Spatial Responses of Vegetation-Soil System to Complex Factors in a Sandy-Rocky Island Chain

The vegetation-soil system is fundamental to island ecosystem and changes considerably across sandy and rocky islands due to different natural and anthropogenic factors. An island chain, which is characterized by the coexistence of sandy and rocky islands, the connection of the islands by bridges, and complex inuencing factors, was used as the study area. The vegetation-soil system was represented using different indicators and three newly-proposed indices, namely, vegetation health index (VHI), soil health index (SHI), and vegetation-soil system health index (VSSHI). The complex factors were identied in aspects of island basic factors, landscape pattern, terrain condition, and ecological indices. Then, the spatial responses of the system to the factors were analyzed at island and site scales. Results indicated that the vegetation-soil system showed similar and different responses to the complex factors across the dual scales. The similarity was represented by the higher sensitivities of VHI and VSSHI compared with that of SHI at both scales, and the difference mainly indicated that the inuences of landscape pattern factors distinctly decreased along the scales from island to site. Island area, sea reclamation proportion, vegetation proportion, and natural ecosystem damaged index were the most important factors at island scale, while the ecological indices showed the highest inuences at site scale. The study revealed the spatial characteristics of the vegetation-soil system across different types of islands, claried the spatial responses of the system to complex factors at the dual scales, and identied the main inuencing factors of the system. (Weigelt 2006; Sfenthourakis and Panitsa, 2012; Triantis et al., 2012; Chi et al., 2019a). In this study, the bridges connected the islands with the mainland and different islands in a chain pattern, and the accessibility of island was determined by the order along the bridge from the mainland. Thus, the PTM was represented using the order number. The SRP was obtained by distinguishing the difference of island outlines between the years of 1984 and 2017, because human exploitations were undeveloped and large scales of sea reclamations were not yet conducted in the 1980s. total of 18 indictors were used to represent different aspects of the vegetation-soil system and three new indices, namely, VHI, SHI, and VSSHI, were proposed to measure the overall characteristics of the system. Four types of factors, including island basic factors, landscape pattern, terrain condition, and ecological indices, were identied to cover all aspects of potential factors of the system. Then, the spatial responses of the system to the complex factors were analyzed at two spatial scales (island and site scales), from two perspectives (single factor and complex factors), and using three approaches (regression analysis, correlation analysis, and CCA ordination). The results revealed the spatial characteristics of the vegetation-soil system across different types of islands, claried the spatial responses of the system to complex factors at the dual scales, and identied the main inuencing factors of the system, which could provide reference for island conservation and contribute to the development of island ecology. The results in the island chain indicated that the species diversity was generally low in tree layer and high in shrub and herb layers. The species sensitively responded to the factors only at site scale in shrub and herb layers, which was caused by the difference of articiality in different layers. At island scale, high VHI and VSSHI were always observed on islands with high VP and PTM and low IA, SRP, NEDI, NP, and AWMSI, and islands with high PTM and LII always possessed high SHI. The factors of SRP, VP, IA, and NEDI were the most important. At site scale, the VHI and VSSHI generally increased with the increase in Al, Sl, NDVI, WI, AWMSI, and VP and the decrease in SI1, BT, BSI, and NEDI. Three ecological indices, that is, BSI, SI1, and NDVI, achieved the highest inuences of all the factors. The vegetation-soil system showed similar and different responses to the complex factors across the dual scales. The similarity was represented by the higher sensitivities of VHI and VSSHI compared with that of SHI at both scales, and the difference mainly indicated that the inuences of landscape pattern factors distinctly decreased along the scales from island to site.

means of eld investigation and remote sensing. Then, the spatial responses of the system to the factors were analyzed at two spatial scales, that is, island and site scales, from two perspectives of single factor and complex factors, and using three approaches, including regression analysis, correlation analysis, and canonical correspondence analysis (CCA) ordination (Fig. 1). The study aims to solve the following questions: How the vegetation-soil system spatially varies across the sandy-rocky island chain? How the vegetation-soil system responds to the complex natural and anthropogenic factors at island and site scales? Which factors contributed the most to the spatial pattern of the vegetation-soil system? 2 Materials And Methods

