Response of Variation of Water and Sediment to Landscape Pattern in the Dapoling Watershed

Abstract

The relationship between water-sediment processes and landscape pattern changes has currently become a research hotspot in low-carbon water and land resource optimization research. The SWAT-VRR model is a distributed hydrological model which better shows the effect of land use landscape change on hydrological processes in the watershed. In this paper, the hydrological models of the Dapoling watershed were built, the runoff and sediment yield from 2006 to 2011 were simulated, and the relationship between landscape patterns and water-sediment yield was analyzed. The results show that the SWAT-VRR model is more accurate and reasonable in describing runoff and sediment yield than the SWAT model. The sub-basins whose soil erosion is relatively light are mostly concentrated in the middle reaches with a slope mainly between 0–5°. The NP, PD, ED, SPIIT, SHEI, and SHDI of the watershed increased slightly, and the COHESION, AI, CONTAG, and LPI showed a certain decrease. The landscape pattern is further fragmented, with the degree of landscape heterogeneity increasing and the connection reducing. The runoff, sediment yield and surface runoff are all extremely significantly negatively correlated with forest, which implies that for more complicated patch shapes of forest which have longer boundaries connecting with the patches of other landscape types, the water and sediment processes are regulated more effectively. Therefore, it can be more productive to carry out research on the optimization of water and soil resources under the constraint of carbon emission based on the SWAT-VRR model.

1. Introduction

The loss of soil and water is one of the most serious environmental problems in the world. It leads to soil degradation and is a threat to food security, and also has a great impact on soil carbon storage and the terrestrial carbon cycle, an important biological process. Many studies have shown that the changes of water and sediment in the watershed not only have a close relation with the composition of land use type, but also are more subject to the influence of the spatial distribution pattern of various landscape types [1,2,3]. Landscape pattern refers to the spatial arrangement of different landscape mosaics, which belong to a geographical complex composed of different ecosystems that effectively reveal the regional ecological status and the spatial variation characteristics [4,5]. Landscape pattern change is a result of land use change by human activities, which leads to intensified soil erosion from natural factors, resulting in loss of soil nutrients [6], fertility [7], and land productivity [8]. Sediment enters rivers and reservoirs, and accumulates in river beds [9,10]. Moreover, nutrients carried by runoff intensify water pollution [11,12]. Hence, it is important to study the various factors concerning water and sediment yield in the watershed and screen out the hazard factors by hydrological simulation to formulate effective soil and water conservation measures. This carries important significance for improving ecological environment, achieving harmonious coexistence between man and nature, and pursuing sustainable development [13,14,15].

At present, water and sediment processes in response to landscape pattern changes have become a research hotspot [16,17,18,19]. Lingling Bin et al investigated the effect of landscape patterns on surface runoff and developed a landscape indicator to establish the synergistic evolution relationship between the landscape and the hydrological cycle [20]. Hyun Woo Kim examined the landscape patterns across the four largest metropolitan areas in Texas, and claimed that the less fragmented and more connected landscape patterns are likely to mediate the mean annual peak runoff [21]. Ludwig et al. found in four different landscape structures that landscape patches have an important impact on landscape function. Landscapes without shrub patches result in 25% more runoff losses than those with patches. Striped and linear shrub mosaic patterns are 8% more capable of intercepting runoff than point-like mosaic patterns [22]. The above research shows that water and soil erosion has different occurrence mechanisms for different landscape patterns, it is possible to effectively reduce surface runoff and maintain soil by adjusting the land use landscape pattern.

Through applying GIS technology and distributed hydrological models, soil erosion can be effectively and quantitatively analyzed, for instance, by the Soil and Water Assessment Tool (SWAT model) [23,24,25]. The SWAT model with a strong physical foundation is mainly used to simulate soil erosion [26,27], non-point source pollution [28,29], agricultural management measures [30,31], etc. Suitable for watersheds with different land use types [32], soil types [33] and topographic conditions [34], it can simulate areas with insufficient data, thus meeting the data richness requirements of traditional hydrological models and satisfying the requirements of simulation accuracy [35,36]. Many investigations are available which have analyzed the effects of land use change on runoff and sediment process simulation by the SWAT model [37,38,39,40,41]. Wiktor Halecki et al proposed a multi-criteria approach for various land use scenarios to assess the loss of topsoil erosion by water in an agricultural mountain catchment of the Mątny stream using the SWAT model [42]. Zdeněk Kliment et al evaluated the sediment load changes in the Blšanka river basin and found that the suspended sediment load was reduced after land use changes [43]. Shreeram Inamdar et al applied the SWAT model to determine streamflow discharge and sediment yields for a mixed-land use watershed and announced that the accuracy and reliability of sediment predictions depends on the accuracy of the input information, especially the resolution of the land use layer [44]. Shanghong Zhang et al analyzed the influence of changes in landscape patterns on runoff and soil erosion in each sub-basin of the Liusha River Watershed by the SWAT model indicating that the variations in the erosion modulus were closely related to changes in landscape metrics [45].

