ISSN: (Print) (Online) Journal homepage:

Transcription

1 Hydrological Sciences Journal ISSN: (Print) (Online) Journal homepage: A comparative analysis of sediment yield simulation by empirical and process-oriented models in Thailand / Une analyse comparative de simulations de l'exportation sédimentaire en Thaïlande à l'aide de modèles empiriques et de processus RABIN BHATTARAI & DUSHMANTA DUTTA To cite this article: RABIN BHATTARAI & DUSHMANTA DUTTA (28) A comparative analysis of sediment yield simulation by empirical and process-oriented models in Thailand / Une analyse comparative de simulations de l'exportation sédimentaire en Thaïlande à l'aide de modèles empiriques et de processus, Hydrological Sciences Journal, 53:6, , DOI: / hysj To link to this article: Published online: 18 Jan 21. Submit your article to this journal Article views: 887 View related articles Citing articles: 15 View citing articles Full Terms & Conditions of access and use can be found at

2 Hydrological Sciences Journal des Sciences Hydrologiques, 53(6) December A comparative analysis of sediment yield simulation by empirical and process-oriented models in Thailand RABIN BHATTARAI 1 & DUSHMANTA DUTTA 2 1 Department of Agricultural and Biological Engineering, 134 W Pennsylvania Avenue, University of Illinois at Urbana-Champaign, Urbana, Illinois 6181, USA rbhatta2@illinois.edu 2 School of Applied Sciences and Engineering, Monash University, Churchill, Victoria 3842, Australia Abstract Although soil erosion has been recognized worldwide as a threat to the sustainability of natural ecosystems, its quantification presents one of the greatest challenges in natural resources and environmental planning. Precise modelling of soil erosion and sediment yield is particularly difficult, as soil erosion is a highly dynamic process at the spatial scale. The main objective of this study was to simulate soil erosion and sediment yield using two fundamentally different approaches: empirical and process-oriented. The revised form of the Universal Soil Loss Equation (RUSLE), along with a sediment delivery distributed model (SEDD) and the Modified Universal Soil Loss Equation (MUSLE), which are popular empirical models, were applied in a sub-basin of the Mun River basin, Thailand. The results obtained from the RUSLE/SEDD and MUSLE models were compared with those obtained from a process-oriented soil erosion and sediment transport model. The latter method involves spatial disaggregation of the catchment into homogeneous grid cells to capture the catchment heterogeneity. A GIS technique was used for the spatial discretization of the catchment and to derive the physical parameters related to erosion in the grid cells. The simulated outcomes from the process-oriented model were found to be closer to observations as compared to the outcomes of the empirical approaches. Key words MUSLE; process-oriented model; RUSLE/SEDD; sediment yield; soil erosion Une analyse comparative de simulations de l exportation sédimentaire en Thaïlande à l aide de modèles empiriques et de processus Résumé Bien que l érosion des sols ait été reconnue mondialement comme étant une menace contre la durabilité des écosystèmes naturels, sa quantification reste un des plus grands défis en matière de gestion des ressources naturelles et de planification environnementale. La modélisation précise de l érosion des sols et de l exportation sédimentaire est particulièrement difficile, l érosion des sols étant un processus fortement variable dans l espace. L objectif principal de cette étude est de simuler l érosion des sols et le bilan sédimentaire à l aide de deux approches fondamentalement différentes: empirique d une part et centrée sur les processus d autre part. La forme Révisée de l Equation Universelle de Perte de Sol (RUSLE) associée à un modèle distribué d exportation sédimentaire (SEDD) et l Equation Universelle Modifiée de Perte de Sol (MUSLE), qui sont des modèles empiriques populaires, ont été mises en œuvre dans un sous-bassin de la Rivière Mun en Thaïlande. Les résultats des modèles RUSLE/SEDD et MUSLE ont été comparés à ceux d un modèle centré sur les processus d érosion des sols et de transport sédimentaire. Ce dernier repose sur la désagrégation spatiale du bassin versant en mailles homogènes afin de tenir compte de l hétérogénéité du bassin. Une technique géomatique a été utilisée pour discrétiser spatialement le bassin et pour estimer les paramètres physiques relatifs à l érosion dans chaque maille. Les résultats de simulation du modèle centré sur les processus apparaissent être en meilleur accord avec les observations que ne le sont les résultats des approches empiriques. Mots clefs MUSLE; modèle centré sur les processus; RUSLE/SEDD; exportation sédimentaire; érosion des sols INTRODUCTION It is a matter of concern that the world population is increasing at a rapid rate and the resources like soil, which are necessary to sustain the population, are steadily declining day by day. The information on the sources of sediment yield within a catchment can be used as a perspective on the rate of soil erosion occurring within that catchment. Estimation of sediment yield from a catchment is important for many reasons. Not only does deposition of sediment transported by a river into a reservoir reduce the reservoir capacity, but also sediment deposition on river beds and banks causes widening of flood plains during floods. Deforestation, urbanization and agricultural intensification are the major factors which influence the rate of erosion and sedimentation. Since it is not possible to monitor the influence of every land-use practice in all ecosystems under all weather conditions, erosion predictions are used Open for discussion until 1 June 29

