2. MATERIALS AND METHODS
2.1 Description of study area
The Kuye River (Fig. 1 ) is a first-order tributary of the middle Yellow River, which flows through Inner Mongolia and Shaanxi and finally flows into the Yellow River. The area of the Kuye River watershed (38°22′–39°50′ N, 109°28′–110°45′ E) is 8651 km2, with a mainstream length of 242 km and an average channel slope of 2.6‰. The elevation ranges from 713 m in the southeast to 1575 m in the northwest. There are two main tributaries (Wu-lan-mu-lun and Bei-niuchuan) in the upper reach of the watershed, and the Wenjiachuan hydrometrical station is located at the outlet of the Kuye River. The watershed has a typical arid to semiarid continental climate, with a multiyear average precipitation and temperature of 415 mm and 7.9°C, respectively. The average annual evaporation is 1788 mm. Intense storms mainly take place in July and August, accounting for more than 52% of the annual rainfall. Therefore, local short-term floods often occur in the watershed during this period. According to the measured data of the Wenjiachuan hydrological station, the sediment load in July and August accounts for 90% of the annual total (Cai et al., 2019).
Due to the sparse vegetation cover, loose soil, terrain fragmentation and dense gullies in this watershed, it has become one of the main sources of sediment in the Yellow River. The average annual sediment load measured at the Wenjiachuan hydrological station from 1954 to 2000 was 1.00×108 t (10800 t/km2/a) (Rustomji et al., 2008). However, the observed sediment load decreased significantly over the past six decades. Compared with the 1960–1999 period, the average annual runoff and sediment discharge from 2000–2016 decreased by 76.72% and 94.50%, respectively (Zhao et al., 2019).
2.2 Datasets
The daily precipitation records of 50 rainfall stations and the daily sediment concentration data of the Wenjiachuan gauge in 1987 and from 2006-2016 were obtained from the “Hydrological Yearbook of the People’s Republic of China - Hydrological Data of the Yellow River Basin”.
The digital elevation map (DEM) came from the Geospatial Data Cloud, with a spatial resolution of 30 m (www.gscloud.cn). The DEM of the Kuye River watershed was extracted with the hydrological analysis tool in ArcGIS 10.2 (ESRI).
The land use scenarios for 1987 and 2006 were derived from the Landsat 5 Thematic Mapper (TM) remote sensing images downloaded from the United States Geological Survey (USGS) (https://glovis.usgs.gov/), while those for 2016 were from the Geospatial Data Cloud website (www.gscloud.cn). Then, a hybrid land use classification technique, which involved unsupervised classification and visual interpretation, was used. Unsupervised classifications were carried out using the iterative self-organizing data analysis (ISODATA) clustering algorithm, while visual interpretation was mainly employed using Google Earth’s historical orthophoto images of corresponding years. To ensure the quality and accuracy of the data, field investigations and in-depth consultations with local elders were undertaken. Seven land use types were identified: forests, grassland, shrubland, bare land, arable land, water bodies and urban and mining areas.
2.3 Model description
The sediment delivery distributed (SEDD) model has been extensively used in different regions of the world. The model is mainly based on the RUSLE model and integrates GIS techniques to predict the sediment yield within a watershed. The SEDD (Ferro and Minacapilli, 1995) model estimates the sediment yield according to the following formula:
\(\text{SY}_{i}=\text{SE}_{i}\bullet\text{SDR}_{i}=R_{i}\bullet K_{i}\bullet\text{LS}_{i}\bullet C_{i}\bullet P_{i}\bullet\text{SDR}_{i}\)(1)
where SEi is the average of annual soil erosion modulus (t/ha/a) on pixel i ; SDRi is the sediment delivery ratio for pixel i ; Rirefers to the rainfall-runoff erosivity factor (MJ·mm/ha·h);Ki denotes the soil erodibility factor (t·ha·h/MJ·ha·mm); LSi expresses the topographic factor; Ci signifies the cover-management factor; and Pi shows the support practice factor.
To explain the relationship between rainfall and erosion, Wischmeier and Smith (1978) proposed the R factor to quantify the impact of rainfall and runoff on soil erosion. Since the daily rainfall data of 50 rainfall stations were available, the method proposed by Zhang et al. (2002) was used to calculate the R factor, which was mainly assessed with the daily rainfall data using a half-month rainfall erosivity model. The annual R factor was generated by kriging interpolation using data from 50 rainfall stations, and the raster of the R factor in 2006 is shown inFig. 2 a.
