Introduction

Habitat loss and fragmentation are key drivers of global biodiversity losses, and also impacts ecosystem function (Haddad et al. 2015). Studies demonstrate that ~ 40% habitat destruction is the maximum permissible habitat loss to avoid high rates of biodiversity loss for most species, and rare species can withstand the loss of even less habitat (Yin et al. 2017). In many regions, restoration of these progressively fragmented landscapes is essential for conservation and the maintenance of ecosystem services (Cotter et al. 2017).

Developing countries often harbor the highest levels of global biodiversity, but with rapid rates of agricultural conversion are in need of effective tools to minimize the impacts of development on their native biodiversity. China is classed as a global biodiversity hotspot (Ministry of Environmental Protection of China 2010; Myers et al. 2000) and over 50% of its species occur in Yunnan province (Gao and Zhu 2014). Yet deforestation continues in this region especially in Xishuangbanna in southwest Yunnan, which is the most diverse part of the province (Xu et al. 2014).

Like most tropical regions, this area is undergoing rapid land-use and land-cover change. Rubber cultivation in Xishuangbanna started in the 1950s and now represents about 30% of China’s rubber production area, replacing many former crops as a major source of income (Sturgeon et al. 2014; Huang et al. 2015). The expansion of rubber plantations has reduced and fragmented the forested area, especially in the tropical seasonal rainforest (Li et al. 2007, 2009; Chen et al. 2016; Zhai et al. 2017). The forest area in Xishuangbanna decreased by approximately 25% between 1976 and 2014, and most of the cleared forested area had been converted into crops such as rubber, tea or banana plantations (Li 2008; Mei 2015). A quantitative landscape analysis of Menglun Township, Xishuangbanna during 1988 to 2003 showed that the average patch-area of forested land decreased at an annual rate of about 431 hm2, and the number of patches increased from 368 to 441, as a result of the expansion of plantations (Liu et al. 2006). Moreover, new plantations are increasingly utilizing higher altitude and latitude areas (Li 2008; Mei 2015; Chen et al. 2016) and continuing to expand into formerly forested areas. The rapid development of the road network has also increased the fragmentation of this area (Fu et al. 2010).

Plantations above 900 m elevation or on slopes > 24° were projected to have a negative Net Present Value (profit) over the next several decades, which simultaneously damages ecosystem service provision and biodiversity, and provides minimal economic gains (Yi et al. 2013; Chen et al. 2016; Zhai et al. 2017). Herein, we refer to these uneconomical rubber plantations as “low-profit rubber plantations.” These plantations are a low economic value at a high cost for biodiversity (by removing remaining forest leading to both direct habitat loss and further fragmentation), as they support fewer species than native rainforest (Zheng et al. 2015; Xiao et al. 2014; Zhai et al. 2017). This lose–lose situation makes these low-profit plantations logical targets for forest restoration (Yi et al. 2013).

The local government is beginning to take action to protect the tropical rainforest and has been exploring effective ecological compensation methods (Zhang 2015). These methods include controlling the expansion of low-profit rubber plantations and encouraging the adoption of environmentally-friendly practices. Local people are also supportive of the concept of restoring unproductive rubber plantations to partially restore rainforest (Ahlheim et al. 2015) by enhancing connectivity in addition to increasing forest coverage. Determining the best practices for effectively utilizing limited land resources to conserve the most important areas for biodiversity and to restore the connectivity of the habitat network represents a key research topic (Saunders et al. 1993; Li et al. 2014; Zhang et al. 2016).

Fragmentation is a particularly problematic issue when not only does it reduce available habitat, but it increases the proportion of edge habitat increasing the edge effect and the accessibility of resources (and species) in the forest. In tropical areas where rates of habitat loss are generally highest, (like those in Xishuangbanna), smaller patches are particularly vulnerable to exploitation, and thus can be expected to show a decreased diversity of native species, and an increased proportion of invasive species and early pioneer species (Lawrence et al. 2018).