However, SWAT divides the area with the same land use type and soil type to form a unique HRU, thus generalizing the slope and the difference of spatial distribution of land use types, as well as adopting each HRU load for separate calculation and summarizing the total load of the watershed. The effect of the flow routing process by forest is not considered. Gassman et al. pointed out that, one area of future research in the SWAT model is more realistic simulation of landscape transport processes [46]. SWAT-VRR (SWAT model with vegetation runoff regulation enhanced) is a distributed hydrological model, modified based on the SWAT model to better show the effect of land use landscape change on hydrological processes in the watershed [35]. It considers the change in the CN value of the land use type with different slopes and flow routing process by forest in different slope zones, thus depicting hydrological characteristics of the study area more accurately [47].

In order to carry out research on the optimization of water and soil resources under the constraint of carbon emission more effectively, this paper builds two hydrological models of the Dapoling watershed based on the SWAT-VRR and SWAT model, simulates the perennial runoff and sediment yield, discusses the applicability of model improvement, analyzes the water and sediment yield of the Dapoling watershed, then analyzes the characteristics of the landscape pattern changes from 2006 to 2011 using transfer matrix, landscape pattern index, etc. The changes of runoff, sediment yield, and surface runoff at sub-basin scale and their response to landscape types are then studied with a view to providing a reference for rational land use planning and scientific management of water resources in the watershed.

2. Materials and Methods

2.1. Study Area

The Dapoling watershed is located at the source of the Huaihe River, at 113°15′–114°46′ E in longitude and 31°31′–32°43′ N in latitude. The perennial average temperature is 15.0 °C, the annual precipitation is 800–1400 mm with great interannual variation and uneven distribution within a year. The precipitation from June to August accounts for more than 50% of the annual total, mostly concentrated heavy rains. The forest is a mixed broadleaf-conifer forest of evergreen and deciduous leaves in the northern subtropical zone. Composed of a variety of mesophytic herbs, the grass is zonal vegetation.

2.2. Data

The DEM (Figure 1) was provided by the “National Science & Technology Infrastructure: National Earth System Science Data Sharing Infrastructure (http://www.geodata.cn/ accessed on 24 November 2019)”. It is worth noting that, the modeler should select fine resolution depending on the watershed size and level of accuracy required because more effort is required to prepare and calibrate the model [48]. Many studies have shown that, the optimal DEM resolution for flow and water quality simulation was 30–200 m, the resolution of the graphs in this study is 30 m [49,50,51].

The soil data are derived from the soil property data of the harmonized world soil database (HWSD). The meteorological data, the measured data on runoff, and the sediment yield come from the Dapoling Hydrological Station. The data series is from 2005 to 2015. Land use data (Figure 2) were acquired by interpreting LandSat remote sensing images in 2005 and 2011. According to the land use type list in the SWAT model, ten land use types were classified and named as: AGRL(Agricultural Land-Generic), AGRC(Agricultural Land-Close-grown), FRST(Forest-Mixed), RNGB(Range-Brush), PAST(Pasture), RNGE(Range-Grasses), WATR(Water), URBN(Residential), URLD(Residential-Low Density), UIDU(Industrial). Then, the parameters of these land use types in the SWAT land cover database or SWAT urban database are loaded [52].

2.3. Description of SWAT-VRR Model

The worldwide applications of SWAT have revealed new research opportunities to develop the specific modules of the model to address the recognized weaknesses [53,54], such as proposing a flexible water and soil salinity module to predict salinity processes [55], applying one traditional response matrix approach to assume hydrological responses [56], coupling with the MODFLOW model to simulate groundwater flow [57], incorporating the rain-on-snow(ROS) model to assess how ROS melt affects winter floods [58], etc. However, the effect of forest runoff regulation in different slope zones has so far still not been considered, as the SWAT model neither adjusts the CN value of SCS for slope [59], nor considers the interaction and spatial distribution among patches. Therefore, the SWAT-VRR model has been improved from the SWAT model in three parts:

(1) The definition of slope grade zone was proposed to reduce the gap between the slope of each land use patch and the average slope of the land use defined by the SWAT model.

(2) The precision of the SWAT model simulation was improved by recalculating the CN value of each land use under different slope grades in each sub-basin. The equation adjusting the curve number to a different slope is defined as:

$${\mathrm{CN}}_{2\mathrm{s}}=\frac{{\mathrm{CN}}_2\times \left\{\exp \left[0.00673\times \left(100-{\mathrm{CN}}_2\right)\right]-1\right\}}{3}\times \left[1-2\exp \left(-13.86\times \mathrm{slp}\right)\right]+{\mathrm{CN}}_2$$

(1)

where ${\mathrm{CN}}_{2\mathrm{s}}$ is the average moisture curve number (${\mathrm{CN}}_2$) adjusted for slope, ${\mathrm{CN}}_2$ is the average moisture curve number (${\mathrm{CN}}_2$) for default 5% slope, and $\mathrm{slp}$ is the average fraction slope of the sub-basin.

