Introduction

Soil erosion is a natural phenomenon and has several implications on agricultural production, livestock, and water quality even in the absence of anthropogenic activities. Erosion rates can be also raised above the natural threshold due to human alterations, which is known as accelerated erosion. Sediment load is the net result of soil erosion and processes of sediment accumulation, so it depends on variables that control water and sediment load in a watershed. Typically, sediment load reflects the influences of climate (precipitation), catchment properties (soil type, topography), land use/cover, and drainage properties (stream network form and density) (Williams 1975; Walling 1994).

Sediment loads vary from low values in humid, low-relief catchments to very high values in arid, mountainous areas (Lavigne and Suwa 2004). Langbein and Schumm (1958) analyzed the relationship between effective rainfall and annual sediment load from 100 gaging stations in the United States (Douglas 1967; Dendy and Bolton 1976). According to this research, sediment yield per unit area was found to reach maximum level at an effective precipitation rate (around 300 mm). Zhang et al. 2015 found positive correlation between sediment yield and rainfall because high rainfall event may increase runoff (Huang et al. 2014). High frequency of individual rainfall events initiate wetter soil moisture and increase runoff which also increase transport capacity and reduce the volume of sediment deposition (Dominic et al. 2015). Surface runoff is eventually gone on to constitute stream flow by the gravity.

Local climate (etc. rainfall, temperature) dominantly influences stream flow (e.g., Leopold et al. 1964; Schumm 1977; Knighton 1998) and obliquely channel width by altering soil moisture and base level, for instance, in a semi-arid climate soil moisture and base level are seasonally increased by more perennial flow through time. Change in channel width depends primarily on discharge and slope, which are eternally affected by regional based factors including local climate, base level, and tectonism (Finnegan et al. 2005). In fact, channel width is naturally raised by higher discharge rate, which is also associated with the size of drainage area (Finnegan et al. 2005; Kirby et al. 2003). Leopold and Maddock (1953) developed a relationship between depended channel properties including channel width and mean annual discharge. The result of the study indicates that channel width rise downstream due to cumulative increase in drainage area.

Sediment fluxes into catchments are more specifically influenced by catchment characteristics associated with open channel hydrology (i.e., discharge rate and channel geometry) (Pinet and Souriau 1988; Summerfield and Hulton 1994; Infante et al. 2006; Babu et al. 2016), topography (i.e., mean slope, elevation, mean channel depth) (Jansen and Painter 1974; Milliman and Syvitski 1992; Ludwig et al. 1996; Yamaguchi et al. 2010), and land use (Dunne 1979; Higgitt and Lu 1999; Nie et al. 2011; Subhash et al. 2016). Summerfield and Hulton (1994) studied the variables controlling mechanical denudation rates in drainage basins exceeding 5 × 105 km² in area and concluded that physical factors (such as basin area, mean annual runoff, temperature, precipitation) have no significant influence on physical denudation processes, although channel gradient and basin relief can be considered as dominant controlling variables. Ayadi et al. (2010) studied sediment yield variability in a semi-arid region of central Tunisia; the research concluded that drainage area is not the main controlling factor of sedimentation. In contrast, previous studies (i.e., De Vente and Poesen 2005; Wuttichaikitcharoen and Babel 2014; Zhang et al. 2015) found a positive correlation between sediment yield and drainage area in different regions, which may occur when the sediment sources origin from stream channel and floodplain (Walling and Webb 1996; Tamene et al. 2006). Other studies (etc. Sytitski et al. 2005; Syvitski and Milliman 2007) indicate that geomorphic and tectonic factors and basin area control sediment fluxes in humid climates. According to Syvitski and Milliman (2007) drainage area size, basin relief, geology, and climate are the controlling factors influencing sediment load from mountainous rivers. Milliman and Syvitski (1992) studied sediment yields for mountainous rivers in North and South America, Asia, and Oceania and found basin topography and area to be the most dominant factors controlling sediment, with climate, land use, and geological factors being second order influences.