The soil erodibility factor was applicable to characterize the sensitivity of soil to water erosion (Rao et al., 2014). Based on the soil map (1:500,000 scale) of the Loess Plateau and the revised erosion/productivity impact calculator (EPIC) model (Williams et al., 1984), the value of the soil erodibility factor was derived (Fig. 2 b).
The LS factor is an index for measuring the impact of topography on soil erosion. As the Kuye River watershed was located on the Loess Plateau, which has complicated terrain, the terrain factor calculation tool 2.0 developed by Fu et al. (2015) for the Loess Plateau was imposed to calculate the LS factor (Fig. 2 c). The DEM was adopted as the main data source for this calculation tool.
The cover-management factor reflects the effect of soil management on soil erosion rates, which are mainly related to land use type, vegetation cover, surface roughness and soil moisture (Renard et al., 1997). In this paper, the method proposed by Cai et al. (2000) was used to evaluate the C factor by integrating the normalized vegetation index (NDVI), which was obtained from remote sensing images of different periods of the Kuye River watershed (Fig. 2 d).
The P factor represents the influence of supporting measures on sediment control, which is generally assigned according to previous research results combined with land use types (Zhou et al., 2019). However, compared with other soil and water conservation activities, check dams played a dominant role in sediment retention (Ran et al., 2008). To quantitatively analyze the influence of the check dam on sediment load, the trapping efficiencies of the dams were applied to evaluate the P factors. The detailed calculation process was based on the method proposed by Zhao et al. (2017).
According to Ferro and Porto (2000), the SDR value of each grid cell is calculated as follows:
\(\text{SDR}_{i}=exp\left(-\beta\bullet t_{i}\right)=exp\left(-\beta\bullet\frac{l_{i}}{k_{i}\sqrt{s_{i}}}\right)\)(2)
where β is a coefficient that is affected by the roughness distribution along the flow path and is related to time.ti is the travel time from cell i to the nearest stream reach, and li is the flow length (m). ki is a coefficient dependent on surface roughness characteristics (m/s), and sirepresents the slope of the cell. The determination of the βvalue followed the method of Fu et al. (2006). To ensure the proper use of formula (2), the minimum value of si was set as 0.003 (Fernandez et al., 2003; Fu et al., 2006). The value ofki was extracted from previous studies (Fernandez et al., 2003; Gashaw et al., 2019).
To calibrate the model, in 2018, we selected two small check dams without a sluicing gate in the Kuye River watershed, and the sediment produced by soil erosion in the dam-controlled watershed was completely intercepted by the dams (Fig. 3 a). Trench excavation was carried out on the dam land and the depth of sediment profile of the two dams reached 4.15m and 4.05m respectively. The flood couplets were then identified, divided and its thickness were measured (Fig. 3 b). Finally, the capacity curve of check dams was reconstructed by the DEM of small watershed, and annual erosion modulus of the two dam-controlled watersheds from 2007 to 2018 were obtained by combining the soil bulk density of each flood couplets. The area-weighted average of the annual erosion modulus of the two dam-controlled watersheds was used as the annual erosion modulus for the whole Kuye River watershed from 2007 to 2018. For details of these two check dams, please refer to Zhang et al. (2020) and Wang (2020). However, it was very difficult to calibrate a model of such a large-scale watershed because the land use and the number of check dams changed continuously during the study period. To obtain a more realistic calculation, the sediment load was calculated for 1987 without considering the effect of check dams because the number of check dams and the total storage capacity were very limited (Fig. 5 ). The calculation of the annual erosion modulus and sediment load from 2006 to 2010 was based on the land use scenario in 2006 and included the effects of the check dams. The same method was applied from 2011-2016, except that the land use scenario was based on 2016. The annual erosion modulus measured from 2007-2016 was used to calibrate the RUSLE model. Additionally, the SEDD model simulation results were validated by using the observed sediment load of Wenjiachuan station in 1987 and from 2006-2016.
2.4 Distinguish the influence of different factors on soil erosion and sediment load
At the watershed scale of the Loess Plateau, soil erosion and sediment yield are mainly affected by climate variation and human activities. Human activities can be divided into land use changes affected by human activities and related soil and water conservation measures. Climate variation is mainly reflected in the change of precipitation, and check dams play a major role in sediment control in the study area (mentioned in the introduction of P factor). Therefore, we assumed that only three main driving factors (precipitation variation, land use changes and check dam construction) affect the soil erosion and sediment yield of the watershed. Taking the soil erosion and sediment load in 1987 as the reference year, the changes of soil erosion and sediment load during 1987-2006 and 1987-2016 were analyzed respectively. The calculation formula was as follows:
\({}_{T}=\left(\text{TSE}_{i}-\text{TSE}_{1987}\right)\text{\ or\ }\left(\text{TSL}_{i}-\text{TSL}_{1987}\right)=_{P}+_{\text{LU}}+_{\text{CD}}\)(3)
where T is the total change of soil erosion or sediment load.TSEi and TSLi represent the total soil erosion and sediment load in ith year, respectively.TSE1987 and TSL1987represent the total soil erosion and sediment load in 1987, respectively. P , LU andCD stand for the effects of precipitation variation (P), land use changes (LU) and check dams (CD), respectively.
