An integrating approach of cellular automata and ecological network to predict the impact of land use change on connectivity

Ecological Indicators 98 (2019) 149–157

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

An integrating approach of cellular automata and ecological network to predict the impact of land use change on connectivity Yun Huang, Tie-Jun Liao

T

⁎

Southwest University, College of Resources and Environment, 2 Tiansheng Road, Chongqing 400716, China

A R T I C LE I N FO

A B S T R A C T

Keywords: Urban expansion Functional connectivity TGRA Least-cost modeling ANN-CA

As a primary concern in biodiversity conservation, understanding the impact of land use/cover change (LUCC) on functional connectivity has the overarching significance. Yet, LUCC simulation models used in previous studies are insufficient to tackle the complex and non-linear interactions of land-use components, and thus may not be able to provide the accurate impact assessments. This paper aimed to fill this gap and to find a solution that avoids significant connectivity loss in the future LUCC. The artificial neural network (ANN) based cellular automata (CA) and graph theory were combined, to explore potential changes in functional connectivity under alternative urban expansion scenarios. Results show that, both urban land and forested land will increase in all of future scenarios; however, functional connectivity of each species will still decrease except under a population change based scenario. The increase in forested land and the contrary connectivity loss proved the strong barrier effect caused by urban expansion. The impacts vary by species: small forest mammals might benefit more from the gain in the stepping-stone-like patches, while the intra-patch connectivity loss is the primary threat for large terrestrial mammals. Finally, we identified the key connecting patches (i.e., stepping-stones) and suggested that those patches should be prioritized for protection.

1. Introduction The arrival of “Anthropocene” is accompanied by the fast population growth rate and marked expansion of human activities in formerly natural areas (Wu, 2013). As a consequence, serious habitat loss and fragmentation happened, threatening the maintenance of biodiversity (Beninde et al., 2015; Thompson et al., 2017). On the other hand, landscape connectivity is crucial to biodiversity conservation, gene flows and many other ecological processes, as it determines to which degree that a landscape facilitates or impedes animals in reaching new suitable habitats (Kukkala and Moilanen, 2017; McRae and Beier, 2007; Taylor et al., 1993; Theobald et al., 2006). Therefore, the connectivity evaluation in a rapidly changing landscape is of great significance, and can be informative for decision makers to understand: (1) how well habitats are connected; (2) which places, if exploited as urban land, have fewer detrimental effects to landscape connectivity; and (3) which urban expansion rate is rational that balances socio-economic development and habitat conservation. Generally, studies around the relationships between urban expansion and landscape connectivity can be separated into two parts: (1) the urban expansion simulation and projection, and (2) connectivity assessment. The former component is generally achieved by applying a ⁎

variety of land-use change simulation models, for example, the model in Baker et al. (2004), the GEOMOD in Pontius et al. (2001), and the STSM in Daniel et al. (2018). However, several limitations exist in these models: some fail to account for the neighborhood effect, which is crucial in affecting landscape dynamics (Batty and Xie, 1994); some can only simulate the changes in two land use types, failing to model the complete land-use system; and some has extremely high biophysical data demand, which can hardly be met in many regions. In this aspect, the cellular automata (CA) model seems more appealing as it can simulate the complex landscape dynamics for multiple land use types with readily accessible spatial and statistical data (Batty and Xie, 1994; Li and Yeh, 2000). As far as the connectivity assessment, connectivity models based on graph theory is widely used (Thompson et al., 2017; Urban and Keitt, 2001; Watson et al., 2011). Recent interests have shifted from the structural connectivity (Bierwagen, 2006; Urban and Keitt, 2001) that only considers the spatial arrangements of habitats, to the functional connectivity that takes account of realistic organism movements and processes. Urban expansion might lead to variations in network connectivity through: (1) loss of nodes (i.e. habitat shrinkage or loss as being encroached by urban land), or (2) even without the node losses, such expansion could also increase the barrier effect in the landscape

Corresponding author. E-mail addresses: [email protected] (Y. Huang), [email protected] (T.-J. Liao).

https://doi.org/10.1016/j.ecolind.2018.10.065 Received 14 May 2018; Received in revised form 19 October 2018; Accepted 30 October 2018 1470-160X/ © 2018 Elsevier Ltd. All rights reserved.

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

connectivity is significant for regional biodiversity conservation and is informative for local decision makers. We studied the functional connectivity of three hypothetical forest mammals with home range sizes of 2 km2, 5 km2 and 8 km2, respectively, to represent the small, medium and large forest mammals in the TGRA. The natal median dispersal distances of the three focal species were estimated by the model in Bowman et al. (2002): the median dispersal distance = 7 times of the linear dimension of the species home range size (i.e., d1 = 7 2 km, d2 = 7 5 km, d3 = 7 8 km ). Besides, multiple data were collected from diverse sources (Table 1).

