Preliminary modelling of sediment production and delivery in the Xihanshui River basin, Gansu, China

Catena 79 (2009) 277–287

Contents lists available at ScienceDirect

Catena j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / c a t e n a

Preliminary modelling of sediment production and delivery in the Xihanshui River basin, Gansu, China Jie Ding, Keith Richards ⁎ Department of Geography, University of Cambridge, Cambridge, CB2 3EN, UK

a r t i c l e

i n f o

Article history: Received 4 June 2008 Received in revised form 15 May 2009 Accepted 15 May 2009 Keywords: Sediment yield Sediment delivery Hillslope sediment supply River bank erosion Soil conservation Mass movement

a b s t r a c t This paper outlines an analysis of the spatial distribution of sediment production, delivery and yield in the Xihanshui River basin, South Gansu, China, using the modelling tools of SedNet (Prosser et al., 2001). This model can assess the delivery efficiency to downstream locations, as well as identifying locations with high rates of sediment production. Preliminary model experiments assist understanding of the spatial dynamics of these sediment processes and evaluation of the effectiveness of soil conservation practices since the mid1980s. Three scenario years (dry, average and wet) from the 1983–2005 record are identified and modelled, and land use and management are represented in the model to reflect known changes since the 1980s. Results show hillslope erosion to be a dominant source of sediment supply, causing the latter to decrease ten-fold between 1984 and 1997/2000. Estimated bank erosion and floodplain deposition rates are sensitive to parameter values, but bank erosion appears less sensitive than hillslope supply to rainfall. The model can be used to assess net changes in floodplain storage; for default parameters, floodplain deposition rates are 25–200 times the rates of bank erosion depending on the climate scenario. Comparing simulation results with measured sediment yields at the three gauging stations indicates encouraging agreement in 2000. In 1984 (the wet year), the model under-predicts, suggesting that additional unmodelled sediment production processes, especially mass movement and gully erosion, may be important in wet years. Mass movement inventory data could close the gap between the high yields measured in the wet scenario year and the estimated yield due to hillslope erosion alone. In 1997 (the dry year), the model over-predicts; this suggests that the land use change parameters required to reflect the effects of conservation may not have been sufficient, implying that conservation has been generally effective, and that evidence of declining sediment yield is not simply a reflection of drier conditions. © 2009 Elsevier B.V. All rights reserved.

1. Introduction The linkages of sediment production, routing, delivery, and downstream yield remain a significant source of theoretical and practical uncertainty in river basin science. Sediment production and routing within river basins involve the activation of multiple sediment sources through the operation of many different processes. These onsite, “upstream” processes include hillslope sheetwash and rill erosion, gully erosion, mass movement, and river bank erosion. Analysis and prediction of this system's behaviour accordingly demand spatiallydistributed approaches and models, rather than the lumped methods often used, such as catchment-averaged delivery ratios. The sediment delivery ratio (SDR, or yield as a proportion of the total production) is a poorly-specified lumped parameter, and there is still debate about the concept, and no general or reliable basis to estimate its value (Parsons et al., 2006).

⁎ Corresponding author. E-mail address: [email protected] (K. Richards). 0341-8162/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.catena.2009.05.014

Sediment yield is the “downstream” output of sediment from a catchment area, and this frequently decreases with basin area because of “off-site” deposition before the catchment outlet is reached (Fig. 1). The relationship between sediment production (by erosion) and sediment yield can be expressed by the equation: Y = SDR · E

ð1Þ

This equation (Walling, 1983) states that sediment yield (Y) is the product of the sediment delivery ratio (SDR) and sediment erosion (E). The sediment delivery ratio is usually <1 (or <100%), and also frequently decreases with the drainage area (Fig. 2), possibly because in larger basins there is more opportunity for deposition between erosional source and basin outlet, and also because rainfall that triggers erosion is less likely to cover the whole of larger catchments. The problem with this formulation of the delivery ratio is that it treats catchments as lumped, black-box phenomena. As Boyce (1975) noted, “external” links in a drainage network structure have quite different sediment delivery characteristics (and delivery ratios) from the “internal” links; in short, then, the spatial variation of sediment delivery within catchments cannot be judged from Figs. 1 and 2 and Eq. (1).

278

J. Ding, K. Richards / Catena 79 (2009) 277–287

catchment, rather than to make explicit forecasts of sediment production and yield for practical purposes. Nevertheless, this exploratory use of the model is facilitated by some comparison with suspended sediment data. SedNet, like many other reduced complexity models (Brasington and Richards, 2007), does not model sediment production and delivery dynamically, but instead requires definition of scenarios for steady state modelling. However, the numerical efficiency that this simplification offers then lends itself to numerical experiments structured in a factorial experimental design to examine different climate and land management combinations and their interactions. Indeed, given the many uncertainties that underlie the component sub-models and parameters of a complex model such as SedNet, it is likely that this exploratory scenariobased and hypothesis-generative approach is its most appropriate use. 2. The study region

Fig. 1. The inverse relationship of sediment yield with basin area (Lu et al., 2003, 2005).

This is an important issue, because sediment management not only requires on-site control of sediment production, but also management of its delivery by the river system to downstream control locations. While on-site conservation measures to manage sediment production may make important contributions to sediment control, they may also be rendered inefficient because much of the sediment produced is in any case not transported to a downstream site (a reservoir, for example), but is deposited off-site, en route. So the critical locations to target for control may be those with efficient delivery within and through a catchment, as much as those with high rates of production. In this context, this paper describes the preliminary application of a spatially-distributed method to quantify patterns and rates of sediment production in the Xihanshui basin, a tributary of the upper Yangtze; and to predict its routing to a downstream point, accounting for losses in deposition en route. This is based on a use of SedNet (Prosser et al., 2001a, b; Lu et al., 2003), which is a distributed model in which the component processes are represented by reduced complexity sub-models to facilitate application at large spatial scales. Ultimately, this model can lead to effective targeting of sediment control, by identifying and managing sites both with high sediment production and with efficient delivery (Lu et al., 2004). However, the aim of this study is also to explore the geomorphological implications of the model's distributed approach to sediment production, delivery and yield. The purpose of this paper is therefore to assess the relative importance of sediment production processes under different climate and land cover conditions, and the controls of the spatial structure of delivery ratios within this large