(3) The SWAT-VRR achieved the effect of vegetation runoff regulation to estimate surface runoff. The equation calculating land use area producing runoff which flows into the forest is defined as:

$${\mathrm{A}}_{\mathrm{i}}^{\prime }=\{\begin{array}{c}\frac{{\mathrm{A}}_{\mathrm{i}}}{2}\times \left[1+\frac{{\mathrm{slp}}_{\mathrm{i}}-{\mathrm{slp}}_{\mathrm{f}}}{{\mathrm{slp}}_{\mathrm{i}}-{\mathrm{slp}}_{\min }}\right] \left({\mathrm{slp}}_{\mathrm{i}}>{\mathrm{slp}}_{\mathrm{f}}\right)\\ {\frac{{\mathrm{A}}_{\mathrm{i}}}{2}}_{\mathrm{i}}\times \left[1+\frac{{\mathrm{slp}}_{\mathrm{i}}-{\mathrm{slp}}_{\mathrm{f}}}{{\mathrm{slp}}_{\max }-{\mathrm{slp}}_{\mathrm{i}}}\right] \left({\mathrm{slp}}_{\mathrm{i}}<{\mathrm{slp}}_{\mathrm{f}}\right)\end{array}$$

(2)

where ${\mathrm{A}}_{\mathrm{i}}^{\prime }$ is the land use area producing runoff that flows into the forest; ${\mathrm{A}}_{\mathrm{i}}$ is the land use area of the slope gradient of sub-basin; ${\mathrm{slp}}_{\mathrm{i}}$ is the average slope of land use (except forest); ${\mathrm{slp}}_{\mathrm{f}}$ is the average slope of forest; ${\mathrm{slp}}_{\max }$ is the maximum slope of the slope gradient of sub-basin and ${\mathrm{slp}}_{\min }$ is the minimum slope of the slope gradient of sub-basin. Equation (2) ensure a high sensitivity of the land use area which produces runoff that flows into the forest, changes to the slope, and area change of land use in different sub-basins.

The Nash–Sutcliffe efficiency coefficient (NSE) and deterministic coefficient R² are used as evaluation indicators to assess the simulation results of the two datasets [60,61]. The performance of the model is treated as being good when NSE is greater than 0.5, and R² is used to evaluate the match of variations between observed data and simulated data, a calculated value closer to 1 suggests a better simulation effect.

2.4. Landscape Pattern Metrics

The ENVI 4.8 and transfer matrix are used to calculate the area change characteristics in various landscape types. The changes of landscape pattern metrics are analyzed using Fragstats 4.2 software. The correlation between water and sediment processes and landscape pattern metrics is determined using spearman correlation analysis. By considering the well-defined and relatively independent landscape pattern metrics with a greater impact on water and sediment yield in the relevant literature, this paper selects Largest Patch Index(LPI), Number of Patches(NP), Patch Density(PD), Edge Density(ED), Contagion Index (CONTAG), Patch Cohesion Index(COHESION), Aggregation Index(AI), Splitting Index(SPLIT), Shannon’s Diversity Index(SHDI), Shannon’s Evenness Index(SHEI) to reflect the landscape area index, fragmentation, shape complexity, connectivity, and diversity, respectively.

3. Results and Discussion

3.1. Model Calibration and Validation

The SWAT-VRR model divides the study area into 33 sub-basins by DEM data, each sub-basin is divided into four slope grade zones (Figure 3) consisting of 0–5°, 5–15°, 15–25° and above 25°. When dividing the HRU, land use, soil and slope thresholds are set to 0%, 20%, and 0% respectively, so that all attributes of land use are taken into the calculation, while reducing the interference of fragmented soil patches. Respectively, SWAT-VRR resulted in a total of 473 HRUs comparing to 197 HRUs by the SWAT model, and the CN value of each HRU was recalculated in the SWAT-VRR model.

The time series of runoff and sediment simulation is 2005–2011, that of meteorological data is 2005–2015. To ensure accuracy and stability in simulation, a one-year warm-up period (2005) is set, and the calibration and validation periods are 2006–2008 and 2009–2011, respectively.

The runoff and sediment yield for 2006–2008 at the outlet of the Dapoling watershed was used to calibrate the SWAT and SWAT-VRR models, the observed data for 2009–2011 was used for model validation [62,63]. In this study, SUIF-2 algorithm in SWAT-CUP (SWAT- Calibration Uncertainty Programs) was used to perform parameter sensitivity analysis, calibration, and validation on the SWAT model and SWAT-VRR model.

Table 1. Evaluation of model simulation results.

	Daily Runoff Simulation				Monthly Runoff Simulation				Monthly Sediment Yield Simulation
Period	SWAT		SWAT-VRR		SWAT		SWAT-VRR		SWAT		SWAT-VRR
	NSE	R2	NSE	R2	NSE	R2	NSE	R2	NSE	R2	NSE	R2
Calibration	0.6	0.63	0.64	0.67	0.71	0.75	0.75	0.8	0.62	0.65	0.67	0.69
Validation	0.61	0.63	0.66	0.69	0.73	0.78	0.78	0.84	0.64	0.67	0.69	0.71

demonstrates the results of model calibration and validation by using observed data, the NSE and R² values for SWAT and SWAT-VRR in monthly and daily simulation are all above 0.6, which show a marginally good or good performance in simulating runoff and sediment yield. During model calibration, the NSE of monthly runoff simulated by SWAT-VRR was increased by 0.04 compared to SWAT, with R² increased by 0.05; the NSE and R² of simulated daily runoff were both increased by 0.04. When simulating sediment yield in the monthly scale, ENS and R² of SWAT-VRR were 0.05 and 0.04 higher compared to SWAT, respectively. During the validation period, the same trend was also demonstrated in the SWAT-VRR model simulation. Accordingly, regardless of the calibration or validation period, the simulation effect of SWAT-VRR is superior to that of the SWAT model, which suggests the SWAT-VRR model is more accurate and reasonable in describing runoff and sediment yield, as it can more effectively display water and sediment processes in the watershed when land use changes.