Controlling further fragmentation and strategic restoration of these systems is essential to effectively conserve local biodiversity (Hughes 2017; Liu et al. 2017). It is both important to know the conservation priorities in current habitat network and the restoration priorities in rubber plantations. Thus understanding and analyzing connectivity to facilitate targeted restoration is a useful approach to maximizing conservation gains for reduced costs. Many studies highlight the importance of maintaining connectivity to maintain species over extended periods, as many species are unable to move easily between patches within a fragmented landscape and thus connectivity is essential to the maintenance of viable populations of any species requiring larger ranges or groups (Hatfield et al. 2018; Mony et al. 2018). Increasing fragmentation also increases the potential for human-wildlife conflict in species with large ranges, such as elephants, which are increasingly forced to travel between patches to access sufficient resources (Shaffer et al. 2018).

To better understand habitat fragmentation landscape connectivity measures the degree to which a landscape can facilitate or impede the movement of organism and other ecological fluxes among habitat patches (Taylor et al. 1993; Tischendorf and Fahrig 2000). Exploring the degree of landscape functional connectivity in conservation approaches can ameliorate negative impacts of fragmentation on biodiversity through targeted restoration and protection (Acevedo et al. 2015). However, identifying where and how to restore connectivity in fragmented landscapes presents a challenge, especially in developing cost effective solutions to reconnecting fragmented landscapes. This challenge requires new and better approaches for analyzing connectivity and assisting in effective decision making to facilitate restoration (Chen et al. 2016; Mitchell et al. 2013; Liu et al. 2017).

Graph theory-based models have been used in landscape connectivity analysis since the end of the last century and have been applied with success in a variety of studies (Shanthala Devi et al. 2013; Wang et al. 2016). These models simplify the complex landscape to links (connections between habitats) and nodes (habitat patches), and their indices can be sensitive to the changes in connectivity induced by the absence of certain landscape elements (Saura and Rubio 2010). Among all graph metrics for prioritizing landscape elements for the maintaining of landscape connectivity, the probability of connectivity (PC) is the only index which fully satisfies the desired properties for decision making. PC is a graph based habitat availability metric which quantifies functional connectivity, it is the probability that two points randomly placed in landscape are interconnected (Saura and Pascual-Hortal 2007). The contribution of landscape elements to overall landscape connectivity can be calculated from the variation of PC caused by removal or adding of individual elements from the landscape (Saura and Pascual-Hortal 2007; Saura and Rubio 2010). Thus the actual and potentially important elements could be both identified. These models therefore offer a simple method to identify the most important patches, not only those in the current habitat network, but also those could be potentially recovered in the landscape to reconnect to maximize connectivity, and the results can be used by local conservationists and planners (Urban and Keitt 2001), especially for conservation and restoration work at extended scales.

However despite the value of these approaches most studies tend to focus on a small number of well-studied species using approaches like radiotracking, and fail to incorporate the majority of species, how they move through the habitat matrix. As gathering sufficient data for any single species requires intense study, other methods are needed to help provide pragmatic solutions to conservation. Because this information is rarely available, we decided to incorporate a range of distances which may be relevant to different species and guilds to use as an indicator of possible scenarios. For the majority of species more detailed information on movement is likely to never exist, and therefore this approach to assaying different measures of connectivity provides a practicable approach to optimize management for more diverse assemblages of species. Thus the study includes both the structural connectivity (see supplements for a discussion on types of connectivity), by measuring fragment size at different time periods, and how they could be connected with restoration; and functional connectivity by encompassing a range of scenarios which assess connectivity of the habitat under different dispersal abilities and distances.

Therefore, in this study: (1) We intend to determine how the landscape functional connectivity in Xishuangbanna has changed over the last 40 years. (2) Identify the patches that contribute the most to functional connectivity within the whole habitat network and the patches that should have be the highest conservation priority by enhancing connectivity and therefore probable value to native species the most. (3) Finally, we also hope to determine which rubber patches in the current low-profit plantations would contribute most to the habitat network if restored to forest.

Methods

Study area

Xishuangbanna prefecture (21° 08′–22° 36′ N, 99° 56′–101° 50′ E), is in southwest Yunnan Province, China and is a mountainous area covering 19,150 km2. It is bordered by Laos to the south and Myanmar to the southwest. The altitude ranges from 475 to 2430 m above sea level. Mountains and hills cover 95% of this region, and the Lancang River (the upper Mekong) winds its way through the prefecture from north to south. The region is dominated by a tropical monsoon climate with an average annual temperature and precipitation of 22.5 °C and 1420 mm. The year can be separated into a rainy season (May–October), a cool dry season (November–January) and a hot dry season (February–April) (Cao et al. 2003). There are seven native forest types in Xishuangbanna: tropical rainforest (including tropical seasonal rainforest in lowlands and tropical montane rainforest on higher elevation), tropical seasonal moist forest, tropical monsoon forest, tropical montane evergreen broad-leaved forest, tropical palm forest, tropical coniferous forest, bamboo forest (Zhu et al. 2015).