The focus of this paper is on one of the headwater basins draining into the Yangtze River. This is obviously of interest because sediment produced by the tributaries of the Upper Yangtze River directly affects the service life-span and normal operation of the Three Gorges Dam, the biggest hydro-power project in the world (Fan et al., 2004). The Jinsha River (the upstream main stem of the Yangtze River) and the Jialing River (the largest tributary of the Yangtze River) are the major sources of this sediment, contributing 72.8% of the sediment but only 48.6% of the runoff (Tan, 2004). The basin area and average annual runoff of the Jialing River account respectively for 15.8% and 15.6% of the total values of these variables for the Yangtze River basin upstream of the Yichang hydrological station, but its sediment production accounts for 25.5% (Zhong, 2001). The scale of basin studied in this research is limited in order to trade off the desirable resolution of the model (the grid cell size) against computational demands. Accordingly, the specific subject of the investigation is the sediment budget of the Xihanshui Basin, a significant tributary of the Jialing River. The Xihanshui Basin (latitude: 33°15′N–34°31′N, longitude: 104°31′ E–106°2′E) is located in the south-east of Gansu Province, and covers an area of 1.02× 104 km2. The Xihanshui River, an upstream tributary of the Jialing River, rises on Qishou Mountain in Tianshui City, Gansu, with its watershed bordering the Yellow River Basin in the north. The river then crosses the south of Gansu Province and enters Shaanxi Province, joining the Jialing River at Lueyang (Fig. 3). This is a basin within which some of the most severe soil erosion in the upper reaches of the Jialing River catchment occurs, especially in its north-east where it drains the southwestern edge of the loess plateau. The north of the basin has a warm temperate, semi-arid, continental monsoon climate, with annual precipitation ranging from 400 to 700 mm, and the climate gradually changes to a warm temperate humid monsoon type, with annual precipitation above 800 mm, in the south. The land use and cultivation methods also change from north to south, from spring wheat and rape to wheat–rice, wheat–corn and rape–corn rotations. Since the 1980s, the Yangtze and its tributaries have seen considerable investment in soil and water conservation, and there is evidence that this has reduced sediment yield (Lu and Higgitt, 1998). However, because these land management changes have co-varied with a decline in rainfall and discharge, it is difficult to be clear about the effectiveness of the conservation. A modelling approach such as that adopted here may therefore help to distinguish the effects of climate and land use changes. 3. The SedNet methodology

Fig. 2. The inverse relationship of the sediment delivery ratio (SDR) with the drainage area (Branson et al., 1981).

The SedNet model used in the research reported here was first developed and applied in Australia (Prosser et al., 2001a); here we use Version 2 (Wilkinson et al., 2004). SedNet involves a set of GIS routines that define river networks and their associated sub-catchments, map the distribution of the rates of various erosional processes, and “route” sediment through the network as a function of river hydrology and morphology, by explicitly estimating sub-catchment sediment budgets.

J. Ding, K. Richards / Catena 79 (2009) 277–287

Fig. 3. The location and a digital elevation model of the Xihanshui Basin.

279

280

J. Ding, K. Richards / Catena 79 (2009) 277–287

Investment prioritization analysis can then be carried out using the results of spatial modelling of sediment budgets (Lu et al., 2004). An application of SedNet begins with discretisation of the study catchment into approximately uniform sub-catchment elements (in this study, about 1/1000th of its total size), each associated with an exterior or interior link. For each of these, a sediment budget is estimated according to Eq. (2): Oi = Ii + ½Bi + Gi + ðEi : hsdriÞ + Mi − Di

ð2Þ

Here, input of sediment from the upstream element is Ii. The sediment generated internally to the element (each component of which is estimated by a simplified model) is in the square brackets in Eq. (2), and includes sediment from bank erosion (Bi), gully erosion (Gi), hillslope erosion (Ei) moderated by a delivery ratio (hsdri, assumed initially to take the SedNet default value of 0.05; Wilkinson et al., 2004), and mass movements (Mi). This production is moderated by a depositional loss term (Di) for the element. The element sediment delivery ratio is then given by the ratio of element output (Oi) to total element sediment supply (both from the upstream element and generated within the element — that is, Ii plus the components in square brackets in Eq. (2)). Thus, sediment delivery is not defined as in Eq. (1) and Figs. 1 and 2, but is explicitly derived from a sediment budget for each element in the catchment. The mean annual delivery of sediment (λik) from a given link element i to any downstream control location k is the within-element sediment supply (Ii) multiplied by the sediment delivery efficiency (γj) through all river links (Mik) along the route to k, according to Eq. (3) (Lu et al., 2004). λik = Ii

Mik Y

γj

ð3Þ

j=1

This represents the chained product of all element delivery ratios on the route from i to k. For all links that are tributary to the control location k, the total yield is then: Tk =

N X

λik

land use scenarios, in order to demonstrate whether land use effects on sediment production and yield differ consistently under different climatic conditions, and vice versa. 4. Initial configuration of SedNet There are several initial decisions required to configure SedNet before undertaking experiments to explore the effects of different climatic conditions and land uses on sediment production, routing and yield, and Table 1 provides a summary of these. 4.1. Division into sub-catchments The first essential parameters are those that define the subcatchment structure for the purpose of modelling. This process begins with a digital elevation model, which for the Xihanshui catchment was derived at a resolution of 50 m to represent a suitable compromise between computational efficiency and the need for typical slope lengths to be represented by several cells. A flow routing algorithm was used to generate a river network, and a threshold catchment area was set at 5 km2. This threshold ensured that a large proportion of the gully systems in the north-eastern loess-covered area were represented as elements of the river network (this is consistent with Chinese usage of the term “gully” in this region, referring to narrow valleys incised into loess). Hillslope sediment supply to, and bank erosion along, these explicitly-defined elements of the river network may in part replace the need for separate assessment of gully erosion, although this assumption remains to be fully tested. Sediment yield from those smaller tributary gullies not represented in the river network can be handled approximately by defining bare soil in cells adjacent to the main gullies, so that sediment yield by hillslope erosion is enhanced to mimic erosion in tributary gullies within headwater sub-catchments. 4.2. Hydrological regionalization A distributed model of sediment production requires spatiallydistributed information on river hydrology, and therefore a method to

