Habitat Loss, Introduction any species face extinction given the current rate Mof habitat loss, fragmentation, and degradation Fragmentation, (Hastings 1980; Ehrlich & Ehrlich 1981; Wilcox & Mur- phy 1985; Wilcove et al. 1986; Hassell et al. 1993; Fahrig 1997; Tilman et al. 1994). Although the implications of and Restoration spatial pattern and arrangement have been studied in terms of species persistence (Dytham 1995a), they have Gary R. Huxel1,2 not been adequately addressed for species recovery 1 plans. Here, using simple spatial models, we argue that Alan Hastings by including spatial processes in restoration manage- ment plans, the effects of habitat loss and fragmentation can be offset (Lewis et al. 1996). The decision as to Abstract which habitat should be restored may be as important The loss and fragmentation of habitat is a major threat as to how much is to be restored. Yet few management to the continued survival of many species. We argue plans take into account spatial effects of habitat conser- that, by including spatial processes in restoration vation/restoration, despite the importance of spatial management plans, the effects of habitat loss and dynamics in species conservation and recovery plans. fragmentation can be offset. Yet few management Tilman et al. (1997) investigated species restoration in plans take into account spatial effects of habitat con- cellular automata, in which habitat in a landscape was servation/restoration despite the importance of spatial randomly selected for restoration. They found a sub- dynamics in species conservation and recovery plans. stantial delay in recovery of the species. Tilman et al. (1997) found a “restoration lag” in simu- Species recovery studies/plans have more frequently lations of species restoration when randomly select- been incorporating realistic spatial dynamics, whereas ing habitat for restoration. Other studies have sug- those that do not do so have come under increasing fire gested that the placement of restored habitat can (Pulliam et al. 1992; Stacey & Taper 1992; Harrison 1993; overcome effects of habitat loss and fragmentation. Harrison et al. 1993; Doak et al. 1994; Dunning et al. Here we report the findings of simulations that exam- 1995; Noon & McKelvey 1996). Additionally, there have ine different regional management strategies, focus- been numerous calls for the inclusion of realistic spatial ing on habitat selection. We find that nonrandom res- dynamics in the discussion of general conservation is- toration practices such as restoring only habitat that is sues (Hanski & Gilpin 1991; May 1993; Gilpin & Hanski adjacent to those occupied by the target species can 1997; Levin et al. 1997; Tilman & Kareiva 1997; Groom dramatically reduce or negate any restoration lag. In & Pascual 1998). fact, we find that the increase in patch occupancy of In response to this need, Tilman et al. (1997) used a the landscape can be greater than two-fold in the ad- spatially explicit model to examine habitat destruction jacent versus the random scenarios after only two res- and restoration. Their model exhibited a restoration lag toration events, and this increase can be as great as that implies a significant number of patches must be re- six-fold during the early restoration phase. Many res- stored before a positive effect on population levels is toration efforts have limitations on both funds and seen. This finding is based upon several key factors, the available sites for restoration, necessitating high po- most important being that the patches in a landscape tential success on any restoration efforts. The incorpo- that are to be restored are randomly selected, and that ration of spatial analyses in restoration management the target species is not reintroduced into the patch. may drastically improve a species’ chance of recovery. However, by including spatial processes in restoration Therefore, general principles that incorporate spatial management plans, the effects of habitat loss and frag- processes and sensible management are needed to mentation may be offset (Kareiva & Wennergren 1995; guide specific restoration efforts. Lewis et al. 1996). This is critical because many species face extinction owing to demographic and/or environ- mental stochasticity given the current rate of habitat 1 Department of Environmental Science and Policy and Insti- loss, fragmentation, and degradation (Hastings 1980; tute of Theoretical Dynamics, University of California, Davis, Ehrlich & Ehrlich 1981; Groombridge 1992; Hassell et al. CA 95616, U.S.A. 1993; Tilman et al. 1994; Fahrig 1997; Huxel & Hastings 2Address correspondence to G. R. Huxel, email [email protected] 1998). The decision as to which habitat should be re- stored may be as important as how much habitat is to © 1999 Society for Ecological Restoration be restored. Fahrig (1997) argues that the effect of frag- SEPTEMBER 1999 Restoration Ecology Vol. 7 No. 3, pp. 309–315 309 Spatial Concerns in Restoration mentation is only seen once a given minimum has been