Ecological Indicators 10 (2010) 192–205

Predicting the trends of vertebrate species richness as a response to wind farmsinstallation in mountain ecosystems of northwest Portugal

Mario Santos a,*, Rita Bastos a, Paulo Travassos a, Regina Bessa a, Miguel Repas b, Joao Alexandre Cabral a

a Laboratory of Applied Ecology, CITAB - Centre for the Research and Technology of Agro-Environment and Biological Sciences, University of Tras-os-Montes e Alto Douro, 5000-911 Vila

Real, Portugalb Strix, Environment and Innovation, Ltd. Rua Pedro Homem de Melo, 55 - room 3.09, 4150-599 Porto, Portugal

A R T I C L E I N F O

Article history:

Received 3 June 2008

Received in revised form 29 April 2009

Accepted 29 April 2009

Keywords:

Stochastic dynamic methodology

Wind farms

Ecological integrity

Vertebrates

Ecological indicators

Species richness

Ecological Impact Assessments

A B S T R A C T

The main objectives of this work were to examine the performance of a holistic stochastic dynamic

methodology (StDM) in predicting the trends of the vertebrate species richness (amphibians, reptiles,

birds and mammals) in response to changes induced by the ongoing wind farm installation in mountain

areas of northwest Portugal. The StDM is a sequential modelling process developed in order to estimate

the ecological status of changed ecosystems that have been damaged by anthropogenic disturbances.

The performance of two complementary temporal approaches was tested, taking into account either

annual or seasonal influences. The data used in the dynamic model construction included true gradients

of environmental changes and was sampled from 2004 to 2006. The dynamic model developed was

preceded by a conventional multivariate statistical procedure performed to discriminate the significant

relationships between the selected ecological components, such as the species richness of each

vertebrate group and the structural changes in habitat conditions. The results show the capacity of the

model in capturing the dynamics of the studied system by predicting consistent trends for the global

vertebrate species richness under complex and variable environmental scenarios. The average annual

approach is considered sufficient for the aims of the most Environmental Impact Assessments while the

seasonal approach is recommended for more detailed studies, namely regarding specific population,

guilds or community dynamics.

� 2009 Elsevier Ltd. All rights reserved.

Contents lists available at ScienceDirect

Ecological Indicators

journa l homepage: www.e lsev ier .com/ locate /ecol ind

1. Introduction

In the last decades, renewable energies, and specially windenergy, have received a huge investment. Wind energy isconsidered as one of the strategies to deal with global warmingand accomplishing the Kyoto Protocol. Between 1997 and 2006, itscapacity grew at 30% per year and in 2006 the world total installedcapacity reached 74 GW (BP, 2007). Recently, the Portuguesegovernment adopted a national planning goal of a yearly windpower generation of 3.8 MW until 2010, implying a substantialincrease from the actual 1.7 MW level installed (Portal do Governo,2007). These policy objectives are well in line with the EuropeanUnion’s current goal to increase the share of renewable electricpower to 22% in 2010 compared to 14% in 1997 (Directive 2001/77/EC). Although wind energy is usually considered as able to generate

* Corresponding author at: Laboratory of Applied Ecology, CETAV - Department of

Biological and Environmental Engineering, University of Tras-os-Montes e Alto

Douro, 5000-911 Vila Real, Portugal. Tel.: +351 259 350 238/239;

fax: +351 259 350 480.

E-mail address: mgsantoss@gmail.com (M. Santos).

1470-160X/$ – see front matter � 2009 Elsevier Ltd. All rights reserved.

doi:10.1016/j.ecolind.2009.04.014

electricity without many of the environmental impacts (toxic airpollution and greenhouse gases, water use and pollution, andhabitat destruction) associated with other energy sources, manydrawbacks have been perceived (Fielding et al., 2006; Gamboa andMunda, 2007). In fact, the entire balance of cost-benefits, includingthe direct and indirect local impacts of wind farms on wildlife andnature conservation, has not been made (e.g. Rabin et al., 2006;Everaert and Stienen, 2007).

On the other hand, ecosystems and landscapes of theMediterranean region (Portugal is, for the most part, included inthis region) have been shaped by a several thousand-year historyof human land use and disturbance, namely by wildfires (Bond andKeeley, 2005). Fire is considered a fundamental process, determin-ing the vegetation dynamics and habitat structure of this region(Wittenberg et al., 2007). Since vegetation cover and diversityplays one of the key factors affecting the animal community inMediterranean ecosystems (Moreira et al., 2001; Cuesta et al.,2006), it’s important to understand the evolution of the plantsuccession after wildfires (Calvo et al., 2008; Capitanio andCarcaillet, 2008). In this region, different communities havedemonstrated their resistance and resilience to fire and commu-nities’ restoration has been traditionally described as an auto-

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 193

successional process where, after a fire, generally it recoverswithout any great changes in the species composition (Ukmaret al., 2007). The major pending question is if these communitiescan cope with new additional disturbances (ex.: wind farmsinstallation, climatic changes) maintaining their recognizedresistance and resilience.