ð4Þ

i=1

This model provides a means of evaluating the spatial variation of contributions to the overall catchment sediment budget, and of disaggregating the total yield at any point into the effects of upstream contributing erosional processes, and the effect of routing from different parts of the basin. As noted above, SedNet is not a fully dynamic model, and does not explicitly route sediment downstream physically with a realistic time scale. Instead it operates in a steady state mode, with the objective being to map the spatial pattern of sediment production under different climate and/or land use scenarios, in order to assess the effects of these different controls. It is applied here for a time period during which some processes, such as sub-catchment in-channel sediment storage and remobilisation, can be neglected. Thus, during an annual hydrological cycle, non-zero fine sediment storage changes in the channels at sub-catchment scales can, as a first approximation, be assumed to be relatively minor components of overall sediment budgets. This can be confirmed by a rough estimate of the sediment volume of an assumed depositional thickness (say, 0.1 m) occurring over the total channel bed area; the estimated load is of the order of 10% of the floodplain deposition, which in turn is less than 10% of the hillslope sediment supply. The aims of modelling in the Xihanshui basin are then to provide a general assessment of how the dominant sediment production and delivery processes contribute to the sediment yield of the Xihanshui under different annual scenarios of climate and land use. Ultimately this could be structured as a factorial experimental design, with the model being run for each climate scenario for a range of

Table 1 Summary of basic parameters and methods used in initial configuration of SedNet application to the Xihanshui catchment. Parameter/configuration Catchment structure DEM cell resolution Threshold catchment area Number of sub-catchments Average sub-catchment area Hydrological regionalization

Bank erosion Bank erosion coefficient Riparian vegetation cover

Bank height

Bank sediment bulk density Bank sediment gravel content Floodplain deposition Floodplain extent Bankfull return period Settling velocity

Value/method 50 m 5 km2 1028 9.77 km2 (i) Runoff coefficient regressed on dryness index; (ii) daily flow variability regressed on rainfall; (iii) bankfull discharge regressed on mean annual flow; (iv) median overbank flow regressed on mean annual flow. Initially 2 × 10− 5 (default value) Derived from land use maps for the 1980s, 1996 and 2000, generated using satellite imagery and field survey 2 m (based on field data which show little systematic variation over three orders of magnitude of catchment area) 1.5 t m− 3 Set initially at 4.1% (average for field samples); also tested at 50% Derived from DEM using gradient threshold for inner edge Set initially at 2.5 years; also tested at 1.5 years Initially 1 × 10− 6 (default value)

J. Ding, K. Richards / Catena 79 (2009) 277–287

regionalize flows from existing gauging station data. SedNet estimates mean annual flow, daily flow variability, bankfull discharge, median overbank flow and the runoff coefficient from the available gauged daily flow time series, then derives regionalization equations to enable prediction of these flow properties for all sub-catchments (Wilkinson et al., 2004). These are empirical relations relating flow properties at the gauging stations to upstream catchment properties (see Table 1), which are then used with gridded dryness index (the ratio of potential evapotranspiration to rainfall) and rainfall data to estimate the flow properties for all sub-catchment river links. There are eight parameters estimated in the regionalization regressions, but with only four sets of gauging data for the Xihanshui, these are all highly uncertain, and the explained variances of the regression fits are low. There are currently no means of testing the validity of these regionalization procedures against additional flow data. Although some model parameters (notably the bankfull discharge) are derived from the 1993–2005 hydro-meteorology data available for analysis, annual scenarios for the Xihanshui are selected as exemplars of the range of climatic conditions. In these scenario analyses, regionalization is undertaken separately for each scenario year, using the relevant annual gridded dryness index and rainfall data. The regionalizations for the scenario years therefore reflect the spatial distributions of climatic controls in those years.

281

length factors were then estimated using Chinese slope factors defined by Liu et al. (1994) and Liu et al. (2000), which provide a compromise between the slope factors in the original USLE and the Revised version (Renard et al., 1997). The slope factor is related by a set of linear functions to the sine of the slope angle, for slopes with angles of <9%, 9%– 25% and >25%, and the slope length factor requires a power function relating the length factor to the measured slope length, with an exponent m that increases in steps from 0.2 to 0.44 for different slope angles from <1% to >30%. As noted above, the default SedNet hillslope delivery ratio of 0.05 was used. Zhao and Shi (2002) have measured sediment delivery ratios on erosion plots, identifying inverse relations with slope length and contributing area that suggest values for the southern edge of the loess plateau of 0.27–0.94, and an average of 0.664. However, data from plots with a maximum length of 60 m cannot give reliable delivery ratio estimates for the much longer slopes in the Xihanshui catchment, and as a first approximation the default value was used. This is an area requiring further investigation, both of averages and of spatially-distributed values reflecting slope length and gradient. Finally, soil erodibilities for the Xihanshui soils were based on the USLE nomographs (Wischmeier and Smith, 1978) and on measurements of soil particle size, soil structure, organic matter content, and saturated hydraulic conductivities sampled in the field in mapped soil units classified and mapped by Chinese soil scientists (National Soil Survey Office, 1995).

4.3. River bank erosion and floodplain deposition 4.5. Rainfall scenarios River bank erosion and floodplain deposition were modelled using the SedNet routines as a first approximation. The bank erosion rate is estimated as a linear function of bankfull stream power, with a coefficient chosen to give bank erosion rates ideally comparable to field-estimated values for unvegetated banks. These rates are then adjusted by multiplying by the factor (1 − PRx), where PRx is the proportion of intact natural riparian vegetation derived from the GIS coverages defining the land cover and the floodplain extent (Table 1; Wilkinson et al., 2004). The floodplain sedimentation model is based on the concept that the proportion of suspended sediment load available for deposition is equal to the fraction of total discharge that goes overbank and the residence time of water on the floodplain. Total deposition (Dx, t) on the floodplain is: Qf − ðTIFx Þ 1 −e Dx = Q

 ! vAf Qf

To examine the effects of rainfall, we selected 3 years to define typical “scenario” conditions. Considering the annual rainfall from 1983 to 2005, we chose 1984 as the wettest year on record for this period, 2000 as an average year, and 1997 as the driest year. Although it is possible to assume that the average rainfall for each of these years occurs over the whole catchment, it is also possible to use the rainfall data for eight stations in and adjacent to the catchment, in order to define the spatial patterns of rainfall in these three scenario years. These spatially-distributed rainfall scenarios were then used to derive estimates of the rainfall erosivity, using a method employing daily rainfall data developed for a Chinese version of the (R)USLE, and based on fortnightly integration of rainfall erosivity which is then cumulated over annual periods (Zhang and Fu, 2003). This rainfall erosivity model is defined as follows:

ð5Þ Mi = α 5

(Table 1; Prosser et al., 2001a,b). Here TIFx is the total suspended sediment supply, Af is the floodplain area and Qf is the floodplain discharge (the median of the set of overbank discharges above bankfull discharge). The settling velocity v was set at the default value (Table 1), but deposition is highly sensitive to this value and experiments are required to refine this choice. Dx is integrated across the frequency spectrum of significant sediment transporting events, to give deposition as a mean annual value (t/y), and after dividing it by the floodplain area, the deposition is expressed as a mean annual rate of floodplain aggradation.