Study area
The island chain is located in the southern waters of Zhejiang Province and adjacent to the East China Sea (Fig. 2). It is the core part of the Dongtou Archipelago and belongs to Dongtou District in Wenzhou City. The 10 islands are named as Is. 1-Is. 10 along the bridges from the mainland. Is. 1 is a sandy island, and the remaining nine islands are rocky islands. The bridges were initially constructed in the late 1990s. Before the time point, the 10 islands were relatively isolated and the communications among the islands were not frequent. After the year of 2000, the bridges were continuously constructed and a series of bridges were formed to connect the 10 islands (Fig. 2). Correspondingly, the accessibility from the mainland was greatly improved, the human activities were increasingly intensi ed, and the communications among the islands were more and more frequent. Is. 1 is the largest island in the archipelago, and Is. 8 is the political, economic, and cultural center of Dongtou District.
The island chain has a homogeneous climate condition because of the small spatial scale, that is, a subtropical, oceanic, and monsoon climate with an average annual temperature and rainfall of 17.9°C and 1410.6 mm, respectively (Chi et al., 2019c). The typhoon frequently happens between July and September. However, vegetation, soil, terrain, and human activities are spatially heterogeneous, especially between sandy and rocky islands. The sandy island is constituted by the sediments from the Oujiang River and continuously extends in recent years. It is covered by various wetland vegetation and agricultural crops. Farming, sea reclamation, pond culture, alongshore industry, and housing construction are main types of island use, of which farmland occupied the largest area of the island. The rocky islands have a similar geological background with the neighboring mainland. Forests, most of which are planted protection forests, cover the largest areas of the islands. Urban and port constructions are the main exploitation types. Thus, the vegetation-soil system on the island chain is in uenced by complex factors and shows distinct spatial heterogeneity. However, current studies on the vegetation and soil of the island chain are very insu cient.

Data source 2.2.1 Field investigation and sampling
We conducted the eld investigation and sampling in September 2018. Based on the island area, terrain condition, and plant community, considering the accessibility and representativeness, a total of 111 sampling sites were set (Fig. 2). A size of 20 × 20 m was designed for each sampling site, and the species in tree layer were investigated in this size. Two quadrates with a size of 10 × 10 m and ve quadrates with a size of 1 × 1 m were set within the sampling site for the species in shrub and herb layers, respectively. Parameters of abundance, height, coverage, and diameter at breast height of each species in tree layers were measured; and parameters of abundance, height, and coverage of each species in shrub and herb layers were measured. Total coverage in different layers was recorded. Terrain factors, including altitude (Al), slope (Sl), and slope aspect (As) were measured. Surface  soils were sampled using a multipoint mixing method. Soil factors, including bulk density (BD), pH, moisture content (MC), salinity (Sa), total carbon (TC), total nitrogen (TN), organic matter (OM), available phosphorus (AP), and available potassium (AK), were then measured in a laboratory. BD was measured using a cutting ring method; pH was measured using a potentiometric method; MC was measured using an oven drying method; Sa was measured using a gravimetric method; TC and TN were measured using an elemental analyzer; OM was measured using a potassium dichromate oxidation method; AP was measured using a sodium hydrogen carbonate solution; and AK was measured using ammonium acetate solution and ame photometry.

Remote sensing
SPOT 6 and Landsat 8 data in 2017 were adopted. The SPOT 6 data, which has a high spatial resolution, was adopted for generating the island outline ( Fig. 2) and landscape types by using a visual interpretation method. The landscape types included 10 types and 24 sub-types, which were shown in a previous study by the authors (Chi et al., 2019c). The Landsat 8 data, which possesses multiple spectra, was used to produce various ecological indices through radiometric calibration, atmospheric correction, and band calculation.

