River sediment load and concentration responses to changes in hydrology and catchment management in the Loess Plateau region of China

Authors


Abstract

[1] To reduce the sediment load of China's Yellow River, soil conservation measures have been progressively implemented across the Loess Plateau region since the 1950s. The effectiveness of these soil conservation measures (which were also coincident with reduced rainfall and streamflow) in controlling sediment movement remains to be ascertained. Here the association between sediment movement, hydrological variability, and the implementation of soil conservation measures is examined for the Coarse Sandy Hilly Catchments region of the Yellow River basin. The hypothesis that the soil conservation measures have reduced suspended sediment yields beyond that associated with rainfall reductions alone, principally by reducing sediment concentration, is examined. Annual sediment yield decreased significantly over time in all subcatchments, and the timing of the change (between 1971 and 1985) was consistent with the timing of change in streamflow. Annual mean sediment concentration in 7 of the 11 catchments exhibited a statistically significant decreasing trend over time, indicating that soil conservation practices reduced the mobilization of sediment in most areas, typically accounting for ∼75% of the observed reductions in annual sediment yield. Lesser reductions in area-specific sediment yield at larger catchment areas after the soil conservation measures were emplaced suggests that larger rivers may be reeroding stored sediment. As these sediment stores are likely to be relatively large given the high historic yields, relatively high area-specific sediment yields may persist at larger catchment areas even with improvements to sediment management in smaller tributaries.

1. Introduction

[2] Soil erosion in the Loess Plateau of China generates some of the highest sediment yields observed on earth, with sediment yields of 3 × 103 to 4 × 104 t km−2 a−1 calculated over recent decades from small (<20 km2 in area) catchments [Zhang et al., 1997, 2006]. Such high area-specific sediment yields are not just a feature of small catchments within the Loess Plateau, with the sediment yield of the entire Yellow River basin (752,000 km2) over the last 2500 years being 1.3 × 103 t km−2 a−1 [Shi et al., 2002]. Subsoil sediments produced by gully incision (and associated processes such as mass slumping) provided 77% of the sediment load in a Loess Plateau headwater catchment [Zhang et al., 1997]; if representative of the region, this suggests that subsoil sediments will dominate the sediment load of the Yellow River. This conclusion is supported by the high drainage density of the Loess Plateau [see, e.g., Zheng et al., 2005] and the large size of many erosion gullies. Basin-wide sediment yields in the decades prior to the 1950s are approximately 2.1 × 103 t km−2 a−1 and much of this increase above the preceding Holocene rate was attributed by Shi et al. [2002] to land use intensification driven by increasing population in the Yellow River basin, though this recent increase was superimposed on a longer-term trend of increasing basin sediment yields driven by nonanthropogenic forces including Quaternary tectonic uplift [Shi and Shao, 2000].

[3] Such high rates of soil erosion and sediment transport have both on-site effects including the loss of soil nutrients [Zheng et al., 2005] and agricultural land, as well as important, negative off-site impacts. The Loess Plateau is the source of 90% of the sediment delivered to the Yellow River above Sanmen Gorge and consequently this sediment export heavily influences downstream conditions [Li, 2003]. Xu et al. [2004], for example, note that the river bed of the lower Yellow River has, because of high rates of sediment supply, aggraded 8–10 m above the surrounding floodplain and this thalweg aggradation poses a major flood hazard to local communities [Shi and Shao, 2000]. Recent periods of low or zero flow have exacerbated sediment deposition within the channel [McVicar et al., 2007b].

[4] To control soil erosion in the Loess Plateau, an extensive series of soil conservation measures have been adopted which can broadly be classified into catchment revegetation and engineering works. Xu et al. [2004] note that in many cases catchment revegetation has proven challenging because of the semiarid climate which contributes to poor survival rates of replanted vegetation. These authors have also questioned the capacity of revegetation to effectively retain sediment in the landscape. Plot-scale studies from the Loess Plateau do show that reduced hillslope erosion rates are associated with increased vegetative cover [e.g., Kang et al., 2001; Li et al., 2003; Pan et al., 2006], however the relevance of these plot-scale findings to the broader catchment wide pattern of sediment delivery, particularly given the suggested importance of gully erosion as a sediment source [Li et al., 2003], is a moot point.

[5] An additional and long-standing approach to reducing sediment yields from the Loess Plateau has been the construction of sediment trapping dams (referred to as “check dams” by Zhang et al. [1997]), across eroding gullies and smaller tributaries. Xu et al. [2004] indicate that 113,500 sediment trapping dams had been constructed prior to 2002 within the Loess Plateau. These dams impede the flow of water and thus induce sediment deposition behind their walls. Given the extremely high sediment yields typical of the Loess Plateau, the sediment trapping dams aggrade rapidly and 3200 km2 of new, highly productive farmland has been created by sediment aggradation behind these structures [Xu et al., 2004]. Failure of sediment trapping dams is also relatively common [see, e.g., Chen and Cai, 2006; Zhang et al., 2006]. Construction of larger dams for water storage on the main rivers has also occurred in recent decades [Zhang et al., 1998].

[6] It has been suggested that reductions in the sediment yield of the Yellow River since at least the 1950s and in particular since the 1980s have been a direct result of the catchment management works described above [Jinze, 1991; Shi et al., 2002; Wei et al., 2006; Wang et al., 2007]. However, as noted by Zhao et al. [1992] and Zheng et al. [2007], climate variability has also impacted river discharge. As the sediment yield from a catchment is the product of the suspended sediment concentration and the discharge, either or both of these variables can contribute to changes in sediment yield. To understand changes in catchment sediment yields, it is therefore necessary to consider both of these terms and the processes that influence them. Zhao et al. [1992] attempted to partition the effects of anthropogenic and climatic changes on catchment sediment yields for a 4161 km2 catchment in the Yellow River basin and concluded each factor contributed to the decline in approximately equal proportions. Jinze [1996] also presented a similar attempt at discriminating between the effects of climatic and land management changes, yet the lack of data presented in this study makes it unclear how the results were obtained.