K   X β5 Dj

ð6Þ

j=1

Here, Mi is half-month rainfall erosivity (MJ mm hm− 2 h− 1) in month i; K is the number of days in a half-month, and Dj is the erosive rainfall (rainfall in excess of 12 mm) on day j in a half-month. The model parameters α5 and β5 are estimated from the following two relationships: −1

−1

β5 = 0:8363 + 18:144 · Pd12 + 24:455 · Py12

4.4. Slope and soil factors

and

In this application of SedNet, we examine the relative contributions to the sediment production and yield by hillslope wash and rill erosion and river bank erosion. Hillslope wash and rill erosion is estimated using a version of the (R)USLE (Renard et al., 1997), which requires the multiplication of factors defining rainfall erosivity, soil erodibility, land cover, and a slope length factor that have been calibrated to be appropriate in a Chinese context. The relatively fixed (R)USLE factors (slope and soil properties) were derived from DEM analysis and soil surveys. Firstly, the slope angles and lengths were derived from a 20 m DEM of the catchment (Fig. 3) using the DEM analysis tools developed by Hickey (2000) and Van Remortel et al. (2004). The slope angle and slope

α 5 = 21:586β5

−7:1891

ð7Þ

ð8Þ

Here, Pd12 is the average daily rainfall (mm) of daily rainfalls >12 mm, and Py12 is the average annual rainfall (mm) of daily rainfalls >12 mm. Daily rainfall data were therefore processed using these three equations to give the half-month rainfall erosivity, then annual rainfall erosivity was calculated by adding the 24 half-month rainfall erosivity values for each year. This is a method that permits use of daily rainfall to estimate erosivity, as a result of correlations having been defined between daily rainfall and event-scale erosivity, permitting subsequent aggregation to the half-month and then annual time scales.

282

J. Ding, K. Richards / Catena 79 (2009) 277–287

Fig. 4. Spatial patterns of rainfall in the three scenario years (expressed as mm per day).

Fig. 4 illustrates the spatial maps of rainfall in the catchment in 1997, 2000 and 1984 (from left to right, driest to wettest). These maps are based on interpolating across a 50 m grid covering the catchment from the point data obtained at the 12 meteorological and gauging stations, using the ArcView spline method with the tension option (ESRI, 2008). The isolines of rainfall, and therefore erosivity, tend to run approximately east–west in orientation in the south of the catchment, reflecting a strong north–south gradient, but have a more north–south orientation in the central and

northern parts. One scenario test that can be performed is to compare the sediment production and yield for spatially-uniform annual rainfall, and to compare this with spatially-variable annual rainfall (see below). 4.6. Land use and land management scenarios A typical land use map for the catchment is that for 1996 shown in Fig. 5. The original was produced by the Data Center for Resources and

Fig. 5. A land use map of the Xihanshui River basin for the late 1990s.

J. Ding, K. Richards / Catena 79 (2009) 277–287

Environmental Sciences, Chinese Academy of Sciences, using satellite image analysis and field checking, but is here simplified from the original twenty classes to five. Before evaluating the (R)USLE, this map must be converted into a GIS layer of C factors using a look-up-table. However, in the absence of regular reliable, detailed land use assessments for the catchment, and given that the land use classes are also unreliable indicators of cover variations arising in a single class, it is necessary to assess the effects of land use and land cover changes by constructing alternative simplified land use scenarios. These could reflect, at one extreme, severe cover degradation relative to the 1996 map, and at another extreme, the effect of soil conservation, and could then form a set of land use scenarios in a two-way factorial experimental design to investigate the interaction of climate and land use. However, to compare modelled sediment production and yield in the selected scenario years with sediment yields measured at the river gauging stations, it is necessary to adjust land cover values to reflect known changes in agricultural and land management practices between these years. This requires manipulation of both cover values (C factors) and land management (P) factors in the (R)USLE, the latter reflecting the changing emphasis on well-maintained structural conservation, such as terracing, since the 1980s. This approach was employed by differentiating land cover classes between 1984 and 1997/2000, so that areas under the “sparse tree” class in 1984 were converted to the “shrubs” class in 1997/2000; with “sparse grass” converted to “medium grass cover” and “medium grass cover” to “dense grass cover”. These resulted in reductions in the C factors with the shift to the improved cover class in the look-up table. In addition, and using evidence from the Office of Soil and Water Conservation (1994, 1995) about the areas subjected to conservation activity at different dates, P factors were changed from 0.5 in 1984 (to reflect the existence of terracing, albeit poorly maintained), to 0.18 (typical for well-maintained terracing) in the sub-catchments covered by Soil and Water Conservation Office projects by 1997/ 2000. These projects have restored and expanded terracing, increased the use of cover plants, expanded areas of conservation woodland, and both planted trees and installed check-dams in gullies. With these changes, it is then possible to evaluate how the combination of rainfall differences and land cover and land management changes between the scenario years affect modelled relative to measured yields. 4.7. Sediment yield data Suspended sediment yield data were collected at three gauging stations along the main Xihanshui River (Shunlixia, Daqiao and Tanjiaba; Fig. 7). These data were used for two purposes. The first was to confirm the choice of scenario years. For this purpose, annual sediment yields for the three gauging stations (left to right, upstream to downstream) are plotted (Fig. 6) as a function of annual discharge. The 1984 data always plot at the top of the scatter, the 1997 at the bottom,