Figure 4 and Figure 5 display the daily runoff curves of simulation of the two models and observed data from 2006 to 2011, respectively (Figure 4 and Figure 5). As can be seen, the runoff curve of the SWAT-VRR model coincides more with the observed value than that of the SWAT model during the calibration and validation period. Compared with the flat SWAT runoff curve, the SWAT-VRR runoff curve has sharper fluctuations, which is more consistent with the observed data at different rainfall intensities. When the rainfall is small, the daily runoff simulation line of SWAT-VRR is slightly lower compared than that of the SWAT model, but the difference is not obvious; as the rainfall gradually increases, the runoff curve of the SWAT-VRR begins to level up and exceeds that of the SWAT model, which is closer to the observed value; in the case of heavy rainfall, the SWAT-VRR has a significantly higher daily runoff value than the SWAT model, which reflects that when the rainfall is very high, the effect of runoff regulation by forest is weak and produces more runoff. This also coincides with the actual flow routing process.

This trend is also manifested in monthly sediment yield simulation (Figure 6). SWAT-VRR does not differ much from the SWAT model in simulation effect during the dry season (from October to April), and the overall simulation value is slightly lower than that of the latter. However, the simulation effect of SWAT-VRR is significantly superior to the SWAT model in the rainy season (from May to September). In particular, when simulating several peak values, the sediment yield is significantly higher compared to the SWAT model and closer to the observed value. Consequently, compared with the SWAT model, the SWAT-VRR model improved simulation accuracy significantly, which more accurately reflect changes in the runoff generation and flow routing process caused by underlying surface changes, and thus more applicable for simulating changes in the water and sediment processes when the landscape pattern of the study area changes.

3.2. Spatiotemporal Pattern of Sediment Yield

The SWAT-VRR model was used to simulate the sediment yield in each sub-basin from 2006 to 2011 (Figure 7 and Figure 8). Annual sediment yield varies greatly for different sub-basins where, each sub-basin had significantly higher sediment yield in 2007, 2008, and 2010 than in other years. One important reason is that the annual rainfall in these three years exceeded 1150 mm; while the rainfall in 2006, 2009, and 2011 was below 1000 mm, with significantly lower sub-basin sediment yield. The soil erosion in sub-basins 20 and 11 is relatively severe. In particular, sub-basin 11 has the highest sediment yield over the years. These sub-basins are mostly concentrated in the upper reaches of the main channel and the largest tributary. Sub-basins 23, 27, 28, 32 and 33 have relatively light erosion. In particular, sub-basin 33 has the lowest sediment yield over the years. Such sub-basins are mainly distributed in the middle reaches of the main channel.

The sub-basins with severe erosion generally have a high ratio of area with a slope above 5°; sub-basins in the lower main channel have a relatively high ratio of 5–15° slope zone (such as sub-basin 11, 13, 15), whose sediment yield is relatively small; most sub-basins with a slope below 5° have slight soil erosion. The low-slope zone has high intense human activities, implying that human activities have a great impact on sediment yield in the Dapoling watershed. When analyzing landscape pattern, more attention should be paid to the changes of landscape types in the low-slope zone.

3.3. Analysis of Land Use and Landscape Pattern

3.3.1. Landscape Type Conversion

With climate change and social economic development, landscape types are becoming prone to a certain scale of interconversion, which will bring structural changes in the partial landscape pattern. The landscape pattern transfer matrix can illustrate the process of mutual conversion between landscape types.

Table 2. Transfer matrixes of landscape types in the watershed from 2006 to 2011 (km²).

2011
		AGRC	AGRL	FRST	RNGB	PAST	RNGE	WATR	URBN	URLD	UIDU	Total
2006	AGRC	19.13	2.31	2.39	0.02	0.00	0.00	0.14	0.76	0.19	0.12	25.05
	AGRL	2.46	618.72	34.26	3.55	2.85	0.42	8.66	3.41	3.35	1.71	679.40
	FRST	2.29	35.12	721.53	2.18	7.30	0.05	1.51	0.06	0.04	0.45	770.54
	RNGB	0.01	3.48	2.05	38.84	0.11	0.00	0.16	0.00	0.01	0.08	44.74
	PAST	0.00	3.26	7.53	0.19	26.88	0.00	0.08	0.00	0.02	0.00	37.94
	RNGE	0.00	0.42	0.00	0.00	0.00	1.31	0.15	0.02	0.00	0.00	1.91
	WATR	0.36	7.18	2.18	0.24	0.20	0.06	20.77	0.06	0.06	0.11	31.21
	URBN	0.02	0.36	0.08	0.00	0.00	0.00	0.02	6.32	0.00	0.00	6.80
	URLD	0.03	3.05	0.14	0.00	0.02	0.03	0.17	0.00	6.67	0.01	10.12
	UIDU	0.08	0.63	0.16	0.05	0.00	0.00	0.06	0.02	0.00	4.17	5.17
	Total	24.39	674.52	770.31	45.06	37.37	1.87	31.72	10.65	10.34	6.65	1612.89