For the second result, the length of the extinction lag reached. Using cellular automata, we test whether in- depends upon the ratio of colonization to extinction corporating spatial processes in regional management and the random instead of directed dispersal of individ- plans can improve the efficacy of restoration efforts. uals (Loehle & Li 1996). Colonization rules that allow for increased dispersal distances may totally negate the lag by allowing individuals to cross lethal areas (see be- Methods low). This model is, therefore, more appropriate for The underlying deterministic model on which we base species that have dispersal distances/rates that are our stochastic, spatially explicit cellular automata is the small compared to the distance between patches. Levins metapopulation model (Levins 1969). Patches Thus, the major results of spatially explicit models may support either populations as in the original model of habitat destruction, fragmentation and restoration, (Levins 1969), or individuals (Tilman et al. 1994). Patches greatly rely on model structure and assumptions. Here that are not “destroyed” are assumed identical, as de- we are only interested in the restoration aspect of these scribed by the simple model in the following equation: models. Thus, we follow the Tilman et al. model (1994, 1997) in that we use a cellular automaton in which dp ------ () patches are linked by dispersal. Tilman et al. (1997) ran- dt 5 mp 1 2 p 2 h 2 ep, (1) domly selected patches to be destroyed until a set pro- where p is the fraction of occupied patches or habitat, h portion of the habitat was left and then patches were re- is the fraction of uninhabitable (destroyed) patches or stored. We followed the same pattern of destroying habitat, m is the migration or colonization rate, and e is habitat at the rate of 1% per time step (one hundred the extinction rate. We assume m (0.20 in all simula- patches) until 25% of the habitat was still suitable, then tions) and e (0.025 in all simulations) are constant and restoring 1% of the habitat per time step until all colonization can occur only in empty but habitable patches were again suitable habitat (Fig. 1). Our spa- sites, as in Tilman et al. (1994). The values were chosen tially explicit model simulated a landscape that was di- for demonstration purposes to ensure that the species vided into a square grid of 10,000 patches (100 3 100). persisted at the minimum number of patches (i.e., when Dispersal was limited to the four-cardinal-direction- the number of destroyed patches was at its maximum). adjacent patches and random so that individuals that These are simplifications in that both colonization and dispersed to uninhabitable sites or off the grid were lost extinction may be dependent on local density (either (i.e., absorbing boundary). Because of an expected ex- within a patch or within a set of adjacent patches) and tinction lag (Tilman et al. 1994), we allowed for the sys- on patch quality. Therefore, within-patch dynamics and tem to approach an equilibrium over a hundred time the rescue effect are ignored in this model. In the non- steps between habitat loss or restoration events. We spatially explicit Levins model, there is no temporal lag performed ten replicates of each scenario. between when habitat is lost and when extinction oc- Because we are interested in the efficacy of various curs due to dispersal being global as opposed to local. management strategies, we restored patches by using In the spatially explicit cellular automata, colonization four strategies: (1) patches to be restored are randomly is local and limited to the four cardinal direction selected (this is our null management strategy); (2) we patches adjacent to an occupied patch (von Neumann randomly select patches to be restored, but we reintro- rule; Durrett & Levin 1994). duce the species into the restored patch; (3) only In spatially explicit model studies of habitat fragmen- patches that are adjacent to occupied patches are re- tation, two important results have been demonstrated stored; and (4) patches are restored in the reverse order and hypotheses concerning these results have been in which they were destroyed (the last patches de- made for natural systems. The two results are: (1) less stroyed are the first to be restored). mobile species may become extinct first (Tilman et al. The four restoration strategies were chosen to repre- 1994; Huxel & Hastings 1998); and (2) there is an extinc- sent the variety of choices that are available to restora- tion lag or hysteresis that occurs when the amount of tion managers and biologists given the various situa- habitable area of a landscape drops below the equilib- tions in which restoration projects occur. Managers of rium threshold level of suitable habitat for a species to individual projects usually do not have all of these persist (Tilman et al. 1994). The first result depends choices; however, we are more concerned with regional upon the manner and pattern in which habitat is lost strategies.