283

with 2000 in mid-scatter. The highest sediment yields, such as those in 1984, involve concentrations of the order of 1000 kg m− 3, which therefore verge on hyper-concentrated flows. However, the main use for these data was as a test of the model predictions. When the model had been set up for a given scenario year, discrepancies between simulated and measured yields at the three gauging stations were identified, and were used in sensitivity analyses to examine the validity of some parameter values, or to attribute them to as yet unmodelled effects. 5. Results, sensitivity analysis, and comparison with data The first step in applying SedNet to an analysis of sediment production, routing and yield is to subdivide the catchment into elements, for each of which the components of erosion can be estimated. Fig. 7 shows this discretisation of the Xihanshui basin and its river network for spatial sediment budgeting purposes. There are approximately 1000 elements averaging about 10 km2 in area (Table 1). These may not be entirely homogeneous, but their size relative to the grid cell size is considered to be sufficiently large that errors in estimating sediment production by hillslope erosion in the individual 50 m grid cells may reasonably be expected to cancel one another approximately at the element scale. The maps in Fig. 8 show the initial estimates of hillslope suspended sediment supply to the sub-catchment elements in Fig. 7 in the dry (1997), average (2000) and wet (1984) years (from left to right). The patterns in these maps reflect the interaction among rainfall erosivity, soil erodibility and slope gradient and length. The DEM in Fig. 3 shows that the approximately east–west trending upper Xihanshui has a broad valley that then turns south and enters a gorge section with steeper slopes; Fig. 8 suggests that these slopes are responsible for high rates of soil erosion, especially noticeable in 1984. A summary of the results is provided in Table 2, which shows how the total sediment supply, the balance between hillslope and river bank erosion, and the depositional losses vary between the 3 years for the whole catchment. Between the wet (1984) and dry (1997) scenario years, a ten-fold decrease in sediment supply is apparent, and, when using the default bank erosion coefficient, distributed hillslope sediment production dominates relative to the linear bank erosion contribution. The hillslope sediment production is strongly dependent on rainfall, while bank erosion is relatively insensitive. There appears to be a net accretion to floodplain sediment storage in the catchment for all climatic scenarios, with depositional input to floodplain storage being between 30 and 200 times higher than bank erosion, and with the higher ratio (proportionally greater deposition) in the wet year. This addition to storage is mainly in the upper (northern) third of the catchment, where the river has created an extensive valley fill before it enters the mid-catchment gorge. However, although the spatial distribution of deposition is realistic, the rates may be highly sensitive to the settling velocity parameter, and further empirical analysis of this is required. Although

Fig. 6. The annual sediment yield as a function of annual discharge at the three gauging stations along the main Xihanshui river: Shunlixia, Daqiao and Tanjiaba (shown from left to right; see Fig. 7).

284

J. Ding, K. Richards / Catena 79 (2009) 277–287

Fig. 7. The Xihanshui basin river network and gauging stations.

caution in interpreting these results is necessary, the bank erosion estimate is of the order of 0.005 m a− 1 over the total length of the river network, and the floodplain deposition is c. 0.01 m a− 1 over the total area of floodplain in 1997 and 2000, but almost an order of magnitude higher in 1984. These are not unreasonable average rates at this spatial scale, and given the localisation of the floodplain areas. Two different approaches have been undertaken initially to evaluate these results. The first is to use sensitivity analysis of some parameters to assess the relative significance of different components of the sediment

budget, and the second is to compare the predictions with measured sediment yields at the gauging stations. 5.1. Sensitivity analysis In Table 2, the first three columns of estimated sediment yields show values based on using the average measured proportion of gravel (the fraction coarser than 2 mm; 4.1%) in bank material samples from the Xihanshui as the percentage of eroded bank material contributing

Fig. 8. Hillslope Suspended Sediment Supply in the Xihanshui River basin in the three scenario years: 1997 (dry), 2000 (average) and 1984 (wet) (shown from left to right).

J. Ding, K. Richards / Catena 79 (2009) 277–287 Table 2 Results of simulation of sediment production and yield (the columns headed 1984a and 1984b repeat analysis for this year with adjusted parameters, “a” for bankfull discharge and “b” for gravel percentage in the bank material). Suspended sediment yield (kt/y) Inputs Hillslope erosion Bank erosion Outputs Reservoir/lake deposition Floodplain deposition Export

1997

2000

1984

1984a

1984b

11,056 45

21,387 47

123,911 49

123,911 52

123,911 26

238 1370 9492

608 1827 18,998

4275 10,261 109,425

4305 11,443 108,216

4275 10,258 109,405

Default parameters are used throughout for the bank erosion and floodplain deposition models (see Table 1).

to bedload rather than suspended load; and a bankfull flood discharge frequency of 2.5 years. The first of these parameter assumptions implies that when bank erosion is estimated, 95.9% of the volume eroded contributes to suspended sediment yield. The column headed 1984a illustrates the effect on the estimates for 1984 of changing the bankfull frequency to 1.5 years; bank erosion and floodplain deposition increase slightly to reflect the greater frequency of overbank events. However the difference is negligible. The last column shows the effect of increasing the bedload percentage to 50%. This almost halves the contribution of bank erosion to suspended sediment yield, but this is a minor effect on the overall sediment yield given the bank erosion coefficient employed. However, these tests indicate that variations in these two model parameters do not materially affect the significance of the between-year differences, or the relative importance of hillslope sediment production. Other sensitivity analyses also suggest a limited impact of some parameters on the simulated suspended sediment yield. For example, when a spatially-uniform rainfall erosivity equivalent to the basin average is employed for two of the scenario years (1997 and 1984), changes in total hillslope sediment supply and basin sediment yield are less than 1%; at the whole catchment scale, individual element values that increase cancel those that decrease. However, the spatial pattern of hillslope sediment supply differs in detail from that in Fig. 8, and this is of specific concern in an analysis designed to examine spatial patterns of sediment delivery within a catchment, and may result in some significant differences between the uniform rainfall and spatially-varying rainfall scenarios. For example, in 1997, the effect of using a uniform erosivity is to increase by about 5% both the hillslope sediment supply and the basin sediment yield. Examining Fig. 4, it is evident that the reason for this is that a uniform rainfall results in higher rainfall erosivity in the north-east of the catchment (where rainfall in 1997 was especially low). In 2000, floodplain deposition is 14% higher for the uniform rainfall scenario, mainly because the rainfall-based method for interpolating discharges results in discharges increasing when a uniform rainfall is employed to the greatest degree in that part of the catchment where floodplains are most extensive. Modelled rates of bank erosion and floodplain deposition are sensitive to, respectively, the bank erosion coefficient and the settling velocity parameter. The bank erosion rate increase is linear with the coefficient, but sensitivity analysis confirms that bank erosion is less sensitive to variation in rainfall between scenario years than is the