describes the transfer matrix of the landscape type changes in the study area from 2006 to 2011. As there is very limited area in landscape type conversion, for better quantitative analysis, the landscape types were reclassified as in the discussion below, where the agricultural type (AGRL+AGRC) was decreased by 5.54 km², accounting for 0.34% of the watershed area, and 5.37 km² was finally transferred to the urban type (URBN+ URLD+ UIDU), accounting for 96.9% of the total reduction. The urban type was increased by 5.56 km², and the area transferred from the agricultural type accounted for 96.6% of the total increase. Therefore, landscape types are mostly transferred from agricultural type to urban type, and there is a very limited area involved in other landscape type transfer. For example, the forest type (FRST) area which accounts for 47.52% of the watershed was eventually decreased by only 0.22 km², accounting for 0.01% of the watershed area. However, up to 49.01 km² area was converted to other landscape types during this period, of which, 37.41 km² was converted into agricultural type; 48.78 km² of other landscape types were converted into forest type, 36.65 km² of which was agricultural type. The area involving mutual conversion between different types obviously exceeds the final result. Such mutual conversion between multiple types increases the change of landscape pattern, which mostly occurs in landscape types with intense human activities, like AGRL, URBL, URBN, etc.

It can be seen from

Table 3. Transfer matrixes of landscape types in 0–5° slope zone from 2006 to 2011 (km²).

2011
		AGRC	AGRL	FRST	RNGB	PAST	RNGE	WATR	URBN	URLD	UIDU	Total
2006	AGRC	15.89	2.22	1.68	0.00	0.00	0.00	0.13	0.76	0.19	0.12	21.00
	AGRL	2.38	530.07	18.02	2.34	1.79	0.28	7.33	3.22	3.16	1.67	570.25
	FRST	1.56	19.10	196.39	0.32	1.41	0.02	1.15	0.02	0.04	0.16	220.17
	RNGB	0.01	2.14	0.30	13.52	0.05	0.00	0.10	0.00	0.01	0.02	16.16
	PAST	0.00	2.05	1.57	0.06	9.85	0.00	0.06	0.00	0.02	0.00	13.60
	RNGE	0.00	0.22	0.00	0.00	0.00	0.97	0.14	0.02	0.00	0.00	1.35
	WATR	0.34	6.07	1.52	0.09	0.15	0.06	17.73	0.04	0.06	0.08	26.15
	URBN	0.02	0.31	0.01	0.00	0.00	0.00	0.02	5.97	0.00	0.00	6.34
	URLD	0.03	2.91	0.14	0.00	0.01	0.03	0.13	0.00	6.31	0.00	9.56
	UIDU	0.08	0.58	0.06	0.05	0.00	0.00	0.03	0.02	0.00	2.96	3.78
	Total	20.33	565.67	219.69	16.38	13.26	1.35	26.82	10.06	9.79	5.00	888.35

that, although the slope zone only accounts for 55.6% of the Dapoling watershed area, the conversion area of the landscape type accounts for more than 90% of the total conversion area of the watershed. Besides, the conversion of various landscape types in the 0–5° slope zone tends to be consistent with the conversion in the watershed. WATR has a conversion area of 0.67 km², accounting for 75% of the total WATR conversion area in the watershed. FRST conversion area in the 0–5° slope zone has the lowest ratio among all landscape types, only accounting for 47.9% of the total area, which relates to the fact that FRST is mainly distributed in the higher slope zones. There are some landscape types whose conversion area exceeds the total conversion area at the watershed scale. For instance, URLD is increased by 0.23 km² in the 0–5° slope zone, while this type only has an increase of 0.22 km² in the watershed, indicating that some area of URLD was converted into other types in the high slope zones. Thus, although there was only a small area converted in the final result at the watershed scale, a large amount of areas participated in the calculation. This reflects that the landscape type conversion is more intense than the final result, thereby causing great changes in landscape pattern.

3.3.2. Analysis of Landscape Pattern Metrics

Table 4. Descriptive statistics of landscape pattern metrics.