The current principles used to guide conservation and manage-ment believe that protection of species richness is an efficient wayto conserve overall biodiversity and sustain key ecologicalprocesses (Pimm and Brow, 2004). Although the observeddiscrepancies in species richness between locations could revealeither true differences in the community composition or differ-ences in the sampling effort adopted (Fleishman et al., 2006), if thesampling effort is standardized then the errors associated may bereduced and species richness might be use as an ecologicalindicator (Pearmam and Weber, 2007). The use of ecologicalindicators presents an important source of information to policymakers and help to guide decision-making (Niemeijer and Groot,2008). For conservation and management purposes, the use of thecorrect ecological indicators may reveal what effect changes ofenvironmental factors have consequences on the integrity ofecosystems (Dale and Beyeler, 2001; Niemeijer and Groot, 2008).By using these indicators it’s possible to predict how anthro-pogenic and natural environmental changes affect the species andthe communities of disturbed ecosystems (Andreasen et al., 2001;Dale et al., 2008).

In many occasions these predictions are simulated by usingmodels, where the indicators trajectory can be followed a priori(Santos and Cabral, 2004; Santos et al., 2007). Since many of theecosystem phenomenological aspects are holistic, whole-systemproperties, the main vocation of the stochastic dynamic metho-dology (StDM) recently developed is a mechanistic understandingof the holistic ecological processes, based on a statistical parameterestimation method. The StDM is a sequential modelling processdeveloped in order to predict the ecological status of ecosystems,from which management strategies can be designed. Our ownrecent research is based on the premise that the general statisticalpatterns of ecological phenomena are emergent indicia of complexecological processes that do indeed reflect the operation ofuniversal law-like mechanisms. In this scope, the StDM was

Fig. 1. (a) Location of the studied area in a mountain of North-western Portugal, ‘‘Serr

(quadrates). The triangles indicate the position of each wind mill.

successfully applied in several types of scenarios, such as agro-ecosystems (Santos and Cabral, 2004; Cabral et al., 2007),mountain running waters (Cabecinha et al., 2004, 2007), estuaries(Silva-Santos et al., 2006, 2008), wildlife conservation (Santos et al.,2007) and bird survey testing (Santos et al., 2009).

In the present study we investigated the potential effects of theinstallation of wind farms in natural ecosystems of westMediterranean mountain areas (Serra do Marao, Portugal), witha focus on the vertebrate species richness and on their habitatstructure and diversity. Actually, plant richness and vegetationstructure have been used as predictors of associated biodiversityand ecological indicators occurrence (e.g. Yost, 2008). Terrestrialvertebrates (amphibians, reptiles, mammals and birds) includeattractive species and species requiring a large amount of habitat,thus they have frequently been used as indicators of disturbance(e.g. Tognelli, 2005; O’Connell et al., 2007; Araujo et al., 2008). Indetail we examined: (1) the performance of a developed holisticStDM predictive model that focus on conceptually isolated keycomponents, explicitly the vertebrate species richness and themain ongoing habitat changes, and (2) the trends of theseindicators in face of different scenarios in the scope of the windfarms installation problematic. The hypotheses to be tested bygeneral applications of a StDM model include: (1) that theindicators selected are representative of the local communities,varying predictably due to changes in the environmental condi-tions of the studied mountains, and (2) that the indicatorsbehaviour can be represented by the state variables used in thedynamic model construction.

2. Methodology

2.1. Study area

The study was carried out in a mountain of North-westernPortugal, ‘‘Serra do Marao’’ (Fig. 1a), with an average altitude of1000 m above the sea level (SW limit, X = 59,000, Y = 4,560,000; NElimit, X = 595,000, Y = 4,567,000). The bio-climatologists classifythe study area as ‘‘supramediterranean superior’’ with precipita-tion values of 1400–2000 mm/year and an average annualtemperature of 10 8C (Molina et al., 1992). Heathlands and

a do Marao’’ (shaded area), and (b) the spatial distribution of the sampling plots