Vegetation indicators
Species composition in tree, shrub, and herb layers were analyzed and the species lists were shown in Tables S1-S3 in the Supplementary Materials. The vegetation was represented by growth condition and plant diversity in the three layers, which comprehensively revealed the quality, vitality, complexity, and stability of the vegetation (Tilman et al., 2006;Chen et al., 2018a). The growth condition was re ected by the total coverage in tree (TCo), shrub (SCo), and herb (HCo) layers. The plant diversity was measured using two frequently-used indices, namely, Shannon-Wiener index (H') and Pielou index (E), which are obtained based on the important value (IV). The IV is calculated using the following equations: where Eq. (1) is used in tree layer, while Eq. (2) is used in shrub and herb layers; IV s,i , Ab s,i , Co s,i , He s,i , and DBH s,i are the IV, abundance, coverage, height, and diameter at breast height of species i in sampling site s, respectively; and Ab s , Co s , He s , and DBH s are the total abundance, coverage, height, and diameter at breast height in sample site s, respectively. The H' and E was calculated using methods reported by Chi et al. (2019a). Then, the H' in tree (TH'), shrub (SH'), and herb (HH') layers, as well as E in tree (TE), shrub (SE), and herb (HE) layers, were obtained. Therefore, TCo, TH', TE, SCo, SH', SE, HCo, HH', and HE were selected as the indicators of vegetation.
Besides, dominant species were identi ed based on the IVs. The species possessing the highest 10 IVs were considered the dominant species in the entire study area. For each of sandy and rocky islands, the species possessing the highest ve IVs were considered the dominant species.

Soil indicators
Nine factors, including BD, pH, MC, Sa, TC, TN, OM, AP, and AK, were used as the soil indicators. BD is a physical parameter that indicates soil porosity and gas; pH in uences soil structure, effectiveness of nutrient elements, microbial activity, and plant growth; MC serves as the main water source for vegetation; Sa represents the soil salinization, which greatly threatens ecosystem health; TC, TN, and OM involve various biogeochemical cycles, of which the latter two, along with AP and AK, are important indicators of soil fertility (Batjes, 1996;Reeves, 1997;Galloway et al., 2008;Cassel et al., 2015).

Vegetation-soil system
The VHI, SHI, and VSSHI were proposed to quantify the overall health conditions of the system. To calculate the three indices, each of the 18 indicators was standardized for realizing the comparability of different indicators. The indicators could be divided into positive, negative, and interval ones according to their properties. The higher the value of the positive indicator is, the healthier the system is, which is opposite for the negative indicator. For the interval indicator, the system is healthy when the indicator value is within an interval and tends to be unhealthy when the indicator value deviates from the interval. In this study, the negative indicators include BD and Sa, the interval indicator refers to pH, and the remaining indicators are positive ones. The standardization method is as follows: 3 where SV i and V i are the standardized and measured values of an indicator in sampling site i, respectively; V lower and V upper are the lower and upper limits of the indicator values, respectively. For positive and negative indicators, the V lower and V upper denote the 5% and 95% percentiles of the indicator values, respectively, to eliminate the effects of extreme values. For interval indicator, that is, pH, the V lower and V upper are given as 6 and 8, respectively. Then, the VHI and SHI are calculated using the following equations: 4 where VHI and SHI are calculated based on the vegetation and soil indicators, respectively. Different indicators indicate different aspects of the system and are all essential to the system. Thus, an equal weight method is adopted. The vegetation and soil indicators are closely interrelated and interacted. Hence, the VSSHI is calculated as follows: Island area (IA), shape, proximity to the mainland (PTM), and proportion of sea reclamation area (SRP) are basic factors for the islands, and determine island geomorphology, carrying capacity, biodiversity, and convenience for human exploitation at a macro scale (Wardle et al., 2003;Whittaker and Fernández-Palacios, 2007;Weigelt et al., 2016;Chi et al., 2019a). The IA was obtained based on the island outline. The shape was represented using an island shape index (ISI), whose detailed calculation method was shown in Chi et al. (2019a) and a higher ISI indicates a more complex shape. In previous studies on island ecology, the PTM was always measured using the distance to the mainland because the studied islands were not connected with the mainland (Panitsa et al., 2006;Sfenthourakis and Panitsa, 2012;Triantis et al., 2012;Chi et al., 2019a). In this study, the bridges connected the islands with the mainland and different islands in a chain pattern, and the accessibility of island was determined by the order along the bridge from the mainland. Thus, the PTM was represented using the order number. The SRP was obtained by distinguishing the difference of island outlines between the years of 1984 and 2017, because human exploitations were undeveloped and large scales of sea reclamations were not yet conducted in the 1980s.