matrix that impedes animal movements. The second mechanism is, however, can only be analyzed from the functional connectivity perspective. For these reasons, the least-cost modeling (Adriaensen et al., 2003) gains increasing attentions (Gurrutxaga et al., 2011; Huang et al., 2018a; Rayfield et al., 2010), in which the distances between pairs of nodes are weighted by the cost of movement, to denote the friction of an intervening landscape matrix. There are a few studies delving into the impacts of land use changes on functional connectivity (Albert et al., 2017; Huang et al., 2018a; Marulli and Mallarach, 2005; Mitsova et al., 2011; Tannier et al., 2012, 2016). However, the land-use simulation methods used in these studies fail to tackle the non-linear and complex interactions of landscape dynamics, so that they may not be able to provide accurate forecasts for the impacts of alternative urban growth scenarios on landscape connectivity (Huang et al., 2018b). To bridge this gap, here we proposed an integrating framework of the artificial neuron network (ANN) based CA model and functional connectivity metrics. To illustrate the performance of our framework, we selected the Three Gorges Reservoir Area (TGRA, SW China) as the study area due to it is one of biodiversity hotspots worldwide (Wu et al., 2003). The research objectives (ROs) were as following: RO1 concerned the future land use projections. RO2 addressed the construction of the ecological network based on least-cost modelling, and evaluated landscape connectivity under different land use scenarios. Finally, RO3 aimed to uncover how each of habitat patches functions in different aspects, and prioritize the key connectivity providers.

2.2. ANN-CA model The transition rule is considered as the primary issue in CA simulations. We opted to apply the artificial neural network (ANN) as the transition rule because it is more effective in dealing with non-linear, complex landscape dynamics (Li and Yeh, 2002). An ANN contains three types of layers: an input layer, the hidden layer(s) and an output layer (Fig. S1). In the input layer that has n different variables X = [x1, x2, ⋯, x n]T , each neuron receives a single value corresponding to an element in X , then generates an output value that will be transferred to the next hidden layer, the signal received by neuron j is: netj (p , t ) = ∑i wi, j × x i (p , t ) , where x i (p , t ) is the input neuron i on grid cell p at training time t ; wi, j is the adaptive weight between the input layer and the hidden layer, which is calibrated during the training process. To build connections between the hidden layer and the output layer, the Sigmoid function is used as the activation function, and thus the output, the initial conversion probability from the existing to the land use type k for grid cell p at time t , denoted as P (p , k , t ) , is given by (Li and Yeh, 2002):

2. Materials and methods 2.1. Study area, target species and data collection

P (p , k , t ) =

The TGRA is located at the middle catchment of the Yangtze River in China (Fig. 1), covering 55,000 km2 and including more than 20 county-level administrative districts. This region is one of the richest areas in biodiversity in China, and the biodiversity of genera and families is even among the highest globally (Wu et al., 2003). Besides, TGRA has also experienced fast urban expansion, mainly due to the expansion in residential land that resettles residents who lost (and/or would have lost) home during Three Gorges Dam Project (He et al., 2017). Thus, studying the impacts of urban expansion on landscape

∑j wj,k ×

1 1 + e−netj (p, t )

(1)

where wj, k is the adaptive weight between the hidden layer and the output layer, and is also calibrated during the training process. Random sample (sample rate: 10%) was conducted, and the number of hidden layers was set as 12. To produce fractal properties that are widely found in real land-use patterns and to reflect the uncertainties in landscape dynamics (Li and Yeh, 2002), a stochastic error term (RA ) is usually incorporated in CA

Fig. 1. The study area. 150

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

Table 1 The data used in this study. All raster datasets were transformed into the same resolution (1 km) and projection (Xian_1980_3_Degree_GK_Zone_36) prior to model implementation. Name

Data type & resolution

Source

Land use data (2010 & 2015) Population raster map (2010) GDP (2010) DEM Soil quality: nutrient availability, oxygen availability to roots, excess salts and workability Traffic network

Raster, 1 km

CAS (http://www.resdc.cn/)

Raster, 30 arcsecond

Harmonized World Soil Database (HWSD) V1.2 (Fischer et al., 2008)

Shapefile

Protected areas Population statistical data (2010–2015)

Shapefile PDF

(Center for International Earth Science Information Network - CIESIN - Columbia University and Information Technology Outreach Services - ITOS - University of Georgia, 2013) The World Database on Protected Areas (WDPA) The statistical yearbook of Chongqing (2011–2016)

to explore all possible situations:

models (White and Engelen, 1993), which is formulated as:

RA = 1 +

(−lnδ )α

(2)

(1) The business-as-usual (BAU) scenario, which was derived from the Markov-chain that controls temporal changes among the land use types according to the transition probability matrix (Table S1 in Supporting Material) (Aburas et al., 2017; Sun et al., 2018). (2) The fast urban growth (FUG) scenario, in which the urban expansion rate is 10% higher than that in BAU scenario. Compared with the BAU scenario, except the difference in the demand of urban land, the amount of other land use types must be adjusted as well, because some of they will be converted to the urban land, and the adjustment formula is as following:

where δ is a uniform random variable within [0, 1], and α is the dispersion factor that controls the degree of stochastic disturbance. Meanwhile, the neighborhood effect should also be considered in the simulation as it determines how a grid cell is affected by the status of neighboring cells (Li and Yeh, 2000). The formula is as following:

Ωtp, k =

∑N × N con (cpt− 1 = k ) (3)

N×N−1

where Ωtp, k

is the neighborhood effect on grid cell p with land use type k at iteration time t ; ∑N × N con (cpt− 1 = k ) is the total number of grid cells occupied by land use type k at the last iteration time t − 1 within the N × N window (N = 3 in this study, the Moore rule). Additionally, a constraint factor Cp is also incorporated. If a grid cell p is belonging to the protected areas where land use transitions are totally prohibited, Cp = 0 ; or Cp = 1 otherwise. Next, multiplying the initial conversion probability by these three terms, the final conversion probability TP (p , k , t ) is obtained:

TP (p , k , t ) = P (p , k , t ) ×

Ωtp, k

× Cp × RA

k urban urban Aadjust = (AFUG − ABAU ) × H k−u

(5)

k is the area subtracted from land use type k in BAU scewhere Aadjust urban urban is the urban land area in FUG and BAU scenario, nario; AFUG and ABAU respectively; and H k − u is the proportion of land loss due to urban en-

croachment on land use type k from 2010 to 2015 (Table S2). (3) The population change based (PCB) scenario, in which the urban expansion rate is in line with the predicted population growth rate. The predicted population growth rate was assumed to equal the average annual growth rate of urban permanent population in TGRA (Table S3). The adjustment of the land use amount is similar urban urban to (2) but replacing AFUG by APCB , and taking the absolute value of the difference.

(4)

Finally, the CA simulation determines whether a grid cell converts its own land-use type or not. This is decided by comparing the values of conversion properties: land-use type will convert from the existing one to another with the highest conversion probability value. If the existing land use type on a given grid cell has the highest conversion probability, the state of it remains unchanged; otherwise, it will transit to another state. We used GeoSOS for ArcGIS (available at: www. geosimulation.cn) for the ANN-CA simulation.

2.5. The construction of ecological network An ecological network is a set of nodes and links, in which nodes represent habitat patches and links represent potential movement corridors of animals. The selection of habitat patch for each species was based on two criteria: (1) the area must be larger than the species minimum home range size, to ensure the relatively long-term species viability; and (2) it should be located more than 200 m from urban land or road (Sunde et al., 1998). Forested areas not identified as habitat patches were considered as areas favorable to the species dispersal. Links were represented by the cost-weighted distance in the leastcost model. Through the cost surface (Table 2), the least-cost model finds an optimal route with the minimum curriculum cost value (Adriaensen et al., 2003). The least-cost model was implemented in Linkage Mapper V1.1 (McRae and Kavanagh, 2011).

2.3. Model evaluation We used the trained ANN-CA to simulate the landscape in 2015 from 2010. The model performance was evaluated through the fuzzy Kappa index (Hagen, 2003) between actual 2015 and simulated 2015. Compared with general Kappa index, the fuzzy Kappa index is more appealing as it allows slight displacements of a simulated landscape compared with the actual one, namely, it takes account of not only cellby-cell agreement, but also the influence of neighborhood cells (Vliet et al., 2011). Our simulation result had a fuzzy Kappa index of 0.83. Generally, a Kappa index higher than 0.75 can be considered satisfactory, and thus it demonstrated that our ANN-CA had considerable power in simulation and forecasting. The fuzzy Kappa index was calculated in Map Comparison Kit software (Visser and de Nijs, 2006).

2.6. Evaluation of functional connectivity To evaluate the overall landscape connectivity, the Probability of Connectivity (PC ) (Saura and Pascual-Hortal, 2007) is a popular metric, and is also supported by empirical studies (Engelhard et al., 2017; Pérez-Hernández et al., 2015). Based on the probabilistic connections model, PC is defined as the probability that two animals randomly are

2.4. Future land use scenarios definitions As uncertainties will always exist in the prediction, we created three future land use scenarios based on different socio-economic conditions 151

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

where dPCk is the node importance of k , denoting how k contributes to landscape connectivity; PCinitial is the PC value of the intact network without removing any node; and PCk, removed is the PC value of the remaining network after removing node k . A node may contribute to connectivity through different aspects, for example, if the patch’s area is large enough, it may be important for the intra-patch connectivity; on the other hand, if a patch with an intermediate area appears to be in a vital location in the network, the contribution might mainly lie in the inter-patch connectivity. Such detailed information is, however, unable to be provided by a simple dPCk value. Thus, we partitioned the dPCk into three parts (Saura and Rubio, 2010): dPCk = dPCintra, k + dPCflux , k + dPCconnector , k , each of which representing a different way that node k contributes to habitat connectivity in the landscape through intra-patch connectivity, area weighted dispersal flux and stepping-stone effects, respectively. We were particularly interested in the last fraction, given that those connectors, if lost, would largely isolate the rest of patches in the landscape, while they may be encroached by urban land easily. dPCconnector , k n n corresponds to a part of ∑i = 1 ∑ j = 1 ai aj pij∗ for each pair of i and j , in which i ≠ k , j ≠ k , and k is part of the maximum probability path between them (Saura and Rubio, 2010). This fraction depends only on the topological position of a node in the network, independent of its area. Besides, patches with a ratio of dPCconnector , k /dPCk > 0.5 were identified as key connecting elements (Dilts et al., 2016).

