Guizhou Province is located in the core area of Southwest China, a region with largest continuous karst landform in the world (Chen et al., 2021), covering an area of about 160,000 km² (Figure 1). This region is one of the most rapidly urbanized in China, growing to 38 million people in the last 20 years (Yang et al., 2022a). The climate is subtropical monsoon (Xue et al., 2023). The suitable hydrothermal environment enriches the plant resources in Guizhou Province, and make a wide distribution of evergreen-deciduous broad-leaved mixed forest. Among them, the evergreen plants mainly include the Quercus, Neolitsea, Sloanea, and the deciduous plants are mostly the Cornus, Cerasus, Carpinuspubescens, and Platycarya (Li et al., 2022b).

The Shamu River Basin, Yangchang River Basin, Nanming River Basin and Hongfeng Lake in Guizhou Province are taken as the study areas of karst natural, countryside, urban and island landscapes (Figure 2). Among them, the Shanmu River Basin is located in the Shibing Nature Reserve, and is a typical karst natural landscape, with forest coverage over 90% and rich biodiversity (Tang et al., 2017). The Yangchang River Basin is located in the rural transitional zone between Guiyang city and Anshun city, with an area of 605.61 km². Vegetation is dominated by crops and shrubs due to agricultural cultivation-led human activities, while forests are concentrated in karst peaks (Pan et al., 2023). The Nanming River Basin is located in Guiyang city and its surrounding areas, with an area of 1131.26 km². Human disturbance in the Nanming River Basin is strong. Construction land and arable land occupy most of dissolved depressions, causing the vegetation mostly grass and shrubs (Li et al., 2022b). Hongfeng Lake is an artificial plateau lake constructed in 1958s and is located between the Yangchang River Basin and Nanming River Basin. Islands in the lake are formed by karst peaks emerging from the lake surface. Vegetation on islands is mainly shrubs and a small amount of forest (Lou et al., 2021).

A total of 40 ground sampling sites were established to obtain the plant diversity data in karst natural, countryside, urban and island landscapes with socioeconomic development from low to high (Table 1).

Plant diversity of each ground sampling site was obtained through the aerial photography of UAV due to the steep terrain of karst peaks. In each of the sampling sites, UAV was used to take low-flying aerial photos, and a total of 10,174 UAV images were obtained. UAV images were processed by Pix4D software to generate the digital orthophoto map with a spatial resolution of about 3 cm. Then, plant species and individuals in the canopy was counted by visual interpretation. According to investigation results, the species accumulation curve of tall plant species tended to converge in the four karst landscapes (Figure 3), showing the sampling for tall plant species could represent the richness of tall plant in landscapes. The evergreen trees, such as Cryptomeria fortune, Juniperus formosana, Juniperus chinensis, Taxus chinensis, Cephalotaxus sinensis, Cunninghamia lanceolata, and the deciduous trees, such as Salix wilsonii, Juglans regia, Platycarya longipes, Castanea mollissima, Broussonetia papyifera, were relatively common in the sampling sites. In addition, the shrubs mainly include Cotinus, Viburnum, Rubus, Rosa, and Chimonanthus.

In the field survey, three 400 m² plots were set at the edge of karst peak in each sampling site to validate the identification accuracy for plant species by UAV. The number of tree and shrub species was identified and recorded in plots, and the number of trees and shrubs individuals was counted with a diameter at breast height greater than 2.5 cm.

Landsat 8 images could provide surface reflectance information from visible light to near-infrared band with a spatial resolution of 30 m. Fractional vegetation coverage (FVC) was calculated by the Landsat 8 images that was the same period as field survey. Landsat 8 images could be downloaded from the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov).

Land cover data were from the China Land Cover Dataset (CLCD) (Yang and Huang, 2021) and could be derived from the Google Earth Engine (https://developers.google.cn/earth-engine). The temporal and spatial resolutions of CLCD were 1 year and 30 m respectively. CLCD divided the land use into 9 types, including arable land, forest, shrub, grassland, water, snow or ice, barren, impervious surface and wetland. Three-phase CLCD images (1990, 2005, 2020) were used to detect the changes in land use and calculate the density of arable land and impervious surface.

The Shuttle Radar Topography Mission (SRTM) DEM data were widely used due to high precision (Bhang and Schwartz, 2008). With a spatial resolution of 30 m, SRTM1 V3.0 DEM data were for two uses in this study. The first was to determine the subbasin in karst landscapes using the Hydrology Tools in ArcGIS 10.6 software. The second was to identify the karst peaks by combining them with Landsat 8 images. SRTM1 V3.0 DEM data were derived from the USGS.

Details of the dataset used in this study was in Table 2.

Plant alpha diversity showed a gradual decrease among karst natural, countryside, urban and island landscapes according to UAV survey (Figure 4A). Plant alpha diversity was the highest in karst natural landscape, due to the Hill number index (¹ H, ² H and ³ H) obviously higher than that of the other karst landscapes. Although plant alpha diversity in karst countryside landscape was slightly lower than that of karst natural landscape, plant diversity was still maintained at relative high level, and the average ¹ H, ² H and ³ H could reach a value of 11.3, 9.7 and 8.7, respectively. In contrast, plant alpha diversity decreased significantly in urban and island landscapes due to the lower component of vegetation in sampling sites.

