Soil detachment by raindrop impacts was estimated by Eq. () (Torri et al., 1987). DR=(1-Cg)kEe-zh, where DR is soil detachment rate by raindrop impact (gm-2h-1) estimated for each time step, k is an index of the detachability of the soil (gJ-1), E is the total kinetic energy of the rainfall (Jm-2h-1), e-zh is the correction factor for water ponding where z depends on soil texture (0.9–3.1), and h is the depth of the surface water layer (mm). Cg is the proportion of soil surface in each grid. Raindrop impacts were categorized into direct rainfall and leaf drip, allowing the total kinetic energy (E) of raindrop to be described by Eq. (). E=(1-CC)EDHDT+CCELHLD, where CC is canopy cover in the model (i.e., in each grid) and was estimated from land use data on a scale of 0.0 to 1.0 (0 for bare land and 1.0 for highly dense forest area). ED is the kinetic energy of direct throughfall drops (Jm-2mm-1), and HDT is depth of direct throughfall drops, for which rain intensity (mmh-1) was used in the model. EL is the kinetic energy of leaf drip (Jm-2mm-1), and HLD is the depth of leaf drip (net rain (mmh-1)), which was estimated by deducting the interception loss of water from the depth of rain intensity (HDT).

The kinetic energy for direct rainfall ED can be described by Eq. () (Brandt, 1989) where I is rain intensity (mmh-1). ED=8.95+8.44log⁡I EL is the kinetic energy due to leaf drip, also as proposed by Brandt (1990) and shown in Eq. (). PH is the effective height of the plant canopy in meters. This study assumed that PH is 1 m following Kabir et al. (2011). EL=15.8PH0.5-5.87 For soil detachment due to overland flow, we used equations derived by Habib-ur-Rehman and Akhtar (2004) and shown as Eqs. (5) and (6). These were used to compute soil detachment based on comparison of critical shear stress and hydraulic shear stress. DF=Kfττc-1forτ>τc,DF=0forτ<τc, where DF is an overland flow detachment (kgm-2s-1), Kf is an overland flow detachability coefficient (kgm-2s-1), τc is critical shear stress for initiation of soil particle motion as obtained from the Shield's curve (Nm-2) and τ is hydraulic shear stress (Nm-2) as given in Eq. (). τ=γhS, where γ is a specific weight of water (Nm-3) and h is depth of overland flow (m). In this study, the depth of overland flow is assumed to be the corresponding surface water depth. S is the slope of the ground surface. In Eq. (5), Kf is subjected to calibration and the critical shear stress values are obtained by the following equation. τC=Nsheildsγs-γDs, where Nsheilds is the value of the dimensionless shield parameter obtained from the Shield's curve, γs is the specific weight of sediment particles (Nm-3), γ is a specific weight of water (Nm-3) and Ds is sediment particle size (µm).

Soil detachments by flow and sediment deposition in rivers are generally considered to occur simultaneously. Flow detachment or deposition can be expressed by Eq. (), as described by Morgan et al. (1998). DFriver=βswvs(TC-Cs), where DFriver is the flow detachment or deposition of sediment (m3s-1m-1) for sediment concentration Cs (m3m-3), TC is the transport capacity (m3m-3), w is the width of the river flow (m) in each subbasin as estimated from the input parameter of the hydrological model and vs is particle settling velocity (m-1) calculated with Stokes's Law. βs is a correction factor used to calculate cohesive soil erosion as shown in Eq. () (Kabir et al., 2011). βs=0.79e-0.85J, where J is the soil cohesion (kPa). Several methods have been developed to estimate TC. This study adopted Eq. (), proposed by Govers (1990), because of its simple structure and available input parameter database. Equation () was only used to estimate SSL, not including bed loads. TC=c(ω-ωcr)η, where ω is the unit stream power (cm-1), ω=10Vs, V is mean flow velocity (cm-1), s is the slope in percentage, and ωcr is the critical value of unit stream power (0.40 cms-1). In this study, 2.67 gcm-3 of soil density was used for the conversion unit in both case studies. c and η are coefficients that depend on the median particle size of the soil (d50 in µm). c=[(d50+5)/0.32]-0.6η=[(d50+5)/300]0.25 The movement of sediment in each grid cell was determined by associating the movement with water discharge, based on the principle of conservation of mass and momentum similar to the flow simulation in the distributed hydrological model. The one-dimensional kinematic wave and finite difference approximation were applied to simulate sediment transport both over land and in the river. On the land grids, the movement of soil and water flow was accumulated at each flow interval with a weighting system that was based on the distance from the main stream. Then, the accumulated flow was streamed into the river as lateral flow. The water discharge (Q) was determined by the one kinematic wave approximation in the river node. The kinematic wave equation shown in Eq. () was also applied to the river routing model to calculate the movement of suspended sediment concentration (Cs) by using the given Q. Using Eqs. () and (), Eq. () was converted to Eq. (). ∂Qs∂x+∂As∂t=qsiflow+DFriverQs=QCs;Q=AV;Qs=AV∂(QCs)∂x+∂(QCsV)∂t=qs(iflow)+βswvs(TC-Cs) where Q is river discharge from the hydrological model (m3s-1), A is the cross-section area (m2) of water and sediment flow, V is stream velocity (m-1), qs is accumulated sediment yield (m3s-1m-1) in flow-interval and iflow is the number of flow intervals in each subbasin. An accumulated sediment yield was considered as the lateral sediment flow and was added at the inlet of the control volume (i.e., the river routing part). In the river routing, the unit of sediment mass (g) was changed to volume (m3) by dividing with the specific weight of sediment (2.67 gcm-3).