Spatial data analysis

Five periods of Landsat MSS, TM, ETM + images (with resolution of 80 m, 30 m and 15 m respectively) from 1976 to 2014 (1976, 1988, 1999, 2007, and 2014), were converted into land-use/land-cover data via supervised classification in ERDAS Imagine software based on former analysis (Li 2008; Mei 2015). Assessment showed that the overall accuracy was 77.3%, 86.4%, 84.7%, 86.5%, and 86.9%, respectively. For this research, we conducted an accuracy assessment for the two land-use types used in this study: rubber plantations and forested areas. The user’s accuracy of the rubber in 2014 is 98.5%, and the forested area from 1976 to 2014 are 93.6%, 93.3%, 89.5%, 92.5% and 96.8% separately. The confusion matrix is listed in the supplement.

The Digital Elevation Model (DEM) was downloaded from the Geospatial Data Cloud (http://www.gscloud.cn) with a resolution of 30 m. Slope, elevation range, and aspect of the study area were all derived from the DEM using surface tools within Arcmap 10.2.

Connectivity analysis

Our graph theory-based connectivity model was implemented with Conefor 2.6 (www.conefor.org). Two types of connectivity analysis were conducted. One was conducted for actual forested area from 1976 to 2014 to quantify the dynamics of landscape connectivity in the last 40 years and identify conservation priorities. The other analysis identified which low-profit rubber plantations would be most effective at increasing forest connectivity if restored.

Indices including Probability of connectivity (PCnum), delta values for PC index (dPC), and number of components (NC) were chosen for the quantification of landscape connectivity and identification of key patches. These indices were selected because they effectively reflect the importance of a patch for maintaining connectivity across the landscape (Baranyi et al. 2011; Saura and Pascual-Hortal 2007; Saura and Rubio 2010). Six dispersal distances were included in the connectivity analysis (i.e., 200 m, 500 m, 1000 m, 2000 m, 3000 m, and 5000 m) to reflect the different dispersal distances of various taxa. We prefer this approach because dispersal distances vary by species and according to the makeup of the matrix, and is unknown for many species. Thus by using a range of distances we can examine how the landscape may be connected for a range of guilds.

$$PC = \frac{{\sum\nolimits_{i = 1}^{n} {\sum\nolimits_{j = 1}^{n} {a_{i}^{*} b_{j}^{*} p_{ij}^{*} } } }}{{A_{L}^{2} }}$$

where N is the total number of habitat nodes in the landscape, ai and aj are the attributes of nodes i and j, AL is the maximum landscape attribute, and P *ij is the maximum product probability of all paths between patches i and j. In this study, the optional parameter AL equals total landscape area (area of the analysed region) remains constant after any node removal or addition if the node attribute is habitat (patch) area (Conefor Sensinode (CS) 2.2 (Saura and Pascual-Hortal 2007), so PCnum was used instead of PC index because the optional parameter AL which equals total landscape area (area of the analysed region) remains constant after any node removal or addition if the node attribute is habitat (patch) area (Conefor Sensinode (CS) 2.2 (Saura and Pascual-Hortal 2007).

Preparing the layers for connectivity analysis

Two types of layers were prepared for connectivity analysis. The first type is forested areas layers, which were extracted from the land-use and land-cover maps of 1976, 1988, 1999, 2007, and 2014, respectively based on Landsat data, with SPOT data to verify accuracy of classification. Only patches over 10 ha were retained for the analysis. The node file and distance file which are needed for the connectivity analysis in Conefor 2.6 were produced in the plug-in component of Conefor 2.6 in ArcGIS, and the attribute chosen was patch area.

The other types of layer showed rubber planted in unsuitable areas, based on productivity and profitability. This data set was extracted from 2014 land-use and land-cover data (the year with the most current available data). The rubber patches were classified as low-profit rubber if they were located on slopes of ≥ 25° or at elevations over 900 m in northerly aspect (including north, northwest, and north east) and 1000 m in a southerly aspect (including south, southwest, and southeast). Rubber plantations that met these requirements were selected, only the patches over 10 hectares were retained and merged with the forest layer as a stock of potential forest patches, and the candidate rubber patches are identified with a “1” in a separate column of the node file to distinguish them from current forest patches.