285

hillslope sediment supply. The response to variation in the settling coefficient parameter is inherently non-linear, but also interacts with the effect of the bank erosion coefficient to a degree that depends on the total sediment supply. For example, in these experiments, there is between 0.5% and 3% more floodplain deposition as the bank erosion coefficient is increased by an order of magnitude, simply because bank erosion adds to the sediment available to be deposited; the percentage decreases from the driest to the wettest year as the total sediment supply increases. However, an order of magnitude increase in the settling velocity parameter results in 2.5- to 4-fold increases in deposition, the increase being highest in the wettest year (again, because sediment supply is highest in the wettest year). One consequence of a higher settling velocity parameter and increased floodplain deposition is that, all else being equal, there is less sediment to be transported downstream, and the downstream yield will be reduced. This means that a higher settling velocity parameter can influence the comparison between modelled and observed yields, potentially reducing the difference in 1997 and 2000, and increasing it in 1984, the wetter year (see Table 3). Finally, the effect of land use can be assessed through scenarios in which arbitrary changes are made to the crop cover factor, such as for example, assuming bare soil (C = 1) everywhere. In this hypothetical scenario, the sediment yield increases by between 6 and 15 times compared to the simulated values in Tables 2 and 3, with the greatest increases in the later, drier years (1997 and 2000) when the land cover and land management has been improved by conservation; the new estimated sediment yield in 1997 approaches 100 times the measured values. In 1984, the bare soil erosion rate implies a sediment yield about 5 times the measured yield. These results demonstrate how sensitive soil erosion and sediment yield are to the land cover and management, and suggest that the C and P values assumed for the three scenario years may be approximately valid. Thus, sensitivity analysis coupled with the evidence of measured sediment yields at different sites at least helps to constrain the parameters to reasonable values. 5.2. Comparison with observed yields Comparison between the SedNet predictions and measured suspended sediment yields requires that the spatially-distributed SedNet estimates are interrogated to define the sediment yields at the locations of the three main gauging stations (the equivalent of the “export” values in Table 2). Fig. 7 shows the locations of the Shunlixia, Daqiao and Tanjiaba gauging stations along the main stem of the Xihanshui, and Table 3 compares the estimated (modelled) and observed (measured) yields at these locations for the three scenario years. Given the uncertainties surrounding the definition and spatial distribution of land cover (C) and land management (P) factors in the three scenario years, it is encouraging that the ratios of estimated to observed yields in 2000 are of the right order. In 1997, the ratio is between 1 and 10, with greater over-estimation of sediment yields by the model in this driest year. This suggests that the adjustments made to the C and P factors to reflect the activities of the Soil and Water Conservation Offices may have been insufficiently generous, leading to the conclusion that soil conservation has indeed been very effective, and that low sediment yields in the late 1990s and early

Table 3 A comparison of observed (measured) suspended sediment yields at three gauging stations on the Xihanshui River with estimated (modelled) yields based on SedNet hillslope and bank erosion sediment production. Suspended sediment yield (kt/y)

1997

2000

1984

Est

Obs

Ratio

Est

Obs

Ratio

Est

Obs

Ratio

Shunlixia Daqiao Tanjiaba

894 3014 8412

114 1105 1169

7.84 2.73 7.20

2635 8278 17,467

898 9340 11,330

2.93 0.89 1.54

20,138 51,218 104,230

31,640 73,711 120,355

0.64 0.69 0.87

286

J. Ding, K. Richards / Catena 79 (2009) 277–287

2000s are not solely because of lower annual rainfalls, but really do reflect the improvements in land management. This is encouraging, given that Lu and Higgitt (1998) have suggested that observed reductions in sediment yield in the Upper Yangtze in the 1990s were largely because of reservoir construction. There are some reservoirs in the Xihanshui basin, but they are small and their impact on sediment yield is negligible (see Table 2). It is notable that the spatial pattern of model over-estimation is similar in 1997 and 2000; the ratio is at its lowest at Daqiao. The most likely explanation for this is that the method for spatially distributing flows in the model has failed to capture the true pattern in these drier years. By contrast with these two scenario years, the magnitudes and pattern of sediment yield in 1984 show an under-estimation by the model of from 36% to 13%, with a spatial pattern in which measured sediment yields exceed the modelled values most noticeably at Shunlixia (upstream) but are closer to the modelled values downstream at Tanjiaba. Thus, observed suspended sediment yields are significantly lower than the SedNet predictions in 1997 and 2000, but are significantly higher in 1984, the wettest year in the 1983–2005 record, suggesting that the catchment behaviour is markedly more non-linear than the behaviour captured by the model. An explanation for this may be that the SedNet predictions only include sediment production by hillslope and bank erosion processes, and that the greater non-linearity evident in the response of measured yields to higher rainfall arises because other processes of sediment supply are activated in years with higher annual rainfall. The discrepancy between estimated and measured yields in 1984 tends to decrease in the downstream direction, which suggests that this additional sediment supply is most evident upstream from Shunlixia. The main loess-covered part of the catchment is located here, and the hillslopes in this area are susceptible to both gully erosion and mass movement. Mass movements, stabilised, recent and active, are evident throughout the north-east loess-covered part of the catchment, with flow-slide features being particularly common (Derbyshire et al., 1991; Meng and Derbyshire, 1998). One of the data sources that can give some insights into this issue, by permitting an assessment of the contribution to sediment supply by mass movement, is the inventory of the volumes of the many landslides that occurred in 1984, created and maintained by the Soil and Water Conservation Offices. This inventory provides the volumes of landslides, of which several are identified as having taken place in 1984. The sediment contribution of these events, based on assuming a bulk density of 1.25 t m− 3, may be a significant additional source of sediment supply in the wettest years. For example, above Shunlixia, there were mass movements in 1984 totalling 19,000 kt, and the cumulative total above Daqiao was 54,000 kt. These figures can easily account for the under-estimation of the observed sediment yield in 1984 by a model only assessing hillslope sediment supply and bank erosion. However, this cannot be regarded as a conclusive test, since there are numerous limitations to the mass movement database, and also to any necessary assumption made concerning the proportion of the volume of mass movements that actually reaches the river network (the mass movement delivery ratio). Nevertheless, it is at least a reasonable hypothesis worthy of more detailed investigation that mass movement (and indeed, gully erosion) both become more important sediment supplies in wetter years. Mass movement is not in fact included in the original formulation of SedNet, so to develop a procedure for modelling this source of sediment supply requires significant development of the original code. There is an additional factor that needs to be considered in evaluating the non-linear catchment response to rainfall, and this is the hillslope delivery ratio. As noted above, Zhao and Shi (2002) suggest a value of 0.664 based on plot experiments. It seems reasonable to conclude that these experiments occur under conditions when nearcontinuous flow paths exist over the plot length. Such flow continuity is unlikely on full-scale hillslopes except in the wettest conditions, but under these circumstances, the hillslope delivery ratio may be higher.

