Conservation planning in a fire-prone Mediterranean region: threats and opportunities for bird species

In response to the processes threatening biodiversity such as habitat loss, effective selection of priority conservation areas is required. However, reserve selection methods usually ignore the drivers of future habitat changes, thus compromising the effectiveness of conservation. In this work, we formulated an approach to explicitly quantify the impact of fire on conservation areas, considering such disturbance as a driver of land-cover changes. The estimated fire impact was integrated as a constraint in the reserve selection process to tackle the likely threats or opportunities that fire disturbance might cause to the targeted species depending on their habitat requirements. In this way, we selected conservation areas in a fire-prone Mediterranean region for two bird assemblages: forest and open-habitat species. Differences in conservation areas selected before and after integrating the impact of fire in the reserve selection process were assessed. Integration of fire impact for forest species moved preferences towards areas that were less prone to burn. However, a larger area was required to achieve the same conservation goals. Conversely, integration of fire impacts for open-habitat species shifted preferences towards conservation areas in locations where the persistence of their required habitat is more likely (i.e. shrublands). In other words, we prioritized the conservation of not only the current distribution of open-habitat birds, but also the disturbance process (i.e. fire) that favours their preferred habitat and distributions in the long term. Finally, this work emphasizes the need to consider the opposing potential impacts of wildfires on species for an effective conservation planning.


Introduction 25
Over the past few decades, habitat loss and degradation arising from land-cover changes have been identified as a prevailing threat to species persistence (Tucker et al 1994;. In the light of the high rate at which habitats and landscapes are being transformed, areas identified as most valuable for the species persistence may be prioritized and managed for conservation actions to ensure the long-term persistence of their required 30 habitats (Margules et al 2002).
The spatial selection of priority areas for conservation is usually performed using shown important changes as consequence of fire impact and land abandonment (Vallecillo et al 2009).
In this context, we selected bird species, whose distributions are strongly affected by 100 changes in forest and shrubland covers, as conservation goals for our study. We found a total of 39 species with higher preference for either forest (24 species) or shrub-like habitats and farmlands (15 species) than for other land-cover classes (Appendix 1 Information on the distribution for each species was sourced from distribution maps developed in the CBBA by means of Generalized Linear Models (GLM). GLM were built using empirical presence/absence data for bird species recorded at 1 km 2 . Predictor variables included in the models were climate, topography, extent of the land-use classes from the land-120 use map of Catalonia 1998 (Viñas and Baulies 1995) and other miscellaneous variables (see Estrada et al 2004 andBrotons et al 2007 for further details). GLM were evaluated by means calibrated the BLM testing six different values (0, 0.005, 0.01, 0.05, 0.1, and 1) to find the optimal BLM that provides a reasonable perimeter/area ratio (Possingham et al 2000).
For each bird assemblage, we ran Marxan 100 times to produce 100 near-optimal 150 solutions to achieve the conservation goals while minimizing the objective function. Multiple runs allow estimating the frequency of selection for each planning unit, which is a measure of the likelihood that an area will be required to meet a given set of targets. Selection frequency in the 100 near-optimal solutions was here used as a surrogate for the conservation value of an area (Cowling et al 1999). According to the selection frequency, we defined as priority 155 conservation areas those selected in more than 50% of the near-optimal solutions (Ardron et al 2008) and as core conservation areas those selected in more than 75% of solutions.

Conservation scenarios
We used two different scenarios to prioritize conservation areas: (a) Reference scenario in which the achievement of conservation targets aimed to minimize the area selected Fire impact = fire risk x vulnerability (forest) + fire risk x vulnerability (shrubland) (Equation 1) Where fire risk at each specific location was obtained from the static map of fire risk, provided by the Catalan Government (available at www.gencat.cat). This map shows ten categories with increasing fire risk (i.e. from 0 to 9), both in frequency and intensity; and was 175 elaborated using data from past fire regime, flammability and fuel models derived from the Ecological and Forest Inventory of Catalonia, topography and climate data (Fig. 1).
#Figure 1 approximately here# Vulnerability was estimated by calculating two different Pearson's correlation coefficients for each bird assemblage, one for forests and one for the shrubland cover, 180 between the conservation value in the reference scenario (i.e. selection frequency of each planning unit) and the proportion of each land-cover within the 1 km 2 planning units.
Information on forest and shrubland covers was obtained from the land-use map of Catalonia from 1998 (Viñas and Baulies 1995). Hence, coefficients of vulnerability are a measure of the degree to which the conservation value at a given location is related to forest and shrubland 185 covers (similarly to Chan et al 2006). Large positive values of the correlation coefficient indicate that areas with high conservation value are favoured by a large extent of the landcover class, either forest or shrubland. If changes occur in the extent of the respective landcover class, conservation value of a given area will be affected suggesting higher vulnerability to future land-cover changes. The coefficients of vulnerability of each bird assemblage were 190 used to weight the fire risk map and spatially explicit estimate the extent to which their conservation areas could be affected by fire in the future.
In this context, the reserve selection algorithm in the fire-impact scenario will achieve conservation targets at the minimum area, while accounting for the potential impact of wildfire on conservation areas. This means that areas with a positive fire impact will be 195 positively selected, as expected for open-habitat birds, whereas those areas with an unfavourable effect of fire will be avoided in the selection of conservation areas for forest birds.
Including fire impact in the reserve selection algorithm as a penalty might result in a reduction in the contiguity of high quality areas and more fragmented reserves (