The effects of precipitation variation and land use changes on soil erosion and sediment load can be distinguished by choosing different scenarios. The detailed calculation formula was shown inTable 1 . Finally, the influence of check dams can be obtained by subtracting the influences of the other two factors from the total effects.
3. RESULTS
3.1 Land use changes
Fig. 4 shows that there were dramatic changes in land use, with the main conversion of arable land and bare land to grassland, shrubland, forestland and construction land. The proportion of vegetation coverage increased from 60% in 1987 to 86% in 2016. In addition, the urban and mining areas expanded rapidly from 9.62 km2 in 1987 to 437.23 km2 in 2016 (Table 2 ). This was mainly reflected in the urban expansion of Shenmu County, Dongsheng district and Yijinhuoqi County, as well as in the development of coal mines, which showed that the population, economy and urbanization had grown rapidly in the past three decades. In contrast, after decades of implementation and promotion of the abandonment of grazing and ‘Grain for Green’ policies, the area of arable land and bare land in the study presented a gradually decreasing trend. The area of arable land decreased from 2182.62 km2 in 1987 to 533.71 km2 in 2016. The proportion of bare land decreased from 13.43% in 1987 to 0.75% in 2016, and the area decreased by 1097.24 km2.
3.2 Check dam construction
Fig. 5 shows the variation of the number and storage capacity of check dams on a time scale. The number and storage capacity of check dams showed similar change trend with time, both increased first and then decreased. In 2005, the annual number of dams and storage capacity reached the maximum, which were 42 and 41.61Mm3 respectively. The accumulated number and storage capacity of check dams increased slowly before 2000, and then increased sharply. The turning point was mainly due to the ‘Hydraulic highlight project’ launched by the Ministry of Water Resources of China, which further increased the construction speed of check dams. Before 1987, only 32 check dams had been built, with a total storage capacity of 35.20 Mm3. This shows that the interception capacity of check dams in 1987 was relatively limited. There were 306 key dams in the Kuye River watershed, with a total storage capacity of 316.64 Mm3 until 2011.
3.3 Model calibration and validation
Fig. 6 a shows the model calibration using the annual erosion modulus, which was obtained through field excavation in 2018. The annual erosion modulus ranged from 39.88 to 64.02 t/ha/a, with an average value of 50.83 t/ha/a. Correspondingly, the simulated soil erosion modulus changed from 40.99 to 58.70 t/ha/a, with an average value of 49.48 t/ha/a. The R2 reached 0.83, which indicated that the results of the model were, overall, in good agreement with the actual measured values.
Fig. 6 b shows the relationship between the measured and simulated annual sediment loads at Wenjiachuan station. The annual sediment loads in 1987 and from 2006-2016 were mainly simulated for model validation. The measured sediment load at Wenjiachuan station varied from 0.05 Mt in 2016 to 33.97 Mt in 1987, while the simulated annual average sediment load ranged from 8.55 Mt to 32.77 Mt. According to the fitting results of the scatter diagram (R2 > 0.90), the simulation results of the model were basically consistent with the measured values, but compared with the measured values in recent years, the model tended to overestimate the sediment load.
3.4 Characteristics of soil erosion and sediment yield
Table 3 shows the classification of soil erosion and sediment yield under three different land use scenarios. According to the Chinese Soil Erosion Classification and Grading Standards (SL190–2007), the soil erosion grade changed from micro erosion to severe erosion. In 1987, the areas of micro, mild, moderate, intensive, extreme, and severe erosion accounted for 29.61%, 11.56%, 12.83%, 9.64%, 12.45% and 23.91%, respectively, of the area affected by erosion. The proportion of micro, mild, moderate, intensive, extreme, and severe erosion in 2016 was 59.68%, 11.38%, 8.80%, 5.44%, 6.08% and 8.61%, respectively. The area affected by mild or less serious erosion accounted for more than 70% of the whole watershed, which indicated that the soil erosion in the watershed had been well controlled by 2016. In 1987, the area of high erosion rates (>intensive) was 3981.29 km2, accounting for 46% of the whole watershed area. In 2016, the area of high erosion rates was 1739.63 km2, accounting for 20.14% of the whole watershed area. Nevertheless, compared with in 1987, the area with high erosion rates decreased by 2241.66 km2 in 2016, and the area with severe erosion shrank by 1325.42 km2. The results showed that the three factors played a remarkable role in reducing erosion.