3 1254 Rabin Bhattarai & Dushmanta Dutta to rank alternative practices with regard to their likely impact on erosion. Assessment of soil erosion as to how fast soil is being eroded is helpful in planning conservation work. Modelling tools can provide a quantitative and consistent approach to estimate soil erosion and sediment yield. Models available in the literature for sediment yield estimation can be grouped into two categories: (a) physically-oriented models; and (b) empirical models. Physically-based models are intended to represent the essential mechanisms controlling the erosion process by solving the corresponding equations. These models are the synthesis of individual components that affect the erosion process, and they are arguably very capable of assessing both the spatial and temporal variability of the natural erosion processes. Examples for physically-based models include the Areal Nonpoint Source Watershed Environment Response Simulation (ANSWERS, Beasley et al., 198), the Water Erosion Prediction Project (WEPP, Nearing et al., 1989), the Kinematic Runoff and Erosion Model (KINEROS, Woolhiser et al., 199) and the European Soil Erosion Model (EUROSEM, Morgan et al., 1998). Although physically-based erosion models emulate the real processes, they suffer from the major drawback of needing many parameters related to each process. This hinders the application of process-based models in many areas that lack the data sets required for the model simulation. Simple empirical methods, such as the Universal Soil Loss Equation (USLE) (Musgrave, 1947; Wischmeier & Smith, 1965), the Modified Universal Soil Loss Equation (MUSLE) (Williams, 1975), and the Revised Universal Soil Loss Equation (RUSLE) (Renard et al, 1991), are frequently used for the estimation of surface erosion and sediment yield from catchment areas (Ferro & Minacapilli, 1995; Ferro, 1997; Kothyari & Jain, 1997; Sadeghi & Mizuyama, 27) because of their simple structure and ease of application. The Erosion Productivity Impact Calculator (EPIC Williams et al., 1984) and the Agricultural Nonpoint Source Pollution model (AGNPS, Young et al., 1987) are examples of watershed models based on USLE methodology and commonly used to compute soil erosion. Although the USLE/RUSLE methodology is based on parameters computed or calibrated on the basis of observations, it has been extensively applied all over the world mainly due to the simplicity in the model formulation and easily available input data sets. Although there has been much research on the performance evaluation of individual models (Montas & Madramootoo, 1991; Chung et al., 1999; Cochrane & Flanagan, 1999; Folly et al., 1999; Nugroho, 23), few comparative studies have been done on the models available for soil erosion and sediment yield estimation. The WEPP predictions were found to be better than those from EPIC and ANSWERS for soil loss quantification for different tillage systems (Bhuyan et al., 22). Similarly, Wolkerstorfer & Strauss (24) compared the performances of MUSLE with the Morgan-Morgan-Finney (MMF) model in terms of absolute value of soil loss, and observed that both models overestimated sediment delivery to the river. Amore et al. (24) applied the USLE and WEPP for sediment deposition computation in reservoirs in three small watersheds in Italy and observed that WEPP estimated sediment volume better than the USLE. An application of RUSLE, EPIC and WEPP models in south-central Chile, a region having a Mediterranean type climate and volcanic soils with distinctive properties compared to non-volcanic soils, showed that WEPP permitted the most adequate parameterization for both climate and volcanic soil, and calculated erosion rates that were in overall agreement with field and USLE values (Stolpe, 25). The models were compared for their consistency with historical rates of erosion and those calculated using the previously calibrated USLE model. Due to the spatial variation in rainfall and catchment heterogeneity, both soil erosion and sediment transport processes are spatially varied. Such variability has promoted the use of data-intensive distributed models for the estimation of catchment erosion and sediment yield by discretizing a catchment into sub-areas each having approximately homogeneous characteristics and uniform rainfall distribution (Young et al., 1987). To encapsulate the spatial variation of the parameters such as topography, soil and land use, etc. in a watershed, the use of a geographical information system (GIS) is well suited. A GIS can be used for the discretization of the catchments into small grid cells, and for the computation of such physical characteristics of these cells as slope, land use and soil type, all of which affect the processes of soil erosion and deposition in the different sub-areas of a catchment.

4 A comparative analysis of sediment yield simulation 1255 The aim of this study was to evaluate soil erosion and sediment yield estimation by two different approaches one empirical and the other a physically-based distributed approach and compare their performances. For the empirical approach, the revised form of the USLE model, RUSLE, was used in conjunction with SEDD model, to predict erosion potential on a cell-by-cell basis and to determine the catchment sediment yield by using the concept of sediment delivery ratio (Ferro & Porto, 2). Additionally, a modified form of the USLE model, MUSLE, was used to compute sediment yield at the basin outlet. The MUSLE is different from the RUSLE in the way that it uses a runoff factor instead of the rainfall erosivity factor used by the latter. The selected process-oriented soil erosion and sediment transport model was developed at the University of Tokyo and contains an overland flow simulation model coupled with a sediment transport model (Mughal, 21). It was applied for the same study areas and, finally, performances of the empirical and process-oriented models were evaluated. Since observations of soil erosion data were not available, the model performances were accessed mainly based on the temporal distribution of sediment discharge. In the case of the process-oriented model, the model outcome was also compared with observed water discharge data. METHODOLOGY Sediment delivery distributed (SEDD) model The SEDD model couples the USLE with a spatial disaggregation criterion of sediment delivery processes. Empirical methods such as the USLE have been found to produce realistic estimates of surface erosion over areas of small size (Wischmeier & Smith, 1978). The USLE is expressed as: A = R K L S C P (1) where A is the average annual soil loss predicted (t ha -1 ), R is the rainfall runoff erosivity factor (MJ mm ha -1 h -1 ), K is the soil erodibility factor (t ha h MJ -1 ha -1 mm -1 ), L is the slope length factor, S is the slope steepness factor, C is the cover management factor and P is the support practice factor (L, S, C and P are dimensionless). The values of USLE factors are computed using the following methods, as described in the Agricultural Handbook 73 (Renard et al., 1996): 1 R = n n m i= 1 j = 1 E j I ( ) 3 j where n is the total number of years, m is the total number of rainfall storms in the ith year, I 3 is the maximum 3-min intensity (mm h -1 ), E j is the total kinetic energy (MJ ha -1 ) of the jth storm of the ith year and is given as: E j = p i= 1 e d k k where p is the total number of intervals of the jth storm of the ith year, d k is the rainfall depth of the kth interval of the storm (mm), e k is the unit kinetic energy (MJ ha -1 mm -1 ) of the kth interval of the storm and is given as (Renard et al., 1996): e k.5 i k =.29(1.72e ) (4) where i k is the intensity of rainfall of the kth interval of the storm (mm h -1 ). If λ is the horizontal projection of the slope length, then the L factor is given as: λ L = 22.1 m where λ is the contributing slope length (m) and m is the variable slope-length exponent. The (2) (3) (5)