The forest and rubber plantation layers were then analysed in Conefor 2.6 to ascertain which candidate patches would have the greatest positive impact on connectivity if restored to forest. This analysis allowed us to determine potential importance of rubber plantations in forest restoration based on the connectivity gained.

Identification and mapping of possible restoration priorities

The dPC value was chosen to determine the potential contribution of the low-profit rubber patches to overall connectivity by iteratively assessing the value of each patch of low profit rubber in terms of enhancing connectivity if converted to forest, and assessing maximal gains. It was calculated for distances of 200 m, 500 m, 1000 m, 2000 m, 3000 m, and 5000 m. The ranks of the important patches are determined by the sums of their dPC values over all distances of each patch. For low-profit rubber plantations, the rank for each patch is a sum of the dPC values over all the distances. All calculations were made in Conefor sensinode 2.6. When calculating the node importance value for current forested area, formula 1 was used, and for the node importance value for the low-profit rubber plantation, formula 2 was used.

dPC for the forest patches, the node for which we are calculating its importance value is removed for each calculation:

$${\text{dPC}} (\% ) = 100 \times \frac{{{\text{PC}} - {\text{PC}}_{\text{remove}} }}{\text{PC}}$$
(1)

where PC is the overall index value when all the initially existing nodes are present in the landscape and PCremove is the overall index value after the removal of that single node from the landscape (e.g. after a certain habitat patch loss).

To calculate node importance we use the following formula for each patch of low-profit rubber:

$${\text{dPC}} (\% ) = 100 \times \frac{{{\text{PC}}_{\text{add}} - {\text{PC}}}}{\text{PC}}$$
(2)

where PC is the overall index value when all the initially existing nodes are present in the landscape (this includes only the nodes that actually exist in the landscape, but not any of the potential nodes) and PCadd is the overall index value after the addition of that new node to the landscape (e.g. after a certain habitat patch creation or restoration).

For each of the six dispersal distances, the importance levels of all the low-profit patches to network connectivity were obtained in Conefor 2.6. Then, the node importance values for the different dispersal distances were summed and ranked for each patch. The 50 patches with the highest importance values were selected and exported, and then their spatial distribution was mapped.

Results

The change in functional connectivity of forested area from 1976 to 2014

Forested area decreased consistently, from 1976 to 2014 (Fig. 1), driving a consistent decrease in the PC value (Fig. 2). This indicates that forested areas are losing connectivity. The number of forest fragments increased accordingly, thus patches in the landscape are divided into more separate groups. In 1976, patches were relatively large, and the total number of patches over 10 hectares was 881, this increased to 1258 in 1988, and it reached 1670 in 2014. The PC index decreased by almost 75%, and the most dramatic decrease occurred between 1976 and 1988.

Fig. 1
figure 1

Decrease in forested areas from 1976 to 2014. This map shows the shrinkage of forested area from 1976 to 2014 by overlay of the forested areas of five research periods, with 1976 on the bottom of the overlay and 2014 at the top. The areas which are not covered by 2014 forested area illustrate forested area that disappeared by 2014

Full size image
Fig. 2
figure 2

The changes of functional connectivity (indexes of PCnum and NC) of forested area from 1976 to 2014. PC value of forested area kept decreasing from 1976 to 2014 (solid line), and the dashed line show that NC value kept increasing all the periods (NC Number of components, at dispersal distance of 200 m)

Full size image

The 10 most important forest patches in different years

The patches which are most important are all consistently large, and these patches include the majority of the forested area (Fig. 3). From 1976 to 2014, the patches with highest contribution to the whole functional connectivity consistently decreased in size. Most of the patches with high contribution rates are distributed in the center and west of the study area. In 2014, the forest patches with high contribution rates to total connectivity significantly decreased in area, and the patches in the western portion of the study area experienced the most significant reductions.