[7] This paper builds upon the changes in streamflow identified by Zhang et al. [2008] in response to coincident land management (soil conservation measures) and climatic variability to examine changes in catchment sediment yield and their causes at a larger spatial scale than the previous research by Zhao et al. [1992]. It examines the tributaries of the middle reaches of the Yellow River that Wang et al. [2007] indicate contribute most of the Yellow River's sediment load. As Zhang et al. [2008] show, both land use modification and climatic change have affected catchment streamflow. Given the goal of understanding the controls on catchment sediment yield, this paper addresses the role the soil conservation measures (including catchment revegetation, terracing and dam construction) have played in influencing the generation and transport of sediment in the Coarse Sandy Hilly Catchments (CSHC) region of the Loess Plateau. We propose the hypothesis that soil conservation and engineering measures have impeded the movement of sediment in the CSHC to a sufficient degree that they have contributed to a reduction in the sediment yield from these catchments. Such a hypothesis is supported by Huang et al. [2003], though only at a relatively small catchment area (∼0.5 km2). If these conservation measures have had an impact on sediment mobility within the catchments, our hypothesis implies that riverine sediment concentrations should be lower after the soil conservation measures have been emplaced. However, before considering this hypothesis, we examine whether there have indeed been statistically significant variations in catchment sediment yields within the CSHC. This is followed by an examination of the degree to which sediment concentrations have varied over time and the role any variations in sediment concentration may have played in influencing catchment sediment yields.

2. Study Site

[8] This study focuses upon ten Yellow River basin subcatchments in the CSHC region, located in the middle of the Loess Plateau as shown in Figure 1, plus the entire CSHC region. The Loess Plateau, covering 620,000 km2, spans the middle reaches of the Yellow River in China and is characterized by an arid to semiarid continental monsoon climate. The combination of erodible, wind-deposited loessic soils, sparse vegetation, intense summer rainfall, a long history of agricultural activity and rapid tectonic uplift [Shi and Shao, 2000] has resulted in a heavily dissected landscape and severe soil erosion. The CSHC cover 113,000 km2 and experience average annual precipitation of approximately 460 mm, though this ranges from 580 mm in the southeast to less than 300 mm in the northwest [McVicar et al., 2007a]. Figure 2 shows the variation in annual rainfall for the CSHC study area along with the variations in streamflow generated within the catchment. About 78% of the annual precipitation occurs during the May to September wet season [Ran et al., 2000]; Zhao et al. [1992] estimate 91% of total annual sediment transport occurs in July and August in the catchment they studied. Topographically, the northwestern part of the CSHC is flatter, while the southeastern region is characterized by a more intensely dissected landscape with gully densities varying from 2 to 8 km/km2 [Ran et al., 2000]. The loess, deposited to an average depth in excess of 100 m during the Quaternary, comprises silty-clay loam textured sediment with sandier sediments occurring in the northwest.

Figure 1.

(a) Map of China showing major rivers and the Loess Plateau (LP) shaded gray. (b) Coarse Sandy Hilly Catchments (CSHC) study region within the Loess Plateau. Study subcatchments are labeled accordingly, and subcatchment 11 represents the whole CSHC region between gauging stations 40103400 and 40104250. The triangles indicate the locations of the stream gauging stations. Note that Huayuankou (referred to in the text) is located approximately 300 km downstream of Sanmen.

Figure 2.

Annual precipitation, streamflow, and percentage of the CSHC treated by soil conservation measures over the period 1959–2000.

[9] As McVicar et al. [2007a] describe, the region has undergone substantial land cover change due to the implementation of soil conservation measures aimed at reducing sediment yields in the Yellow River. Table 1 lists changes in the area of each of the ten study subcatchments of the CSHC region and also for the entire CSHC region that have been affected by soil conservation measures including terracing, revegetation with trees, pasture reestablishment, and the surface area of sediment trapping dams constructed within each catchment (note that sediment trapping dams do not include larger reservoirs constructed for water supply management). As is evident for the CSHC as a whole (Figure 2), the treated area has increased between 1959 and 1996 from approximately 2% to 31% of the total catchment area.

Table 1. Cumulative Areas of Different Conservation Practices in Each Catchment During the Period of the 1950s–1990sa
Catchment Number and NameYearsTerrace (km2)Trees (km2)Pasture (km2)Check Dams (km2)Total Area Treated (%)
1 Huangfu1959242201
196912122614
19791526728410
198922573531321
199631529512420
2 Gushan195927701
19692221514
197927543329
19893113158618
199642118721019
3 Kuye19595272201
196933975222
19796641511087
19896710043531217
19969911843801920
4 Jialu195949212
19692742247
19796798101017
1989104294131338
1996141295161642
5 Wuding19595044825103
19692081237194346
1979421413658710618
1989745870889414635
19969891035282315842
6 Shiwang195951100
19691731212
19793968315
19895915111110
19967423313215
7 Xinshui19591414431
19694144652
197987104885
1989138355321214
1996217630321622
8 Sanchuan19592313431
19698133953
197913219813159
1989220639243022
1996334942323933
9 Weifen195975101
19692513313
19793395449
1989531605615
199610426316827
10 Zhujia1959763102
196915138415
197923221729
1989464447517
1996826877627
11 CSHC19592361251320212
19697812909332985
19791492769890227211
1996335621567199548130

3. Methods

[10] Streamflow and sediment data were obtained from the Water Resources Committee of the Yellow River Conservancy Commission. The annual precipitation data were obtained from the State Meteorology Bureau. Wet season sediment concentration observations made at river gauging stations in the CSHC listed in Table 2 are analyzed (see Li et al. [2005] for a description of the sediment sampling regime). The river sediment sampling normally took place daily, but in some periods the sampling frequency decreased to once every 10 days. Catchment boundaries used to define the study catchments within the CSHC study area are those derived by Yang et al. [2007].