Landscape pattern
Landscape pattern denotes the composition of different landscape types and the con guration of landscape patches, which in uence species number and ow, habitat suitability, soil quality, and many other ecological processes (Thies and Tscharntke, 1999;Zheng et al., 2018;Chi et al., 2018Chi et al., , 2019b. In this study, ve common landscape pattern factors, namely, proportion of vegetation area (VP), proportion of construction area (CP), number of patches (NP), areaweighted mean shape index (AWMSI), and landscape isolation index (LII) were used to represent the vegetation coverage, urbanization degree, landscape fragmentation, shape complexity, and patch isolation, respectively. The VP and CP were calculated using the data of landscape types, and the calculation methods for NP, AWMSI, and LII were shown in Chi et al. (2019b). Furthermore, a natural ecosystem damaged index (NEDI) was proposed in a previous study by the authors (Chi et al., 2019c). The NEDI could accurately measure the negative effects of human activities based on the landscape types, sizes, levels, and processes, and was also used as a factor of landscape pattern.

Terrain condition
Terrain condition affects local habitat, micro climate, and geological stability, and involves the in uences from the sea (Kura et al., 2014;Gao et al., 2016;Ding et al., 2017). Al, Sl, As, and distance to the shoreline (DTS) were used to represent the terrain condition. The former three factors were measured in the eld investigation. As was processed using a south-direction principle; the processed As ranged from 0 to 1 and a higher As indicates a more southerly direction. The DTS was obtained based on the island outline using the tool of Euclidean Distance in ArcGIS 10.0.

Ecological indices
Different types of ecological indices, which represent different characteristics of the island ecosystem, were generated through band calculation based on spectral re ectance. Normalized difference vegetation index (NDVI), salinity index 1 (SI1), salinity index 2 (SI2), brightness temperature (BT), wetness index (WI), and bare soil index (BSI) were adopted, of which NDVI is advantageous for rapidly monitoring the vegetation changes, SI1 and SI2 are effective in representing the soil quality, and BT, WI, and BSI, which are fundamental physical quantities, represent the heat, humidity, and aridity degrees, respectively. The aforementioned factors possessed natural and anthropogenic attributes and covered all aspects of potential factors of the vegetation-soil system ( Table 1).  the system at the dual spatial scales. At island scale, the landscape pattern was analyzed using the extent of island outline. At site scale, the landscape pattern was analyzed using the extent of surrounding area of each sampling site. The surrounding area was considered as a circle that is determined using the sampling site and a certain length as the center and radius of the circle, respectively. To explore the optimum scale for analyzing the in uence of landscape pattern on the vegetation-soil system, the radius was respectively given as 50 m, 100 m, 150 m, and 200 m, and ve of the landscape factors, that is, VP, CP, NP, AWMSI, and LII, were calculated for four times at site scale.

Two perspectives
The spatial responses of the vegetation-soil system to the factors were analyzed form perspectives of single factor and complex factors. From the perspective of single factor, the spatial responses to each of the factors were analyzed to reveal the changes of vegetation and soil indicators, VHI, SHI, and VSSHI under the in uence of each factor. From the perspective of complex factors, the spatial responses to all of the factors were discussed to clarify the spatial characteristics of the system under the comprehensive in uences of the factors and to identify the contributions of different factors to the spatial characteristics.