Fig. 3
figure 3

The ten patches with highest dPC value in five periods (at the distance of 200 m). The colors from dark green to dark red indicate the rank of dPC value from first to tenth. (Color figure online)

Full size image

The ten most important patches are largely the same patches from 1976 to 2014, and they are only differ in the rank from first to tenth due to changes in dispersal distances (Fig. 4 and Supplement 1, Fig. 4a and b show the maximum and minimum connectivity based on maximum and minimum dispersal distances used in analysis). The first to third most important patches remained the same for each year. These patches were all large and accounted for a large proportion of the total area in each the year.

Fig. 4
figure 4

a The ten patches with highest dPC value of 1976 in six dispersal distances. The colour from dark green to dark red means the rank of dPC value from first to tenth. The dispersal distance are 200 m (a), 500 m (b), 1000 m (c), 2000 m (d), 3000 m (e), 5000 m (f) separately for each dispersal distance, with dPC value ranks color coded. For other years see Fig. S1. b The ten patches with highest dPC value of 2014 at six dispersal distances are shown in the maps. The color from dark green to dark red means the rank of dPC value from first to tenth. The dispersal distance are 200 m (a), 500 m (b), 1000 m (c), 2000 m (d), 3000 m (e), 5000 m (f) separately. Though the same patches remain important the relative rank varies depending on the dispersal distance considered. (Color figure online)

Full size image

In 1976, when forest patches were least impacted by rubber plantations, the impact of dispersal distance only differed for distances < 1000 m (Fig. 4a), because forests had good connectivity and were almost all connected at distance 1000 m in 1976. While in 2014 (Fig. 4b), the impact of dispersal distance changed at all distance levels, which showed that dispersal distance needed for all the forest patches to be connected increased significantly. Dispersal distance does have effects when determining which habitat patch is the most important, as indicated by the changing patterns of connectivity based on the buffer distances utilised. In this study few large patches dominated the whole landscape, so the most important patches were usually those with greatest area (Fig. S1).

The identification of restoration priorities in low-profit rubber plantations

The expansion of rubber plantations is the main cause of forest fragmentation within Xishuangbanna, and by 2014 rubber plantations covered approximately 20% of the total area (Mei 2015). Most rubber plantations are located in areas with low elevation and flat slopes. The majority of rubber plantations exist in areas with elevations between 600 and 800 m and slopes between 5° and 15° (Mei 2015).

In 2014, 839 patches covering 17340 ha of low-profit rubber plantation existed within Xishuangbanna, with nearly 70% of this planted < 7 years ago. As low-profit rubber is often that planted on steeper slopes this means that over time there has been an increasing use of steeper slopes and higher elevations, thus more recently planted rubber plantations use less suitable areas (low profitability) as the more suitable areas have already been converted to plantations. The largest rubber patch size is 183 ha, and the mean patch size is around 20 ha, but low-profit patches are very small in size, and the majority of such patches are very close to the forest edge (Fig. 5), as they frequently use marginal sloping land.

Fig. 5
figure 5

The distribution of low-profit rubber plantations of Xishuangbanna in 2014. The red little patches are the low-profit rubber plantation, and green area is forested area. (Color figure online)

Full size image

Low-profit rubber plantations represent only 3.4% of the total rubber plantation area. The restoration of this small portion of rubber plantations represents a more attainable goal than complete restoration of the area. The locations of rubber patches with the highest dPC values (from rank 1st to 50th) are shown in Fig. 5. Few important rubber plantation patches are distributed in the western portion of the study area because most rubber plantations are concentrated in the middle and east of the study area, and forest fragments in the west are generally smaller. This limits the contributions of these rubber patches to the total landscape connectivity, especially when the area of the patches is the only habitat attribute considered when conducting the connectivity analysis (Fig. 6).

Fig. 6
figure 6

The distribution of the low-profit rubber patches with the greatest potential for recovering landscape connectivity. The size of their symbol is correlated with patch importance instead of actual patch size

Full size image

Discussion

The connectivity of forested area in Xishuangbanna decreased between 1976 and 2014 with potential connectivity decreasing by 75% across this period. Case-studies for birds, bats and insects in Xishuangbanna and Thailand (Aratrakorn et al. 2006; Phommexay et al. 2011) report that natural forest sustains higher biodiversity than monocultural rubber plantations (Xiao et al. 2014; Zheng et al. 2015; Zhang et al. 2017).This study provides further evidence of the negative environmental effects of rubber plantations replacing natural forest in this area (Hu et al. 2008; Tan et al. 2011; Li et al. 2012). While Rubber plantations contribute significantly to the local economy by increasing the income of farmers and the revenue to the government. Therefore it is unrealistic to restore all the rubber plantations to forested area, even though local people acknowledge environmental problems caused by rubber plantation and may be willing to improve environmental management, the policy is still limited to forbidden new rubber plantation from taking forested area, not to convert current rubber plantation back to forest (Zhang 2015). Starting with conversion of low-profit rubber provides a practical conservation solution to achieve a relative balancing between development and conservation.