5 1256 Rabin Bhattarai & Dushmanta Dutta slope-length exponent m is related to the ratio β of rill erosion (caused by flow) to inter-rill erosion (principally caused by raindrop impact) by the following equation: m = β/(1 + β) (6) For moderately susceptible soil in both rill and inter-rill erosion, McCool et al. (1989) suggested the equation: (sin θ /.896) β = (7).8 3.(sin θ) +.56 where θ is the slope angle in degrees. The slope steepness factor S is evaluated from the following equations (McCool et al., 1987): S = 1.8 sin θ +.3 for s < 9% (8) S = 16.8 sin θ.5 for s 9% where s is the slope in percentage. The factors C and P are assigned to different grid cells according to land cover, while the K factor is estimated using the soil data. In a catchment, part of the soil eroded in an overland region deposits within the catchment before reaching its outlet. The values of the ratio of sediment yield to total surface erosion, which is termed the sediment delivery ratio (D R ), for an area, are affected by catchment physiography, sediment sources, transport system, texture of eroded material, land cover etc. (Walling, 1983, 1988). However, variables such as catchment area, land slope and land cover have been mainly used as parameters in empirical equations for D R (Hadley et al., 1985; Williams & Berndt, 1972; Kothyari & Jain, 1997). Ferro & Minacapilli (1995) and Ferro (1997) hypothesized that D R in grid cells is a strong function of the travel time of overland flow within the cell. The travel time is strongly dependent on the topographic and land cover characteristics of an area and therefore its relationship with D R is justified. Based on their studies on probability distribution of travel time, the following relationship was assumed herein for a grid cell lying in an overland region of a catchment: D = exp( γt ) (9) R i where t i is the travel time (h) of overland flow from the ith overland grid to the nearest channel grid down the drainage path, and γ is a coefficient considered as constant for a given catchment. The travel time for grids located in a flow path to the nearest channel can be estimated if the lengths and velocities for the flow paths are known. In grid-based GIS analysis, the direction of flow from one cell to a neighbouring cell is often ascertained by using an eight-direction pourpoint algorithm. Once the pour-point algorithm identifies the flow direction in each cell, a cell-tocell flow path is determined to the nearest stream channel and thus to the catchment outlet. If the flow path from cell i to the nearest channel cell traverses m cells and the flow length of the ith cell is l i (which can be equal to the length of a square side or to a diagonal depending on the direction of flow in the ith cell), and the velocity of flow in cell i is v i, the travel time t i from cell i to the nearest channel can be estimated by summing up the time through each of the m cells located in that flow path: t i = m i= 1 li v i For the present study, the method of determination of the overland flow velocity proposed by the US Soil Conservation Service was chosen, due to its simplicity and the availability of the information required (SCS, 1975). The flow velocity (v i ) is considered to be a function of the land surface slope and the land cover characteristics: i i b i v = a S (11) (1)

6 A comparative analysis of sediment yield simulation 1257 Data sets required Spatial Temporal Digital Elevation Model (DEM) Soil Land cover/use Rainfall Sediment Preparation of USLE parameter (R, K, L, S, C, P) layer on grid basis Erosion estimation (A = R K L S C P) Sediment yield computation using erosion and sediment delivery layers Comparison with observations Fig. 1 Flow chart showing the processes involved in the USLE-based SEDD model to compute soil erosion and sediment yield. where b is a numerical constant equal to.5 (SCS, 1975; Ferro & Minacapilli, 1995), S i is the slope of the ith cell and a i is a coefficient related to land use (Haan et al. 1994). Introducing equations (1) and (11) into equation (9) gives: m l = i DR exp γ (12).5 i= 1 aisi It should be noted that l i /S i.5 is the value of travel time used by Ferro & Minacapilli (1995). Values of the coefficient a i for different land uses were adopted from Haan (1994). If S E is the amount of soil erosion produced within the ith cell of the catchment estimated using equation (1), then the sediment yield for the catchment, S y, is obtained as follows: S y = n i= 1 D R S E where n is the total number of cells over the catchment and the term D R is the fraction of S E that ultimately reaches the nearest channel. Since the D R of a cell is hypothesized as a function of travel time to the nearest channel, it implies that the gross erosion in that cell multiplied by the D R value of the cell becomes the sediment yield contribution of that cell to the nearest stream channel. The D R values for the cells marked as channel cells are assumed to be unity. Figure 1 shows the steps involved in the USLE-based model to compute soil erosion and sediment yield. Modified Universal Soil Loss Equation (MUSLE) One version of the USLE, which is widely used to estimate sediment yield, is the Modified USLE (MUSLE) (Williams & Berndt, 1977). The MUSLE is intended to estimate sediment yield for a single event. The model was developed from the data collected from 778 individual storm events in 26 catchments with areas ranging from 15 to 15 ha in Texas. The best fit equation for the data is expressed as: (13)

