge-weighted trapping efficiency of the basin based on discharge, Q m is discharge at the basin outlet, n is the number of reservoirs in each regulated sub-basin j, and m is the number of regulated sub-basins in the basin

For the Irrawaddy RB, different T E b a s i n values were calculated for each General Circulation Model (GCM) under RCP 2.6 and RCP 8.5 for the two future periods considered (i.e., 2046–2065 (mid-century) and 2081–2100 (end-century)). Since there are no planned reservoirs for the Kalu basin, reservoir effects are not considered in the fluvial sediment load estimations. Furthermore, a zero trapping efficiency was considered while implementing the BQART model for both river basins to explicitly represent the climate-driven impacts on fluvial sediment supply. The calculation of T E b a s i n (Eq. 11) is based on the volume of reservoirs, discharge from reservoirs, and discharge at the basin outlet. Based on three GCMs and two RCPs, the average basin-wide trapping efficiency ( T E b a s i n ) of the Irrawaddy RB was computed as 42%. The river basin and reservoir data used by Vörösmarty et al. (2003) include some Asian River Basins, including Mekong (0%–20%), Chao-Pharya (20%–40%), Ganges and Brahmaputra (0%–20%), Krishna (80%–100%), Yangetz Basin (60%–80%), and Indus (40%–60%). The number of large reservoirs within each basin varies from 3–20. Furthermore, large reservoirs (633) and small reservoirs (>40,000) in the world contribute 30% and 23% of sediment trapping, respectively (Vörösmarty et al., 2003). Considering this wide range of computed trapping efficiencies, the above calculated T E b a s i n of the Irrawaddy RB appears to be reasonable.

Table 1 summarizes the input parameters of the BQART model. The river discharge (Q) values were obtained from the SWAT simulation results presented in Sirisena, 2020; Sirisena et al., 2021a; Sirisena et al., 2021b over the baseline and mid-and end-century periods for RCP 2.6 and RCP 8.5. The mean temperature (T) over the basins was obtained from the three selected GCMs/RCMs (discussed in Sirisena et al., 2021a; Sirisena et al., 2021b for the same RCPs. The human-induced erosion factor ( E h ) in Eq. 7 is based on land use, socio-economic situation, and population density. The suggested optimum range of E h is 0.3 ≤ E h ≤ 2.0 (Syvitski and Milliman, 2007). Similar to the assumption made for the land-use change in SWAT simulations, for future simulations with the BQART model, the human-induced erosion factor ( E h ) is assumed to remain constant throughout the 21st century (Table 1).

The parameter ‘B’ in the BQART represents the geology and human activities in the basin. The human-induced erosion factor ( E h ) is one of the main influencing factors and is linearly correlated with B and consequently with the fluvial sediment load computed by BQART; Syvitski and Milliman (2007) have derived the E h factor based on land use, socio-economic situation and population. However, recent studies by Balthazar et al. (2013); Bamunawala et al., 2020b; Bamunawala et al., 2020a; Bamunawala et al., 2021 have shown that the human footprint index (HFPI) presented by Wildlife Conservation Society-WCS and Center for International Earth Science Information Network-CIESIN-Columbia University (2005) can be a better representation of the anthropogenic influences on fluvial sediment load estimation via BQART. The HFPI was developed based on several global datasets, including population, urban areas, land use, navigable waterways, roads, and electrical infrastructure (Sanderson et al., 2002).

As a further test of SWAT and BQART projected fluvial sediment loads, the BQART projections were re-calculated with the updated E h (using HFPI) and re-compared with the corresponding SWAT projections. To do this, HFPI values, which range from 0–100, need to be rescaled to the global range of E h (0.3 ≤ E h ≤ 2.0) estimated by Syvitski and Milliman (2007) for use in the BQART. HFPI values across the Irrawaddy and Kalu River Basins were obtained from 0.25° × 0.25° resolution data (Wildlife Conservation Society-WCS and Center for International Earth Science Information Network-CIESIN-Columbia University, 2005) and rescaled accordingly. Subsequently, the basin average rescaled HFPI values were used with BQART to re-compute the fluvial sediment load at the outlets of the two basins. The comparison of projected sediment loads from SWAT and BQART (with the two different E h factors; one derived from HFPI and one following Syvitski and Milliman (2007) with climate change only (i.e., no reservoirs) is shown in Table 2. With the HFPI-based E h , BQART projects lower sediment loads in both basins than its previous estimations. This is because the HFPI-based human-induced erosion factors E h for the two basin areas are almost half the E h values suggested by Syvitski and Milliman (2007) (Table 1).

The sediment load projections by SWAT and BQART are different in the two river basins. In Irrawaddy RB, SWAT simulations project a higher sediment load than those predicted by BQART and vice versa in the Kalu RB. For example, when using the human-induced erosion factor ( E h ) obtained from Syvitski and Milliman (2007), for the baseline period (1991–2005), simulated sediment loads by the models differ by 18%–22% in the Irrawaddy RB and 315%–331% in the Kalu RB. During the end-century period under RCP 8.5, those differences in projected sediment loads are 15%–33% in the Irrawaddy RB and 116%–145% in the Kalu RB. During the same period and under the same RCP, the projected sediment supply is further reduced in the Irrawaddy RB due to the proposed reservoirs. Therefore, the difference in sediment loads between BQART and SWAT projections is 48%–59%. Compared to the baseline period (1991–2005) model estimations, BQART projections generally show lower increments than SWAT projected increments in the Irrawaddy and Kalu River Basins when the reservoirs are not considered (i.e., climate change only).

When adopting the human-induced erosion factor ( E h ) derived from the human footprint index (HFPI), the baseline period (1991–2005) projected sediment loads by BQART and SWAT differ by 46%–50% and 93%–101% in the Irrawaddy RB and the Kalu RB, respectively. For the end-century period under RCP 8.5, those differences in the sediment load projections are 70%–72% and 1%–14% in the Irrawaddy RB and the Kalu RB, respectively.

BQART does not provide information on sediment erosion, transport, and deposition in the flood plains, as it only estimates the sediment delivery rate at/near sea level of the basin outlet (Syvitski and Milliman, 2007). The empirical BQART equation is developed based on 488 global river datasets comprising catchments sizes spanning a large range (160 km²–5,853,804 km²) with high accuracy in calibration (R ² = 0.96 for 292 basins) and validation (R ² = 0.94 for 196 basins) (Syvitski and Milliman, 2007). However, an analysis by De Vente et al. (2013) summarized that non-linear regression models like BQART might provide more accurate results than distributed models such as SWAT would for sediment yield in basins larger than 10,000 km². Nevertheless, a study of the Blue Nile and Atbara river systems showed that a global flux model such as BQART is less suited for capturing highly spatially varied sediment yields ranging from thousands of ton/km²/year in a basin (Balthazar et al., 2013). Although, spatially distributed models such as SWAT demand more input data and high calibration efforts they are more suitable for assessing environmental change scenarios such as those due to climate change, land use, and management practices (Neitsch et al., 2011). Therefore, both models have their advantages and disadvantages in sediment load estimation for a selected region under diverse environmental and geographical conditions.