This study explores how to effectively target the conservation and restoration priorities in an agriculture landscape which need balancing the development and conservation. Here we show that the connectivity changes of any given area can be tracked over time, and the greatest importance rank has changed over time. Furthermore, when different dispersal distances are considered, patterns of connectivity change and this has fundamental implications for managing the system more effectively. By converting some low profit rubber into forest we could improve the potential connectivity of the region, and potentially minimize some of the negative human-wildlife interactions which may otherwise occurs (i.e. human elephant conflict: Nyhus and Tilson 2004).

The identification of conservation and recovering priorities

Most studies about the forest conservation in Xishuangbanna focus on finding and conserving the most valuable and important areas (Fu et al. 2010; Zhang et al. 2016), but this research tries to explore how to restore areas of low economic value but high in biodiversity cost, such as the low-profit rubber plantations to most effectively restore forest connectivity.

Spatially explicit results provide a reference for land-use management and protected area planning and assists in conservation and recovering forested area in Xishuangbanna (Zhang et al. 2015; Yi et al. 2013). Whilst we only assessed the value of restoring low-profit rubber in this analysis, pragmatic and practicable approaches like this are needed to develop real-world solutions for conservation. Low elevation areas are both prime areas for species connectivity and the most economically valuable. However human decisions will always include an economic component assessing the potential restoration value and reflecting this in solutions can boost connectivity and biodiversity retention in an agricultural landscape. Thus only low-profit rubber was considered as areas which could potentially be restored in relation to their ability to restore connectivity to forest remnants.

The results allow targeted interventions which maximizes the effectiveness of conservation outcomes at the minimum cost and are applicable for local land-use planners and conservationists. It answers the two main concerns which face conservationists dealing with the structural fragmentation: the conservation of remnant patches and the effective use of limited resources to maximize conservation value (Li et al. 2014; Zhang et al. 2016). The decrease in connectivity found by this study and others studies for the forested area in Xishuangbanna is consistent despite different approaches and distances (Fu et al. 2010; Liu et al. 2014).

Comparatively, this study covers a longer temporal scale (about 40 years) and a bigger spatial range (whole of Xishuangbanna instead of part of it), and also provides more recent data than similar studies. Over 1500 nodes were considered in this study, including forest patches, and the low-profit rubber plantation patches which could be restored to forest were identified to facilitate conservation (higher than any former similar study (i.e. Fu et al. 2010; Shanthala Devi et al. 2013; Li et al. 2014).

Generic species vs single species

In the functional connectivity analysis literature, studies largely fall into two categories. The first category is species-specific analysis which analyzes the actual habitat patches based on species specific field-based analysis (Fletcher et al. 2016). Given limited resources and the challenges of interpolating connectivity results for a single species to inform broader landscape planning goals, other approaches are needed to provide coarser mechanisms for managing landscapes for biodiversity. Therefore, a second category of studies with more predictive and theoretical approaches has evolved, i.e., the use of “generic species” to explore the potential connectivity of a landscape be used on a range of possible behaviors and abilities.

We used a range of distances because for the majority of species no data exists on their ability to move through landscape matrices, and how distance and land-cover impact on their ability to move between fragments. Whilst many approaches favor carnivores, or other large mammals these have been found in some studies to be poor indicators of diversity patterns and are unlikely to be indicative of the movement ability of other taxa (Andelman and Fagan 2000; Jones et al. 2016). Thus by encompassing a range of distances we allow for finer scale analysis for any given guild, as is required in a landscape scale approach to conservation. Given that the majority of species in most environments are little known, and cannot be easily monitored by current technologies such as individual movement and habitat use (most species are too small for radiotracking, and ringing may impact on fitness), using generic approaches which examine tradeoffs between different guilds is a more pragmatic and practical way to conserving species and ecosystems effectively, by balancing the needs and abilities of different guilds and taxa.