M. Santos et al. / Ecological Indicators 10 (2010) 192–205194

pastures are the dominant habitats, which result from traditionalgoat herdsman practices, such as induced fires and animal grazing.The main heathland community present is classified as anassociation of the Ericion umbellatae with Erica australis subsp.aragonensis as the dominant species (Willk) (Rivas Martınez et al.,1987). Pinus pinaster, Pinus nigra, Betula celtiberica and Quercus

pyrenaica are the major tree species, although distributed in smallpatches. In contrast, shrubs prevail, namely E. australis, Erica

umbellata, Pterospartum tridentatum, Halimiun alyssosides, Hali-

mium ocymoides and Ulex minor, while the widespread herbaceousare Agrostis sp., Pseudarrhenaterum sp. and Pteridium aquilinum

(Strix, 2007). The vertebrate community is characterized bygeneralist mountain species, even though some rare species occursuch as the Golden eagle (Aquila chrysaetus), the Iberian Wolf (Canis

lupus signatus), the Pyrenean Desman (Galemys pyrenaicus) or theGolden striped Salamander (Chioglossa lusitanica) (Strix, 2007). Thecomplete list of species identified during field work for the studyarea is presented in Appendix 1 (available on line as electronicSupplementary material). The study was carried out in four windfarms (Penedo Ruivo, Mafomodes, Seixinhos and Teixeiro) thatwere installed in this mountain throughout the years 2004–2006(Fig. 1b). The turbines are located on the mountain ridge whichranges from 800 to 1300 m above sea level. The total number ofturbines is 27 (power of 4 MW), distributed by two models(tubular steel tower of 60 and 80 m and rotors of 62 and 91 m,respectively).

2.2. General methodology

One hundred and ninety eight quadrates (plots of250 m � 250 m) were randomly surveyed (Random NumberGenerator Pro 1.241) from May 2004 to March 2007 with aseasonal periodicity. The quadrates considered were chosen withthe intention of collecting data from disturbed and non-disturbedlocations, varying the distance from the nearest turbine (orassociated structure) between 0 and 2000 m (Fig. 1b). Thesequadrates include some structural heterogeneity, in terms of thedominant local habitat characteristics, an important requirementfor the unbiased comprehension of the indicators response. Eachquadrate descriptions and surveys were made in its centre, sited byusing a Global Position System device (Magellan GPS 3201).Transects between the quadrates centres were made for com-plementary data.

2.3. Vegetation sampling

The list of species by quadrate was compiled in order to coversatisfactorily the flora present in different sites at different timesof the year. At each quadrate, the slope, aspect and topographywere recorded. The relative species density was then calculatedfrom the abundance data for each quadrate. For each species, theabundance-dominance was registered in 2 m � 2 m plots using a 9level scale (Van der Maarel, 1979): (1) rare, (2) a few, (3) many and<5% of cover, (4) abundant and<5% of cover, (5) 5–12.5%, (6) 12.5–25%, (7) 25–50%, (8) 50–75% and (9) >75% of cover. This scaleshould prevent an over-emphasization of very abundant species(Noest et al., 1989). The relative species abundance was thencalculated from the abundance-dominance data for each quadrate(Elzinga et al., 2001).

2.4. Vertebrates sampling

A survey of all species of herpetofauna in an area requiresmore than one technique (e.g. Corn and Bury, 1990; Ryan et al.,2002). Time-constrained searches were executed for amphibiansand reptiles, which were immediately collected by hand (and

released after identification). Equal effort was expended in eachquadrate by 10 min time limited searching. Two pitfall trapswere placed by quadrate, spaced by 10 m (Corn and Bury, 1990).Trap locations were marked using a GPS device and wereremoved after 12 h. Systematic complementary searches weremade in predetermined wet locations of the study area (Ryanet al., 2002).

For bird surveys, considered the most conspicuous of allvertebrates, one 10 min point count (unlimited radio point count)(Ralph et al., 1995; Bibby et al., 2000) was completed by quadrate.Although information related with bird’s phenologic state wascollected, the most important data was the community composi-tion by quadrate (number of species and number of individuals byspecies). Complementary transects were executed for additionalinformation. The same ornithologist achieved the constantperformance of the records.

Mammals have been sampled using techniques based eitheron direct observations of the animal itself or on indirectobservations of the animal’s activities, such as scats, tracks andbreeding dens (Gitzen et al., 2007). Time-constrained searchesfor indirect evidences of the species presence were carried outduring a 10 min period. Additionally a Sherman trap and twopitfall traps were located by quadrate, spaced by 10 m(Flowerdew et al., 2004). Traps (marked with a GPS) wereoperating during one night and baited with a mixture of meat,peanut butter, flour, seeds, fruit, and were set under the cover ofherbs or rocks when possible to provide camouflage and thermalinsulation (Torre et al., 2007). Traps were checked within 12 h,and animals captured were identified to species and released atthe point of capture. Camera traps, with a built in flash and abuilt in infrared motion sensor, were used to identifying speciesof medium sized mammals and to monitor relative abundanceand studying activity patterns (Yasuda, 2004). Given that thespecificities associated to bat monitoring are not easily fitted inour experimental design the results obtained for this group werenot considered in this work.

