Study on the influence of vegetation change on runoff generation mechanism in the Loess Plateau, China

In recent years, the amount of water and sediment in the Yellow River Basin has dropped drastically. This paper selected 125 rainfall and flood data points from 1965 to 2015, combined hydrological methods and mathematical statistics to analyze the hydrological factors and runoff generation mechanism, and combined the underlying surface conditions of the Gushanchuan Basin. The characteristics of change revealed the temporal and spatial variation characteristics and related factors of the runoff generation mechanism in the basin. The results showed that the Gushanchuan Basin is still dominated by HOF runoff, but the runoff generation mechanism has also changed with changes in the underlying surface, which are reflected in increased runoff components, the reduced proportion of HOF runoff, and the increased proportion of saturation-excess overland flow (SOF) runoff and mixed runoff. We analyzed the variation law of underlying surface in the basin, which indicated that the increase in the forest grass area was the main factor affecting changes in the watershed runoff generation mechanism. This research will enable a deeper understanding of the runoff generation mechanism of the main soil erosion areas in the Loess Plateau, and reveal variations in the runoff generation mechanism in the Yellow River.


INTRODUCTION
With continued population expansion, the scale of the economy is expanding, and humans are overly interfering with the natural ecosystems. This interference has resulted in a series of ecological and environmental problems in the Loess Plateau, which is located in the arid and semi-arid regions of the northwest, and has a prominent soil erosion problem (Fu et al. ). For a long time, water shortages and soil erosion have been sources of restriction of economic development in the Loess Plateau. The area is known as having 'frequent drought and hardship rank first among the world,' and one-fifth of the country's impoverished counties are located in the Loess Plateau (Shi & Shao ). Vegetation is an important part of soil and water conservation, it is a core aspect of ecological environmental construction in the Loess Plateau, and it is also one of the primary ways to reverse the ecological environment of the Loess Plateau (Wang et al. ; Zhao et al. ). In the past 50 years, this country has successively implemented a series of ecological construction projects, such as key construction of soil and water conservation, construction of the Three-North Shelterbelt System, protection of natural forest resources, returning farmland to forests and grasslands, and construction of silt dams in the Loess Plateau. Most of these major initiatives are based on vegetation restoration and reconstruction, to promote the benign development of the regional water cycle as well as the balance of the socioeconomic-ecological complex system. The improvement of soil erosion in the Loess Plateau through vegetation restoration has two aspects: (1) changing the water circulation path to prevent soil erosion, as perennial vegetation can reduce surface runoff by increasing effective vegetation coverage; and (2) increasing soil water content and increasing vegetation productivity (Wang et al. ).
The Loess Plateau is the main runoff area of the Yellow River Basin. After the 1980s, the Yellow River water volume showed a notable downward trend (Mu et al. ; Wang et al. a, b). The cause of this sharp decline in the Yellow River water volume has been a significant focus of academic debate. Guzha et al. () found that land-use change may lead to important changes in runoff generation processes and water storage. The runoff generation mechanism has three types of characteristics: (1) infiltration-excess (Hortonian) overland flow (HOF), (2) saturation-excess overland flow (SOF), and (3) a combination of HOF and SOF (mixed runoff).
HOF is generated when the precipitation rate exceeds the infiltration capacity of the soil or land surface. This can be a dominant process in urbanized or otherwise disturbed areas as well as in areas that typically receive high-intensity precipitation and have a low permeable crust at the soil surface (Horton ). SOF is generated when the soil becomes saturated to the extent that additional precipitation cannot infiltrate. Saturation-prone areas include those with a high water table and shallow soils that provide little additional storage for water (Dunne & Black a). The Loess Plateau is characterized by sparse rainfall, high rainfall intensity, and low vegetation coverage, and the thickness of the aerated zone reaches tens of or even hundreds of meters, which is typical of an HOF area (Jiao et al. ).
This study identified the Gushanchuan Basin in the middle reaches of the Yellow River as the study area. On the basis of an analysis of the underlying surface changes in the Gushanchuan Basin, we distinguished the runoff mechanism of the floods in the basin according to long-term sequence rainfall runoff data and flood data for the Gushanchuan Basin. This study provided insight into the changes in the runoff generation mechanism of the Gushanchuan Basin over the past 40 years, examined reasons for the reduction of the Yellow River water volume from the perspective of the runoff generation mechanism, and provided the necessary scientific basis for the effectiveness of the ecological construction of the Loess Plateau and the formulation of regional sustainable development countermeasures.