“Generic species” approaches are frequently used in potential connectivity studies and stand in contrast to methods which quantify species-specific information on movement (Fletcher et al. 2016; Sahraoui et al. 2017). In Xishuangbanna, both approaches have been explored. These various studies have highlighted the need to account for the different dispersal capabilities of various species (Liu et al. 2014; Zhang et al. 2016). Recent research has attempted to utilize clustering analysis to identify conservation planning strategies that could be suitable for several species (Lechner et al. 2016). Though optimizing landscape connectivity for the greatest number of species would entail minimizing the distance between patches (i.e. 200 m in this study), this may not be practical due to both linear infrastructure and costs. However, utilizing a range of different distances allows the evaluation of relative gain from incorporating different distances, for example Fig. 4b shows that relative connectivity changes relatively little for major patches with dispersal distances of 500 or 1000 m, yet the cost of such approaches may vary radically, thus using a range of approaches allows more holistic planning or gains relative to costs in complex landscapes.

In this study, six dispersal distances from 200 to 5000 m were analyzed and the results were integrated to determine the importance of the patches. The results therefore conservation aims for species with different dispersal capabilities, and the relative merits of different distances compared. One of the limitations of using a generic species method is that habitat suitability is modeled with a nonspecific value, e.g., area. This results in large patches being identified as the most important patches. If more specific attributes are used, e.g., habitat quality or population density, the method would produce more nuanced results, but cannot be easily generalized (especially for species for which little information exists).

The consequences of this shortcoming increases as habitat patch sizes decrease in area and connectivity, and species-specific critical habitat requirements become less likely to be met (such as specific foraging, denning areas) by coarse level planning. Thus this approach provides a sensible compromise for the majority of species and regions for which only a limited selection of data is available with which to make decisions for conservation and management. Whilst such approaches are less detailed than when more data is considered it is easier to derive answers which should provide effective conservation measures for a wider selection of lesser known species.

We caution that addressing the critical habitat needs of single species is still needed and should be addressed whenever possible. Rare species are not considered in our approach, and could change the prioritization of habitat patch preservation and restoration where rare species critical habitat needs are being compromised. Single specialist species may continue to be lost if their needs are not met, and thus factors like minimum viable population size, and other species specific constraints may need to be considered for rarer species, and such considerations becomes increasingly important as habitat patches become smaller and more fragmented.

Whilst single species conservation is important, it is frequently limited to large “charismatic” species, which in many cases can move more effectively through the landscape than smaller, harder to study and track species. In our approach we use a range of distances between patches to account for this spatial variance, and to at least some degree allow for species with different dispersal abilities. As funding is frequently allocated on a species rather than a habitat or ecosystem basis, it is likely that approaches like ours would act as a complement to single species conservation to ensure that that funding has ecosystem wide applications and benefits whilst still benefiting more fundable species.

The scope of graph theory applications to understanding fragmentation

Statistical approaches such as graph theory can support land-use planning for prioritization of areas for conservation and restoration (Foltête et al. 2014; Sahraoui et al. 2017). In this study graph theory-based models offer a quick and sensible method for quantifying landscape connectivity and identifying conservation priorities. The data demand is flexible. Patch area can be used as a habitat attribute when no further data is available, thus conservation and recovery priorities can be assessed.

Subsequent research, including the habitat evaluation, restoration and conservation of local endangered species, the potential effects of certain development to biodiversity conservation could effectively use and expand upon methods utilized here to provide a dynamic complement to those normally applied.

Conclusion

Our approaches highlight the value of temporal assessments of landscape connectivity using graph theory, to allow targeted and effective interventions for landscape scale conservation and restoration. Such approaches can increase the cost effectiveness of achieving habitat connectivity, and provides a practicable mechanism for landscape restoration planning in increasingly fragmented landscapes.

Ultimately conservation needs to be holistic, integrating social and cultural dimensions to biodiversity and management scenarios. Using approaches like those detailed here we can maximize conservation gains, whilst simultaneously reducing any negative economic implications of those approaches, and by understanding patch history we also provide the ability to understand species specific dynamics. Such approaches if combined with population genetic or more individual based approaches (ringing, camera traps) then minimum viable population size, and dispersal dynamics in these landscapes could be better understood, thus providing a range of conservation and management outcomes from applying patch dynamic and connectivity analyses, such as those shown here to conservation and management.