In addition, karst natural and countryside landscapes maintained a good similarity of plant species compositions and biological information exchange, while differences in species composition was obvious in karst urban and island landscape (Figure 4B). The average beta value increased from 0.42 to 0.47 in karst natural and countryside landscapes to 0.60 in karst urban landscape, and reached the maximum value of 0.61 in karst island landscape.

The strong correlation between FVC and SHDI was found (Figure 5A). The coefficient of determination (R ²) reached 0.85, showing FVC could well describe the variations in SHDI. p-value was lower than 0.01, indicating the significant positive linear correlation between FVC and SHDI due to passing the 99% significance test. Then, the spatial pattern of SHDI could be estimated by FVC (Figure 5B). Although the proportion of area with high plant diversity in karst countryside landscape was lower than that of karst natural landscape, the spatial fragmentation remained low. In contrast, the areas with high plant diversity were serious spatially fragmented in karst urban landscape, leading to overall decreases in plant diversity. In karst island landscape, vegetation was only scattered on the island. Therefore, the integrity of karst natural and countryside landscape was better than that of karst urban and island landscape in terms of the spatial pattern of areas with high plant diversity.

Arable land had positive impact on plant diversity, while impervious surface had negative impact, according to the conversion relationship of land use in karst natural, countryside, urban and island landscapes (Figure 6). During the last 30 years, plant diversity maintained rich in spite of a small amount of forest converting to arable land in karst natural landscape. Although arable land was dominant, plant diversity still remained relative rich in karst countryside landscape. Plant diversity was low in karst urban landscape, because a large amount of arable land was converted into impervious surfaces.

In addition, karst peaks and river networks had positive impacts on plant diversity according to Table 3. Karst peak density and river network density both decreased from karst natural, countryside to urban landscapes, which was consistent with the decreases in plant diversity among karst natural, countryside and urban landscapes.

The spatial distance for plant diversity diffusing around karst peaks was about 400 m (Figure 9A). The average SHDI showed a significant decreasing trend with the increasing distance from karst peaks. Distance for plant diversity diffusing around karst peaks reached the limit between 300 m and 400 m in karst countryside landscape, and between 400 m and 500 m in karst urban landscape. Changes in SHDI were no longer obvious with the increasing distance after exceeding the limit.

According to the slope of changes in SHDI of karst peaks in different river network buffer zones (Figure 9B), river network could enhance the ecological connectivity among karst peaks until reaching the distance of 200–300 m, 300–400 m and 300–400 m in karst natural, countryside and urban landscapes, respectively. Therefore, the effect of river network on enhancing the ecological connectivity was about 300 m.

Similarly, there was also a distance limit about 450 m that arable land could influence the ecological connectivity among karst peaks. Because decrease in SHDI of karst peaks tended to 0 when the distance reached about 400–500 m (Figure 9C). For impervious surface, the SHDI of karst peaks gradually increased with the increasing distance from impervious surface, showing a weaker effect of impervious surfaces on ecological connectivity (Figure 9D). When the distance from karst peak to impervious surface reached about 350 m, changes in SHDI of karst peaks were no longer significant.

Plant diversity could diffuse around karst peak. The impacts of river network, arable land, and impervious surface on plant diversity were reflected by changing the ecological connectivity among karst peaks. All these factors influenced the ecological connectivity with a spatial distance limit. Therefore, the barrier distance of ecological connectivity was formed according to karst peak and the combination of river network, arable land, and impervious surface in landscapes, and thus revealed the barrier risk to the ecological connectivity of plant diversity.

The barrier distance of ecological connectivity was calculated in Table 5. The combination of arable land and river networks could influence plant diversity by enhancing ecological connectivity among karst peaks within about 710 m, which was significantly higher than the other combinations with impervious surface, such as the combination of impervious surfaces and river networks (0 m), the combination of impervious surfaces and arable land (120 m) and the combination of impervious surfaces, river networks and arable land (390 m) (Figure 10A).

20 plant diversity aggregation regions were determined as the validation regions by the local Moran’s I index (Figure 10B). The combination type of river networks, arable land and impervious surfaces in each validation region was identified (Figure 10C). In karst natural and countryside landscapes, high plant diversity aggregation was dominant, while low plant diversity aggregation of karst peaks was dominant in karst urban and island landscapes.

When the distance among karst peaks was lower than the barrier distance of ecological connectivity, ecological connectivity among karst peaks maintained well, and formed high plant diversity in landscapes. Otherwise, plant diversity tended to be low (Figure 10C). In validation regions with high plant diversity aggregation, the distance among karst peaks was generally lower than the barrier distance of ecological connectivity, indicating a well ecological connectivity among karst peaks. In contrast, karst peaks tended to be isolated in validation regions with low plant diversity aggregation due to the overall distance among karst peaks higher than the barrier distance of ecological connectivity, thus resulting in the low plant diversity of karst peaks.