Basin elevation and morphology have great influences on river sediment fluxes, according to Pinet and Souriau (1988). They found that sediment fluxes are linearly correlated with mean basin elevation based on sediment yields in large world rivers. Relief is also a major controlling variable, which indicates greater mechanical erosion (Montgomery and Brandon 2002). Chakrapani (2005) found that weathering may cause higher erosion rates in young continental crust than relatively more resistant old continental crust. Even though many previous studies have tried to investigate the significance of relief, only a few mathematical relationships have been proposed. In the early 1960s, Schumm (1963) recommended a linear relationship between erosion rate and drainage basin relief for basins, which average 3885 km² and predominantly underlain by sedimentary and metamorphic rocks in the United States. Other researches (Ahnert 1970; Pazzaglia and Brandon 1996) found out a linear correlation between denudation rate and drainage basin relief, as well. Another field based study found evidence that downstream narrowing channel locally associated with uplift rate and erodibility (Montgomery and Gran 2001).

Morris and Fan (1997) reported that anthropogenic activities are the most predominant factors controlling land degradation. Siakeu et al. (2004) also studied suspended sediment concentration in central Japan with a special reference to human impact. They concluded that anthropogenic activities, especially in industrial countries, create a large variation in the pattern of suspended sediment along relatively small basins.

Examining the relative importance of controlling variables on sediment load contributes to better understand sedimentation processes in the stream channel, which are also, enable to identify the major factors driving sediment. Numerous studies have been done for predicting the sediment yield; however few researches have focused on controlling factors on suspended sediment load at a catchment scale.

The main purpose of this study is to determine the relative importance of several potential control 68 variables on sediment load that was addressed by testing the following objectives: sediment load correlates most strongly with topography, and to a lesser extent with stream discharge, channel geometry, land cover, and drainage characteristics. Studying in this particular catchment is important and would contribute to a better understanding of the physical processes related to sediment transport and deposition in the research area. It also can help to design sediment control strategies within the catchment.

Study Area

The study site is located in the upper portion of Ankara River basin (Fig. 1). Annual mean air temperature over the past 30 years is 10 °C and annual mean precipitation is 448 mm. Most of precipitation occurs during winter and spring. The Ankara River is generally characterized by flashy Q regime in which stream flow rises sharply after rainfall and then falls more gradually. Discharge varies seasonally, with higher flow in the winter and spring and lower flows during summer and early fall. This subbasin morphologically consists of plateaus, alluvial plains, and low-relief areas ranging from 856 to 1164 m in altitude. The stream has deposited alluvial materials (gravel, sand, clay, and silt) to a depth of 25–30 m and a width of 1–1.5 km along the channels, thus many sandpits have been excavated within the floodplain (Yasti 2008).

Fig. 1
figure 1

Location of the Ankara River basin and its network

Full size image

The main soil texture in the basin is heavy clay (59%), which tends to be more resistant to erosion than sand or silt as the clay helps bind soil particles together (Mirsal 2008). Besides the heavy clay materials, northern portions of the catchment have more sand, and the northeast side has silt and clay (Fig. 2). This site also contains different land use and land cover types. The land use map was obtained from the Global Land Cover 2000 (GLC 2000), which is the harmonization of all the regional products into a full resolution global product with a generalized legend. A standard classification resulted in 13 types of land use within the site. Land cover types present in a small percentage were summarized as generic classes such as all agricultural crops summarized as “agriculture”, three types of forests represented by one mixed forest, and one class representing two types of urban classes in the detailed land use/cover map. In the Ankara River basin, the most common land use types and land cover are irrigated land (57%), grass land (13%), and forest land (3%) (Fig. 3). Three types of land uses (arable land, grassland, forest) are particularly widespread in the catchment.