7 1258 Rabin Bhattarai & Dushmanta Dutta Y = 11.8(Q q p ).56 K C P L S (14) where Y is the sediment yield from an individual storm (t), Q is the storm runoff volume (m 3 ), q p is the peak runoff rate (m 3 s -1 ), K is the soil-erodibility factor (t h t -1 m -1 cm -1 ), L is the slope length factor, S is the slope gradient factor, C is the crop management factor, and P is the erosion-controlpractice factor. Peak runoff (q p ) is computed using the relationship: q p =.288 A (R/T) (Q t /R.2S t ) (15) where A is the watershed area (ha), R is the rainfall depth (cm), T is the rainfall duration (h), Q t is the runoff depth (cm), and S t is the soil retention parameter (cm). Equations (14) and (15) are not dimensionally consistent and special care should be given to the units while using these relationships. Runoff depth (Q t ) can be computed using the relationship: Q t = (R.2S t ) 2 /(R +.2S t ) (16) where S t = 254/CN 25.4 (cm), CN being the curve number. Process-oriented distributed model The process-oriented distributed model was developed using the physically-based governing equations of overland flow and soil erosion and sediment transport mechanisms (Mughal, 21). The overland flow simulation model is coupled with a soil erosion and sediment transport model for grid-based simulation. The one-dimensional (1-D) form of the Saint Venant continuity and momentum equations is used for overflow routing. The momentum equation is used with a kinematic wave approximation. The continuity equation is represented as: Q A + = (17) x t and is applied between the centre points of the two consecutive grids. Similarly, the kinematic wave approximation of the momentum equation can be represented as (Chow et al., 1988): S f = S o (18) where t is time, x is the distance along the longitudinal axis of the water course, A is the crosssectional area, Q is discharge through A, S f is the friction slope and S o the bed slope. The soil erosion and sediment transport is modelled as the detachment of soil by raindrop impact, leaf drip impact, detachment by overland flow over the entire grid and 1-D transport or routing of the eroded material by overland flow on the regular square grid discretized system. Detachment due to raindrop impact process is modelled based on relationships between detachment and kinetic energy of the rainfall due to both direct throughfall and leaf drip impact as a function of their kinetic energies. This enables the effects of different heights of vegetation and canopy and residue to be simulated explicitly. The rainfall energy reaching the ground surface as direct throughfall (KE(DT)) is assumed the same as that of the natural rainfall. It is estimated as a function of rainfall intensity from an equation derived by Brandt (1989). KE(DT) = log I (19) where KE(DT) is the kinetic energy of direct throughfall (J m -2 mm -1 ) and I is rainfall intensity (mm h -1 ). The energy of leaf drainage is estimated from the following relationship developed experimentally by Brandt (199): KE(LD) = 15.8(PH) (2) where KE(LD) is kinetic energy due to leaf drip (J m -2 mm -1 ) and PH is the effective height of the plant canopy (m).

8 A comparative analysis of sediment yield simulation 1259 The total kinetic energy of the rainfall can be calculated by multiplying energies obtained from equations (19) and (2) by their respective depths of direct throughfall and leaf drainage received and summing the two values: KE = (1 C c ) KE(DT) H DT + C c KE(LD) H LD (21) where KE is the total kinetic energy of the rainfall (J m -2 ), C c is the canopy cover in the model square grid, H DT is the depth of direct throughfall (total rain in mm), and H LD is the depth of leaf drips (net rain in mm). Detachment due to rainfall impact is estimated for each time step using the following equation (Torri, 1987), which relates the detachment due to raindrop impact with the total kinetic energy of the rainfall: D R = (1 C g )k KE e -z H (22) where D R is the soil detachment by raindrop impact (g m -2 ), k is an index of the detachability of the soil (g J -1 ) and depends on the soil texture (Morgan, 1995), KE is the total kinetic energy of the rain (J m -2 ), z is an exponent, for which a working value of 2. is representative of a range of values between.9 and 3.1, H is the depth of the surface water layer (mm) and C g is the proportion of ground cover in each processing cell or flow element to consider the non-erodible surfaces, such as rock outcrops, surface rock fragments, thick grass and surface vegetation less than.5 m height, concrete and tarmac, occurring within the flow element. For modelling soil detachment due to overland flow, the following equation (Ariathurai & Arulanandan, 1978) is used: D F = K f (Τ/T c 1) for Τ > T c (23) D F = for Τ T c (24) where D F is the overland flow detachment (kg m -2 s -2 ), K f is the overland flow detachability coefficient (kg m -2 s -1 ) and can be determined experimentally, T c is the critical shear stress for initiation of motion from the Shield s curve and T is the hydraulic shear stress (N m -2 ) as given by: T = γ h S (25) where γ is the specific weight of water (N m -3 ), h is the depth of overland flow (m) and S is the slope of the ground surface. The term K f is best regarded as a calibration coefficient, to be determined by fitting the simulated variation of sediment discharge to be measured. Total potential detachment at any cell (x) and time (t), e(x,t), is then calculated as the sum of splash and flow detachment as follows: e(x,t) = D R (x,t) + D F (x,t) (26) Transportability of the detached material depends on the amount of the detached material and the remaining transport capacity of the flow (transport capacity minus existing sediment discharge from upstream). When transport capacity of the flow is greater than the sediment load, the actually detached load (erosion) is estimated as described in equation (26). If the transport capacity of the flow in that particular cell at time t will be less than the sediment load, then excess material will drop as deposition and the actually detached load will be zero from that cell at that time step, and the load carried by the flow will be equivalent to the transport capacity. For 1-D forward sediment transport routing, the kinematic mass balance equation can be applied between the centres of two consecutive grids considering the flow direction. Total detachments are calculated as the sum of the splash detachment and detachment due to overland flow. After considering the transport capacity of the flow, the total actual detached load is determined, and this load is considered as the lateral sediment flow and is added at the inlet of the control volume. ( AC) ( QC) + = (27) t x