Table 2 Movement costs characterizing each land use type in the cost surface (Gurrutxaga et al., 2011). “H” stands for the preferable habitat, “HM” is the hospitable matrix, and “IM” is the inhospitable matrix. We didn’t take into account the linear elements such as roads and highways, as they are already included in “Urban land”. Land use type

Landscape matrix type

Cost

Cropland Forest Meadow Water body Urban land Unused land

HM H HM IM IM HM

30 1 15 10,000 10,000 40

positioned within the habitat patches which are connected (Saura and Pascual-Hortal, 2007). More importantly, one remarkable merit of PC is that it takes account of the stepping-stone effect, which is of great importance in connectivity evaluation, while often ignored in other direct dispersal models (Saura et al., 2014). Two nodes in a landscape can be directly connected by a single path if they are close enough, or can be indirectly connected by paths made up of a set of steps in which no node is traversed more than once in case they are more distant. The PC index is formulated as (Saura and Pascual-Hortal, 2007): n

PC =

n

∑i = 1 ∑ j = 1 ai aj pij∗ Al 2

3. Results

(6)

where n is the number of habitat patches; Al is the total area of the landscape; ai and aj denote the area of patch i and j ; pij∗ is the maximum product probability of all potential migration paths between i and j . The migration probability is usually formulated as an exponential function of distance units (Clark et al., 1999):

pij = e−αdij

3.1. Future land use scenario projections In 2020, the area of urban land in the FUG scenario has an increase of 47.2% compared with that in 2015, and is the largest one among three scenarios (Table 3). The new urban land is mainly allocated to the current urban fringe, as well as places near the Yangtze River (Fig. 2). On the contrast, the cropland will suffer the most significant loss in the FUG scenario (−1.8%), whereas the variation in the PCB scenario is rather slight. The land use types that will expand in the future include urban land, forest land and water body, while cropland, meadow and unused land decreases in all scenarios.

(7)

where pij is the migration probability from i to j ; dij is the cost-weighted distance unit in the least-cost model case; α is the coefficient reflecting the dispersal abilities and varies with species. The median natal dispersal distance of each species was multiplied by the median value of movement cost in the cost surface (median cost: 30), and their products corresponded to a 0.5 migration probability for each species (Gurrutxaga et al., 2011). For example, for the large forest mammal, its 3 α value was parameterized to make e−α × 7 8 × 10 × 30 equal 0.5. We employed Conefor 2.6 (Saura and Torne, 2009) to calculate these functional connectivity metrics.

3.2. Functional connectivity in current and future scenarios All of three future scenarios are forecasted to have forested land expansion; however, the habitat area of each target species in all alternative scenarios is expected to shrink (Fig. 3). Besides, the changes in frictions of the landscape matrix also alter the least-cost corridors connecting each pair of core habitat patches (Fig. 4). Based on the costweighted distances along these least-cost corridors, PC index was calculated (Fig. 5). Curves of 2020 BAU and FUG scenarios are below the curve of 2015, while the curve of 2020 PCB is slightly higher than that of the current. Taking the comparisons among different species in a given land use scenario, we found that functional connectivity would decrease with the increasing dispersal distance, which was contrary to our intuitions, and reasons would be explained later.

2.7. Mapping key connecting nodes Based on PC , the node importance can be identified by removing a given node from a network and calculating the variation in PC . Hence, the node k 's importance is represented by:

dPCk =

PCinitial − PCk, removed × 100% PCinitial

(8)

Table 3 The amount of each land use type (unit: km2) in current (2015) and the future (2020) with different development scenarios (BAU, FUG and PCB). Land use type

Cropland Forest Meadow Water body Urban land Unused land

2015

21,368 28,132 6237 1144 1646 5

2020 BAU

Variation (%)

FUG

Variation (%)

PCB

Variation (%)

21,042 28,408 5434 1294 2343 3

−1.5% 1.0% −12.9% 13.1% 42.3% −40.0%

20,982 28,393 5431 1293 2423 2

−1.8% 0.9% −12.9% 13.0% 47.2% −60.0%

21,330 28,472 5445 1297 1977 3

−0.2% 1.2% −12.7% 13.4% 20.1% −40.0%

152

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

Fig. 2. The land use maps of the actual 2015, the simulated 2015 and projected future land use scenarios.

(intermediate and long dispersers), they were likely to suffer the most severe loss in their key connecting nodes under the FUG scenario (Table 4).