Fig. 2
figure 2

Land use classification of the year 2000

Full size image
Fig. 3
figure 3

Soil classification and distribution in the catchment

Full size image

Data Collection

Sediment load, as indicated by total sediment outflow from a sub-basin measured was obtained from the physically based model. The spatial distribution of sediment load may reflect special variability in watershed characteristics and anthropogenic activities (drainage area, elevation, mean channel slope, runoff, channel width, depth, and land cover/use types etc.). Table 1 indicates data sources and methods of data collection for spatial variables controlling sediment in the subbasin. The data associated with the amount of sediment in the catchment was obtained from (1239 Ova Cayi–Eybek) sediment gaging site (Fig. 1) draining approximately 322 km². Observed monthly sediment data were used for calibration and validation for the Soil Water Assessment Tool (SWAT). Simulated sediment data of monthly suspended sediment load are within the acceptable range of the Nash-Sutcliffe efficiency (NSE), used to assess the predictive power of the model; the relative error (RE), often used to compare approximation of varying size of numbers; and coefficient of determination (R²), used to indicate how well data fit to the hydrological model.

Table 1 Investigated catchment properties, data sources, methods of data collection in Ankara
Full size table

Each land use type has various responses to soil erosion because these plant covers protect the ground surface from the impacts of raindrops and reduce the runoff and velocity by influencing infiltration capacity (Thornes 2001; Descheemaeker et al. 2006). Land use/cover types were reclassified in five main categories including percentage of cultivated areas, percentage of forest areas, percentage of grass lands, percentage of industrial areas, and percentage of surface water for each subbasin. The categorization of land use type may significantly correlate with sediment load because widespread cultivated areas correspond to more lands with seasonal farming or other farming methods that result in soil erosion, especially in semiarid regions.

The collinearity in a regression model illustrates inter-correlation among the independent variables. The non-collinearity assumption can be explained through the values of the tolerance and the evidences inflation factor (VIF). The tolerance values ranged from 0 to 1 and VIF values ranged from 0 to 10. The cutoff value is 0.2 for tolerance and for VIF values fewer than 4. “Durbin Watson” statistics was also used to test the hypothesis that there might be autocorrelation in the data. If the Durbin Watson value is within the acceptable range of 1.5–2.5, meaningful serial correlation is absent in the dataset (Khorasani and Zeyun 2014).

Method

SWAT Model

The Soil and Water Assessment Tool is a continuous, long term, physically based distributed model designed to predict the impact of land management practices on the hydrology, sediment load, and water quality in agricultural watersheds (Arnold et al. 1998). The current version of the model routes the maximum amount of sediment in a reach as a function of the peak channel velocity and estimates sediment load for each HRU using MUSLE (Williams 1975). SWAT can be a tool in predicting sediment load, especially at the catchment scale, because it considers spatial and temporal variation based on different potential physical variables. The simulation can also provide a better understanding of sediment transport and deposition processes by overland flow and allows reasonable prediction and forecasting. Sediment load is the sum of sediments produced by overland flow, gully, and stream channel erosion in a catchment. Model performance was tested using the long term measured discharge and suspended sediment data from the gage of Ova Cayi–Eybek (1239), and the model statistics were considered highly acceptable. The most sensitive input parameters in the watershed are ground water delay time and base flow regression for stream flow, and soil depth and Manning’s coefficient of channel roughness for sediment. The simulation results demonstrate the uniqueness of the corresponding response, and reflect the model performance based on the sediment statistics in the calibration period, such as NSE, RE, and R² of 0.81, −1.55, and 0.93, respectively, and for the validation period NSE, RE, and R² of 0.77, −2.61, and 0.87, respectively. Then, the simulated data were used in multivariable regression analysis due to lack of spatially distributed sediment data in the basin.

Multivariable Regression Analysis

Multivariable regression analysis is a statistical tool for estimating the relationship between dependent and independent variables. The analysis computes the conditional expectations of controlling variables against sediment load in this study, contributes to better understanding of which controlling factors dominantly affect sedimentation processes, and allows exploring the form of this relationship in the catchment.