There is a simple linear dependency of hillslope sediment supply on the delivery ratio, all else being equal. However, if the continuity of flow paths on a hillslope increases with catchment wetness, and as a result increases connectivity to the basal river, this may mean that the delivery ratio should increase as a function of wetness, and in doing so, will produce the non-linear relationship between sediment yield and rainfall evident in the variation of observed sediment yields. 6. Discussion: issues and future work There are several areas for further research implied by this study. There remain uncertainties about the choice of bank erosion and settling velocity coefficients, and these will have to be resolved by empirical studies of bank erosion rates, based on archived aerial photography, and floodplain deposition rates which could be constrained by studies of radioactive isotopes (Zhang et al., 2006). However, it seems unlikely that either of these processes will significantly alter the relationship between hillslope sediment production and sediment yield. Nevertheless, it would also be valuable to acquire empirical data to test the individual process predictions (hillslope sediment production, gully erosion) to validate this assumption. As noted, there also needs to be a new approach to modelling the hillslope delivery ratio as a spatially- and temporally-variable phenomenon. Instead of assuming a constant value (of 0.05), the delivery ratio should vary spatially in relation to slope angle, slope length, and land cover (and hence, surface roughness). Zhao and Shi (2002) suggest some simple empirical relationships with slope length and contributing area, albeit based on relatively short (plot) slopes. Alternatively, the slope properties could define a hillslope travel time which is inversely related to the delivery ratio (Ferro and Minacapilli, 1995; Ferro, 1997; Ferro et al., 1998). Also required is a means of estimating and modelling the dependence of hillslope delivery ratio on rainfall, to capture this non-linear effect. Areal and linear erosion have been represented in the simulations outlined above (in the form of hillslope erosion and river bank erosion), but discrete sediment sources such as gully erosion and mass movement have been omitted. These processes cause relatively localised sediment supply, and require an approach in which sediment supplies are “positioned” stochastically, dependent on environmental conditions such as relief, soils, and rainfall; this is particularly necessary when the modelling is at an annual timescale. This is quite different from the other processes represented in SedNet, which are treated deterministically, with the implication that the spatial structure of sediment delivery is roughly constant. At the catchment scale, a discrete process such as a mass movement is not likely to occur at a deterministically predictable location. Instead, the probability of occurrence is likely to vary spatially with topographical and lithological controls, and with the occurrence of rainfall. To evaluate the effect of mass movement thus requires a Monte Carlo approach (similar to that developed by Benda and Dunne, 1997), in which a series of probabilistic realisations is simulated, and an assessment is made of the ensemble implications for sediment yield. This is more complex to deal with in the SedNet routing procedures, because depending on the precise locations of the randomly-drawn events, the downstream yield consequences will differ. This will make assessment of routing efficiency more difficult to handle, and will require an evaluation of the sensitivity of total yield to different spatial locations of mass movements. 7. Implications and conclusions This investigation has provided some valuable insights into the sediment production, delivery and yield processes and rates in the Xihanshui catchment. It has shown that SedNet is a powerful tool to explore sediment supply in a large catchment, providing rich and detailed insights. Since it is a model consisting of a number of linked reduced complexity sub-models, each of which embodies heuristic assumptions, there will always be uncertainty about its quantitative

J. Ding, K. Richards / Catena 79 (2009) 277–287

predictions. To some degree, this uncertainty can be evaluated through undertaking sensitivity analyses of the model's response to varying certain parameter values. However, the model is most useful in suggesting hypotheses that can be the subject of further testing. In the present study, for example, pertinent suggestions are (a) that there is net accretion to floodplain storage; and (b) that mass movement sediment supply is important in wet years. In the analysis outlined above, it is nevertheless instructive and encouraging that the necessary adjustments of land cover (C) and land management (P) factors required between the scenario years leads to the conclusion that in spite of reduction in rainfall between 1984 and 1997/2000, soil conservation seems to have had a detectable influence in reducing hillslope erosion and sediment yield. An additional advantage of SedNet is that it may be possible to adapt it to incorporate additional processes, and new or alternative reduced complexity models can be constructed to estimate sediment supply from particular processes. These can be implemented outside SedNet, and their outputs entered using the existing input routines. However, it would be valuable if an open source version were to be available, for which researchers could contribute alternative reduced complexity formulations to add to the process suite that it simulates, and to diversify the available sub-models so that they can be intercompared. Finally, a key advantage of using SedNet is that it lends itself to a factorial experimental design (for example, of interacting climate and land use change effects), and could be adapted to Monte Carlo simulation of stochastically-varying sediment supply processes. These together demand large numbers of experimental runs, but allow both sensitivity analysis and uncertainty assessment. These have often not been undertaken as rigorously as is desirable in the application of spatially-explicit reduced complexity models. Acknowledgements The authors acknowledge financial support provided by a Dorothy Hodgkin Postgraduate Award (DHPA/GS/10020424), an RGS-IBG Hong Kong Research Grant, and the University of Cambridge Philip Lake Fund II. Thanks are extended to a number of collaborators in China including: Yu Su, Jianjun Chen, Xinling Wang and Hanxu Lin (Department of Geography, National University of Singapore) for help with fieldwork; Feng Zhang and Wenyu Li for showing us soil conservation activity in the catchment; Liping Zhou (College of Environmental Sciences, Peking University, China) for providing laboratory facilities and Visiting Student status for JD; Jianqing Ji (Department of Geology, Peking University, China) for assistance in the laboratory; Baoyuan Liu (School of Geography, Beijing Normal University, China) for discussions on Chinese soil erosion estimation and access to relevant literature; Zhuo Zhen, Chengbo Zhang, Heping Zou, Bo Yan and others (Department of Earth Science, Sun Yat-sen University, China) for fieldwork support; Xiaojian Li (Department of Geography, Henan University) for assistance in acquiring soil data; and in Cambridge, Steve Boreham and Chris Rolfe for laboratory assistance; and Philip Stickler for cartographic support. We are also grateful to eWater CRC (eWater Ltd.), University of Canberra, for providing us with access to SedNet, especially to the updated Demo version. References Benda, L., Dunne, T., 1997. Stochastic forcing of sediment routing and storage in channel networks. Water Resources Research 33, 22865–22880. Boyce, R.C., 1975. Sediment routing with sediment delivery ratios. Present and Prospective Technology for Predicting Sediment Yields and Sources. US Department of Agriculture, pp. 61–65. Publication ARS-S-40.