9 126 Rabin Bhattarai & Dushmanta Dutta where C is sediment concentration, A is cross-sectional area of flow and Q is discharge or volume flow rate. Since there is only one unknown in the sediment mass balance equation, that is sediment concentration at any time and space, the above equation can be rewritten in terms of sediment discharge as: ( Qs / V ) ( Qs ) + = (28) t x where V is the mean velocity of flow and Q s is sediment discharge. Using a finite difference approach, sediment discharge Q s can be obtained since other parameters in the equation are known. The detailed algorithm for the process-oriented soil erosion and sediment transport model is shown in Fig. 2. Splash detachment Flow detachment Upstream sediment discharge = sedq Potential detachment =ee Total sediment load = (ee+sedq) Compare Transport capacity = Tc No IF Tc > sedq Yes Actual detachment = erosion = deposition = (sedq Tc) Actual detachment = Erosion = (Tc sedq) ee Route actually detached load to the forward grid Fig. 2 Flow diagram describing the algorithm for soil erosion and sediment transport routing in the process-oriented model. STUDY AREA The Mun River basin lies between N and E (Fig. 3). The Mun River is the largest right-bank tributary of the Mekong River, situated in the northeastern part of Thailand. The Chi River joins the Mun River at about 1 km upstream of the confluence with the Mekong River. Chi-Mun basin covers 15% of the Mekong basin area and the discharge contribution of the basin is 6.1% in the dry season and 4.7% in the rainy season. The total draining area of Mun basin is approx. 69 km 2. In an average year, the contribution of Chi-Mun to the Mekong is approx.

10 A comparative analysis of sediment yield simulation 1261 Fig. 3 Study area (M91 sub-basin). 25 hm 3 (million m 3 ), which is equivalent to an annual runoff of 21 mm or 8 m 3 s -1. Roughly two thirds of this comes from the Mun River. The average annual rainfall in the basin is 12 mm, varying between 16 mm in the east and 1 mm in the western part of the basin. It covers five provinces (Nakhon Rathchasima, Buri Ram, Surin, Sisaket and Ubon Ratchathani) entirely and three (Maha Sarakham, Rio Et and Yasothom) are partly within the basin. Between 199 and 1995, the average deforestation rate in the Lower Mekong basin was 1.6% per year one of the highest rates in the world. The erosion in the basin is mainly rainfall based runoff erosion subject to the effects of land use (MRC, 23). Chi-Mun basin contains more than 2 dams and, the deposition of sediment transported by river into the reservoirs is reducing the capacity of the reservoirs. The average annual loading of suspended sediment during the 199s at the Chi-Mun/Mekong Junction was t year -1 (Al-Soufi, 24). Based on the locations of the flow and sediment gauging stations, several upstream sub-watersheds of the Mun River basin

11 1262 Rabin Bhattarai & Dushmanta Dutta were identified for modelling. Due to the similarity of size, land cover and hydrogeological characteristics of these watersheds and modelling outcomes, the modelling outcomes in the M91 sub-watershed are presented in this paper. The size of the M91 sub-watershed is 128 km 2 with an average annual sediment yield of t for the period Its outlet is the sediment gauging station M91, which is not affected by the reservoir located downstream. Monthly average sediment yield for the gauging station was obtained from weekly depth-integrated suspended sediment sampling. The elevation in the sub-watershed varies from 183 to 483 m a.m.s.l., with an average slope of 3.9%. Agricultural land is the major land use which covers 62% of the subwatershed, while forest covers the remaining 38%. Sandy loam soil covers 93% of the subwatershed area, while the remaining 7% is covered by silty clay loam soil. DATA PREPARATION AND SIMULATION The rate of soil erosion from an area is strongly dependent upon its soil, vegetation and topographic characteristics, besides rainfall and runoff. These factors are found to vary greatly within the various sub-areas of a catchment. Therefore, the catchment needs to be discretized into smaller homogeneous units before making computations for soil loss. A grid-based discretization is found to be the most reasonable procedure in both process-oriented models as well as in other simple models (Beven, 1996; Kothyari & Jain, 1997). For modelling, 3-hourly rainfall data for the period were obtained from the Thailand Meteorological Department. The R value for the RUSLE model was computed using equations (2), (3) and (4) by evenly distributing each 3-hourly rainfall event into 3 min intervals. The long-term annual averaged R value for the Tha Thum station was computed to be 968 MJ mm ha -1 h -1. Topographical parameters (L, S) were extracted from a 9-m resolution DEM obtained from NASA ( Equation (5) was used for the L factor calculation, while the S factor was computed using equation (8) for each cell. While computing the L factor, the contributing slope length, λ, was set to a fixed value of 9 m when the flow was in the cardinal direction (flow direction values 1, 4, 16 and 64), and m for flow in the diagonal direction (flow direction values 2, 8, 32 and 128), for a grid size of 9 m for the entire basin and modified accordingly for 3-m grid size. The values for the factors K, C and P were estimated for different grids using the soil and land cover data. The spatial data of land-use and soil characteristics were obtained from the digital database (CD-ROM Thailand on a disc ) provided by the Department of Land Development, at the scale of 1:25. The K values were assigned on the basis of soil texture (Schwab et al., 1981) and are presented in Table 1.The C value, which depends on land use, was derived from several in the literature (Schwab et al., 1981; Morgan, 1995) and is shown in Table 2. The value of the P factor was taken as.5 for agricultural land where soil conservation practices such as contour farming were applied, and 1. for rest of the land-use classes, where farmers did not apply any soil conservation practices (Schwab et al. 1981). The a i values used to compute SDR for different land-use classes are presented in Table 2. Table 1 Soil erodibility factor (K) by soil texture in t ha h MJ -1 ha -1 mm -1. Textural class Organic matter content (%) Sandy loam (Group A) Silty clay loam( Group D) Table 2 Cover management factor (C), curve number (CN) and a i value on the basis of land-use type. Land use C value basis C value CN a i Cultivated land Crops, disturbed land.4 62 (Group A) 81 (Group D) 1.55 Forest land Forest.2 3 (Group A) 77 (Group D).76