	Minimum	Maximum	Mean	Standard Deviation	CV
NP	5	123	42.1692	28.50963	67.6%
PD	0.24	12.25	1.6781	2.2589	134.6%
LPI	29.28	90.66	61.2942	19.13789	31.2%
ED	12.14	60.73	26.5245	10.73646	40.5%
CONTAG	38.07	85.82	68.6958	12.01938	17.5%
COHESION	95.08	99.78	99.0432	1.03607	1.0%
SPLIT	1.22	5.54	2.6012	1.17923	45.3%
SHDI	0.36	1.35	0.8665	0.29576	34.1%
SHEI	0.23	0.97	0.5132	0.19502	38.0%
AI	91.29	98.15	96.0243	1.41799	1.5%

shows the mean value and dispersion degree of the 10 selected landscape metrics. Statistics indicate quite different landscape characteristics of the 33 sub-basins, where the coefficients of variation (CVs) of ED, SPLIT, NP, and PD are 40.5%, 45.3%, 67.6%, and 134.6%, respectively, showing great differences. These indexes are mainly used to indicate the degree of landscape fragment and separation, so the landscape fragment degrees vary greatly between various sub-basins. The CVs of CONTAG, LPI, SHDI, and SHEI are relatively small, with the values are concentrated between 15% and 40%. The CVs of COHESION and AI are small, so the aggregation degree of the landscape patches is similar between different sub-basins.

Table 5. The difference of landscape metrics in each sub-basin and watershed during 2006–2011.

Subbasin	NP	PD	LPI	ED	CONTAG	COHESION	SPLIT	SHDI	SHEI	AI
1	2	0.023	−0.157	0.169	−0.077	0.000	0.006	0.002	0.001	−0.025
2	5	0.099	20.135	0.006	−0.151	0.191	−0.746	0.005	0.003	0.001
3	1	0.030	−0.640	0.831	−0.651	−0.021	0.018	0.018	0.010	−0.120
4	−5	−0.071	4.295	−0.265	−0.146	0.102	−0.561	0.007	0.004	0.042
5	−3	−0.073	0.368	0.365	−0.067	0.005	−0.016	0.000	0.000	−0.053
6	−2	−0.013	0.352	0.281	0.095	0.009	−0.017	−0.005	−0.003	−0.043
7	−9	−0.086	0.033	0.260	0.105	0.002	−0.005	−0.006	−0.003	−0.042
8	7	0.130	−0.493	−0.346	1.895	−0.025	0.026	0.007	−0.030	0.057
9	0	0.000	−0.191	−0.709	0.102	0.032	0.003	0.001	0.000	0.100
10	22	0.333	0.422	0.313	0.070	0.002	−0.010	−0.004	−0.002	−0.049
11	−1	−0.019	−19.305	0.258	−2.201	−0.275	1.411	0.003	0.035	−0.044
12	1	1.545	−0.278	−3.709	−0.734	−0.187	0.257	0.032	0.023	0.594
13	8	0.274	−0.594	0.228	3.188	0.016	0.017	0.007	−0.052	−0.039
14	3	2.914	1.749	1.915	−0.091	0.020	0.061	−0.011	−0.008	−1.145
15	−10	−0.317	0.329	−0.816	−0.213	−0.152	0.534	0.011	0.007	0.034
16	2	0.042	0.719	−0.201	0.147	−0.057	0.004	−0.002	−0.002	0.028
17	−6	−0.094	−0.088	−0.090	0.046	0.013	0.072	−0.001	−0.001	0.013
18	9	0.213	0.117	0.349	0.063	−0.018	−0.003	−0.004	−0.003	−0.055
19	−5	−0.200	0.579	−0.250	−0.189	0.001	−0.010	0.007	0.004	−0.093
20	3	0.081	−0.926	0.415	−2.531	−0.074	0.186	0.101	0.049	−0.013
21	18	2.696	−52.248	42.359	57.682	−1.644	2.542	1.353	0.695	−5.825
22	−7	−0.071	0.158	0.330	1.198	0.001	−0.004	0.011	−0.020	−0.038
23	−4	−0.078	−14.302	−0.088	0.039	−0.195	0.572	−0.001	0.000	0.014
24	−2	−0.111	0.815	−0.863	−0.033	0.012	−0.013	0.006	0.004	−0.022
25	−8	−3.074	1.951	−12.310	2.893	0.215	−0.075	−0.001	−0.001	1.758
26	2	0.027	0.096	−0.049	0.236	0.010	−0.003	−0.010	−0.005	0.002
27	4	0.096	0.028	0.268	0.031	0.000	−0.001	−0.002	−0.001	−0.055
28	−12	−0.398	0.961	−1.679	−2.071	−0.047	−0.021	0.005	0.042	0.026
29	−2	−0.326	−0.880	−2.470	−4.207	−0.041	0.077	−0.002	0.078	0.232
30	−1	−0.128	−1.261	1.309	−0.806	−0.055	0.061	0.020	0.012	−0.309
31	14	0.091	0.381	1.335	−0.894	−0.031	0.042	0.030	0.014	−0.224
32	−13	−0.224	0.246	−0.967	0.455	0.020	−0.009	−0.009	−0.005	0.045
33	7	0.111	−0.132	0.160	2.153	−0.013	0.005	0.006	−0.036	−0.028
Total	18	0.011	−4.300	0.135	−0.342	−0.019	0.650	0.014	0.006	−0.038

shows the difference between the Dapoling watershed and each sub-basin in landscape pattern metrics. It can be seen that the NP value of the watershed is increased by 18 and the PD value is increased by 0.011. It should be noted that the NP value is the sum of NP values of all sub-basins, rather than the NP values calculated at the watershed scale. The reason is that the impact of landscape patches on water and sediment processes is often limited to the scope of the sub-basin. If a patch is located in multiple sub-basins, the patch should be divided into multiple patches and added together. The small increase in NP and PD indicates that the landscape at the watershed scale is further fragmented, with the degree of landscape heterogeneity increasing. The landscape spatial structure gradually turns more complex, with the landscape being subjected to increasing human disturbance.