287

Branson, F.A., Gifford, G.F., Renard, K.G., Hadley, R.F., 1981. Rangeland Hydrology. In: Range Science Series, vol. 1. Society for Range management, Denver, CO. Brasington, J., Richards, K., 2007. Reduced-complexity, physically-based geomorphological modelling for catchment and river management. Geomorphology 90, 171–177. Derbyshire, E., Wang, J.T., Billard, A., Egels, Y., Jones, D.K.C., Kasser, M., Muxart, T., Owen, L., 1991. Landslides in the Gansu loess of China. In: Okuda, S., Rapp, A., Zhang, L. (Eds.), Loess: Geomorphological Hazards and Processes: Catena, Supplement, vol. 20, pp. 119–145. ESRI, 2008. ArcGIS Desktop Help 9.3. ESRI Support Centre; http://support.esri.com/. Fan, J.R., Zhang, J.H., Zhong, X.H., Liu, S.Z., Tao, H.P., 2004. Monitoring of soil erosion and assessment for contribution of sediments to rivers in a typical watershed of the upper Yangtze River Basin. Land Degradation & Development 15, 411–421 (in Chinese). Ferro, V., 1997. Further remarks on a distributed approach to sediment delivery. Hydrological Sciences Journal 42 (5), 633–647. Ferro, V., Minacapilli, M., 1995. Sediment delivery processes at basin scale. Hydrological Sciences Journal 40 (6), 703–717. Ferro, V., Porto, P., Tusa, G., 1998. Testing a distributed approach for modelling sediment delivery. Hydrological Sciences Journal 43 (3), 425–442. Hickey, R., 2000. Slope angle and slope length solutions for GIS. Cartography 29, 1–8. Liu, B.Y., Nearing, M.A., Risse, L.M., 1994. Slope gradient effects on soil loss for steep slopes. Transactions of the ASAE 37 (6), 1835–1840. Liu, B.Y., Nearing, M.A., Shi, P.J., Jia, Z.W., 2000. Slope length effects on soil loss for steep slopes. Soil Science Society of America Journal 64 (5), 1759–1763. Lu, X.X., Higgitt, D., 1998. Recent changes of sediment yield in the Upper Yangtze, China. Environmental Management 22, 697–709. Lu, H., Moran, C.J., Prosser, I.P., Raupach, M.R., Olley, J., Petheram, C., 2003. Sheet and rill erosion and sediment delivery to streams: a basin wide estimation at hillslope to medium catchment scale. Tech Rep.,15/03, CSIRO Land and Water, Canberra. Lu, H., Moran, C.J., Prosser, I.P., DeRose, R., 2004. Investment prioritization based on broadscale spatial budgeting to meet downstream targets for suspended sediment loads. Water Resources Research 40, W09501. doi:10.1029/2003WR002966. Lu, H., Moran, C.J., Sivapalan, M., 2005. A theoretical exploration of catchment-scale sediment delivery. Water Resources Research 41, W09415. doi:10.1029/2005WR004018. Meng, X.M., Derbyshire, E., 1998. Landslides and their control in the Chinese Loess Plateau: models and case studies from Gansu Province, China. In: Maund, J.G., Eddleston, M. (Eds.), Geohazards in Engineering Geology. In: Engineering Geology Special Publications, vol. 15. Geological Society, London, pp. 141–153. National Soil Survey Office, 1995. Chinese Soil Genus Records, Volume 1–Volume 6. Chinese Agriculture Press. (In Chinese). Office of Soil and Water Conservation, 1994. Compilation of Completed Outcomes of the Soil and Water Conservation Project, Phase One. Office of Soil and Water Conservation, Gansu. Office of Soil and Water Conservation, 1995. Compilation of Completed Outcomes of the Soil and Water Conservation Project, Phase Two. Office of Soil and Water Conservation, Gansu. Parsons, A.J., Wainwright, J., Brazier, R.E., Powell, D.M., 2006. Is sediment delivery a fallacy? Earth Surface Processes and Landforms 31, 1325–1328. Prosser, I.P., Rutherfurd, I.D., Olley, J.M., Young, W.J., Wallbrink, P.J., Moran, C.J., 2001a. Large-scale patterns of erosion and sediment transport in river networks, with examples from Australia. Marine and Freshwater Research 52 (1), 81–99. Prosser, I.P., Young, B., Rustomji, P., Moran, C., Hughes, A.O., 2001b. Constructing river basin sediment budgets for the National Land and Water Resources Audit. Tech. Rep. 15/01, CSIRO Land and Water, Canberra. Renard, K., Foster, G., Weesies, G., McCool, D., Yoder, D., 1997. Predicting soil erosion by water: a guide to conservation planning with the Revised Universal Soil Loss Equation (RUSLE). USDA Agriculture Handbook 703. 384pp. Tan, Y., 2004. Applications of mathematical modelling to sediment research in the Three Gorges Project. US–China Workshop on Advanced Computational. Modelling in Hydroscience & Engineering, Oxford, Mississippi, USA. Van Remortel, R.D., Maichle, R.W., Hickey, R.J., 2004. Computing the LS factor for the Revised Universal Soil Loss Equation through array-based slope processing of digital elevation data using a C++ executable. Computers & Geosciences 30, 1043–1053. Walling, D., 1983. The sediment delivery problem. Journal of Hydrology 65 (1–3), 209–273. Wilkinson, S., Henderson, A., Chen, Y., Sherman, B., 2004. SedNet User Guide, Version 2. Client Report, CSIRO Land and Water; Canberra. 108pp. (www.toolkit.net.au/sednet). Wischmeier, W.H., Smith, D., 1978. Predicting rainfall erosion losses: a guide to conservation planning. USDA-ARS Agriculture Handbook No. 537, Washington DC. 58 pp. Zhang, W.B., Fu, J.S., 2003. Rainfall erosivity estimation under different rainfall amount. Resources Science 25 (1), 35–41 (in Chinese). Zhang, X., Walling, D.E., Yang, Q., He, X., Wen, Z., Qi, Y., Feng, M., 2006. Cs-137 budget during the period of 1960s in a small drainage basin on the Loess Plateau of China. Journal of Environmental Radioactivity 86 (1), 78–91. Zhao, X.G., Shi, H., 2002. The change of sediment delivery ratio from the slope and watercollected plot in the gully region of loess plateau. Journal of Mountain Science 20 (6), 718–722. Zhong, S.N., 2001. A remote sensing and GIS based dynamic soil erosion monitoring system in the major soil erosion areas in Upper Jialingjiang Watershed. Remote Sensing for Land & Resources 47, 1–8 (in Chinese).