Geographical information for the Chao Phraya River Basin (e.g., topography, soil type, and land use) was collected for the development of a hydrological model. A DEM was obtained from the Shuttle Radar Topography Mission (URL: http://dds.cr.usgs.gov/srtm/version2_1/SRTM3/). The model has 90 m spatial resolution. In the study area, the resolution was aggregated to 1 km for simulation. Soil type classification relied on the Digital Soil Map of the World (version 3.6) from the Food and Agriculture Organization of the United Nations (FAO/UN). The dominant soil is clay and sand in the upper region and sandy silt in the lower region. Land use data (2001) were obtained from the Land Development Department of Thailand (http://www.ldd.go.th/). The land use categories are paddy field, farm land, forest, grassland, bare land, urban area, and water body. Daily precipitation data were collected from two kinds of rain gauge network systems. The first is open-source rain gauge network data provided by the Hydrology and Water Management Center for the upper northern region of the Royal Irrigation Department (RID). The other is managed by the Thai Meteorological Department (TMD). Both data sources have daily temporal resolution. Daily dam operational data such as water level, storage, inflow, and outflow (release) for the Sirikit Dam and Bhumibol Dam were gathered from RID and the Electricity Generating Authority of Thailand (E-GAT).

River discharge and SSL were calibrated and validated in combination with dam operation for the period from 2001 to 2010 at four stream gauges in the upper region (P73-Ping River, W3A-Wang River, Y37-Yom River, and N13A-Nan River) and one stream gauge in the outlet (C2-Chao Phraya River) (Fig. 2). Taking into account the availability of data, the monthly river discharge and SSL data for 2001 were used for the calibration model at all five stream gauges, whereas the observation data from 2002–2010 were used for validation. For the parameter calibration, a semi-automatic calibration method was used. It was the Shuffled Complex Evolution (SCE) algorithm (Duan et al., 1992). It was implemented in 2001 to identify suitable parameters. The dominant factors affecting the hydrological process and soil erosion, such as land use and soil characteristics, were considered for parameter calibration, as listed in Table 1. As for parameters related to sediment transport and soil erosion, the FAO global soil dataset was used to consider spatial distribution of soil properties. The parameters of sediment detachability from rain drop (k) and from sheet flow (Kf) in the basin were calibrated respectively based on the observed SSL at Khong Chiam. Soil cohesion (J) was determined in each subbasin. In the sediment model, the sediment particle size (d50) was assumed to be 50 µm based on the suspended sediment distribution in a river near Chiang Mai (unpublished data). To demonstrate the applicability of the proposed model, the model was calibrated and validated with two efficiency criteria: the Nash–Sutcliffe efficiency coefficient (NSE) and correlation coefficients (r) for 2001–2010. Lastly, the eleven soil types in the Chao Phraya Basin were reclassified into three types: clay, sand, and silt.

The monthly calibration for the hydrological and sediment process was implemented with SCE in Chao Phraya in 2001 (Table 1). The parameters for sediment (k, J) were calibrated to be larger than the reported values (Morgan et al., 1998), but they are within the reasonable range for the Chao Phraya River Basin (Bhattarai and Dutta, 2005).