Three approaches
Regression analysis, correlation analysis, and CCA ordination were adopted to reveal the spatial responses at the two spatial scales from the two perspectives.
Regression analysis was used at island scale from the perspective of single factor. Regression equations of vegetation and soil indicators, VHI, SHI, and VSSHI with each factor were generated through Excel. Four functions, including linear, exponential, logarithmic, and power functions, were attempted, and the one with the highest coe cient of determination (R 2 ) was selected.
Correlation analysis was used at site scale from the perspective of single factor. Correlation coe cients (CCs) of the indicators and indices with each factor were generated through IBM SPSS 18.
CCA ordination was used at the two spatial scales from the perspective of complex factors. Two types of CCA ordinations were conducted. The rst type was aiming to reveal the species spatial pattern in uenced by the complex factors. The species IVs in the three layers were successively used as the input data, and the complex factors, as well as the vegetation and soil indicators, were used as the environmental data. In the other type of CCA ordination, the vegetation and soil indicators were considered as the input data, and the complex factors were used as the environmental data to clarify the spatial responses of the vegetation The dominant species are shown in Tables 2 and 3. In the entire study area, most of the dominant species in tree layer were arti cial plantation species; parts of the dominant species in shrub layer, including Citrus reticulata, Pittosporum tobira, and Rhododendron simsii, were arti cially planted; and none of the dominant species in herb layer were arti cially planted. On different types of islands, Cinnamomum camphora, Casuarina equisetifolia, and Acacia confusa were common dominant species in tree layer on sandy and rocky islands; Pittosporum tobira is the only common dominant species in shrub layer; and the dominant species on sandy and rocky islands were totally different.

Vegetation and soil indicators
The vegetation and soil indicators on different islands are shown in Table 4

VHI, SHI, and VSSHI
The spatial distributions of the three indices among different islands and different sampling sites are shown in Figs. 5 and 6. Of all the islands, the sandy island possessed the lowest VHI and VSSHI, as well as the second lowest SHI. In the rocky islands, Is. 4 and Is. 7 showed the lowest and highest SHIs, respectively, and Is. 6 exhibited the highest VHI and VSSHI (Fig. 5). At site scale, the VHI showed distinct spatial heterogeneity among different sampling sites, whereas the SHI was not as heterogeneous as the VHI. The spatial distribution of VSSHI combined the characteristics of VHI and SHI. In addition, the spatial heterogeneity within the sandy island was lower than that within the rocky islands (Fig. 6).
3.2 Spatial responses to single factor

Island scale
The regression equations of the factors with each indicator and index at island scale are shown in Table 5. The response was considered as insensitive, sensitive, and very sensitive when R 2 was < 0.3, ≥ 0.3 and < 0.6, and ≥ 0.6, respectively. The equation trends were classi ed as increasing and decreasing trends, which indicated that the dependent variable monotonically increased and decreased with the increase in the independent variable, respectively. The

Site scale
CCs of factors with each indicator and index at site scale are shown in Table 6. We considered the response as insensitive, sensitive, and very sensitive when the P value ≥ 0.05, < 0.05 and ≥ 0.01, and < 0.01, respectively. For the vegetation and soil indicators, 10 and two indicators very sensitively and sensitively responded to Al, respectively; 12 and two indicators very sensitively and sensitively responded to Sl, respectively; one and two indicators very sensitively and sensitively responded to As, respectively; three and two indicators very sensitively and sensitively responded to DTS, respectively; 13 and two indicators very sensitively and sensitively responded to NDVI, respectively; 14 indicators very sensitively responded to SI1; ve and ve indicators very sensitively and sensitively responded to SI2, respectively; 12 and one indicators very sensitively and sensitively responded to BT, respectively; 11 and four indicators very    decreased from bottom to top along Axis 2. Shrub species were also distributed in a concentrated spatial pattern, yet to a lesser degree than tree species. In herb layer, TCo, TH', SCo, SH', SE, OM, TN, Al, Sl, WI, NDVI, and VP distinctly increased and HCo, pH, Sa, AK, SI1, BT, BSI, and NEDI distinctly decreased from left to right along Axis 1; Sa and VP distinctly increased and HH', HE, BD, BT, and NEDI distinctly decreased from bottom to top along Axis 2. The species were dispersedly distributed in the diagram. In the dominant species, species 1, 3, 5, 8, 9, and 10 were located closely in the third quadrant, species 4, 6, and 7 were located closely in the rst quadrant, and species 2 was in the edge of the diagram and far from the other dominant species.