Karst peaks could affect plant diversity because terrain heterogeneity was highly correlated with plant diversity (Stein et al., 2014). The topographic relief of karst peaks caused spatial heterogeneity in soil physical and chemical properties, such as soil water content, pH value and soil calcium (Zhang et al., 2014). Differences in spatial distribution of soil physical and chemical properties affected the growth suitability of plants, leading to the spatial heterogeneity of plant species and richness (Zhang et al., 2013), and thus the increase in plant diversity. In addition, human disturbance was an important driver for plant diversity decrease (Li et al., 2022b). The steep slope prevented the human disturbance on karst peaks to some extent. Meanwhile, karst peaks were generally considered one of the regions with the richest plant diversity in karst landscapes (Yang et al., 2021). The above factors indicated karst peaks maintained the rich plant diversity in karst landscapes.

Rivers could affect the ecological connectivity to influence plant diversity by environmental suitability and riparian vegetation zone. On the one hand, a suitable hydrothermal environment was necessary for plant life activities (Yao et al., 2021). Hydrological conditions could change the direction and grade of plant succession by the available water for plant (Zou et al., 2022). Rivers not only controlled the surface water supply to plants but were also the main migration carriers of necessary nutrients for plants (Li et al., 2022a). On the other hand, the impact of rivers on riparian vegetation was significant, because plant diversity along rivers was generally higher than that in areas far from rivers (Zhang et al., 2022a). In addition, rivers were also a positive factor for constructing the resistance surface to derive ecological corridors (Cui et al., 2020). This showed a positive influence of rivers on ecological connectivity and plant diversity.

Arable land and impervious surfaces had the opposite impacts on plant diversity due to the differences in spatial pattern and vegetation composition. First, strong human interference in impervious surfaces made plant species and richness subject to human control. In contrast, the mosaic structure of fallow land, hedges, and shrubs in arable land increased the heterogeneity of the landscapes (Chen and Zhang, 2021), and contributed to maintaining rich plant species diversity (Turtureanu et al., 2014). Second, impervious surfaces were generally distributed in a spatial aggregation pattern (Yang et al., 2022b). Large impervious surface patches cut off the ecological connectivity between habitats (Crouzeilles et al., 2021). Arable land was distributed mainly in the transition zones among karst peaks or between karst peaks and impervious surfaces. It helped to weaken the negative impacts of impervious surfaces, and enhance ecological connectivity among karst peaks (Yang et al., 2021). Thus, arable land had a positive impact on plant diversity by influencing ecological connectivity among karst peaks, while the impacts of impervious surface were opposite.

Plant dispersal among habitats was of great significance for plant diversity formation and maintenance (Liccari et al., 2022). According to current research, the dispersal distance of acorn species ranged from 3 m to 550 m in the eastern Iberian Peninsula, Spain (Pons and Pausas, 2007). 15 tree and shrub species were evaluated in Saipan, and their dispersal distance maintained at about 500 m (Rehm et al., 2019). In addition, the dispersal distance of more than 200 plant species was within 1 km investigated by Thomson et al. (2011). Even in tropical regions where seeds spread farther away, the farthest dispersal distance of tree species was mostly at the level of hundreds meters (Chen et al., 2019). As shown in Table 5, the barrier distance of ecological connectivity also remained at the level of hundreds of meters, which was close to the dispersal distance of most plants. Therefore, it was feasible to use the barrier risk to ecological connectivity to indicate the ecological connectivity of plant diversity among habitats.

Plant dispersal ability was limited in space (Morgan and Venn, 2017). However, the dispersal distance of plants could be influenced by environment factors, such as karst peak, river network, arable land and impervious surface, leading to differences in the barrier risk to the ecological connectivity of plant diversity. In countryside landscape in Guizhou Province, the average distance was about 400 m among karst peaks with rich plant diversity (Yang et al., 2021), within the barrier distance of the four different combination types in Table 5. In the same regions, areas with high plant diversity rarely occurred within about 500 m from towns (Yang et al., 2021), consistent with the 400 m and 520 m barrier distance that included impervious surface in Table 5. For river networks, the gradient difference in plant diversity was greater than the distance of 100 m in riparian zones (Zhang et al., 2022a), within the 300 m ecological connectivity enhancement distance among karst peaks. In summary, karst peaks tended to form high plant diversity when the distance among karst peaks was lower than the barrier distance of ecological connectivity, and vice versa.

Although the main limitation of this study was the difficulty to sample short plants by UAV due to the difficulties in plot investigation at karst peaks and the occlusion of tall plant canopy, the dispersal distance of tall plants was generally longer than that of short plants (Thomson et al., 2017). The barrier distance of tall plants could cover the influence distance by karst peak, river network, arable land, and impervious surface on short plants. It showed the maximum spatial response distance of plant diversity to the barrier risk to ecological connectivity. Therefore, it was reliable to evaluate the barrier risk to the ecological connectivity of plant diversity by the tall plants which could be sampled by UAV.