Model evaluation revealed that the river discharge simulation performed satisfactorily, as shown by NSE and r in Table 2 (refer to Fig. A1 for hydrographs). The values of NSE and r at P73 were closer to 1 than in other drainage basins and were the lowest at C2 among the gauges. The reason for the lowest NSE occurring at the downstream gauge is related to the flooding situation in the Chao Phraya River Basin. Normally, the river discharge overflows every year during rainy season in the lower region because the discharge capacity around C2 is low. Therefore, the overestimated discharge is likely to overflow to land in real situations. Moreover, the average slope is 1.3 % in the lower basin, whereas it is 3.1 % in the upper mountainous region where surface water can inflow smoothly to the river channel. Thus, the condition lengthens retardation time and river discharge gets stuck in the lower regions. In addition, water withdrawal for irrigation canals, which was not modeled in this study, also has an effect on the lowest hydrological simulation in the lower region.

Regarding SSL, NSE for all stream gauges was larger than 0.5 except at C2 (Table 2). The simulation results captured the high peak of SSL during rainy season, as shown in Fig. 3. Overall, the performance of the model in the Chao Phraya River Basin indicated sufficient accuracy for long-term simulation (Table 2). The results from C2, located at the lowest reach, were not as good as in other stream gauges. At C2, the results indicate underestimation, even though the total simulated river discharge at the lower reach was overestimated and supposedly resulted in higher simulated SSL in the lower reach than in the upper reach. However, it appears the simulation error may have been larger at the lower reach because of accumulating uncertainty. The average total annual SSL was estimated to be 1.28×106 tyr-1 showing a ten-year increase from 5.72×106 to 8.10×106 tyr-1. This estimate was slightly lower than the reported estimate of the average total annual SSL in the Chao Phraya River Basin (11×106 tyr-1), as reported by FAO/AGL (2005). The different locations of the control points could be a reason for these different estimates of SSL.

It is inferred that the process of soil loss was strongly influenced by rainfall intensity. This was clearly shown by the simulated SSL at the Nan River (Fig. 3), where rainfall is higher (1341.8 mm) than at other tributaries. In contrast, the simulated SSL was lowest at the Wang River due to that area having the lowest annual average rainfall (1181.3 mm). Walling (2009) reported that the annual SSL in Chao Phraya declined from around 28×106 tyr-1 in the 1960s and early 1970s to around 6×106 tyr-1 in the 1990s. In this study, the average annual SSL was estimated to be 1.28×106 tyr-1 over 10 years based on simulation in the 2000s. In fact, the observed SSL shows a decreasing trend with a decline in annual runoff, primarily reflecting the trapping of sediment by a large number of small dams and irrigation structures and also by the larger Bhumibol and Sirikit dams (Walling, 2009). But for the 10 years targeted in this study, the observed SSL at C2 shows no decreasing trend. Nevertheless, climate change, population growth, land clearance, land use change, reservoir construction, and other infrastructure development can be expected to cause some changes in the SSL over the longer-time scale of 50 years in large river basins like the Mekong River Basin (Walling, 2011).

The simulated SSC also shows good correlations with observed data at all upper stream gauges, indicated by r larger than 0.5 except at the watershed outlet, C2 (Table 2). Possible errors in simulated SSC at the outlet could be related to river discharge simulation. The overall average relative mean square error (RMSE) between simulated and observed SSC ranged from 0.07 to 0.09 kgm-3 in the upper basins (P73, W3A, Y37, and N13A) and 0.09 kgm-3 in the outlet, C2. Basically, we confirmed two peaks every year in both observed and simulated SSC in four stream gauges: P73, W3A, Y37, and N13A (Fig. 4). The first peak occurred in May, which is the beginning of the rainy season. Beginning in June, the concentration fell while river discharge increased from the upper stream, due to the starting rainy season. The second peak occurred in the main monsoon periods (August, September, and October), which have heavy rainfall that increases the volume of river discharge.

The sensitivity of modeled SSL was also investigated for the reasonable ranges of the input parameters. The target parameters for this sensitivity analysis were soil detachability from rain drop (k), soil detachability from sheet flow (Kf), and soil cohesion (J). The target period for this analysis was one year 2005 at P73.