Spatial responses of vegetation and soil indictors
The CCA ordination diagrams of vegetation and soil indicators with the complex factors are shown in Fig. 9

Spatial responses of VHI, SHI, and VSSHI
The CCA ordination diagrams of islands and sampling sites with the complex factors are shown in Fig. 10. The relationships of the complex factors with axes are the same as for Fig. 9. At island scale, the islands with high VHI and VSSHI always possessed low IA, SRP, NEDI, NP, and AWMSI and high PTM and VP. The high SHIs were observed on islands with high PTM and LII. At site scale, the sampling sites with high VHI and VSSHI were generally in areas with high Al, Sl, NDVI, WI, AWMSI, and VP and low SI1, BT, BSI, and NEDI, whereas different values of SHI did not show clear spatial inclination.

Contributions of different factors to the spatial pattern
The contributions of the vegetation and soil indicators to the species spatial pattern are shown in Table 7  The abbreviations for the indicators are the same as for Table 4.
The contributions of the complex factors to the spatial pattern of vegetation-soil system are shown in Table 8. At island scale, the contributions of PTM, VP, NEDI, SRP, ISI, and IA to the species spatial pattern were distinctly higher than those of the remaining factors. To the vegetation and soil indicators, the contributions of SRP, VP, IA, and NEDI were the highest, those of AWMSI, NP, LII, and PTM were intermediate, and those of CP and ISI were the lowest. At site scale, DTS, VP, and SI2 contributed the most to species spatial pattern in tree layer, SI1, VP, and LII contributed the most in shrub later, and BSI, NDVI, and SI1 contributed the most in herb later. To the vegetation and soil indicators, the contributions of BSI, SI1, NDVI, and Al were much higher than those of the other factors, and those of As and LII were lowest.  In this study, the species in tree layer were mostly arti cially planted with speci c species, of which Casuarina equisetifolia is the most dominant tree species in the study area and one of the important plantation species in coastal areas of Southeast China (Huang et al., 2003). It is characterized by fast growth rate and high adaptability and has been planted in a large scale since the 1960s. In recent decades, it greatly helped resist the disasters of wind and storm surge and improve the coastal ecosystem stability (Huang et al., 2012). Correspondingly, the species diversity was generally low in tree layer. The shrub and herb species were partly or barely arti cially planted and grew well in the warm and wet climate, thereby showing a high species diversity. The dominant species in tree layer were similar, yet those in shrub and herb layers changed and greatly changed, respectively, between sandy and rocky islands. Humans planted the trees using speci c several species over the study area, resulting in the homogeneity of dominant species in tree layer. The herb species were mostly naturally developed. Thus, the considerable differences in natural conditions between sandy and rocky islands generated distinct heterogeneity of dominant species. The shrub species were partly arti cially planted and partly naturally developed, thereby resulting in the coexistence of similarity and difference of dominant species between sandy and rocky islands.
The species composition was compared with that in the entire Wenzhou City and the Yandang Mountain. The former covers the study area and was used to represent the regional characteristics (Xiong et al., 2017); the latter is located in the alongshore mainland and was used to represent the species composition of the neighboring mainland (Chen et al., 2018b). At family level, the common families were similar in the three regions, that is, Composita occupied the most species number and Gramineae was one of the common families. At genus level, the similarity of common genera could still be observed, i.e., Polygonum was the common genus in the three regions. At species level, the species lists were generally identical because of the close geographical position and frequent species communication (Chi et al., 2019a). However, species endemic to the study area existed due to the unique natural conditions of the sandy and rocky islands. For instance, Tamarix chinensis, which is a typical wetland plant in North China, was observed in several islands of the study area and was absent in the other areas in Zhejiang Province.

