Sediment is a natural constituent of rivers, but excess sediment loading in rivers is a leading cause of the degradation of water quality and impairment of aquatic ecosystems. Remedial actions require the identification of the sources of sediment loading and the factors that control this loading (Belmont et al., 2011). Despite extraordinary efforts, sediment resulting from soil erosion remains one of the most difficult non-point source pollutants to quantify (Wu and Chen, 2012). Soil erosion is contingent on multiple factors, including climate, soil, topography, and land use (Wei et al., 2012); thus, soil erosion is typically episodic and highly localized (Trimble and Crosson, 2000). Moreover, eroded sediment may exit the watershed quickly or be stored for very long periods of time (Bracken et al., 2013). In recent decades, numerous process-based erosion models have been developed (de Vente et al., 2013). However, the application of these models suffers from the need for extensive parameterization and calibration, which is often problematic because of the low quality of available input data (Jetten et al., 2003). It is not surprising that many studies on the estimation of sediment yield are based on empirical models of soil erosion and require a scalar reduction factor to estimate sediment yield as a fraction of erosion, e.g., the sediment delivery ratio (SDR) (Lu et al., 2005). The concept of the sediment delivery problem was introduced into the literature by Walling (1983). According to this concept, only a fraction of the gross soil erosion within a catchment will reach the outlet and be represented as sediment yield. However, few studies have provided information on how to quantify this reduction factor, and available observations indicate diverse and highly nonlinear scaling with respect to watershed characteristics (Belmont et al., 2011; de Vente et al., 2007).
Our ability to manage landscape sediment routing system and its response to anthropogenic pressure depends on our ability to estimate sediment yield accurately and to identify the controlling factors (Ali and De Boer, 2010; Fu et al., 2009). With readily available digital datasets such as digital elevation models, remote sensing image data, and soil databases, investigators have come to rely on watershed characteristics (e.g., topography, land uses, soil types) as predictors of sediment yield (Hassan et al., 2008; Kuhnert et al., 2012; Ouyang et al., 2010; Tramblay et al., 2010; Xin et al., 2011). The reliability of sediment source estimates can be improved by using multiple overlapping methods of measurement within geographic information systems (GIS). The use of linkages between watershed characteristics and the dynamics of sediment yields has exciting potential as an inexpensive alternative to ground-based monitoring. This approach is strengthened by a suite of new research tools that allow for the rapid and precise dating of land surfaces and high-resolution measurement of topography.
Despite the great potential of watershed characteristics analyses and indicator approaches, they also present particular analytical challenges. Multivariate approaches are commonly used to relate various watershed characteristics to the sediment yield at the micro- and meso-watershed and river basin scales (Ouyang et al., 2010; Xin et al., 2011). However, watershed characteristics, including topography, land use, geology, and soil, are highly collinear or codependent and not independent predictors. This lack of independence can confound correlative analyses and yield potentially misleading results. Moreover, land use types within a watershed also tend to be patchy and spatially autocorrelated. Spatial autocorrelation may be particularly problematic in watershed studies because the locations of these land use types often correspond to an underlying pattern in the landscape (King et al., 2005). Consequently, the apparent relationships between land use and sediment yields within watersheds could just as easily be explained by physiographic factors that necessarily co-vary with land use patterns. Thus, many apparent linkages between land use and sediment yields in watersheds may be spatially confounded.
The inherent limitations of traditional multivariate approaches in handling multi-collinear and noisy data can be overcome by applying techniques based on multivariate statistical projection. For example, principal component analysis is one of the most widely used techniques for reducing redundancy and the dimensionality of input data. Partial least-squares regression (PLSR) is a new technique that combines features of principal component analysis and multiple linear regression and generalizes these two analytical approaches (Abdi, 2010; Wold et al., 2001). PLSR can handle highly correlated noise-corrupted datasets by explicitly assuming dependency among the variables and estimating the underlying structures, which are essentially linear combinations of the original variables (Carrascal et al., 2009). Another striking feature of PLSR is that it is particularly suitable for multivariate problems when the number of observations is less than the number of possible predictors (Onderka et al., 2012; Shi et al., 2013).
We previously developed quantitative relationships between sediment yield and land cover changes and land cover patterns within the Upper Du River watershed in China (Shi et al., 2013; Yan et al., 2013). However, soil erosion and the resulting sediment export result from the joint effects of soil, topography, and land use under stochastic rainfall events (Wei et al., 2009). The quantification of the effects of watershed characteristics on sediment yield is essential to effective watershed management. Therefore, in this study, the Upper Du River watershed was chosen as the case study area. PLSR was used to explore the relationship between watershed characteristics and specific sediment yield and SDR. The objectives of this study are the following: (i) to determine how the specific sediment yield is related to catchment size, topography, land use composition, and land use patterns at the sub-watershed scale; (ii) to identify which of these characteristics exert major and minor influences on the specific sediment export; and (iii) to develop an empirical model for SDR as a function of watershed characteristics.
Source: Essay UK - http://www.essay.uk.com/free-essays/environmental-studies/soil-erosion.php