First, results were obtained by changing the detachability of soil (k) from 7.0 (for clay and silt) and 9.1 (for sand) (gJ-1) by +50, -50, and -75 % from those calibrated values. The theoretical range of this soil detachability index is 0.01–10 gJ-1, where the minimum is for clay, and the maximum is for sand (Gumiere et al., 2009; Morgan et al., 1998; Morgan, 2001). The peak of SSL increased as the detachability increased (Fig. 5a). The calibrated parameter k for dominant clay soil (7.0 gJ-1) was larger than one for the dominant sandy soil in the Chao Phraya River Basin (3.5 gJ-1, Bhattarai and Dutta, 2005) and one for a wide range of soil texture that are commonly used for agriculture in Europe (2.0 gJ-1, Morgan et al., 1998). Basically, the soil detachability is associated with soil texture, showing a higher detachability with a lower clay content. Thus, soils having a high clay content are difficult to detach by raindrops (Sharma et al., 1994). In the Chao Phraya River Basin, detachability (k) was relatively high, indicating high soil detachment and resulting in high SSL transport into the river. The presented results reveal the importance of raindrop detachment for different type of soils, especially for clay soil in the Chao Phraya River Basin.

Second, we focused on Kf, which indicates soil detachability from sheet flow (Eq. 5). The initial values of Kf (0.6 (clay), 1.0 (silt), and 1.1 mgm-2s-1 (sand)) were shifted by the factors 100, 10, 0.1, and 0.01 (Fig. 5b) in each type of soil; clay, silt and sand. The results show that the simulated suspended sediment peaks (August to October) increased slightly as Kf increased, although the changes were small and invisible in Fig. 5b. Thus, SSL is less sensitive to Kf than k, which is possibly because the precipitation is the main agent for sediment yield.

Third, soil cohesion (J) was shifted by +25, -25 and -50 % from the calibrated value (3.0 kPa). Each model output was confirmed to understand the degree of net soil detachment in streams influenced by transport capacity. The peak of SSL in September increased as J increased (Fig. 5c). In this case, the lateral inflow of sediment was the same as the initial results with 3.0 kPa. Equations () and () infer that soil erosion increases SSC as higher J under saturated SSC condition (i.e., TC < Cs) contributes to less deposition. Generally, J needs to be adjusted considerably to properly predict the measured net soil loss, since J is related to erodibility and limits detachment within river sediment. The simulated SSL from the LISEM model consistently increased with measured SSL and with increasing J (range from 2 to 7 kPa) (Nearing et al., 2005).

Sensitivity analysis was conducted for three parameters to evaluate the reliability of the model for simulating sediment dynamics. Overall, the two input parameters (k and J) that describe soil erodibility showed the significant influence on SSL in the Chao Phraya River Basin. The range of input parameters used in this model could be a useful reference for sediment-related research in the Chao Phraya River Basin.

The input data for the model include weather data, topography data, soil properties and land cover. In this study, the GTOPO30 global DEM data with a horizontal grid spacing of ∼20 km2 (grid area: 2min×2min) resolution was used to delineate the Mekong River Basin. The land cover and soil type for the basin were obtained from Global Land Cover 2000 (http://edc2.usgs.gov/glcc/eadoc2_0.php) and the FAO soil map of the world (FAO, 2003), respectively. The elevation data was first converted to 3.6km×3.6km resolution, and land cover and soil data were aggregated by reclassifying the land use data for nine classes and the soil data for eight classes. Daily precipitation and air temperature data from 65 station weather stations were obtained from the Mekong River Commission (MRC).

Annual records of river discharge and SSC in the study were extracted from the historical record published by the MRC (Mekong River Commission, 2005). The historical record tabulated measurements of river discharge, SSC, water quality, and other physical characteristics at gauging stations located along the Mekong River Basin and those of the river's tributaries. In this study, river discharge and SSC records from the three targeted stations were identified and used to calibrate and validate SSL simulation. The stations were selected based on their relative locations and the completeness of river discharge and sediment records at the station. Unlike river discharge, which was measured daily, SSC was monitored monthly. SSC samples were collected near the surface of the river (0.3 m depth) in the middle of the main stream (MRC, 2000). In this study, monthly observed SSL was computed from the monthly measured SSC and the corresponding measured daily river discharge.