The correlation of observed data and temporal control variables are examined at monthly time intervals dictated by the availability of information on land use and land cover. Relationship between sediment data and all 19 explanatory variables (catchment area, min–max elevation, mean channel width, elevation, stream channel depth, relief ratio, drainage density, hydrometric integral, land use/cover, and land use/cover types) for upstream of Ankara river basin are evaluated, and then explanatory variables in the model were used respect to their statistical significance with a confidence interval of 90% for the 74 studied catchments in the Central Anatolia of Turkey (Table 2).

Table 2 Catchment properties for the 74 studied sub-catchments in central Turkey
Full size table

Model performance was tested using the ANOVA F-test statistics, and results were tested with the coefficient of determination (R²) to determine whether there are statistically significant linear relationships among the variables. The value of R² can range between 0 and 1, with higher values reflecting more accurate regression models. A strong relationship exists between several variables and sediment load with a value of R² above 0.5 by using a 90% confidence interval (CI). The SPSS software started with the multiple R, adjusted R² of the variability of sediment load, and adjusted R², which is somewhat lower than the original R² as it takes into account sample size. The standard error of estimation is associated with R², not adjusted R² in the analysis.

Principal Component Analysis (PCA)

To simplify model performance analysis, covariance matrix can be diagonalized to decouple the multivariate time series into component univariate models. PCA is a widely used component extraction technique to extract components that are mutually uncorrelated. This method focuses on providing a representation of a multivariate data set using the information contained within the covariance matrix. These components are generally referred to as principal components (PCs) or eigenvectors. Each PC is a linear combination of the original variables and describes a different source of variation (information). The largest or 1st PC is oriented in the direction of the largest variation of the original variables. The 2nd largest PC lies in the direction of the next largest variation, and so forth.

The PCA extracted from the 19-by-19 correlation matrix (i.e., 19 explanatory variables explained above) was used to examine dominant patterns of intercorrelation among the variables to identify subsets of indices that describe the major sources of variation while minimizing redundancy.

Results

Physical Variables Controlling Sediment Load

In order to assess factors controlling suspended sediment load in the upper portion of Ankara river basin, series of statistical methods of multivariable regression analysis and PCA were performed using data from 74 subbasins (Table 2). Pearson correlation was computed for all controlling factors respect to catchment properties. A stepwise regression was performed for the factors with the aim of exploring whether a unique combination of local variation on sediment load, which can be expressed by a combination of several controlling factors. The stepwise method chooses an independent variable that has the largest Pearson correlation value with the dependent variable, and puts that into the regression analysis. Then it sequentially does the same for the next highest independent variable until it finds a non-significant predictor. The model found two statistically significant variables with R² above 0.9, which indicates that more than 90% of variability of sediment load can be accounted for by two controlling variables: drainage area and flow (Table 3).

Table 3 The model summary of multivariable regression analysis by using SPSS
Full size table

In terms of the tolerance values, drainage area and stream flow were within the acceptable range (below 0.2) for the statistically significant parameters. The VIF values of drainage area and stream flow were slightly over the cutoff number, with a value of 5.3 (Table 4). Therefore, the assumption of multicollinearity was not met for these two controlling factors. Table 5 illustrates the multivariable regression estimates including the intercept and significance level if the parameters. In the stepwise multivariable regression analysis, a highly significant intercept and highly significant values of stream flow, drainage area, channel width, water surface, and relief ratio were found.

Table 4 Collinearity statistics of multiple linear regression estimates
Full size table
Table 5 Correlation matrix between catchment properties and sediment load
Full size table

Figure 5 presents the two-dimensional ordination illustrating the major patterns of inter-correlation among the 19 explanatory variables for the combined set of 74 subbasins. First and second PCs explain 54% of the variability in the factors. In this figure, the correlation between any two factors is related to the cosine of the angle between their index-axes. Therefore, two factors separated by a small angle are highly positively correlated; two factors separated by a large angle are highly negatively correlated; and, two factors separated by a right angle are uncorrelated or independent. This figure demonstrates that the degree of correlation among the explanatory factors varies greatly. A large number of factors are situated in the upper left and upper right quadrants of the figure, illustrating that within these two quadrants the variables are highly positively correlated. The upper left cluster contains a number of factors of location describing specific characteristics of the basin with the highest effect on sediment load (SED_OUT). These factors include stream flow, catchment area, mean channel width and percent of industrial land uses. The results are consistent with the results obtained from multivariable regression analysis reported above.