Runoff generation mechanism
Due to the uneven spatial distribution of rainfall and underlying surface conditions, the runoff mechanism is changing during the rainfall process, and nine types of different runoff mechanisms are combined under specific aerated zone structures and rainfall characteristics (Dunne & Black b) (Table 1). However, according to the impact where R is runoff depth generated by rainfall (mm); P is the total amount of precipitation (mm); E is watershed evapotranspiration (mm); W 0 is initial water storage capacity (mm); i is rainfall intensity (mm/h).
Because of the spatial and temporal distribution nonuniformity of rainfall and the complexity of the underlying surface, for a specific watershed the runoff generation mechanism is not static and it is difficult to quantify with simple indicators. This study analyzed the composition of watershed runoff components from the flood flow process line (Birtles ). According to the principle of runoff generation (Kirkby ), the main components of the runoff in the three sections were surface runoff, interflow, and ground runoff. We used Dunne's flow theory (Dunne & Black b) and a flow pattern comprehensive analysis table (Table 2) combining rainfall, rainfall intensity, runoff depth, and pre-influence rainfall to classify the runoff pattern into HOF and SOF from the perspective of quantitative, qualitative, and runoff impact factors. The mixed-flow pattern included both HOF and SOF in a rainfall event, which we comprehensively analyzed and judged.
We used the slope method to separate the baseflow for the No rainfall occurred for 20 days before the flood. In the early stage of the flood, the API was very small, the flow process line showed a multipeak type, the main runoff generation mechanism was the surface runoff (R s ), the rate of water retreat was fast, and the flow process line steeply rose and fell. With continuous rainfall, soil water content increased continuously, and the interflow (R int ) and groundwater runoff (R g ) were revealed after point B. Because of different confluence speeds, the flood process line presented a steep rise and slow fall in the later stage (Hu et al. b). The runoff mechanism was converted from R s to R s þ R int þ R g , and thus this flood had a mixed runoff mechanism.

Rainfall and runoff
We calculated surface rainfall using the Thiessen polygon method. Rainfall duration was an effective duration and did not take intermittent time into account. The ratio of  surface rainfall-to-rainfall duration was the average rainfall intensity. The rainfall runoff correlation map was a quanti- where Δt is the period length (h); Q i is the i-period flow (m 3 /s); n is the number of periods; A is the basin area (km 2 ); and K is the storage coefficient.

Antecedent precipitation index
The antecedent precipitation index (API) reaction was the amount of rainfall that was retained in the soil during the previous rainfall. In this study, we used the recursive formula method (Heggen ) to estimate the amount of rainfall affected in the early stage of the flood. According to its definition From this, the formula for calculating the amount of rainfall in the early stage is derived: where API t is the previous impact rainfall (mm) on day t; API tþ1 is the previous impact rainfall (mm) on t þ 1 day; n is the number of previous rain days affecting the runoff, which was usually about 15 days; P t is the t-day rainfall (mm); P tÀ1 , P tÀ2 … is the rainfall (mm) on 1 day before t day, 2 days before …; and k is the daily decrease coefficient of the soil water content.

Catchment water storage capacity
We derived the catchment water storage capacity (W m ) from successive floods that included multiple floods (Llorens & Gallart ). The calculation selected the rainfall floods that saturated the water content of the basin and produced a large flood process. The process is as follows: Let W m, 0 ¼ I m , where the loss is I m ¼ P À R, P is the total amount of rainfall, and R is the total amount of runoff depth generated by rainfall (Hu et al. ). The calculation process of W m follows: and where k e is the evapotranspiration capacity of the basin, the ratio of E m to water surface evaporation; K j is the constant coefficient; P i is the effective rainfall for the i day before the flood; i is the power; j is the number of iterations; and W m,0 is the assumed value for the first iteration. After several iterations, W m tends toward a stable value, which represents the water storage capacity of the basin.

Curve estimation model
Curve estimation theory could use one variable to predict another variable. The mathematical models are presented in Table 3. When an optimal model cannot be determined based on observations immediately, a simple and more suitable model can be established from many regression models using the curve estimation method. Curve estimation requires that the independent and dependent variables belong to numeric variables.

Goodness-of-fit test
We used the determination coefficient (R 2 ) as the criterion for a goodness-of-fit test to select the following curve equation: where SSR is the regression sum of squares; SSE is the sum of squares of residuals; and SST is the sum of total deviations squared. The determination coefficient is a comprehensive measure of the goodness of fit of the regression model.
The larger the evaluation coefficient, the higher the model's goodness of fit.

Significance test of the regression equation
We used an F-test to test the significance of the established regression equation. The F-test is the ratio between the average regression sum of squares and the average residual sum of squares: where n is the sample numbers; and k is the number of independent variables.

Data analysis
We statistically analyzed rainfall characteristics, runoff, and sedimentation. We evaluated data using SPSS for Windows Linear function of one variable 16.0 (SPSS Inc., Chicago, IL, USA). We used one-way analysis of variance (ANOVA) to assess the influence of runoff for sedimentation and the relationship between runoff and rainfall, and rainfall density.

Characteristics of rainfall and flood
As time has passed, the annual runoff and sedimentation have shown a decreasing trend in Gushanchuan Basin.
The peak runoff corresponded to the peak sedimentation.
By 2015, the runoff reduction rate was 81%, and the annual runoff reduction was 1.35 billion m 3 (Figure 3(a)), and the correlation between annual runoff and sediment was significant (Figure 3(b)). The results showed that soil and water conservation has played an obvious role in recent decades, and as a result, soil erosion has decreased sharply.
Because of the influence of rainfall and underlying sur- Among the 125 flood events, there were 19 light rains, 42 moderate rains, 36 heavy rains, and 28 rainstorms. Therefore, we determined that the Gushanchuan Basin suffered primarily from moderate rains and heavy rains. The flood duration also showed a first increasing and then a decreasing trend, and the peak flow showed a significant decrease over the year (Table 4).