2.5. Data analysis

2.5.1. Determining the indicators responses

In the StDM, the dynamic model construction was precededby a conventional multivariate statistical procedure for para-meters estimation. A stepwise multiple regression analysis (Zar,1996) was used to test for relationships between ecologicalindicators (vertebrates) and environmental variables (vegeta-tion, topography and wind farm structures). A step downprocedure was used so that the effect of each variable in thepresence of all other related variables could be examined firstwith the least significant variable being removed at every step.The analysis stopped when all the remaining variables had asignificant level P < 0.05 (Zar, 1996). Although the lack ofnormality distribution of the dependent variables was not solvedby any transformation (Kolmogorov–Smirnov test), the linearityand the homoscedasticity of the residuals were achieved by usinglogarithmic transformations (X0 = log[X + 1]) in each side of theequation, i.e. on both the dependent and independent variables(Zar, 1996). The lack of substantial intercorrelation amongindependent variables was confirmed by the inspection of therespective tolerance values. In order to understand and comparethe indicators responses to the environmental variables, twotemporal scales were considered by the regression, the annualscale (using all the available information together: ‘‘Base model—B equations’’) and the seasonal scale (using the informationseparately by season of the year: ‘‘Alternative model—Aequations’’). All the statistical analysis was carried out usingthe statistical software SYSTAT 8.01.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 195

2.6. Conceptualisation of the model

Since the previous statistical procedure was supported on adatabase which included true gradients of disturbance (namely byfire and wind farm installations), over space and time, thesignificant partial regression coefficients were assumed as relevantholistic ecological parameters in the dynamic model construction.This is the heart of the philosophy of the StDM. In a holisticperspective, the partial regression coefficients represent the globalinfluence of the environmental variables selected, which are ofsignificant importance on the indicators, namely on severalecological complex processes associated with species richness(Santos and Cabral, 2004). These processes were not includedexplicitly in the model, but were implicitly related to the statevariables (or indicators) under consideration. For the developmentof the dynamic model the software STELLA 9.0.3.1 was used.

2.7. StDM performance and simulations

For StDM performance assessment, independent environmen-tal data from two areas of Marao mountain were used to confrontthe simulated responses (by inserting the environmental data intothe model) with the observed real responses (contemporaneous tothe environmental data). A regression analysis was performed(Fitted line) to compare the observed real values of the indicatorswith the expected values obtained by the model simulations. Atthe end of each analysis, the 95% confidence limits for the interceptand the slope of the regression were determined, which, togetherwith the results of the respective analysis of variance (ANOVA),allowed us to assess the proximity of the simulations producedwith the observed values (Sokal and Rohlf, 1995). When the resultsof the regression analysis were statistically significant the modelsimulations were considered credible (Sokal and Rohlf, 1995;Oberdorf et al., 2001).

Thereafter, the tendency of total richness (composed by thesum of all vertebrate species) was simulated using realisticscenarios of fire cycles and wind farm installation (15 years, usingthe week as unit of time). Although the model is prepared to workin a totally random mode we used a possible pathway consideringa straightforward comparison of observed scenarios (Strix, 2007).The scenarios considered, for the same hypothetical quadrate(based on realistic data) and for academic demonstration, were: (a)a quadrate with no fire occurrence and no wind mills installation;

Table 1Specification of the key variables considered in this study.

Variables Specification

Environmental variables

Flora richness Number of species

Flora diversity Shannon Index

Altitude Meters above sea level

Slope Inclination in degrees

Vegetation height Maximum vegetation height

Vegetation cover Total cover in percentage

Shrubland cover Percentage

Herbaceous cover Percentage

Distance mill Distance to the nearest mill

Num mills by quadrate Number of mills inside a qua

Average distance of mills in the quadrate Average distance mills inside

Num mills by buffer Number of mills inside a circ

Average distance mills in the buffer Average distance mills inside

Degraded area by wind farm installations Percentage

Indicators

Amphibian richness Number of species

Bird richness Number of species

Mammal richness Number of species

Reptile richness Number of species

Total richness Number of species of vertebr

(b) a quadrate with fire occurrence but no wind mills installation;(c) a quadrate with no fire occurrence but with wind millsinstallation; (d) a quadrate with fire occurrence and with windmills installation. All scenarios start off with an early series of asecondary post-fire succession, considering a wildfire occurrencein time = 0 with the aim of initiating plant succession in time = 1.The same succession path was considered in all scenarios for aneasier perception of the comparisons.