3.3. Loss of key connecting nodes in 2020 We plotted the distribution maps of important connectivity providers, which indicate the node importance by dPC, dPCconnector and their ratio, respectively (Fig. 6). Figures show that the distribution patterns of nodes with large dPC and dPCconnector values are distinct: rather few nodes have high dPC values, the largest one is located in the eastern TGRA, which is the largest habitat patch; however, this patch only has a moderate stepping-stone effect; the node with the largest dPCconnector value appears in the central-southern TGRA. Regarding the key connecting nodes, the amount of them decreases with the increasing dispersal distance, and these nodes are mainly located in the urban fringe and the central TGRA. By overlapping the projected 2020 land use maps on the distribution map of key connecting nodes, we found that, for small mammals (i.e., short dispersers), the amount of encroached key connectors was the same under all scenarios; however, for medium and large mammals

4. Discussions The methods used and results obtained raised several points for discussions. Regarding the urban expansion simulation, due to the model development over almost twenty years, diverse transition rules for CA simulation have been proposed, including the logistic regression (Wu, 2002), the agent-based model (Li and Liu, 2008) and many others. We opted to use the ANN as the transition rule because it does not require much information about individual’s preference to infer the decision process as the agent-based model does, yet it is robust and effective in dealing with the non-linear relationships between different land use types than the regression approach (Li and Yeh, 2002). The ANN-CA model performance in landscape dynamics simulation was Fig. 3. The areas of forested land, habitats and nonhabitat forested land in different land use scenarios. ‘Non-habitat (small)’ stands for the non-habitat forested land of small mammals, and so forth. The label above each group of bars denotes the total forest area in the corresponding scenario (Total forest area = Habitat area + Non-habitat forest area).

153

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

Fig. 4. The examples of cumulative cost surfaces and the corresponding least-cost paths for large forest mammals in different land use scenarios.

specific estimate (Koen et al., 2014), so we studied the functional connectivity of forest mammals with a wide range of home range size, as well as dispersal abilities, aiming to represent most species in the TGRA. According to the overall connectivity assessments, even though forested land increases in all of future land use scenarios, the functional connectivity of each species still decreases in the BAU and the FUG scenarios (Fig. 5). The reason is that, the amount of forested patches that are qualified as habitats declines with the urban expansion (Fig. 3). For example, a patch with an area of 8 km2 in 2015 could be considered as a habitat patch for large mammals, but once part of it is converted to the new urban land, it can no longer support the long-term viability of large mammals; instead, it can only be a land use type favorable to the animal movements. Similarly, now we can answer why the functional connectivity of the large mammals (long dispersers) is the lowest whereas that of small mammals is the highest. Intuitively, benefited from the strong dispersal abilities, more habitats are reachable for long dispersers, and thus they are supposed to have a higher connectivity degree (Fu et al., 2010; He et al., 2018; Huang et al., 2018a; Liu et al., 2014). Yet, these two studies overlooked the habitat size requirement of long dispersers. Such requirement filters out many forest mosaics that can provide habitats for small or medium mammals. Therefore, compared with the other two, less intra-patch connectivity is provided in the habitat network of large mammals, and further leads to the low overall connectivity value of large mammals. Another thing worth noting is that, in the PCB scenario, the amount of habitats for each species is even smaller than that in other two scenarios, but surprisingly, the connectivity degree of each species is even higher, regardless of the shrinkage of habitat area. This suggests that, in this scenario, the increased forested area that are favorable to animal movement eases the friction of landscape matrix, and also increases the inter-patch connectivity, which outweighs the loss in intra-patch connectivity. Besides, between the PCB scenario and the current status, the difference of connectivity of small mammals is larger than that for middle-sized or large mammals, indicating that short dispersers might benefit more from the gain in inter-patch connectivity. Such difference also provides a guide to the spatial conservation prioritization (SCP) issue in GFGP:

Fig. 5. Functional connectivity (indicated as PC) that varies with different species in alternative land use scenarios.

validated, and the calibrated ANN-CA was then used to project future land use scenarios based on three alternative development modes. Projections results show that, forest area and urban land area both increase in all future scenarios. In the BAU and PCB scenarios, urban expansion represents the pattern of gap-filling, the new urban land mainly concentrates in the city core proper. Nevertheless, in the FUG scenario, there are also some new urban land allocated along the Yangtze River. The increase in forested land is due to the Grain-forGreen Project (GFGP), which converts the cropland in the steep slope to the forest, with the primary goal of bringing soil erosion and frequent flooding in the upper reaches of the Yangtze River under control (Long et al., 2006). The functional connectivity evaluation is most often a species154

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

Fig. 6. The importance of each node (the centroid of the habitat patch) in 2015 indicated by dPC, dPCconnector and the ratio between them. Panels (A)-(C) were mapped for small forest mammals, (D)-(F) for medium forest mammals, and (G)-(I) for large forest mammals.