Analysis of underlying surface changes
The grassland area in the Gushanchuan Basin accounted for the largest proportion of land (61-63%), which was followed  Table 5).

Runoff and meteorology factors, underlying surface conditions
We obtained three principal components (P, API, and V c ) based on the PCA. We obtained the expressions between each independent variable and dependent variable according to curve estimation. We took the runoff coefficient (α) as the dependent variable, and rainfall (P), antecedent precipitation index (API), and vegetation coverage (V c ) as independent variables. The expression was as follows: where a, b, c, d, and e are the fitting parameters, and f is the constant. Other parameters have the same meanings as noted earlier. We obtained fitting parameters by nonlinear regression (Tables 6 and 7).
In this study, we combined the curve estimation method with the multivariate nonlinear regression analysis, which we used to build the runoff coefficient prediction model.
The main factors affecting runoff coefficient included rainfall (P), antecedent precipitation index (API), and vegetation coverage (V c ). The runoff coefficient prediction model had high precision and a certain theoretical significance and research value, which provided additional references for the runoff simulation ( Figure 6).

The runoff generation mechanism
From 1965 to 2015, the Gushanchuan Basin was dominated by HOF, accounting for 62.4% of the flood events. Mixed   1965-1979 1980-1998 1999-2015 Mean rainfall duration ( 1965-1979, 1980-1998, and 1999-2015, respect-ively. The proportion of HOF events decreased significantly from 1965 to 2015. The proportion of flood events with mixed runoff as the main mode of runoff in each of the three stages was 28, 40, and 47%. In general, the HOF was dominant in each stage but the proportion of events decreased, whereas SOF and mixed runoff increased (Table 8).    The effect of land-use cover changes on runoff

Groundwater outflow
In the Gushanchuan Basin, forest area increased rapidly.
Previous studies found that discharge capacity had an influ- Loch ). As confluence time expanded, the flood process changed: forest areas increased 73%, which resulted in an 87% increase in groundwater outflow (Table 9).

Infiltration capacity
The forest areas increased by 27%, rainfall required before runoff generation increased by 192%, and the duration of flood subsidence increased by 22% (Table 9). This is because increasing forest area will increase rainfall interception and soil infiltration capacity. In the early stage of a storm, rainfall is consumed by the infiltration. Flood subsidence is an important aspect of runoff generation and confluence. A certain amount of infiltration occurs during the process of flood subsidence. Generally speaking, with an increase in rainfall and rainfall intensity, the runoff coefficient will also increase. As the years passed, more rainfall was needed in the Gushanchuan Basin to produce runoff, which indicated that infiltration had increased gradually. The duration of flood subsidence also increased gradually (Table 9). Averaged rainfall before generated flow and catchment water storage capacity increased from 1965 to 2015, which indicated an increase in infiltration, resulting in an increase in soil moisture. Equations (1) and (2)  Our previous study also found that flood events with SOF increased as flood events with HOF decreased because of increased forest areas (Li ).    1965-1978 1979-1998 1999-2015 Forest areas (

Water storage capacity
The water storage capacity of the Gushanchuan Basin increased year by year (Table 9). Previous studies have shown that the water storage capacity of woodland was higher than other land-use covers and that an increase in forest areas would result in the increase of catchment water storage capacity and reducing runoff (Jian et al. ). The forest areas increased by 37% as the water storage capacity increased by 54% (Table 9); other studies have found similar results (Mascha et al. ). In Gushanchuan Basin, both SOF and HOF were evident. Our previous study found that the flood events were mainly HOF, and the SOF flood events showed an increasing trend in Gushanchuan Basin (Li ).

CONCLUSIONS
The effect of forest vegetation coverage on runoff generation mechanism is very significant in the Gushanchuan basin.
This information enabled us to more deeply understand the water production law of the main soil erosion areas in the Yellow River Basin and revealed the mechanism of water quantity change in the Yellow River. The outcomes are of great significance for the local government to carry out vegetation restoration, a preliminary understanding is as follows: (1) it is necessary to fully understand the ecological benefits of forest vegetation in conserving water resources and reducing flood peak flow, establish longterm water resources strategic; (2) It is necessary to return cultivated land to forests, establish a complex system of agriculture, forest and water, and realize a virtuous cycle of ecological environment in the basin; (3) it would establish an industrial structure that combines agriculture with forestry to achieve sustainable economic development.
Land-use change is an important factor that cannot be ignored in causing runoff changes, particularly in the flood season. Because of the impact of land-use change on water resources and ecosystem health in the Gushanchuan Basin, appropriate consideration should be given to the role played by land-use change. In fact, climate change should be considered when assessing the impact of future land-use cover changes. The combined effects of climate and land-use cover changes are complex, however, and are beyond the scope of this paper. Future research will consider land-use prediction models based on cellular automaton and Markov chains and will be combined with regional future climate scenarios to quantitatively predict future land-use and climate change impacts on runoff.