12 A comparative analysis of sediment yield simulation 1263 In the MUSLE case, computation was carried out on an event basis. The sediment yield was computed for each event and monthly yield was computed taking the sum of yield generated by each event in that month. The K, C, P, L and S values were assigned using the same procedure that was used for RUSLE simulation. The average value of each of these factors was taken while computing the sediment yield in one storm event. The CN value was assigned on the basis of soil and land-use types. In the case of the process-oriented model, it is necessary to calibrate and verify the model for water discharge before applying it to the sediment yield comparison. The model was calibrated for monthly mean discharge at M91 by varying the runoff coefficient. Daily discharge data for for M91 sub-watershed were obtained from the Royal Irrigation Department, Thailand. The land-use parameters used during the calibration and verification are presented in Table 3. Soil water properties in the study area were obtained from the study of Department of Soil Science, Table 3 Different land-use parameters (source: Mugal, 21). Land-use type Manning s roughness coefficient, n Canopy cover (frac.) Canopy height (m) Ground cover (frac.) Cultivated land Forest Leaf area index (a) Discharge (m 3 s -1 ) Rainf all obs sim Rainfall (mm) Jun-9 Jul-9 Aug-9 Sep-9 Oct-9 Nov-9 25 (b) Discharge (m 3 s -1 ) Rainf all Obs sim 4 6 Rainfall (mm) Jun-91 Jul-91 Aug-91 Sep-91 Oct-91 Nov-91 Time (months) Fig. 4 Process-oriented model: (a) calibration (June November 199) and (b) verification (June November 1991) for water discharge 1

13 1264 Rabin Bhattarai & Dushmanta Dutta Kasetsart University, Thailand (Suntaree, 1993). The model calibration was performed for the period June November 199 and verification was done for the same period in Since the model computes only the surface component of the total river flow, the baseflow was separated from total river discharge before model calibration and validation for water discharge. The results obtained from the model calibration and verification were compared with the mean observed discharge and the comparisons are shown in Fig. 4(a) and (b), respectively. Discharge is generally overestimated by the model during the model calibration and verification for water discharge, except for June 199. The overestimation of discharge may result because of the higher value of runoff coefficient (.8) considered during model simulation as the model lacks sub-surface flow components. RESULTS AND DISCUSSION The sediment contribution of each grid cell to the outlet was computed with the help of an erosion potential map and a SDR map. The simulated sediment yield at the outlet was compared with the measured field data obtained from the Royal Irrigation Department, Thailand. The simulation was carried out for two DEM resolutions: 9 and 3 m (re-sampled from 9 m). Using the ArcInfo program, the slope for a cell is calculated from the 3 3 neighbourhood using the average maximum technique. The technique is effective in preserving topographical variation while resampling a DEM into finer resolution to some extent. The values of average annual sediment yield at the catchment outlet computed by the RUSLE method are presented in Table 4, together with the observed data. In the case of 3-m resolution, the simulated yield is closer to the observed than the value obtained using 9-m DEM resolution. Table 5 shows the effect of DEM resolution on different RUSLE parameters and SDR values; it can be seen from this table that the L and S factors vary significantly for the two DEMs of different resolution. Changes in grid size affect the slope values and ultimately affect the values of L and S. The L factor is dependent on grid size and slope, whereas the S factor depends on slope alone. The time series of computed and observed sediment yields at the monthly scale are shown in Fig. 5. Improved results were obtained for the DEM resolution of 3 m compared to those obtained from 9-m resolution. From the results and analysis, it is found that the RUSLE-based SEDD model greatly overestimated sediment yield and the simulated results are strongly influenced by the resolution of the DEM. The model prediction may have been improved if the γ coefficient had been calibrated using the measured sediment yield values at mean annual scale for SDR computation. During the SDR calculation, the sensitivity analysis of the parameter γ showed that the computed S y was not very sensitive to γ in equation (12). The variation of γ by a value within the range changed the S y value by only 1%. Since large variation in γ affected S y insignificantly during sensitivity analysis, the γ value was taken as 1 in the computation, for simplicity. The sensitivity analysis supports the findings of Jain & Kothyari (2) who reported that S y was not very sensitive to γ. Figure 6 shows the results of simulation by the MUSLE method for different CN values, together with the observations. It may be seen that the MUSLE computation predicted a higher Table 4 Computed and observed value of annual average sediment yield using RUSLE. Station Observed sediment yield Computed sediment yield (t km -2 ): % error: (t km -2 ) 9-m DEM 3-m DEM 9-m DEM 3-m DEM M Table 5 DEM effect on USLE parameters and SDR in RUSLE. DEM resolution Range of L factor Range of S factor Range of SDR 9 m m

14 A comparative analysis of sediment yield simulation 1265 Sediment yield (t) Rainf all 9m DEM 3m DEM Obs Rainfall (mm/month) Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec Time (year 199) Fig. 5 Time series of observed and simulated yield by the RUSLE-based model. 4 9 Sediment yield (t) Rainf all Obs CN =69.16 CN = 65 CN = Rainfall (mm/month) 1 4 Jan Feb M ar Apr M ay Jun Jul Aug Sept Oct Nov Dec Time (year 199) Fig. 6 Time series of observed and simulated yield by the MUSLE-based model. value than the observed monthly sediment yield. The curve number (CN) is a very important parameter in sediment yield computation in this approach. A higher value of CN produces more runoff and, subsequently, results in greater sediment yield compared to a lower CN value. A high value of peak discharge was observed during the computation, which may be the result of the large watershed area. In several other past studies, the MUSLE model was found to be applicable for small watersheds of area up to 15 ha. In this study, it was applied to a watershed of 128 km 2 area. The outcomes of this study agree with this general notion that the use of the MUSLE method should be limited to small watersheds. The MUSLE model prediction may have been improved by using lower CN values, but these can only be varied in a small range for given land use and soil type. Process-oriented model outputs for the period June November 199 are shown in Fig. 7(a), together with the observed monthly sediment yields. As may be seen from Fig. 7(a), the simulated results show better agreement with the observed monthly sediment yield at the catchment outlet. The different soil parameters used in the model simulation are shown in Table 6, and the error statistics of the model in Table 7. The Nash-Sutcliffe coefficient, or efficiency index (EI) value of.78 and R 2 value of.92 show that the model results have high correlation with the observed values. Sediment yield was also simulated for the six months from June to November 1991, shown in