The sediment yield was artificially divided into three grades: low, moderate and high sediment yield (Table 3 ). The distribution area of sediment yield under the three land use scenarios was low > moderate > high. Moreover, over time, the area occupied by the low sediment yield rate increased (from 4957.94 to 7276.03 km2), while the others decreased. This indicated that sediment control in 2016 greatly improved under the influence of precipitation variation, land use changes and check dams.
The average SDR values in 1987, 2006 and 2016 were 0.32, 0.28 and 0.23, respectively, across the whole watershed, showing a decreasing trend. The results expounded that with the implementation of the ‘Grain for Green’ policy, land use changes had reduced the SDR.
Considering the check dams, we applied hot spot analysis (Getis-Ord Gi∗) in ArcGIS software to analyze the soil erosion and sediment yield of the Kuye River watershed in 1987 and 2016 (Fig. 7 ). The distribution of cold and hot spots of soil erosion in 1987 and 2016 was consistent with that of sediment yield, and we found that the hot spots of soil erosion and sediment yield were mainly clustered around Shenmu County (Fig. 1 ). This indicated that the high-risk regions of soil erosion and sediment yield were mainly concentrated in the middle reaches of the watershed. However, the hot spot region in 2016 decreased compared with that in 1987. On the other hand, most of the cold spots were identified in Yijinhuoqi County, especially the sediment yield cold and hot spot maps of 2016 were the most obvious. This signified that the erosion degree in this area was relatively low.
3.5 The influence of different factors on soil erosion and sediment yield
Table 4 represents the contribution of three factors (precipitation variation, land use changes and check dams) to soil erosion and sediment yield in different periods. Compared with 1987, the total soil erosion in 2006 decreased by 57.74 Mt, and the contribution rates of the three factors were 8.61%, 20.44% and 70.95%, respectively. In 2016, the total soil erosion decreased by 60.42 Mt, and the contribution rates of the three factors were 34.22%, 8.01% and 57.77%, respectively. This signified that check dams were the dominant factor, but its contribution rate to different years was diverse. In the past three decades, the impact of precipitation variation on the reduction of soil erosion had increased (from 8.61% in 2006 to 34.22% in 2016), while the impact of land use changes and check dams on soil erosion had decreased.
When compared with 1987, the total sediment load in 2006 and 2016 decreased by 21.50 Mt and 24.22 Mt, respectively, and the sediment load reduction in 2016 was greater. From 1987 to 2006, precipitation variation, land use changes and check dams reduced sediment load by 11.91%, 34.32% and 53.77% respectively. From 1987 to 2016, the contribution of these three factors to sediment reduction was 29.10%, 40.09% and 30.81%, respectively. The check dams were still the dominant factor in 2006, but the three factors all had obvious influence on sediment load in 2016. The influence of precipitation variation and land use changes on sediment load was increasing, while the influence of check dams on sediment load was decreasing.
4. DISCUSSION
4.1 Reasons for the change of contribution of three factors to soil erosion and sediment load reduction
By using the model to simulate different scenarios, we quantitatively distinguish the impact of three different factors on soil erosion and sediment load in different periods. In order to further explore the reasons for the change of contribution rate of precipitation variation to soil erosion and sediment load reduction, we separately analyzed the influence of precipitation variation on soil erosion and sediment yield (Fig. 8 ). The abscissa from low to high corresponds to the average R in 2016, 2006 and 1987 (903.58, 1117.52 and 1140.64 MJ·mm/ha·h). With the increase of precipitation, soil erosion modulus and sediment yield increased. Due to the largest reduction in precipitation in 2016, the impact of precipitation variation on soil erosion and sediment load also reached the maximum (Table 4 ). According to the slope of the graph, the impact of precipitation variation on erosion modulus is greater than that on sediment yield (Fig. 8 ). Therefore, compared with the impact on total sediment load reduction, precipitation variation has a greater impact on total soil erosion reduction (Table 4 ). In this study, the contribution of precipitation variation to sediment load reduction mainly changed from 11.91% to 29.09%, which was similar to the previous studies (Mu et al., 2012; Yang et al., 2018).