cases, urban planners might not be interested in knowing which patch is important for its size, as it is very easy to compare; instead, they might be more interested in identifying patches that, if lost, would seriously isolate the remaining habitats. Therefore, partitioning dPC into three components and scrutinizing the dPCconnector fraction can help us understand which node dominantly contributes to connectivity by its stepping-stone effect (Saura and Rubio, 2010), and we suggested that these key connecting nodes should be put under strict protection. Results show that, although the loss rates of key connecting nodes for short dispersers are equal in three alternative scenarios, they would be much serious for middle and long dispersers under the FUG scenario, and thus the FUG mode would not be recommended considering the conservation of large mammals. Besides, the amount of key connecting nodes decreases with the increasing dispersal abilities, indicating that when dispersal abilities are large enough, species can move directly from one patch to another without using the intermediate stepping stones, as a consequence, most intermediate patches only have moderate contributions to facilitate the large mammal’s movements, while relatively fewer patches could act as key connecting elements, which agrees with the findings in Garcia-Feced et al. (2011) and Saura and Rubio (2010). Our research also had some limitation. We recognized that the spatial resolution (1 km), is quite coarse. The barrier effect of urban land, as well as the edge effect of forested land, are sensitive to the spatial resolution, and can be better delineated with a finer resolution (Toger et al., 2016). However, we argued that our analysis would still be reliable enough, as this resolution is rather common in other related studies (please see the review in Sawyer et al. (2011)).

Table 4 The amount of key connecting nodes entirely encroached by urban expansion in 2020 under different scenarios. Species

The amount of key nodes in 2015

Scenario

The amount of key nodes encroached by urbanization

Key nodes loss rate (%)

Small mammals

35

BAU FUG PCB

6 6 6

17% 17% 17%

Middle-sized mammals

10

BAU FUG PCB

1 2 1

10% 20% 10%

Large mammals

7

BAU FUG PCB

2 3 1

29% 43% 14%

depending on the specific conservation objectives, agricultural patches prioritized for reforestation might either be the ones that can potentially act as the stepping-stones (for short dispersers), or the ones close to a forest patch whose area is only slightly smaller than the home range size of a given species (for long dispersers), so that the adding reforested patch can increase the habitat amount, as well as the intrapatch connectivity. As far as the prioritization of key connectivity providers, we found that, those nodes with high dPC share two common characteristics: (1) having large area and (2) appearing in the vital location in a network. After all, dPC value itself is a joint result of patch area and topological location (He et al., 2018). However, even two nodes have the similar dPC value, it does not necessarily indicate that they can equally contribute to landscape connectivity (Bodin and Saura, 2010). In many 155