Table 2. Catchment Characteristics and Streamflow Records
Gauging Station NumberCatchmentArea (km2)Streamflow and Sediment RecordsAnnual Average
Precipitation (mm a−1)Streamflow (mm a−1)Sediment Yield (106 t a−1)Sediment Yield (103 t km−2 a−1
  • a

    Annual precipitation and annual PET data for catchment 5 are missing for years 1997–2000, so the following analysis related to streamflow and precipitation and PET in the Wuding catchment only encompassed years 1957–1996.

  • b

    The catchment area, streamflow, and sediment data for CSHC are calculated from two hydrology stations, one located at the entrance of the Yellow River into the CSHC and the other located at the exit of the Yellow River out from the CSHC (see Figure 1). Streamflow and sediment yield data did not include years 1987–1989 or 2000.

40600900(1) Huangfu32111954–200038545.848.715.2
40601500(2) Gushan13041954–200042564.420.515.7
40705000(3) Kuye92891954–200038972.5100.410.8
40603000(4) Jialu12791957–200039559.114.811.6
40801200(5) Wudinga301111957–199639639.8119.84.0
40607200(6) Shiwang23271959–200057135.12.20.9
40605800(7) Xinshui40691955–200055135.216.44.0
40604100(8) Sanchuan41231957–200049757.419.34.7
40602300(9) Weifen15481956–200048041.37.95.1
40601700(10) Zhujia29561957–200045010.212.84.3
40103400, 40104250(11) CSHCb1296541954–200045641.97245.6

[11] The Mann-Kendall rank correlation coefficient [Mann, 1945; Kendall, 1975], commonly used to test for trends in hydrometeorological time series [see, e.g., Chiew and McMahon, 1993; Burn and Hag Elnur, 2002; Xiong and Guo, 2004; Zhang et al., 2008], has been used in this study. This method is a nonparametric trend detection algorithm recommended for use by the World Meteorological Organisation [1988] as a standard procedure for detecting trends in hydrological data and assumes serially independent data. The Mann-Kendall test statistic (S) is given by

equation image

where

equation image

and n is the data set record length; xj and xk are the sequential data values. When S is positive, a positive trend is present and vice versa. Under conditions of no serial correlation in the data, existing formulae can be used to assess the significance of the trend using standard z score methods. However, serial correlation is common in many hydrologic data sets including those examined here (the lag 1 serial correlation coefficients for annual sediment yield and concentration are listed in Tables 3 and 4, respectively). Serial correlation has the effect of increasing the variance of the Mann-Kendall test statistic [Yue et al., 2002], which in turn leads to rejections of the null hypothesis at a rate above what would be expected to occur by chance for a given significance level [Bayazit and Önöz, 2007]. The method of “prewhitening” has been proposed to deal with the effect of serial correlation [see, e.g., Burn and Hag Elnur 2002], however this has been criticized by Yue et al. [2002] and Yue and Wang [2002] on the grounds that prewhitening alters the detectability of trends. Yue and Wang [2004] have proposed an alternative method for removing the effect of serial correlation however the task is complicated by the fact that trends and serial correlation are related.

Table 3. Analysis of Annual Sediment Yield Data Using the Mann-Kendall, Sen, Pettitt, and Kolmogorov-Smirnov Testsa
CatchmentAnnual Sediment Yield
Lag 1 CorrelationMann-Kendallβ (103 t km2 a−1)Change PointKS Test
SBS PercentYearpDp
  • a

    Statistically significant values at the α = 0.05 significance level are listed in bold. The column BS Percent shows the percentile corresponding to S within the bootstrapped sampling distribution where the lag 1 serial correlation is within ±0.05 of the lag 1 serial correlation listed in the second column. Values less than 5% indicate a statistically significant reduction in annual sediment yield.

1−0.11−2730.15−0.2519820.090.400.04
2−0.07−2490.91−0.3019790.050.48<0.01
3−0.05−2241.27−0.1919790.110.350.12
40.32−3630.01−0.421977<0.010.68<0.01
50.32−3410.18−0.091971<0.010.69<0.01
60.16−4480.01−0.031982<0.010.83<0.01
70.13−4040.01−0.121979<0.010.70<0.01
80.23−3860.01−0.121978<0.010.73<0.01
90.29−4690.01−0.181981<0.010.87<0.01
100.14−3280.10−0.101982<0.010.58<0.01
11 (whole of CSHC)0.19−3140.17−0.111979<0.010.53<0.01
Table 4. Analysis of Annual Sediment Concentration Data Using the Mann-Kendall, Sen, Pettitt and Kolmogorov-Smirnov Testsa
CatchmentAnnual Sediment Concentration
Lag 1Mann-Kendallβ (103 t km−2 mm−1 a−1)Change PointKS Test
CorrelationSBS PercentYearpDp
  • a

    Note that the year in brackets under the change point column shows the change point identified from the annual sediment yield data. Statistically significant values at the α = 0.05 significance level are listed in bold. The column BS Percent shows the percentile corresponding to S within the bootstrapped sampling distribution where the lag 1 serial correlation is within ±0.05 of the lag 1 serial correlation listed in the second column. Values less than 5% indicate a statistically significant reduction in annual sediment concentration.

  • b

    Note that a highly significant difference between the means of the before and after annual sediment concentration distributions was identified by a t test (p = 0.007) for the entire CSHC.