Patch size greatly impacts material migration in the ecosystem, especially the loss of soil and nutrients [64]. LPI declines slightly at the watershed scale. The value of LPI in some sub-basins changed obviously. For example, the change values of LPI in sub-basin 21, 11, 23, and 30 are all less than −1, which reaches the minimum value of −52.248 in sub-basin 21; while the change values in sub-basin 14, 25, 4, and 2 are all greater than 1, which achieves a maximum of 20.135 in sub-basin 2. In the comparison of the distribution of river networks (Figure 1), it was found that sub-basins with great changes are mainly distributed along the main channel and the upper and middle reaches of the largest tributary. These areas are often low-slope zones with intensive human activities. Human activities thus greatly influence the landscape pattern of the Dapoling watershed.

The ED value has an increase of 0.135 at the watershed scale, indicating that the edge density and total length of the landscape patches are increasing, the boundary between the patches is growing, and the shape of the landscape patches is becoming more irregular [65]. Comparison of ED values of the sub-basins reveals that the sub-basins with the larger reduction in ED value are mainly distributed in the middle and lower reaches of the main channel, such as: sub-basin 25, 12, and 29; while sub-basins 30, 14, and 21 with the largest increase are mainly distributed in the upper reaches of the main channel.

COHESION, AI, and CONTAG, which characterize the degree of landscape aggregation in the watershed [66,67], show a certain decrease, while SPIIT, which characterizes the degree of separation, shows a certain increase, implying that the landscape distribution is more dispersed. Combining the sub-basin distribution, we can see that the sub-basins with big changes in landscape aggregation index are mainly distributed near the main channel. For sub-basins with FRST as the main landscape type, changes in the above indexes suggest that the forest fragment degree is increasing, the forest shape is irregular, the connectivity is reducing, and the soil and water conservation capacity is declining.

SHEI and SHDI that characterize landscape diversity in the watershed show a certain increase, indicating that the patch types tend to be equally distributed with growing heterogeneity [68]. The continuous balanced distribution of multiple landscape types reveals the purposeful interference of human activities. While the landscape is constantly fragmented, more attention should be paid to balanced distribution of different landscape types.

In summary, despite the changes of landscape areas in the Dapoling watershed being small, the landscape pattern metrics of each sub-basin are relatively much changed, the different slope zones and different sections of the main channel have different values, leading to diversified landscape pattern changes in the watershed.

3.4. Effect of Landscape Pattern on Water and Sediment Yield

3.4.1. Relationship between Landscape Types and Water-Sediment Yield

Pearson correlation analysis was used for the landscape type and water and sediment processes at the watershed scale, with results shown in

Table 6. The correlation coefficients among landscape types and runoff, sediment yield, surface runoff.

	Land use	AGRC	AGRL	FRST	RNGB	PAST	RNGE	WATR	URBN	URLD	UIDU
Water–shed	runoff	0.518	0.554 *	−0.450 **	−0.024	−0.51	−0.32	−0.186	0.514	0.036	0.002
	Sediment yield	0.394	0.226 *	−0.57 **	0.354	0.396 *	0.44	0.362	0.55	0.368 *	0.014
	Surface runoff	0.334	0.435 *	−0.678 **	−0.02	0.382 *	0.385 *	0.504	0.507 **	0.364 *	0.234
0–5 slope grade	runoff	0.392	0.382	−0.458 **	−0.088	−0.382 *	−0.42	−0.15	0.420 *	0.434	0.114
	Sediment yield	−0.432	0.244 *	−0.578 **	0.286	0.425 *	0.484	0.328	0.654	0.341 *	0.04
	Surface runoff	0.316	0.416 *	−0.686 **	0.026	0.403 *	0.375 *	0.482	0.526 **	0.684	0.21

Note: *: significance at p < 0.05 level, **: significance at p < 0.01 level.

. The runoff is significantly positively correlated with AGRL, extremely significantly negatively correlated with FRST, and not significantly correlated with other types. Sediment yield is extremely significantly negatively correlated with FRST, and significantly positively correlated with AGRL, PAST, and URLD. Surface runoff is extremely significantly negatively correlated with FRST, significantly positively correlated with AGRL, PAST, URLD, RNGE, and extremely significantly positively correlated with URBN. It suggests that FRST can effectively reduce surface runoff and maintain soil. AGRL, PAST, and URLD play a negative role in water and soil conservation in the watershed, where AGRL and URLD have a close relation with human activities. Comparison with Figure 3 shows that more than half the area of PAST is located in the slope zone above 15°, and some AGRL degrades into PAST after soil structure destruction with reduced rate of runoff infiltration and soil water content, making PAST unable to inhibit soil erosion in the Dapoling watershed.