Species spatial pattern
The CCA ordination diagrams for the species showed that the dominant species were generally distributed in a concentrated form near the origin in the three layers at island scale, indicating the insensitivity of species to the environmental factors at this scale (Fig. 7). Though great differences in dominant species exited between sandy and rocky islands, those among the nine rocky islands were not distinct, resulting in the insensitivity. At site scale, the sensitivities of the species to the environmental factors increased along tree, shrub, and herb layers according to Fig. 8. This could also be explained by the arti ciality. In the environmental factors, vegetation indictors were inherently related to the species data, and the remaining factors, including soil indicators and the complex factors, could be considered the habitat factors. Since the species only exhibited sensitivity in shrub and herb layers at site scale, the environmental factors for the two layers at this scale were discussed. In shrub layer, SH', SE, and SCo made the highest three contributions in the vegetation indicators, and Sa, MC, and AK were the most important soil indicators. In herb layer, SH', TCo, and SE contributed the most in the vegetation indicators, whereas pH, TN, and Sa were relatively important soil indicators. Sa was an effective indicator for salinization and greatly in uenced the species spatial pattern in the study area, especially on the sandy island. In both layers, ecological indices and landscape pattern made more contributions than terrain condition did, of which BSI, NDVI, SI1, VP, and NEDI were the most important.
The species-area relationship is one of the core issues in biodiversity study, and the island provides a natural laboratory for this type of study, thereby herb layer was much higher than that in tree and shrub layers, and the slope of trend line increased along the tree, shrub, and herb layers (Figs. 11a, 11b, and   11c). The R 2 and slope of trend line for all species in the three layers combined the characteristics of those in different layers, that is, they were higher than those in tree and shrub layers and lower that those in herb layer (Fig. 11d). It indicated that herb species responded more sensitively to IA that tree and shrub layers, which was in accordance with the study by Chi et al. (2016). From the perspective of single factor, the sum of the R 2 for a factor was considered as its total in uence, and the results were highly identical to the results from the perspective of complex factors (Tables 5 and 8) to NEDI. The VHI and VSSHI were sensitive and exhibited negative trends, whereas the SHI was insensitive. The IA was closely related to SRP and SESI. The sea reclamations were mainly conducted in large islands, including Is.1, Is.2, Is. 5, and Is. 8, of which Is. 1 possessed the highest IA and SRP. Besides, larger islands always carried higher human activity intensity and suffered from more damage, which was proven in the previous study (Chi et al., 2019c). Therefore, similar to the results for SRP and NEDI, most of the indicators, as well as the VHI and VSSHI, showed clear decreasing trends responding to IA. The VP referred to the total coverage of forest, grassland, and wetland vegetation. It is a common parameter to judge the overall ecological condition in a region and involves a series of ecological processes (Zhou et al., 2006). In this study, 11 indicators were sensitive to VP, most of which showed increasing trends. The increase in VP resulted in the increase in VHI and VSSHI. The ISI and CP exhibited little in uence on the vegetation-soil system. The low in uence of ISI was in accordance with the previous studies by Chi et al. (2018Chi et al. ( , 2019a, and indicated that the ISI was not a key factor for islands that were occupied by human activities. The low in uence of CP could be explained by the small difference of CP across islands.
Different vegetation and soil indicators exhibited different sensitivities to the factors. In term of the sum of R 2 from the perspective of single factor, vegetation indicators showed higher sensitivities than soil indicators, and the TE and AP possessed the highest and lowest R 2 , respectively. The VHI, VSSHI, and SHI were in the descending order of R 2 , which was in accordance with the results from the perspective of complex factors in Fig. 10, that is, the VHI and VSSHI responded more sensitively to the factors than the SHI did. Generally, high VP and PTM and low IA, SRP, NEDI, NP, and AWMSI indicated high VHI and VSSHI, and high PTM and LII resulted in high SHI.