The river discharge and SSL were simulated by considering an existing dam in the Chinese section of the main stream (Manwan Dam). The model simulated river discharge, SSL, and SSC for 10 years from 1991 to 2000, and three stream gauges along the main stream were adopted for calibration and validation (Fig. 6). The daily river discharge and sediment data for the period from 1991 to 1995 were used for calibration, whereas the data from 1996 to 2000 were used for validation. For the sediment particle size (d50), 50 µm were adopted in this study area because all the sediment is commonly <62 µm in diameter (i.e., silt and clay), and sediments less than 2 µm make up 45 % of the section near Vientiane (Ahlgren and Hessel, 1996). The five parameters shown in Table 1 were initialized with empirical values and then calibrated according to the observed river discharge and SSL at the three gauges. All the parameters were calibrated by SCE (Duan et al., 1992). Observation at Chiang Sean was used for calibration of parameters that reflect only the upper basin, whereas observation at Khong Chiam was used for calibration for the middle basin and observation at Phnom Penh was used for the lower basin. The fit between simulated and observed results (river discharge and SSL) was evaluated using the NSE (Nash and Sutcliffe, 1970) and correlation coefficients (r) at monthly intervals from 1991 to 2000.

The river discharge of the Mekong was well simulated at all three stations (Table 3) (refer to Fig. B1 for hydrographs). The NSE values for river discharge at Chiang Sean, Khong Chiam and Phnom Penh were larger than 0.7 for calibration and validation from 1991 to 2000. The average correlation r between observation and simulation discharge was equally high (r>0.8). These indicators imply the GBHM satisfactorily describes the seasonal cycle and spatial distribution of the hydrology processes in the Mekong River Basin, although the simulated discharge showed slightly higher peaks.

For the sediment model, we calibrated the same parameters as in the Chao Phraya River basin. In the Mekong River, wider ranges were determined for k and Kf, whereas the range of J narrowed slightly due to different physical characteristic such as topography (e.g., rill, interill, and gully) and structure of soil (e.g., erodibility and type of soil content). For example, the range of k for the Mekong (7.0–100.0 kgJ-1) is wider than that for the Chao Phraya (7.0–9.1 kgJ-1) (Table 1). This is probably because of the soil structure and content. The Chao Phraya River Basin is mostly covered with sandy soil, which can be easily detached. In contrast, the Mekong River Basin is dominantly covered by clay soil. This may be why SSL is not sensitive to k, and k is not important for sediment yield in the Mekong River Basin.

Figure 7 shows the simulated results of monthly SSL compared with measurements at three gauging stations from 1991 to 2000. The simulated results are in good agreement with observations, as summarized in Table 3. NSE was larger than 0.6 for the upper, middle, and lower stations in the calibration (1991–1995) and validation (1996–2000) periods, except in the case of the validation period for the upper station (Chiang Sean) (NSE =0.51). The model simulation underestimated at the upper (Chiang Sean) station (Fig. 7), and this error may have been caused by the effect of the Manwan Dam on the main stream in the upper basin of the Mekong River Basin (Lu and Siew, 2006). Nevertheless, the linear correlation coefficient (r) between simulated and observed SSL was in the range of 0.80–0.86 for all three stations. Generally, the SSL was fairly well simulated at the three stations (Fig. 7). The simulated results describe the seasonal pattern of SSL in the Mekong River Basin, and higher SSL is expected during the rainy season, as described also by Walling (2008). The model results reveal that, for the entire period of 1991–2000, the average annual SSL values were the highest (8.9×107 tyr-1) at the middle region, whereas the upper and lower regions showed average annual SSL values of 3.4×107 and 5.6×107 tyr-1, respectively. Walling (2009) reported that the annual SSL is high in the middle region before increasing further downstream. Our simulated results also showed higher SSL in the middle region (i.e., Khong Chiam station) than in other regions. This is probably due to the large tributary drainage area. The annual SSL at Lower Mekong (after the Chinese boundary, including the middle region) tends to increase as basin area increases.

The simulated total annual SSL at Phnom Penh fluctuated over the period 1991–2000 (average =6.86×107 tyr-1). This study shows a 57 % lower SSL than the reported average annual SSL (16.0×107 tyr-1) (FAO/AGL, 2005). Lu and Siew (2006) reported the average annual SSL for the period 1962–2003 in the Mekong River Basin was about 14.5×107 tyr-1 based on rating curve estimation. The difference in SSL between those reports and our estimate here is probably due to the different locations of the control stations. In FAO/AGL (2005), SSL was estimated for the whole Mekong River Basin area, including the Mekong delta. This study covered only the area to Phnom Penh. In addition, different methods for SSL estimation could explain the variances.