In the 0–5°slope zone, the FRST shows a very significant negative correlation with runoff, sediment yield, and surface runoff, and the difference existing in the 0–5° slope zone of the watershed is quite small. Other landscape types have approximately the same results compared to the the watershed scale, some results are even higher. For instance, URBN shows a significant correlation with runoff in the 0–5° slope zone, and its correlation coefficient with surface runoff is also above the watershed scale value. In conclusion, FRST is the most important landscape type for soil and water conservation. Low-slope zones with intense human activities, have a landscape pattern that thus greatly affects the water and soil conservation function of the watershed.

3.4.2. Relationship between Landscape Pattern Metrics and Water-Sediment Yield

Correlation analysis was performed between the landscape pattern metrics and runoff, sediment yield, and surface runoff in the Dapoling watershed, where, PD was negatively correlated with runoff, and extremely significantly positively correlated with sediment yield and surface runoff (

Table 7. The correlation coefficients among landscape metrics and runoff, sediment yield, surface runoff.

	NP	PD	LPI	ED	CONTAG	COHESION	SPLIT	SHDI	SHEI	AI
runoff	−0.314	−0.432	0.623	−0.563 *	0.636 *	0.446	−0.479	−0.562	−0.437	0.417
Sediment yield	0.256	0.568 **	−0.479 **	−0.431 *	−0.715 **	−0.635 *	0.517 *	−0.352	0.296	−0.398
Surface runoff	0.195	0.768 **	−0.332 **	−0.752 *	−0.823 **	−0.791 **	0.429	0.366	0.349	−0.462 *

Note: *: significance at p < 0.05 level, **: significance at p < 0.01 level.

). LPI is extremely significantly negatively correlated with sediment yield and surface runoff. Comparison with Figure 2 reveals that LPI reduction mainly derives from FRST. FRST plays an important role in regulating the hydrological functions of the watershed, which can promote rainfall redistribution, affect the soil water movement, and change the conditions of runoff generation and the flow routing process, thereby playing a role in water and soil conservation. FRST is the dominant landscape type in the Dapoling watershed; more fragmented landscape means more scattered patches and less ability to inhibit the process of water and soil erosion, and vice versa. ED indicates the degree of irregularity in the landscape shape, which has significant negative correlation with runoff, sediment yield, and surface runoff. This suggests that for more complicated patch shapes of FRST which has a longer boundary connecting with the patches of other landscape types, more runoff and sediment will flow into FRST, thus regulating the water and sediment processes more effectively.

The spatial organization and connectivity of landscape patches reflect the ability of material flow and the interaction strength between heterogeneous landscape patches, which will directly affect the ecological processes and have an important impact on the migration and transformation of pollution elements. The metrics indicating landscape convergence and dispersion are generally significantly related to sediment yield and surface runoff, where, CONTAG is extremely significantly negatively correlated with sediment yield and surface runoff; COHESION is significantly negatively correlated with sediment yield and significantly negatively correlated with surface runoff; AI is significantly negatively correlated with surface runoff; SPLIT is significantly positively correlated with sediment yield. This implies that higher contagion and better cohesion of patches are more favorable for improving the water and soil conservation function of the watershed. Other landscape pattern metrics such as NP, SHEI, SHDI have no significant correlation with the hydrological processes of the watershed.

4. Conclusions

In this paper, hydrological models of the Dapoling watershed were constructed based on the SWAT-VRR and SWAT models, the runoff and sediment yield from 2006 to 2011 were simulated, and the relationship between landscape pattern and water-sediment yield was analyzed. The results show that:

(1) Compared to the SWAT model, the SWAT-VRR is more accurate and reasonable in describing runoff and sediment yield, and the runoff curve of the SWAT-VRR model can reflect changes more accurately in runoff generation and the flow routing process caused by underlying surface changes.

(2) The annual sediment yield varied greatly for different sub-basins from 2006 to 2011. These sub-basins whose soil erosion is relatively severe are mostly concentrated in the upper reaches of the main channel and the largest tributary. Analyzed with the slope, the sub-basins whose sediment yield is small, are mainly located in the 0–5° slope zones. These low-slope zones having high intense human activities, imply that human activities have a great impact on sediment yield in the Dapoling watershed.

(3) The landscape type conversion in the Dapoling watershed is very limited, but the conversion area of landscape type in the 0–5° slope zone accounts for more than 90% of the total conversion area. The landscape pattern at the watershed scale is further fragmented, with the degree of landscape heterogeneity increasing and the connection reducing. The patch shape is irregular, and the patch types tend to be equally distributed with growing heterogeneity. Human activities thus greatly influence the landscape pattern of the Dapoling watershed, and the soil and water conservation capacity is declining.

(4) The runoff, sediment yield and surface runoff are extremely significantly negatively correlated with FRST. The runoff is significantly negatively correlated with ED, the sediment yield and surface runoff are extremely significantly negatively correlated with LPI, CONTAG, COHESION, and ED. This indicates that for more complicated patch shapes of FRST which has a longer boundary connecting with the patches of other landscape types, the water and sediment processes are regulated more effectively.

Therefore, based on the SWAT-VRR model, we can more effectively carry out research on the optimization of water and soil resources under the constraint of carbon emission, which is of great significance to further understand the law of the carbon cycle and the stability of the ecosystem, and to build a low-carbon land use pattern, and promote the low-carbon development of the economy and society.