Site scale
The In the terrain condition, Al showed a relative high in uence, and the vegetation-soil system was generally healthier when the Al was higher. In the study area, the areas with low Al were always occupied by urban constructions and farming activities, and those with high Al were covered by vegetation, thereby rendering the vegetation-soil system sensitive to Al. The similar situation also occurred on the Miaodao Archipelago in North China, that is, the terrain condition in uenced the vegetation through affecting the spatial pattern of human exploitations (Chi et al., 2016). Although VP and NEDI showed intermediate in uences, the landscape pattern factors were generally weak in in uencing the system at site scale. Different from the other factors, the landscape pattern factors exert in uences at different scales, and the in uences may change with the change in the scale, which is called the scale effect (Wu, 2000;Buffa et al., 2018). In this study, the landscape pattern factors were calculated at island and site scales; at site scale, they were calculated at four scales of 50, 100 m, 150 m, and 200 m. The spatial responses at different scales showed that the in uences of landscape pattern factors were higher at island scale than at site scale, and at the 200 m scale than at the other scales within the site scale, which indicated that the vegetation-soil system responded more sensitively to the landscape pattern at a larger scale.
The sensitivities of the vegetation and soil indicators showed consistent characteristics at island and site scales. The vegetation indicators in tree and shrub layers possessed high CCs in Table 6 and showed distinct spatial inclinations in Fig. 9(b), and the vegetation indicators in herb layer and most of the soil indicators were not sensitive to the factors. Similar to the results at island scale, the VHI and VSSHI exhibited higher sensitivities than SHI, and the SHI responded insensitively to the factors from the perspective of complex factors at site scale. The VHI and VSSHI generally increased with the increase in Al, Sl, NDVI, WI, AWMSI, and VP and the decrease in SI1, BT, BSI, and NEDI.

Conclusions
A sandy-rocky island chain in the Dongtou Archipelago in South China was selected to demonstrate the study on the spatial responses of the vegetation-soil system to complex factors across different types of islands. A total of 18 indictors were used to represent different aspects of the vegetation-soil system and three new indices, namely, VHI, SHI, and VSSHI, were proposed to measure the overall characteristics of the system. Four types of factors, including island basic factors, landscape pattern, terrain condition, and ecological indices, were identi ed to cover all aspects of potential factors of the system. Then, the spatial responses of the system to the complex factors were analyzed at two spatial scales (island and site scales), from two perspectives (single factor and complex factors), and using three approaches (regression analysis, correlation analysis, and CCA ordination). The results revealed the spatial characteristics of the vegetation-soil system across different types of islands, clari ed the spatial responses of the system to complex factors at the dual scales, and identi ed the main in uencing factors of the system, which could provide reference for island conservation and contribute to the development of island ecology.
The results in the island chain indicated that the species diversity was generally low in tree layer and high in shrub and herb layers. The species sensitively responded to the factors only at site scale in shrub and herb layers, which was caused by the difference of arti ciality in different layers. At island scale, high VHI and VSSHI were always observed on islands with high VP and PTM and low IA, SRP, NEDI, NP, and AWMSI, and islands with high PTM and LII always possessed high SHI. The factors of SRP, VP, IA, and NEDI were the most important. At site scale, the VHI and VSSHI generally increased with the increase in Al, Sl, NDVI, WI, AWMSI, and VP and the decrease in SI1, BT, BSI, and NEDI. Three ecological indices, that is, BSI, SI1, and NDVI, achieved the highest in uences of all the factors. The vegetation-soil system showed similar and different responses to the complex factors across the dual scales. The similarity was represented by the higher sensitivities of VHI and VSSHI compared with that of SHI at both scales, and the difference mainly indicated that the in uences of landscape pattern factors distinctly decreased along the scales from island to site.

Declarations
Funding and Con icts of interests Figure 1 Framework for identifying the spatial responses of vegetation-soil system to complex factors in a sandy-rocky island chain   Spatial distributions of soil indicators at site scale: The abbreviations for the indicators are the same as for Table 4. The legends were divided using a quantile method in ascending order of indicator values.

Figure 10
Page 25/25 CCA ordination diagrams of islands and sampling sites with the complex factors at the dual scales: Different colors of circles indicate different index values of as for Fig. 6. (a), (c), and (e) refer to VHI, SHI, and VSSHI at island scale, respectively; (b), (d), and (f) denote VHI, SHI, and VSSHI at site scale, respectively. Abbreviations for the factors are the same as for Table 1.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.