1−0.10−6625.72−0.00071984 [1982]1.060.260.44
2−0.23−4431.50−0.00051977 [1979]0.980.310.22
3−0.25−9713.69−0.00081979 [1979]0.840.280.31
40.24−2033.86−0.00361975 [1977]0.010.52<0.01
50.25−2252.88−0.00111971 [1971]0.020.56<0.01
60.12−3060.05−0.00071978 [1982]<0.010.60<0.01
70.17−2681.48−0.00151977 [1979]0.020.400.05
80.19−3010.25−0.00141978 [1978]<0.010.55<0.01
90.44−4240.04−0.00331982 [1981]<0.010.74<0.01
100.10−4360.01−0.00591982 [1982]<0.010.92<0.01
11 (whole of CSHC)0.03−1408.05−0.00071979 [1979]0.170.380.10b

[12] Rather than attempting to remove the effects of serial correlation on the data prior to calculating the Mann-Kendall statistic, which can be difficult when a trend exists, we have adopted a bootstrap-based procedure that modifies sampling distribution of the Mann-Kendall test statistic to reflect the degree of autocorrelation present in the data. For each data set of length n, n samples were drawn at random with replacement and the lag 1 serial correlation calculated. If the lag 1 correlation coefficient was within ±0.05 of the lag 1 serial correlation coefficient of the original data, then the Mann-Kendall test statistic S* was calculated (the asterisk denotes a bootstrapped statistic). This process was repeated until 10,000 S* values were obtained for each data set. The empirical cumulative distribution function was then calculated from these 10,000 S* values. This distribution represents a bootstrapped sampling distribution of S under conditions of lag 1 serial correlation approximately equal to the lag 1 serial correlation of the original data (as distinct to the standard sampling distribution under the assumption of no serial correlation). Finally, the percentile corresponding to S = S* is read off and these percentiles are reported in Tables 3 and 4 under the BS percent (i.e., bootstrap percentile) column. Here, we test the hypothesis (H0) that there has been no negative change in either annual sediment yield or annual sediment concentration. Under normal hypothesis testing procedures, if S is less than the 100 × α percentile value of the bootstrapped distribution (for a one sided test), H0 can be rejected and a the conclusion of a statistically significant negative trend being present in the data can be drawn.

[13] The nonparametric median–based linear model method proposed by Sen [1968] has been used to fit trend slopes, β, to the data:

equation image

for all (xj, xk) pairs, where 1 ≤ k < jn and n is the number of known values. The slope estimate β is essentially the median of all possible combinations of pairs for the whole data set.

[14] The nonparametric method developed by Pettitt [1979] was used in this study for detecting a change point in the data. This algorithm was assessed by Kundzewicz and Robson [2004] in their methodological review as being “robust to changes in distributional form and relatively powerful” and examples of its use in hydrological studies can be found in work by Tu et al. [2005], Gilles et al. [2006], Zhu et al. [2008]. Pettitt's test detects a significant change in the mean of a time series when the exact time of the change unknown. The test used a version of the Mann-Whitney statistic Ut,N, that examines whether two samples x1, …, xt and xt+1, …, xN are from the same population. The test statistic Ut,N is given by

equation image

where sgn(θ) = 1 if θ > 0; sgn(θ) = 0 if θ = 0; sgn(θ) = −1 if θ < 0.

[15] The test statistic counts the number of times a member of the first sample exceeds a member of the second sample. The null hypothesis of Pettitt's test is the absence of a changing point. Its statistic k(t) and the associated probabilities used in the significance testing are given as

equation image

and

equation image

[16] Additionally, the nonparametric Kolmogorov-Smirnov test has been used to examine whether the distributions of area-specific sediment yield and annual concentration differ either side of change points identified by the Pettitt [1979] test. Let Fx be the empirical probability density function for X where X is the set of n observations in the before period. Fx is defined as

equation image

[17] If Fy is the set of Y observations in the after change point period, defined in the same way as for Fx, then a two sided Kolmogorov-Smirnov test is calculated as

equation image

where D is the maximum deviation between the two data sets and for which a p value can be calculated to assess the significance.

4. Results

4.1. Temporal Trends in Annual Sediment Yield

[18] The Mann-Kendall trend test identified statistically significant negative trends in annual sediment yield in all 11 catchments (Table 3). The rate of change of annual sediment yield (given by β) varied from −0.03 × 103 to −0.42 × 103 t km−2 a−1 for the ten subcatchments while for the whole CSHC region, β was −0.11 × 103 t km−2 a−1. The change points for annual sediment yield occurred between 1971 and 1982 and were significant (note that the p ≤0.05 significance level is used throughout) for all subcatchments except subcatchment 3. This range in the change points is consistent with the timing of the reductions in streamflow, with values for the latter clustering in the 1970s to early 1980s [Zhang et al., 2008]. As an additional test of the variation in annual sediment yield, each subcatchment's area-specific sediment yields were divided into “before” and “after” change point periods and the distribution of values for both periods is plotted in Figure 3. For all subcatchments, the distribution of sediment yields in the before period is clearly higher than after the change point, with the ratio of annual sediment yield in before to after periods typically in the range 2 to 7, though some higher values are noted for subcatchment 6. The Kolmogorov-Smirnov test results, comparing the before and after sediment yield distributions indicate statistically significant reductions in area-specific yield either side of the change point for all subcatchments except for number 3, which had the highest p value (i.e., least significant result) for the Mann-Kendall test.

Figure 3.

Distribution of annual sediment yields for each subcatchment divided into pre-change-point and post-change-point time periods. Each catchment's change point year is shown in brackets. The Kolomogorov-Smirnov test statistic D is shown along with the p value of this statistic. The vertical bars above the “before” curve show the ratio of the “before” to “after” change point sediment yields at 0.1 proportion interval increments.