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

5. Conclusion

He, J.H., Huang, J.L., Li, C., 2017. The evaluation for the impact of land use change on habitat quality: a joint contribution of cellular automata scenario simulation and habitat quality assessment model. Ecol. Model. 366, 58–67. He, J.H., Huang, J.L., Liu, D.F., Wang, H., Li, C., 2018. Updating the habitat conservation institution by prioritizing important connectivity and resilience providers outside. Ecol. Indic. 88, 219–231. Huang, J.L., He, J.H., Liu, D.F., Li, C., Qian, J., 2018a. An ex-post evaluation approach to assess the impacts of accomplished urban structure shift on landscape connectivity. Sci. Total Environ. 622, 1143–1152. Huang, Y., Huang, J.-L., Liao, T.-J., Liang, X., Tian, H., 2018b. Simulating urban expansion and its impact on functional connectivity in the Three Gorges Reservoir Area. Sci. Total Environ. 643, 1553–1561. Koen, E.L., Bowman, J., Sadowski, C., Walpole, A.A., 2014. Landscape connectivity for wildlife: development and validation of multispecies linkage maps. Methods Ecol. Evol. 5, 626–633. Kukkala, A.S., Moilanen, A., 2017. Ecosystem services and connectivity in spatial conservation prioritization. Landscape Ecol. 32, 5–14. Li, X., Liu, X.P., 2008. Embedding sustainable development strategies in agent-based models for use as a planning tool. Int. J. Geogr. Inf. Sci. 22, 21–45. Li, X., Yeh, A.G.O., 2000. Modelling sustainable urban development by the integration of constrained cellular automata and GIS. Int. J. Geogr. Inf. Sci. 14, 131–152. Li, X., Yeh, A.G., 2002. Neural-network-based cellular automata for simulating multiple land use changes using GIS. Int. J. Geogr. Inf. Sci. 16, 323–343. Liu, S., Dong, Y., Deng, L., Liu, Q., Zhao, H., Dong, S., 2014. Forest fragmentation and landscape connectivity change associated with road network extension and city expansion: a case study in the Lancang River Valley. Ecol. Indic. 36, 160–168. Long, H.L., Heilig, G.K., Wang, J., Li, X.B., Luo, M., Wu, X.Q., Zhang, M., 2006. Land use and soil erosion in the upper reaches of the Yangtze River: some socio-economic considerations on China's Grain-for-Green Programme. Land Degrad. Dev. 17, 589–603. Marulli, J., Mallarach, J.M., 2005. A GIS methodology for assessing ecological connectivity: application to the Barcelona Metropolitan Area. Landscape Urban Plann. 71, 243–262. McRae, B.H., Beier, P., 2007. Circuit theory predicts gene flow in plant and animal populations. Proc. Natl. Acad. Sci. U.S.A. 104, 19885–19890. McRae, B.H., Kavanagh, D.M., 2011. Linkage Mapper Connectivity Analysis Software, 1.1 ed. The Nature Conservancy, Seattle WA. Mitsova, D., Shuster, W., Wang, X.H., 2011. A cellular automata model of land cover change to integrate urban growth with open space conservation. Landscape Urban Plann. 99, 141–153. Pérez-Hernández, C.G., Vergara, P.M., Saura, S., Hernández, J., 2015. Do corridors promote connectivity for bird-dispersed trees? The case of Persea lingue in Chilean fragmented landscapes. Landscape Ecol. 30, 77–90. Pontius, R.G., Cornell, J.D., Hall, C.A.S., 2001. Modeling the spatial pattern of land-use change with GEOMOD2: application and validation for Costa Rica. Agric. Ecosyst. Environ. 85, 191–203. Rayfield, B., Fortin, M.J., Fall, A., 2010. The sensitivity of least-cost habitat graphs to relative cost surface values. Landscape Ecol. 25, 519–532. Saura, S., Bodin, O., Fortin, M.J., 2014. Stepping stones are crucial for species' longdistance dispersal and range expansion through habitat networks. J. Appl. Ecol. 51, 171–182. Saura, S., Pascual-Hortal, L., 2007. A new habitat availability index to integrate connectivity in landscape conservation planning: comparison with existing indices and application to a case study. Landscape Urban Plann. 83, 91–103. Saura, S., Rubio, L., 2010. A common currency for the different ways in which patches and links can contribute to habitat availability and connectivity in the landscape. Ecography 33, 523–537. Saura, S., Torne, J., 2009. Conefor Sensinode 2.2: a software package for quantifying the importance of habitat patches for landscape connectivity. Environ. Model. Software 24, 135–139. Sawyer, S.C., Epps, C.W., Brashares, J.S., 2011. Placing linkages among fragmented habitats: do least-cost models reflect how animals use landscapes? J. Appl. Ecol. 48, 668–678. Sun, X., Crittenden, J.C., Li, F., Lu, Z.M., Dou, X.L., 2018. Urban expansion simulation and the spatio-temporal changes of ecosystem services, a case study in Atlanta Metropolitan area, USA. Sci. Total Environ. 622, 974–987. Sunde, P., Stener, S.O., Kvam, T., 1998. Tolerance to humans of resting lynxes Lynx lynx in a hunted population. Wildl. Biol. 4, 177–183. Tannier, C., Foltete, J.C., Girardet, X., 2012. Assessing the capacity of different urban forms to preserve the connectivity of ecological habitats. Landscape Urban Plann. 105, 128–139. Tannier, C., Bourgeois, M., Houot, H., Foltete, J.C., 2016. Impact of urban developments on the functional connectivity of forested habitats: a joint contribution of advanced urban models and landscape graphs. Land Use Policy 52, 76–91. Taylor, P.D., Fahrig, L., Henein, K., Merriam, G., 1993. Connectivity is a vital element of landscape structure. Oikos 68, 571–573. Theobald, D.M., Crooks, K.R., Sanjayan, M., 2006. Exploring the functional connectivity of landscapes using landscape networks. In: Crooks, K.R., Sanjayan, M. (Eds.), Connectivity Conservation. Cambridge University Press, Cambridge, pp. 416–444. Thompson, P.L., Rayfield, B., Gonzalez, A., 2017. Loss of habitat and connectivity erodes species diversity, ecosystem functioning, and stability in metacommunity networks. Ecography 40, 98–108. Toger, M., Malkinson, D., Benenson, I., Czamanski, D., 2016. The connectivity of Haifa urban open space network. Environ. Plann. B-Plann. Des. 43, 848–870. Urban, D., Keitt, T., 2001. Landscape connectivity: a graph-theoretic perspective. Ecology 82, 1205–1218.