In contrast, the lower right cluster contains mainly elevation factors indicating that lower sediment loads were observed from the upstream subbasins. This was also supported by the results of the multivariable regression analysis. In addition, the upper right cluster included parameters pertaining to the shape and topography of the watershed (eg., relief ratio, drainage density, and depth) which have very low or no correlation with the sediment generation. This was confirmed by the regression analysis as well.

Results suggest that sediment load is likely to reflect some of the controlling variables consisting of stream flow, drainage area, channel width, etc. (Table 5). No other variables can be considered as prime control factors on sediment load in the region. The multivariable regression analysis shows that mean annual precipitation (PCP), elevation (Elv), mean slope (Slope), mean channel depth (Depth), relief ratio (RR), drainage density (DD), hypsometric integral (HI), long path (LgnPth), and land cover are not significant variables on sediment load. Results of multivariable linear regression estimates of ANOVA test are presented in Table 6. According to these results a strong relationship exists between suspended sediment load and drainage area and stream flow, as well as channel width. Also, in central Ankara, an increase in channel width usually corresponds to: (1) an increase in stream flow; (2) an increase in sediment load; and (3) a decrease in mean elevation. The table of Sig-1 tailed matrix indicates a significant p-value for controlling variables such as drainage area, stream flow, mean elevation, minimum elevation, and maximum elevation, mean slope, mean width, grassland, and industrial lands (Table 7). All these variables are statistically significant considering a 90% confidence interval.

Table 6 ANOVA table of multiple linear regression estimates
Full size table
Table 7 Sig -1 Tailed matrix between catchment properties and sediment load
Full size table

Trend in Sediment Load and Land Use/Cover

Anthropogenic and natural changes, including construction of new dams and climate change, influence sediment transport or sediment deposition in a stream channel because these changes have direct and indirect effects on hydrology, geomorphology, riverine environment, and aquatic habitats. Sediment station of Ova Cayi Eybek (1239) was initiated in 1967 in the upstream portion of the Ova Creek catchment. The data collection indicated that there was a large flood in 1991 in the surrounding area, which covered a total area of 322 km² and had an annual sediment load of 6697 ton/year. Accelerated erosion results from surface runoff in the catchment caused by deforestation, poor cultivation techniques, and overgrazing (Ongwenyi et al. 1993). In the catchment, monthly suspended sediment load from the result of the SWAT simulation indicates that monthly suspended sediment load increases progressively from upstream to downstream. Results from multivariable regression analyses, rainfall does have an insignificant influence on sediment load (R² = 0.08), because upstream areas have less sediment load but higher precipitation. All tributaries have higher suspended sediment load downstream (Fig. 4). On the other hand, that might give different results if a correlation was made between sediment loads from each subbasins.

Fig. 4
figure 4

Distribution of sediment load along the Ankara River (ton/ha)

Full size image

Discussion

The research assessed the relative importance of watershed characteristics on sediment load in 74 sub-basins with various characteristics and land use/cover in the Central Anatolia Plateau. The Cubuk River is characterized by flashy streams in which stream flow rises sharply after rainfall and then falls more gradually. Discharge varies seasonally, with higher flow in the winter and spring and lower flows during summer and early fall. This stream water is mainly used for irrigation purposes during dry session. PCA was performed to reduce multicollinerity of predictor variables (Afifi et al. 2012; Dormann et al. 2013). The PCA raised the effect of stream flow, drainage area and mean channel width on sediment load (Fig. 5). The result of the PCA and multivariable analysis indicate that these controlling factors are consistent (Fig. 5).