4.2. Temporal Trends in Annual Mean Sediment Concentration

[19] Annual sediment concentrations have been calculated by dividing the area-specific annual sediment loads by the area-specific annual runoff values with resulting units of tons of sediment per square kilometer per millimeter of runoff per year. Note that annual sediment concentrations expressed in such a manner can be converted to a mass per unit volume by multiplying by an appropriate catchment area. The Mann-Kendall trend test indicated that all catchments experienced a negative trend in annual mean sediment concentration, though the result was statistically significant for only seven of the eleven catchments (Table 4). For the seven catchments with statistically significant negative trends in annual sediment concentration, the rate of change ranged from −0.0007 × 103 to −0.0059 × 103 t km−2 mm−1 a−1. Change points in the annual sediment concentration data fell within the range 1971 to 1982 for stations with statistically significant trends in annual concentration and, with the exception of station 6, the change points for concentration were within ±2 2 years of the change point identified for annual sediment yield (station six lagged by 4 years). The distribution of the annual sediment concentration data is shown for each subcatchment in Figure 4, along with the Kolmogorov-Smirnov test results comparing the before and after change point distributions (based for consistency on the sediment yield change point year). The Kolmogorov-Smirnov test results are consistent with the Mann-Kendall trend results in terms of identifying statistically significant changes in annual sediment concentration (note that for consistency, the change point year adopted for each subcatchment are those used for the annual sediment yield tests listed in Table 3). Annual sediment concentration distributions for stations 1, 2 and 3 are very similar in the before and after periods, indicating that the soil conservation works appear to have had minimal impact on the ability of a unit of runoff to entrain sediment within these catchments. For the whole CSHC study area (station 11), no significant reduction in sediment concentration was detected by the Kolmogorov-Smirnov test (p = 0.10), despite the differences in annual sediment concentration values in the before and after periods evident in Figure 4. The lack of a significant reduction in annual sediment concentration for the whole CSHC region appears, in part, to be an artifact of the lower statistical power of the nonparametric Kolmogorov-Smirnov test relative to its parametric alternatives. The annual concentration data for the entire CSHC were fairly normally distributed, suggesting the parametric t test would be a valid alternative to the nonparametric Kolmogorov-Smirnov test. When the before and after period concentration data are tested with a t test, a highly significant difference is detected (p = 0.007) between the means of the two time periods. Nonsignificant t test results were also obtained for subcatchments 1–3, while significant differences in the mean values were also detected by t tests for catchments 4–10. That is, apart from subcatchment 11, the t test and Kolmogorov-Smirnov test results were consistent. In cases where the test results conflict, the greater statistical power of the t test relative to the Kolmogorov-Smirnov test implies that the results of the former should take precedence.

Figure 4.

Distribution of annual mean sediment concentration for each subcatchment divided into pre-change-point and post-change-point time periods. The change point year adopted is shown in brackets. Note that the change points are those of sediment yield which generally agree closely with the change points for sediment concentration. The Kolomogorov-Smirnov test statistic D is shown along with the p value of this statistic. The vertical bars are as in Figure 3.

4.3. Daily Sediment Concentration

[20] Thus far it has been demonstrated that, on an interannual timescale, there have been statistically significant reductions in sediment yield for all subcatchments within the CSHC study area and the timing of these sediment yield reductions are consistent with the timing of streamflow reductions. While all catchments show a negative trend in annual mean suspended sediment concentration, the reduction in concentration is not in all cases statistically significant. Thus two scenarios exists, where, at the interannual timescale, some catchments show (1) a significantly lower sediment yield which can be attributed to significant reductions in both streamflow and sediment concentration and (2) a significant reduction in annual sediment yield associated only with a significant decline in streamflow.

[21] Previous research has suggested that sediment transport in catchments of the Loess Plateau is strongly event driven [Zhang et al., 1997; Shi and Shao, 2000], implying that the annual data examined above may potentially be masking important event-based detail. To explore this, sediment yield and concentration data for catchments two (the Gushan River) and nine (the Weifen River), illustrative of the two scenarios listed above, are examined at a daily sampling resolution. Both are observed to have experienced a statistically significant reduction in annual sediment yield in their after periods. While the Mann-Kendall, change point and Kolmogorov-Smirnov tests all indicated that station nine experienced a statistically significant decline in annual sediment concentration, no significant change in annual sediment concentration was identified for station two by any of the statistical tests. As sediment concentration is widely known to be positively related to discharge (at least at the macroscale), it is necessary to consider the distribution of sediment concentration with respect to streamflow.

[22] Figure 5a shows the distribution of daily sediment concentration measurements with respect to streamflow for the May to October wet season period for station two on the Gushan River for a selection of years occurring prior to and including the change point year of 1979, and after the change point (5(b)). As is commonly observed, sediment concentration and streamflow are positively related and each varies by several orders of magnitude. At moderate streamflow (100–101 m3 s−1) and for both time periods, the variability in sediment concentration is highest, spanning 4 orders of magnitude. However, at larger streamflows, both before and after distributions converge within the range 102 to 103 kg/m3 of sediment.

Figure 5.

(a) Mean (solid line) and median (dashed line) suspended sediment concentrations for station 2 at 40 discharge classes for the period up to 1979. (b) As for Figure 5a but for the period after 1979. (c) Mean and median sediment rating curves for before and after the 1979 period. The inset box plots show before and after sediment concentration for days where discharge ≥10 m3 s−1. Also shown (right-hand axis) is the proportion of the total May–October wet season load transported in the 5 days of highest flow within each year and the corresponding range in discharge for these 5 days for both the before (circles) and after (crosses) periods.

[23] For both before and after periods, the mean and median sediment concentration (calculated using nontransformed concentration data) at 40 equally spaced discharge classes (defined on the log-discharge scale) have been calculated. These curves represent a form of sediment rating curve for the catchment. If the soil conservation measures implemented in the CSHC catchments had reduced the capacity for sediment to be mobilized in the landscape (as the hypothesis proposed in the Introduction would suggest), a reduction in the sediment concentration at a given streamflow would be expected in the after period. However, Figure 5c shows that the mean sediment concentration curves are very similar for both time periods. The median concentration in the after period is actually higher in the range 100 to 101 m3 s−1, but at higher streamflows the median sediment concentration trends converge for the two time periods. Thus for station two, even when examining daily data, there is no evidence that catchment management actions have reduced the mobility of sediment within this catchment. This is consistent with the lack of any significant reduction in annual sediment concentration identified above.