15 1266 (a) 4 35 (b) Sediment yield (t) Sediment yield (t) Rabin Bhattarai & Dushmanta Dutta Rainfall Obs Sim Jun-9 Jul-9 Aug-9 Sep-9 Oct-9 Nov-9 Rainfall Obs Sim Rainfall (mm/month) Rainfall (mm/momth) Jun-91 Jul-91 Aug-91 Sep-91 Oct-91 Nov-91 Time (months) Fig. 7 Comparison of observed and simulated sediment yield by process-oriented model: (a) June November 199; and (b) June November Table 6 Soil parameters used in process-oriented model simulation. Soil texture Soil detachability index, k (g J -1 ) Overland flow detachability coefficient, K f (kg m -2 s -1 ) Density of particle (kg m -3 ) Sandy clay loam Loamy sand Median particle diameter (μm) Table 7 Error statistics of process-oriented model simulation. Year Efficiency index (EI) Root mean square error (RMSE) Mean absolute error (MAE) Mean percent error (MPE) R 2 Table 8 Comparison of performance of the three models. Date Simulated sediment yield (t) Observed % Error RUSLE MUSLE Process-oriented RUSLE MUSLE Process-oriented June July August September October November

16 A comparative analysis of sediment yield simulation 1267 (Fig. 7(b)) together with the observed data at a monthly time scale. These simulated results also agree well with the observed monthly sediment yield at the catchment outlet. In this simulation, EI was.93 and R 2 =.93, showing that the model results are highly correlated with the observed values. A comparison of the simulated results obtained from RUSLE/SEDD, MUSLE and processoriented model is presented in Table 8, together with the observed values. The results reflect that the RUSLE/SEDD computed values were higher than the observations from the period August October 199. In the case of the MUSLE-based model, the computed sediment yield values are lower than the observations in the low rainfall months and higher in the case of high rainfall months, such as September and October 199. Unlike the RUSLE and MUSLE based approach, the process- oriented model results showed good agreement with the observations for the same period. One of the reasons behind the process-oriented model outperforming empirical models is that the process-oriented model was calibrated before application, while the empirical models were not. The empirical models may have performed better if proper calibration had been carried out. However, the calibration process for the empirical models may prove more rigorous compared to a process-orientated model, since most empirical models are data-intensive and require a longer duration of observations. CONCLUSIONS This study was an attempt to estimate soil erosion and sediment yield in a river basin using empirical (RUSLE/SEDD and MUSLE) and process-oriented approaches in a distributed manner, and then compare their performances. The empirical models did not perform well and the outcomes were influenced by the DEM resolution in the case of the RUSLE-based SEDD model. The error between the computed and observed annual average sediment yields was found to be 411% in the case of 9-m DEM resolution. After resampling the 9-m DEM into 3-m resolution, the computed error was reduced by almost 45%. The improvement was due to the effect of DEM resolution on L, S and SDR factors. The variation in the result may be due to certain assumptions made during the analysis, such as computation of soil erodibility value on the basis of soil texture and use of constant instead of time varying C values. In the MUSLE-based analysis, the observed error was very high. We know that the formulation of the MUSLE model was for small watersheds (up to 3 ha), and it was applied here in a sub-basin of 128 km 2. The result reflects the fact that the use of the MUSLE method is limited to small watersheds. In time series computation, the performance of the process-oriented model was better than that of the empirical (RUSLE/SEDD and MUSLE) models. From June to October 199 (peak sediment discharge period), the error between results simulated by the process-oriented model and observations was within 7%. Although there are many input parameters for the process-oriented model, it mimics the processes of detachment, erosion and transportation of sediment and hence produces better results than the empirical approach. Though empirical (RUSLE/SEDD and MUSLE) models are economical in terms of computational resources and data requirement, compared to the process-oriented model, their application for the temporal analysis of sediment transport is found to be less useful. Another reason for the process-oriented model outperforming the empirical models is that the processoriented model was calibrated before application, while the empirical models were not. Empirical models may have performed better if proper calibration had been carried out. REFERENCES Al-Soufi, R. (24) Soil erosion and sediment transport in the Mekong basin. Proc. Second APHW Conference (Singapore), Amore, E., Modica, C., Nearing, M. A. & Santoro, V. C. (24) Scale effect in USLE and WEPP application for soil erosion computation from three Sicilian basins. J. Hydrol. 293, Ariathurai, R. & Arulanandan, A. D. (1978) Erosion rates of cohesive soils. J. Hydraul. Div. ASCE 14,