Compactness = 1 / [boundary length / (2√(pi x Area)] (Equation 2)
Reserves are more compact as compactness index approaches to 1, becoming close to 0 for highly fragmented reserves. 210 The compactness of conservation areas is closely related to the role of the BLM, which was previously determined by a calibration process (see the Reserve selection algorithm section). In the reference scenario, we found 0.005 to be the optimal BLM (i.e. reasonable area/perimeter ratio) for both bird assemblages. In the fire-impact scenario we obtained a BLM of 0.05 and 0.01, for forest and open-habitat species respectively, yielding 215 conservation areas with comparable compactness to the conservation areas in the reference scenario (Table 1).
#Table 1 approximately here# Then, we evaluated the differences between the conservation areas selected for both conservation scenarios, before and after integrating fire impact, by analysing: (1) Total 220 conservation area required for the targeted species, estimated as the average of the number of 1 km 2 planning units selected in the 100 independent Marxan runs.
(2) Efficiency estimated as the inverse of the conservation area required per species (i.e. averaged across 100 solutions).
The smaller the area per species needed to meet the conservation goals, the more efficient the reserve is (Pressey and Nicholls 1989) (Equation 3). 225 Efficiency = [1 / (area/species)] x 100 (Equation 3) (3) Core area was considered as the number of planning units selected in more than 75% of the near-optimal solutions. (4) Mean fire risk in conservation areas was estimated within all 100 near-optimal solutions using the fire risk map described above.
Finally, we also evaluated to what extent burned areas were selected as priority 230 conservation areas for open-habitat birds. With this purpose, we calculated the overlap of the priority areas (i.e. those selected in more than 50% of the near-optimal solutions) for openhabitat birds with the areas burned between 1980 and 1999. Information on fire occurrence was provided by the Centre for Ecological Research and Forestry Applications and the Catalan Government (Fig. 2). Note that the forest assemblage was not considered here since 235 there was small overlap between burned areas and their priority conservation areas.

Conservation requirements 240
Comparison between conservation areas selected in both the reference and the fireimpact scenarios for two bird assemblages showed that open-habitat species needed larger area to achieve the conservation goals than forest species (Table 2).
#Table 2 approximately here# Furthermore, conservation areas for open-habitat species showed lower efficiency than 245 for forest species. That is, the open-habitat bird assemblage, in spite of having a smaller number of species than the forest assemblage, required more than twice the extent of the conservation area per species than the forest ones (Table 2). Moreover, larger core areas (i.e. those selected in more than 75% of near-optimal solutions) for open-habitat species than for forest birds showed smaller flexibility to achieve conservation goals for this bird assemblage 250 compared to forest species (Table 2; Fig. 3).

Vulnerability of conservation areas
Coefficients of vulnerability to changes in the forest cover were larger than for the shrubland ones, for both forest and open-habitat bird assemblages (Table 3). This indicates 255 that changes in the extent of forest cover have larger influence on the conservation value of a given area than changes in the shrubland extent. As expected, forest species showed to be favoured by increases in forest cover, whereas open-habitat species resulted disadvantaged, as shown by the negative sign of the correlation coefficient (Table 3).

#Table 3 approximately here# 260
Conversely, the low coefficients of vulnerability to changes in the shrubland extent showed smaller dependence between the conservation value of a given area and the shrubland extent compared to forest cover. Furthermore the positive sign for both bird assemblages revealed that conservation value of a given area can be favoured by certain extent of shrubland cover, for both forest and open-habitat birds. But open-habitat species resulted 265 favoured at larger extent by increases in the shrubland cover than forest species (Table 3).

Fire impact in conservation planning
Integrating fire impact led to larger and less efficient conservation areas for both bird assemblages. Forest species resulted more negatively affected by considering the impact of fire than open-habitat birds as conservation areas for forest species underwent a larger 270 increase in area and decrease in efficiency than open-habitat birds (Table 2). However, integration of fire impact for the forest assemblage yielded a reduction of its core area (Table   2; Fig. 3), leading to higher flexibility for the achievement of conservation targets.
Importantly, we demonstrated how to integrate the estimated fire impact as a constraint in the reserve selection algorithm at the required level to produce significant 275 changes of fire risk in conservation areas (Table 2). Conservation areas for forest species were selected in zones with lower fire risk than in the reference scenario, whereas areas prone to be affected by fire were chosen for open-habitat birds.
Finally, from about 4,800 planning units affected by fire between 1980 and 1999, 21% were selected as priority conservation areas for open-habitat birds in the reference scenario, 280 increasing to 49% after integrating fire impact in the conservation planning (see Fig. 2 and 3 for a graphical comparison). In this last scenario, burned areas constituted the 51% of the priority areas (i.e. selected in more than 50% of the near-optimal solutions).