[24] It must be recognized however that much of the sampled data presented in Figure 5 is from low-flow, ambient conditions. It is commonly observed that a large proportion of total fluvial sediment loads move during relatively short-duration floods when both streamflow and sediment concentration are highest. In such cases, much of the low-flow detail evident in Figure 5 is of little consequence for the estimation of annual sediment loads; the most critical examination should be on variations in high flow sediment concentrations as these are the most critical determinants of total sediment yield. Figure 5c also demonstrates that high flows do dominate the entire wet season (and hence annual) sediment yield. On the right-hand vertical scale, the percentage of the wet season sediment load transported in the top 5 days of each wet season is shown while on the horizontal scale, the streamflow range associated with these top 5 days is indicated. More than 50%, and in some cases more than 90% of the total wet season sediment load is transported within 5 days of maximum streamflow, which typically exceed 10 m3 s−1 for both before and after periods. Clearly, attention needs to be focused on changes in sediment concentration at streamflow >10 m3 s−1 to establish whether variations in sediment concentration can account for any reductions in total sediment yield for this station. Indeed, at streamflow exceeding 10 m3 s−1, the before and after sediment rating curves (either mean or median) are indistinguishable and the univariate distribution (box plots) of sediment concentration for these high-flow samples are also similar. Thus, for station two there is no evidence that the mean or median discharge-specific sediment concentration has improved in the post-1979 period either under ambient, low-flow conditions, or at the high streamflows that transport the majority of the wet season sediment load. This result, drawn from the higher temporal resolution daily data, confirms the conclusion drawn from the annual sediment yield and concentration data. That is, the reduced sediment yield of the post-1979 period for catchment two appears to be largely a product of the reduction in streamflow, though it is worthwhile to note that 61% of the observed reduction in streamflow is estimated to be due to the impact of soil conservation measures on catchment hydrology [Zhang et al., 2008].

[25] The same analysis has been repeated for station 9, the Weifen River catchment, but using the sediment load change point year of 1981, with the results shown in Figure 6. A moderate increase in both mean and median sediment concentration is apparent for streamflow in the range 100 to 101 m3 s−1 in the post-1981 period, however, beyond 2 × 101 m3 s−1, which are the streamflows where most of the sediment load in each wet season is transported, the two mean rating curves are broadly similar though there is some evidence that the median rating curve may be lower in the after period. Viewed on the log scale, this suggests that water quality (as represented by either the mean or median fitted curves) has remained broadly constant in the before and after periods (or at least has not varied by an order of magnitude for example) and that any changes in sediment load are primarily related to changes in streamflow. However, this conclusion, which is counter to the finding listed in Table 4 that suggested there was a statistically significant reduction in annual sediment concentration, is incorrect. The univariate analysis of sediment concentration for streamflow >101 m3 s−1 (i.e., those streamflows associated with the majority of annual sediment transport) shows that for the post-1981 period, these suspended sediment measurements are from a more compact, lower-magnitude distribution. The median concentration of these high-flow data is ∼40% of the pre-1981 median value; this result is consistent with that drawn from the annual sediment concentration distribution shown in Figure 4 where annual concentrations in the before period were 2 to 4 times greater than those in the after period. Both the mean and median rating curves fail to capture (on account of the skew in the distributions) the important cluster of high-concentration observations (>102 km/m3) at the main flood carrying streamflows evident in the before-period data. The conclusion of a reduction in sediment concentration drawn from the annual data can, when viewed in this manner, be seen to be concordant with that drawn from the daily observations. The conclusion that the soil conservation measures implemented in catchment nine have contributed to reducing total catchment sediment yield by both reducing river flow and also by reducing sediment mobility within the landscape is thus supported.

Figure 6.

As for Figure 5 but with change point set at 1981.

4.4. Soil Conservation Measures and Changes in Annual Sediment Concentration

[26] Figure 7 examines the relationships between percentage area treated by the four main management actions (and the total treated area) and whether or not a statistically significant reduction in annual sediment concentration has been observed for the eleven study catchments. Of the four “treatments” and total treated area indices, statistically significant variations between the means of the yes and no classes were only detected by a Mann-Whitney U test for the terrace and pasture management classes (with p = 0.024 in each case). When considering the yes/no data distributions (as opposed to the class means), a statistically significant difference was only detected for the pasture treatment (p = 0.03) according to a two sided Kolmogorov-Smirnov test; the p value for the terrace treatment was marginally not significant at 0.067.

Figure 7.

The percentage area of each subcatchment affected by different (and total) land management actions in 1996 plotted against whether a statistically significant reduction in annual sediment concentration was observed (at p = 0.05).

[27] The finding that lower-percentage area of pasture revegetation is associated with significant reductions in sediment concentration is surprising and somewhat counterintuitive, as presumably pasture reestablishment would have been undertaken to reduce soil erosion. This unexpected result may reflect intercatchment landscape variations as the area of terracing and pasture reestablishment are negatively (though not significantly) correlated (r = −0.35, p = 0.29), suggesting that catchments more suited to terracing were perhaps less suitable for pasture reestablishment in the first place. However, the general lack of significant differences in the percentage area treated between the yes and no sediment concentration reduction classes suggests two things. Firstly, with n = 11 samples, there is a clear need for a more extensive database to more robustly (in a statistical sense) characterize the importance of management effects. Secondly, it may not solely be the area treated by a particular management action that is the key determinant of whether sediment concentrations are reduced, but perhaps the geographical distribution of the treatments within a catchment, or alternatively other catchment characteristics that have not been measured here may be stronger influences. For example, Xu and Yan [2005] have shown that area-specific sediment yields vary downstream as Loess thickness varies; influences such as this may also impact on the differing efficacy of treatment actions in the eleven study catchments. Clearly this area is one worthy of future research.