17 1268 Rabin Bhattarai & Dushmanta Dutta Beasley, D. B., Huggins, L. F. & Monke, E. J. (198) ANSWERS: a model for watershed planning. Trans. ASAE 23(4), Beven, K. J. (1996) A discussion of distributed modeling. In: Distributed Hydrological Modelling (ed. by M. B. Abbott & J. C. Refsgaard), Kluwer, Dordrecht, The Netherlands. Bhuyan, S. J., Kalita, P. K., Janssen, Keith A. &. Barnes, P. L. (22) Soil loss predictions with three erosion simulation models. Environ. Modell. Software 17, Brandt, C. J. (1989) The size distribution of throughfall drops under vegetation canopies. Catena 16, Brandt, C. J. (199) Simulation of size distribution and erosivity of raindrops and throughfall drops. Earth Surf. Processes and Landf. 15, Chow, V. T., Maidment, D. R. & Mays, L. W. (1988) Applied Hydrology. McGraw-Hill, New York, USA. Chung, S. W., Gassman, P. W., Kramer, L. A., Williams, J. R. & Gu, R. (1999) Validation of EPIC for two watersheds in southwest Iowa. J. Environ. Qual. 28, Cochrane, T. A. & Flanagan, D. C. (1999) Assessing water erosion in small watersheds using WEPP with GIS and digital elevation models. J. Soil Water Conserv. 54(4), Ferro, V. (1997) Further remarks on a distributed approach to sediment delivery. Hydrol. Sci. J. 42(5), Ferro, V. & Minacapilli, M. (1995) Sediment delivery processes at basin scale. Hydrol. Sci. J. 4(6), Ferro, V. & Porto, P. (2) Sediment delivery distributed (SEDD) model. J. Hydrol. Engng ASCE 5(4), Folly, A., Quinton, J. N. & Smith, R. E. (1999) Evaluation of the EUROSEM model for the Catsop watershed, the Netherlands. Catena 37, Haan, C. T., Barfield, B. J. & Hayes, J. C. (1994) Design Hydrology and Sedimentology for Small Catchments. Academic Press, New York, USA. Hadley, R. F., Lal, R., Onstad, C. A., Walling, D. E. & Yair, A. (1985) Recent Developments in Erosion and Sediment Yield Study. UNESCO (IHP) Publication, Paris, France. Jain, M. & Kothyari, U. C. (2). Estimation of soil erosion and sediment yield using GIS. Hydrol. Sci. J. 45(5), Kothyari, U. C. & Jain, S. K. (1997) Sediment yield estimation using GIS. Hydrol. Sci. J. 42(6), McCool, D. K., Brown, L. C., Foster, G. R., Mutchler, C. K. & Meyer, D. (1987) Revised slope steepness factor for the universal soil loss equation. Trans. ASAE 3, McCool, D. K., Foster, G. R., Mutchler, C. K. & Meyer, L. D. (1989) Revised slope length factor for the Universal Soil Loss Equation. Trans. ASAE 32, Montas, H. J. & Madramootoo, C. A. (1991) Using ANSWERS model to predict runoff and soil loss in soutwestern Quebec. Trans. ASAE 34(4), Morgan, R. P. C. (1995) Soil Erosion and Conservation. Longman, UK. Morgan, R. P. C, Quinton, J. N., Smith, R. E., Govers, G., Poesen, J. W. A., Auerswald, K., Chisci, G., Torri, D. & Styczen, M. E. (1998) The European soil erosion model (EUROSEM): a process-based approach for predicting sediment transport from fields and small catchments. Earth Surf. Processes Landf. 23, MRC (Mekong River Commission) (23) State of the basin report. Mekong River Commission, Phnom Penh, Cambodia. Mughal, H.-U.-R. (21) Regional scale soil erosion and sediment transport modeling. PhD Thesis, The University of Tokyo, Tokyo, Japan. Musgrave, G. W. (1947) The quantitative evaluation of factors in water erosion a first approximation. J. Soil Water Conserv. 2(3), , 17. Nearing, M. A., Foster, G. R., Lane, L. J. & Flinkener, S. C. (1989) A process based soil erosion model for USDA water erosion prediction project technology. Trans. ASCE 32(5), Nugroho, S. P. (23) Application of the Agricultural Non-Point Source Pollution (AGNPS) model for sediment yield and nutrient loss prediction in the Dumpul sub-watershed, Central Java, Indonesia. In: Erosion Prediction in Ungauged Basins (PUBs): Integrating Methods and Techniques (ed. by D. H. de Boer, W. Froehlich, T. Mizuyama & A. Pietroniro), IAHS Publ IAHS Press, Wallingford, UK. Renard, K. G., Foster, G. R., Weesies, G. A. & Porter, J. P. (1991) RUSLE, Revised Universal Soil Loss Equation. J. Soil Water Conserv. 46(1), Renard, K. G., Foster, G. R., Weesies, G. A., McCool, D. K. & Yoder, D. C. (1996) Predicting soil erosion by water: a guide to conservation planning with the Revised Universal Soil Loss Equation. US Dept Agriculture, Agricultural Research Services, Agricultural Handbook 73. Sadeghi, S. H. R. & Mizuyama, T. (27) Applicability of the Modified Universal Soil Loss Equation for prediction of sediment yield in Khanmirza watershed, Iran. Hydrol. Sci. J. 52(5), Schwab, G. O., Frevert, R. K., Edminster, T. W. & Barnes, K. K. (1981) Soil and Water Conservation Engineering. John Wiley & Sons Inc., New York, USA. SCS (Soil Conservation Service) (1975) Urban hydrology for small watersheds. US Dept Agriculture Tech. Release no. 55. Stolpe, N. (25) A comparison of the RUSLE, EPIC and WEPP erosion models as calibrated to climate and soil of southcentral Chile. Acta Agricul. Scandinavica B 55(1), 2 8. Suntaree, Y. (1993) A Catalogue of Water Retention Functions of Major Soil Series of Thailand. Department of Soil Science, Kasetsart University, Thailand. Torri, D. (1987) Splash detachment: runoff depth and soil cohesion. Catena 14, Walling, D. E. (1983) The sediment delivery problem. J. Hydrol. 65, Walling, D. E. (1988) Erosion and sediment yield research some recent perspectives. J. Hydrol. 1, Williams, J. R. (1975) Sediment routing for agricultural watersheds. Water Resour. Bull. 11, Williams, J. R. & Berndt, H. D. (1972) Sediment yield computed with universal equation. J. Hydraul. Div. ASCE 98(HY12), Williams, J. R. & Berndt, H. D. (1977) Sediment yield prediction based on watershed hydrology. Trans. ASAE 2 (6), Williams, J. R., Jones, C. A. & Dyke, P. T. (1984) A modeling approach to determining the relationship between erosion and soil productivity. Trans. ASAE 27,