Fig. 5
figure 5

Ordination from the principal component analysis of the 74 subbasins based on 19 explanatory variables

Full size image

SWAT model has a limited ability of rigorously simulating ground water flow, which is significant in this region. Groundwater recharge, groundwater river interaction, and interflow may play important role on flow processes as well as some springs stream water that are supplying irrigation water in the region. Due to the lack of data in the spatial distribution of springs and irrigation channels in the basin, the model may fail simulating interaction between precipitation and stream discharge during dry session. Therefore, the model may propose physical variables controls sediment load rather than climatic factors in the region.

Drainage density and drainage area are the two main variables that might affect sediment yield at a catchment scale. Drainage area is positively correlated with sediment yield (Walling 1983; Verstraeten et al. 2003). The relationship between drainage area and sediment yield can also be positively correlated as long as the main sediment source is from channel and floodplain deposition (Walling and Webb, 1996). The correlations are quite weak, with the value of R² ≈ 0.08 among precipitation and sediment load, which means that precipitation alone, cannot describe accurately the complex physical processes underlying sediment load as a result of local geology and hydrology (Dearing and Jones 2003). However, precipitation has indirect effect on stream flow, which is one of the predominant factors controlling sediment load in the region. Also, wider channels correspond to more stream discharge, which creates more stream power for sediment transport. It can be also inferred that sediment yield increases with higher stream flow and channel width because higher discharges equate to more potential energy to transport more suspended sediment.

Although there has no study found testing controlling variables on sediment load in the region, other studies concluded that drainage area (Dendy and Bolton 1976; Zhang et al. 2015), stream flow (Xu 2009; Zhao et al. 2013) are positively correlated factors affecting sediment load in basins with similar characteristics around the world. In terms of channel width, another factor in the interchange of sediment load along active channel and floodplain, takes place as discharge increase at a particular cross section in alluvial rivers (Leopold and Maddock 1953). Even though these studies highlights predominant factors affecting sediment, the researchers should also consider a combination of other local environmental factors for better understanding of this short of complex phenomenon.

Conclusions

Factors controlling sediment load in a watershed in central Anatolian Plateau were indentified in the study. Various topographic variables and to lesser extent with stream discharge, drainage characteristics, channel geometry, land cover were analyzed to determine dominant factors controlling sediment load. Analysis of sediment load and morphometric, hydraulic, and meteorologic variation from upstream portion of Ankara river basin represents that the dominant processes controlling sediment as a function of basin scale are stream flow, drainage area, main channel width. These dominant controlling variables are statistically significant, based on the coefficient interval of 90%, although the proposed physically distributed model does not explain all observed variation in stream flow and sediment load.

Multivariable regression analysis and PCA were performed to assess the role of various drainage characteristics in the sediment load. Multivariable regression analysis for the prediction of sediment load indicates that sediment load can be predicted by computing these three variables with an efficiency of 0.95. In terms of the monthly suspended sediment load, three principal components were identified to be statistically significant explaining 54% of the variability in the factors. Two factors separated by a small angle are highly positively correlated; two factors separated by a large angle are highly negatively correlated; and, two factors separated by a right angle are uncorrelated or independent. The upper left cluster contains a number of factors of location describing specific characteristics of the basin with the highest effect on sediment load. These factors include stream flow, catchment area, mean channel width and percent of industrial land uses. These results are consistent with the results obtained from multivariable regression analysis reported above.

The results indicate that stream flow, drainage area, and channel width explain most of the variability in sediment load among the catchments. The implications of the results, efficient siltation management practices in the catchment should be performed to stream flow, drainage area, and channel width. Prediction of sediment load for non-point sources and planning watershed management require multidisciplinary approaches. This study established a basis for future multidisciplinary studies on modelling sediment loads and for better understanding of correlations between potential control variables and the resulting sediment loads.