4.5. Relative Impacts of Precipitation Changes and Soil Conservation Works on Sediment Yield

[28] To examine the relative importance of soil conservation works versus rainfall variation on the changes in sediment yield from the eleven study catchments, linear regression models were fitted to the square root (for normalization purposes) of the annual sediment yield data (QS) as a function of rainfall (rain) and a dummy variable (epoch) indicating before (epoch = 0) and after (epoch = 1) conditions respectively (as based on the change points listed in Table 3):

equation image

[29] The variable epoch is taken to be a surrogate variable representing the combined effects of the soil conservation works. While the soil conservation works were introduced progressively, for the purposes of this analysis they are treated as a binary variable. In cases where a negative change in sediment yield occurs, b < 0. Note that in this case the intercept for equation (8) is by default 0 when epoch = 0 (i.e., the before period). The fitted curves and parameters for equation (8) are shown in Figure 8. As expected, the yield in the after period is significantly lower (i.e., b is negative and significantly different from zero) in all cases for a given rainfall. The question examined now is the magnitude of the vertical offset between the epoch = 0 and epoch = 1 curves (i.e., b) relative to the change in mean annual rainfall (▵equation image) between the before and after periods. To express this in consistent units, the offset b is converted to units of precipitation equivalent (ΔPeq) by setting equation image = 0 and solving equation (8) for rain when epoch = 1:

equation image
Figure 8.

Linear regression models for annual sediment yield incorporating before and after change point periods as a [0,1] factorial variable. The R2 value pertains to the fitted model, and r0 and r1 are the Pearson correlation coefficients for the before and after periods, respectively. The solid triangle shows the magnitude of the change in mean annual rainfall between the two time periods.

[30] Here ▵Peq represents the reduction in rainfall required to obtain a yield reduction equivalent to b assuming epoch = 0 (i.e., assuming no soil conservation practices). This term is indicated graphically in Figure 8 by the rainfall at the point where the after curve intersects the x axis. The value of ΔPeq can be compared to the magnitude of the observed reduction in mean annual rainfall, also shown in Figure 8 by the solid triangle. The percentage contribution of the soil conservation works on the observed changes in sediment yield (SC%) can be calculated by taking the ratio of these terms and converting to a percentage:

equation image

[31] The values of Δequation image, ΔPeq, SC% are listed in Table 5. In most catchments, the soil conservation measures can be seen to account for the majority (64% to 89%) of the reduction in annual sediment yield. This result is consistent with the finding of Wang et al. [2007] who concluded that decreases in precipitation are responsible for 30% of the reduction in sediment yield at Huayuankou, downstream of the CSHC study area. However, there are three catchments where the effects of the soil conservation measures appear to be lower than the reduction attributable to changes in rainfall, namely catchments 2, 7 and 11. In the case of catchment 2, none of the other statistical tests suggested that there had been any significant change in annual sediment concentration, reinforcing the notion that any changes in yield from this catchment were largely due to rainfall variations. For catchments 7 and 11, the results of the Kolmogorov-Smirnov test for significant differences in the distribution of annual sediment concentration were equivocal as to whether differences existed, again this is consistent with the roughly equivalent effects of soil conservation measures and rainfall in affecting sediment yields in these catchments.

Table 5. Percentage Influence of the Soil Conservation Measures on the Observed Reductions in Annual Sediment Yieldsa
CatchmentΔequation imageΔPeqequation imageSC %
  • a

    SC% is soil conservation measures. Note that 1 − SC% represents the percentage influence of precipitation variations. See text for notation definition.

113.7108.77.987
263.6101.11.637
321.5100.44.779
453.7177.93.370
545.4126.72.864
668.1266.33.974
792.6168.91.845
835.0174.75.080
929.7271.49.189
1028.6224.87.987
1151.996.21.946

5. Discussion

[32] Walling [1983] observed that worldwide, the sediment yield per unit catchment area normally decreases as catchment area increases. The rate of decline represented a drainage basin's sediment delivery ratio. This negative trend was explained by the erosion of sediment from upland areas and the subsequent storage of this mobilized sediment upon downstream floodplains, and in more recent times, reservoirs and other flow impoundments. Walling [1983] noted that the relationship between area-specific sediment yield and catchment area for the middle Yellow River, using data drawn from Gong and Xiong [1980], did not show a negative trend. More recently, Xu and Yan [2005] have shown that a negative trend does indeed exist for the Yellow River basin, but only for catchment areas greater than approximately 2000 km2. Figure 9 shows the relationship between mean area-specific sediment yield and catchment area for the 11 CSHC subcatchments reported here for the 1960–1969 period. The curve fitted to these data plus the 1950s to 1970s Sanmenxia station yield is shown in black. The Sanmenxia yield is indicative of the maximum basin yield, as downstream of this point, the river progressively loses sediment onto the North China Plain. The 1990–1999 period yields from this study are also shown, along with a curve fitted to these data and the yields of the 1990s and 1980s from Huayuankou and Sanmenxia stations respectively. For comparison, sediment yields from other Yellow River basin studies presented by Xu and Yan [2005] are also shown. When considering the 1960s period (i.e., the black curve), the CSHC data do indeed show a negative trend of slope −4.15 × 103 t km−2 a−1 km−2, consistent with the 1950–1970 yield data of Xu and Yan [2005]. The linear trend fitted to the 1990s data (along with the later period Huayuankou and Sanmenxia yields) illustrates the magnitude of the sediment yield reduction in recent decades, though there is clearly some overlap with the 1950s–1970s yields of Xu and Yan [2005]. The slope of the regression curve is also lower in the later period, at −1.10 × 103 × 103 t km−2 a−1 km−2. This reduction in slope reflects a greater reduction in area-specific sediment yield for smaller catchment areas relative to larger catchment areas in both the CSHC study area and also apparently for the entire Yellow River basin when comparing the 1960s period with the 1990s. Notably, the yields at Sanmenxia and Huayuankou gauging stations appear to be close to or below the area-specific sediment yield of the last 2500 years, a point previously noted by Wang et al. [2007].