The monthly SSC simulation at the stream gauging stations for the period 1991–2000 is shown in Fig. 8. The correlation coefficient r was larger than 0.5 between monthly observed and simulated SSC at the three gauging stations (Table 3). The overall average RMSE between the observed and simulated SSC was 0.31 at Chiang Sean, 0.25 at Khong Chiam and 0.14 kgm-3 at Phnom Penh. The results reveal a decreasing trend in the SSC along the three regions from upstream to the downstream (Fig. 8). The average monthly SSC was the highest at Chiang Sean station (estimated to be 0.33 kgm-3). The average SSC was lowest at Phnom Penh (estimated to be 0.13 kgm-3). The low value at Phnom Penh was due to the sediment deposition in the lower region. This trend was due to the decrease in the main stream water velocity, which promotes sediment deposition and decreases SSC. In fact, a decreasing trend in mean monthly SSC was observed along the entire length of the Mekong River since water quality measurement began in 1985 (Lu and Siew, 2006). The model results also show that the SSC was higher in the rainy season (July, August, and September) than the dry season at all three stations (Fig. 8). This is due to the intensive soil erosion mainly caused by heavy precipitation in the rainy season. The high simulated SSC in July at the upper station showed a good agreement with the observed SSC, which recorded the highest concentrations occurring early in rainy season, in mid-July. After mid-August, the observed SSC began to decline, continuing to decline through early September. This trend matches the simulated SSC trend, which showed a decline in August and September.

A sensitivity analysis was applied for SSL in the Mekong River Basin in the same manner as in the Chao Phraya River Basin. First, all the parameters were set to calibrated values, which were k=7 gJ-1, Kf=1 mgm-2s-1, and J=3 kPa. The same factor was used in the Mekong River Basin and Chao Phraya River Basin in order to compare the sensitivity of the parameters for SSL in the both basins. The SSL was simulated at Khong Chiam station for all the parameters in 1999.

The results revealed that SSL increases slightly in September when k decreases by 50 % from the initial (Fig. 9a). In addition, the peak SSL kept increasing and showed smaller changes with further decreases (at a factor of 25 %) from the initial value. Thus, in the Mekong River Basin, SSL is less sensitive to k than in the Chao Phraya River Basin. The subtle response of SSL for different k implies that soil strength and clay content are not important, although studies of soil strength for the different group soils of sands, loams, and clays (Sharma, 1999) show that k decreases as soil strength increases. Such results from a sensitivity analysis suggest that soil detachment by raindrop contributes little to SSL generation in the Mekong River Basin.

Regarding Kf, the peaks in SSL in 1999 decreased by 40 % with the multiplying a factor of 100 % (Fig. 9b) due to the increase of soil detachability from sheet flow. The simulated SSL also drastically decreased with decreasing Kf (with factors 0.1 and 0.01). The simulated SSL reveals the opposite trends from that reported by Bhattarai and Dutta (2005), who found the simulated SSL peaks from August to October increased with increasing Kf values (0.4 to 0.6 mgm-2s-1). This result implys that soil detached from sheet flow is important input for SSL transport in the river in the Mekong River Basin. The literature does not contain conclusive results on the sensitivity range of Kf (Lukey et al., 2000; Bathurst et al., 2002). For example, sediment modeling using Kf with a small range from 0.0019 to 0.0045 mgm-2s-1 in New Zealand (Russell and Sandy, 2006) showed the inverse trend of the result shown in Mekong for the same range, where the simulated SSL increased with increasing Kf.

The analysis of soil cohesion (J) shows that the SSL peak in September increases by 150 % with a factor of 1.25 for J (Fig. 9c). In contrast, the SSL peak decreases by 70 and 80 % by decreasing the soil cohesion factor by 0.75 and 0.5, respectively. The change in SSL is more sensitive to soil cohesion than k and Kf, as soil cohesion indicates soil detachability within rivers. In the equations for net soil detachment in rivers (Eqs. 9 and 10), soil cohesion limits the detachment of sediment. Soil cohesion is recognized to be related to erodibility, and no unique relationship exists even for a single size of soil (Govers et al., 1990).

The three input parameters that describe soil erodibility have a significant influence on the output of SSL. Generally, soil cohesion (J) is the most sensitive parameter in both river basins. The SSL change was sensitive to soil detachability over land (Kf) in the Mekong, whereas SSL change was more sensitive to detachability by raindrop (k) in the Chao Phraya.