Discussion 285
We have described a reasonable straightforward method for the integration of potential impacts of future land-cover changes into the selection of priority conservation areas with clear benefits for conservation and land use planners. This method is one of the first attempts to explicitly quantify the impact of future land-cover changes and implementing this in a static conservation planning approach. Methods applied until now have assumed that species implementation, and are difficult to apply for large number of species.

Comparison of conservation areas for forest and open-habitat birds
Priority conservation areas (i.e. those selected in more than 50% of the solutions) for forest birds were located in the northern mountainous regions of Catalonia, at about 1,780 m a.s.l. Conversely, priority conservation areas for open-habitat species were situated at roughly 300 450 m a.s.l., mostly in the centre of Catalonia (Figure 3).

Definition of conservation areas in Catalonia for bird species with contrasting habitat
requirements suggests that open-habitat birds require special conservation attention. First, this assemblage hosts a larger proportion of threatened species than the forest ones in the study area (Appendix 1). Furthermore conservation targets for open-habitat birds may be more 305 difficult to achieve since their conservation areas showed smaller efficiency than forest birds and hence, larger conservation area per species will be required ( Table 2) The vulnerability to land-cover changes estimated for both bird assemblages 315 confirmed the variable impact of fire depending on the targeted bird assemblage. This is, the large vulnerability of the conservation areas to changes in the forest cover confirmed the key Tables   Table 1. Description of conservation areas for two bird assemblages selected after integrating fire impact in the reserve selection process (Fire-impact scenario) with increasing levels of spatial aggregation of reserves according to the Boundary Length Modifier (BLM). Values of area, perimeter and compactness of reserves were averaged for 100 near-optimal solutions. Compactness was compared with that obtained before integrating fire impact (Reference scenario) at the optimal BLM (0.005) to choose the BLM at which compactness between both scenarios was analogous  Table 2. Comparison between priority conservation areas for two bird assemblages selected under two conservation scenarios: (1) Reference scenario in which the achievement of conservation goals aims to minimize the area selected for conservation under the current conditions of species distributions and (2) Fire-impact scenario in which the reserve selection process includes the potential impact of fire, separately estimated for the two bird assemblages, as a penalty for the selection of planning units  Accipiter gentilis Northern Goshawk least concern 0.71

Dendrocopos major
Great Spotted Woodpecker least concern 0.84

Dendrocopos minor
Lesser Spotted Woodpecker least concern 0.91

Parus ater
Coal Tit least concern 0.91

Parus caeruleus
Blue Tit least concern 0.86

Parus cristatus
Crested Tit least concern 0.84

Parus palustris
Marsh Tit least concern 0.94

Sitta europaea
Wood Nuthatch least concern 0.90

Open-habitat bird assemblage
Alauda arvensis Eurasian Skylark least concern 0.88

Sylvia undata
Dartford Warbler least concern (near threatened) 0.86 *BirdLife International (2012)  The final aim of Marxan is to adequately represent a set of species at the 595 required level by selecting as few planning units as possible. However to find an optimal reserve Marxan tries to minimize an objective function where penalties, spatial design and cost trade-offs are considered. After creating a random initial reserve system, planning units are added or discarded from the reserve system in an attempt to minimise the objective function: 600 Objective function = ∑Cost + ∑ SPFxSpecies penalty + BLM∑Perimeter + Fire Penalty In its simplest form, the Marxan objective function is a combination of the total cost (∑Cost) of the reserve system and a penalty for any of the ecological targets that are not met (∑ SPFxSpecies penalty).
The total cost (∑Cost) of the reserve in our case was the number of 1 km 2 605 planning units (i.e. total area selected). Because the final solution proposed by Marxan might fail to meet adequate conservation objectives (i.e. species distributions) at the required level, there is the species penalty factor (∑ SPFxSpecies penalty). In this work the SPF for not fully representing all the species at the targeted level was set up to 10.
However, Marxan also allows taking into account the fragmentation of the 610 reserve system, so that it will generally be desirable for a reserve system not to be too fragmented. More fragmented reserves will have a greater overall boundary length.
Thus this boundary length, plus a weighting on its importance were included in the objective function (BLM ∑ Perimeter). The BLM was calibrated in order to find the optimal solution where the area/perimeter ratio was minimized.