Figure 9.

Area-specific sediment yield plotted against catchment area for the Yellow River basin. Data shown are mean annual yields for the eleven CSHC examined here for the decades 1960–1969 and 1990–1999; “other” 1950–1970 mean annual yields for Yellow River catchments with areas <50,000 km2 from Xu and Yan [2005]; whole-of-basin sediment yield estimates derived from floodplain aggradation measurements from Shi et al. [2002] for the last 2.5 thousand years; the 1950s to 1970s sediment yield from the Sanmenxia gauging station (note that this value is that described by the authors as that “commonly cited”; we have been unable to identify the definitive source of this value [Walling and Fang, 2003]); the 1980s Sanmenxia yield from Walling and Fang [2003]; and the Huayuankou gauging station yield from the 1990s from Liu et al. [2006]. The black regression line is fitted to the 1960–1969 CSHC data and the 1950s–1970s Sanmenxia point. The gray regression line is fitted to the 1990–1999 CSHC data and the Huayuankou 1990s and 1980s Sanmenxia yields.

[33] The greater yield reduction for the smaller catchments arguably reflects the greater ease of trapping sediment at smaller catchment areas in structures such as terraces and sediment trapping dams, as well as the influences of catchment revegetation, notwithstanding the apparent poor correlations noted above. For larger catchments, remobilization of existing in-channel sediment stores can represent an important sediment source which could continue to maintain high fluvial sediment loads even if the nonchannel sediment sources (i.e., hillslope and gully erosion) were reduced substantially. Indeed, the very high sediment yields of Loess Plateau rivers in the “presoil conservation era,” coupled with the steeper specific sediment yield versus catchment area curve of Figure 9, implies that considerable sediment storage occurred in river system during the middle of the twentieth century. If remobilization of stored sediment was occurring, it should manifest as systematic channel enlargement in recent years.

[34] Yang et al. [2002] examined the temporal variations in sediment concentration of the Yangtze River, between 1951 and 2000. After an increase in mean-decadal suspended sediment concentration from the 1950s to the 1960s at the Datong gauging station (located on the river's lower reaches), suspended sediment concentration declined, with the 1990s value being approximately 60% of the 1960s concentration. This ratio is broadly consistent to that observed for the entire CSHC study area when comparing the before and after period concentrations shown in Figure 4. Yang et al. [2002] attributed the lower annual sediment concentration in the Yangtze River to increased river discharge (i.e., an opposite hydrologic trend to that observed in the Yellow River). As total sediment loads also declined, this implies that sediment mobility in the Yangtse River basin become increasingly restricted, which Yang et al. [2002] suggest was related to a large increase in reservoir capacity in the catchment since the 1970s. Other landscape management effects such as terracing and revegetation were not examined in their study. The conclusion that human management actions have played a role in reducing both the sediment yields and concentrations is becoming increasingly clear in these two major Chinese rivers.

6. Conclusions

[35] This study examined the role that changes in streamflow and land management have played in altering the sediment yields of selected tributaries to the Yellow River originating in the Loess Plateau over the period ∼1950 to 2000. Statistically significant negative trends of −0.03 × 103 to −0.42 × 103 t km−2 a−1 in annual sediment yield were detected in the ten study catchments, plus the entire Coarse Sandy Hilly Catchments region. Statistically significant change points in annual sediment yield occurred between years 1971 and 1982 in nine of the eleven catchments. To discriminate between reductions in sediment yields due to reduced precipitation and catchment management actions aimed at reducing soil erosion, annual sediment concentration data were also examined. For seven of the ten subcatchments plus the entire Coarse Sandy Hilly Catchments region (if a change in mean is considered), statistically significant reductions in annual sediment concentration were detected and it was estimated that catchment management practices aimed at reducing soil erosion contributed 64–89% of the reduction in annual sediment yield for these subcatchments. Change point years for annual sediment concentration were generally consistent with the change point years identified for annual sediment yield.

[36] Examination of daily sediment concentration and yield data from two catchments (one each that did and did not experience significant reductions in annual sediment concentration) revealed that 50–99% of each year's sediment load was typically transported by the 5 days per year with maximum stream flow. Variations in the sediment rating curves for the two catchments showed little difference between before and after change point periods as the bulk of the suspended sediment data were from low-flow conditions. However, when suspended sediment concentrations under the high-flow conditions that transport most of the sediment load were examined, the daily data supported the conclusions about variations in suspended sediment concentration drawn from the annual data.

[37] Correlations between the percentage area affected by different land management actions and the existence of statistically significant reductions in annual sediment concentration were generally not significant. This arguably indicates that the spatial organization of land management actions may be more important than simply the magnitude of the area affected, or that other metrics representing the soil conservation works may need to be considered. Lesser reductions in area-specific sediment yield at larger catchment areas potentially indicate that larger rivers are accessing in-channel sediment stores, which even if other diffuse catchment sources were reduced substantially, could potentially sustain high fluvial suspended sediment yields for many years into the future.

Acknowledgments

[38] Tim McVicar and three anonymous reviewers are thanked for comments on the manuscript. Francis Chiew is thanked for the idea of redefining the sampling distribution of the Mann-Kendall statistic to account for autocorrelation. X. Zhang would like to thank the Chinese Academy of Sciences for funding obtained through “The impact of land use/land cover change on the streamflow and sediment in He-Long region in the Loess Plateau, China” (B183) project and the Chinese Scholarship Council for their support. Lu Zhang would like to acknowledge the Chinese Academy of Sciences for their support through the Outstanding Overseas Chinese Scholars Fund.

Ancillary