To further ascertain the primary factors leading to soil erosion and sediment yield reduction, we analyzed the contributions of different land use types to soil erosion and sediment yield (Fig. 9 ). In order to eliminate the impact of precipitation variation, we simulated all scenarios using the 1987 rainfall conditions. Without considering the effects of check dams, the amount of soil erosion in 1987 was mainly contributed by grassland, arable land and bare land, accounting for 40.90%, 26.09% and 13.04% of the total erosion, respectively. As of 2016, due to the increasing grassland area (Table 2 ), the amount of soil erosion caused by grassland increased slightly, and this land use type became the main source of soil erosion. In addition, as the area of arable land and bare land decreased rapidly, the contributions of these two land use types to soil erosion were significantly reduced by 21.72 Mt and 12.82 Mt, respectively. Therefore, without considering the roles of check dams, the reduction in soil erosion was mainly due to the change in the land use pattern, especially the conversion from arable land and bare land to vegetation land. Considering the action of check dams, the erosion amounts from grassland, arable land and bare land decreased by 52.77%, 57.12% and 42.92%, respectively, without land use changes in 1987. Under the three land use scenarios, the construction of check dams reduced the erosion of all land use types to varying degrees, and grassland had the largest erosion reduction.
In terms of sediment load, without considering the effect of check dams, the main sources of sediment in 1987 were arable land and bare land, accounting for 67.17% of the total sediment load. Nevertheless, grassland was no longer a major sediment source, indicating low sediment connectivity in the area. This development further led to less sediment that could be transported to the watershed outlet. Compared with 1987, the sediment contribution of arable land and bare land decreased significantly in 2016, with a reduction of more than 80%. Although the sediment loads of all land use types decreased to different degrees under the condition of check dams, the conversion of land use played a more important role in the reduction in sediment load over time. This was due to the decrease of SDR caused by land use changes (mentioned above), so that the contribution of land use changes to sediment load reduction increased slightly over time (Table 4 ).
4.2 Impact of land use changes and check dams on sediment load
Many studies have applied various models and methods to clarify the impact of land use changes or check dams on sediment load in different regions of the world (Table 5 ). The land use of the Kuye River watershed has undergone dramatic changes in the past few decades, especially since the implementation of the Grain for Green Project in 1999. Nevertheless, land use change will significantly affects soil erosion and sediment transport (Fang, 2017). In this study, the conversion of arable land and bare land into vegetation land reduced the erosion of the source area and further decreased the sediment export by diminishing the SDR of the whole watershed. This was consistent with the results of Zhou et al., (2019). In contrast, when natural vegetation is artificially transformed into other land use types, such as cultivated land and urban land, the opposite result is obtained (Aneseyee et al., 2020; Sushanth and Bhardwaj, 2019).
As the main channel construction and important soil and water conservation engineering measures, check dams play a significant role in channeling watersheds, which can conserve water and soil, provide fertile farmland and prevent the downstream channel from scouring, reducing the downstream sediment load. Previous studies (Borja et al., 2018; Polyakov et al., 2014) have confirmed that check dams can maintain a significant proportion of the sediment load. Based on data from previous studies, Ran et al. (2008) indicated that in the Kuye River watershed, the reductions in sediment mass caused by check dams were 37.2% from 1970 to 1996. According to the results of this study, check dams reduced sediment export by 53.77% in 2006 and 30.81% in 2016. With the continued restoration of vegetation in the watershed, the contribution rate of check dam construction to sediment control may decrease slightly.
The SEDD model in this study overestimated the sediment load in recent years, which may be partly because only the key dams were considered but the small dams, terraces, concrete roads and other soil and water conservation measures were not considered. According to the results of the hot spot analysis, there were still some high-risk regions of soil erosion and sediment yield in this watershed, which may be mainly due to the rapid expansion of Shenmu County and the large-scale mining of coal resources in the surrounding area in recent decades. Then, these hot spot areas should become the focus for implementation of effective soil and water conservation measures to intervene with soil erosion. In addition, the vegetation coverage in this study area reached more than 80% in 2016 (Table 2 ), further reforestation will be difficult with the limitation of soil moisture. This implies that the impact of dams on sediment reduction will be very important in the future. On the other hand, the service life of the check dams was limited. Under extreme rainstorm conditions, check dams are prone to collapse and damage (Bai et al., 2020), releasing more sediment and producing serious disasters. In view of the existing circumstances, the operation status of check dams in the Kuye River watershed should be considered, and effective measures (such as spillway construction and dam reinforcement) for protecting check dams must be taken to guarantee the maximum efficiency of sediment interception.