To conclude, we argued that our coupled approach could be helpful in studying the impacts of land use changes on functional connectivity, and could be supportive for decision makers, as well as conservation practitioners, to understand which land development mode could balance the socio-economic development, and to designate the right places for conservation. Conflict of interest The authors declare no conflicts of interests. Acknowledgements This study is funded by the Humanities and Social Science Fund in Ministry of Education of the People’s Republic of China (Grant number: 15XJA790002). Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.ecolind.2018.10.065. References Aburas, M.M., Ho, Y.M., Ramli, M.F., Ash'aari, Z.H., 2017. Improving the capability of an integrated CA-Markov model to simulate spatio-temporal urban growth trends using an analytical hierarchy process and frequency ratio. Int. J. Appl. Earth Obs. Geoinf. 59, 65–78. Adriaensen, F., Chardon, J.P., De Blust, G., Swinnen, E., Villalba, S., Gulinck, H., Matthysen, E., 2003. The application of ‘least-cost’ modelling as a functional landscape model. Landscape Urban Plann. 64, 233–247. Albert, C.H., Rayfield, B., Dumitru, M., Gonzalez, A., 2017. Applying network theory to prioritize multispecies habitat networks that are robust to climate and land-use change. Conserv. Biol. 31, 1383–1396. Baker, J.P., Hulse, D.W., Gregory, S.V., White, D., Van Sickle, J., Berger, P.A., Dole, D., Schumaker, N.H., 2004. Alternative futures for the Willamette River Basin. Oregon. Ecol. Appl. 14, 313–324. Batty, M., Xie, Y., 1994. From cells to cities. Environ. Plann. B-Plann. Des. 21, S31–S38. Beninde, J., Veith, M., Hochkirch, A., 2015. Biodiversity in cities needs space: a metaanalysis of factors determining intra-urban biodiversity variation. Ecol. Lett. 18, 581. Bierwagen, B.G., 2006. Connectivity in urbanizing landscapes: the importance of habitat configuration, urban area size, and dispersal. Urban Ecosyst. 10, 29–42. Bodin, O., Saura, S., 2010. Ranking individual habitat patches as connectivity providers: integrating network analysis and patch removal experiments. Ecol. Model. 221, 2393–2405. Bowman, J., Jaeger, J.A.G., Fahrig, L., 2002. Dispersal distance of mammals is proportional to home range size. Ecology 83, 2049–2055. Center for International Earth Science Information Network - CIESIN - Columbia University, Information Technology Outreach Services - ITOS - University of Georgia, 2013. Global Roads Open Access Data Set, Version 1 (gROADSv1). NASA Socioeconomic Data and Applications Center (SEDAC), Palisades, NY. Clark, J.S., Silman, M., Kern, R., Macklin, E., HilleRisLambers, J., 1999. Seed dispersal near and far: patterns across temperate and tropical forests. Ecology 80, 1475–1494. Daniel, C.J., Sleeter, B.M., Frid, L., Fortin, M.J., 2018. Integrating continuous stocks and flows into state-and-transition simulation models of landscape change. Methods Ecol. Evol. 9, 1133–1143. Dilts, T.E., Weisberg, P.J., Leitner, P., Matocq, M.D., Inman, R.D., Nussear, K.E., Esque, T.C., 2016. Multiscale connectivity and graph theory highlight critical areas for conservation under climate change. Ecol. Appl. 26, 1223–1237. Engelhard, S.L., Huijbers, C.M., Stewart-Koster, B., Olds, A.D., Schlacher, T.A., Connolly, R.M., 2017. Prioritising seascape connectivity in conservation using network analysis. J. Appl. Ecol. 54, 1130–1141. Fischer, G., Nachtergaele, F., Prieler, S., van Velthuizen, H.T., Verelst, L., Wiberg, D., 2008. Global agro-ecological zones assessment for agriculture (GAEZ 2008). IIASA, L. Austria and FAO, Rome, Italy. Fu, W., Liu, S., Degloria, S.D., Dong, S., Beazley, R., 2010. Characterizing the “fragmentation–barrier” effect of road networks on landscape connectivity: a case study in Xishuangbanna, Southwest China. Landscape Urban Plann. 95, 122–129. Garcia-Feced, C., Saura, S., Elena-Rossello, R., 2011. Improving landscape connectivity in forest districts: a two-stage process for prioritizing agricultural patches for reforestation. For. Ecol. Manage. 261, 154–161. Gurrutxaga, M., Rubio, L., Saura, S., 2011. Key connectors in protected forest area networks and the impact of highways: a transnational case study from the Cantabrian Range to the Western Alps (SW Europe). Landscape Urban Plann. 101, 310–320. Hagen, A., 2003. Fuzzy set approach to assessing similarity of categorical maps. Int. J. Geogr. Inf. Sci. 17, 235–249.

156

Ecological Indicators 98 (2019) 149–157

Y. Huang, T.-J. Liao

modeling approach to the evolution of urban land-use patterns. Environ. Plann. A 25, 1175–1199. Wu, F.L., 2002. Calibration of stochastic cellular automata: the application to rural-urban land conversions. Int. J. Geogr. Inf. Sci. 16, 795–818. Wu, J.G., 2013. Landscape sustainability science: ecosystem services and human wellbeing in changing landscapes. Landscape Ecol. 28, 999–1023. Wu, J.G., Huang, J.H., Han, X.G., Xie, Z.Q., Gao, X.M., 2003. Three-Gorges Dam - experiment in habitat fragmentation? Science 300, 1239–1240.

Visser, H., de Nijs, T., 2006. The map comparison kit. Environ. Model. Software 21, 346–358. Vliet, J.V., Bregt, A.K., Hagen-Zanker, A., 2011. Revisiting Kappa to account for change in the accuracy assessment of land-use change models. Ecol. Model. 222, 1367–1375. Watson, J.R., Siegel, D.A., Kendall, B.E., Mitarai, S., Rassweiller, A., Gaines, S.D., 2011. Identifying critical regions in small-world marine metapopulations. Proc. Natl. Acad. Sci. U.S.A. 108, E907–E913. White, R., Engelen, G., 1993. Cellular-automata and fractal urban form - a cellular

157