3. Results

3.1. Effects of environmental variables on the selected indicators

3.1.1. Annual influences (Base model, B equations)

A total of four dependent variables (log amphibian richness B,log reptile richness B, log bird richness B and log mammal richnessB) and fourteen independent variables (Table 1) were considered inthe multiple-regression analysis to search for significant relation-ships between these components by using a mixed database of allannual conditions. The regression equations, and their significance,for all the combinations performed, are exposed in Table 2.

3.1.2. Seasonal influences (Alternative model, A equations)

Taking into account a seasonal criteria, a total of four dependentvariables (log amphibian richness A, log bird richness A, logmammal richness A and log reptile richness A) and fourteenindependent (Table 1) variables were considered to test anypossible correlation between them by using separated databases,each one representing the main conditions of winter, spring,summer and autumn. The amphibian richness (log amphibianrichness A) and reptile richness (log reptile richness A) wereremoved from the summer and winter analysis, respectively,because no occurrences of species from these groups wererecorded during these periods. The regression equations, andtheir significance, for all the combinations performed are alsoshown in Table 2.

3.2. Construction of the model and equations

The diagrams of the sub-models presented in the Figs. 2–4 arebased on (a) the relationships detected in the multiple regressionanalysis (Table 2), (b) the expected evolution of the vegetation inthis type of mountains in north Iberia, highly influenced by fire

Model codes

Flora richness A; flora richness B

Flora diversity A; flora diversity B

Altitude

Slope

in centimetres Vegetation height

Total cover

Shrubland cover

Herbaceous cover

in meters Dist mill

drate of 6.25 ha Num mills qua

a quadrate of 6.25 ha Ave dis mills qua

le of 28.27 ha Num mills buf

a circle of 28.27 ha ave dis mills buf

Degraded area

Amphibian richness A; amphibian richness B

Bird richness A; bird richness B

Mammal richness A; mammal richness B

Reptile richness A; reptile richness B

ates Total richness A; total richness B

Table 2The regression equations, degrees of freedom (DF), coefficient of determination (R2), F-value and their significance level (*P < 0.05, **P < 0.01 and ***P < 0.001) for all the

combinations selected as significant by the stepwise multiple regression. The specification of all variables are available in Table 1.

Equations DF R2 F

Annual (B equations)

log amphibian richness = �0.1264 + 0.069 log flora richness B + 0.035 log

total cover � 0.0061 log ave dis mills buf

240 0.061 5.13**

log bird richness = 0.6151 + 0.160 log vegetation height � 0.258 log total

cover + 0.069 log slope � 0.171 log degraded area

240 0.085 5.51***

log mammal richness = 0.4016 � 0.138 log degraded area 240 0.016 3.85*

log reptile richness = �2.268 + 0.76 log altitude + 0.200 log flora richness

B + 0.068 log vegetation height � 0.145 log total cover + 0.038 log slope

240 0.139 7.58***

Seasonal (A equations)

Winter

log amphibian richness = �0.09328 + 0.39 log flora richness A � 0.68 log

flora diversity A

40 0.308 8.45**

log bird richness = 1.0870 � 0.39 log herbaceous cover � 0.095 log ave

dis mills buf

40 0.310 8.53***

log mammal richness = 0.1804 + 0.160 log vegetation height 40 0.094 4.03*

Spring

log amphibian richness = �0.15907 + 0.147 log flora richness A 51 0.126 7.23*

log bird richness = �0.071 + 0.25 log flora richness A + 0.147 log

vegetation height + 0.143 log slope � 0.301 log degraded area

51 0.330 5.78**

log mammal richness = �0.1554 + 0.40 log flora richness A 51 0.089 4.89*

log reptile richness = 0.170 + 0.36 log degraded area 51 0.113 6.34*

Summer

log bird richness = 0.7388 � 0.62 log flora richness A + 0.20 log vegetation

height � 0.25 log shrubland cover + 0.99 log flora diversity A + 1.20 log

num mills qua � 0.22 log ave dis mills qua

79 0.180 2.67*

log mammal richness = �1.2008 + 0.61 log altitude � 0.65 log flora

diversity A � 1.03 log num mills buf + 0.228 log ave dis mills buf

79 0.181 4.14**

log reptile richness = �2.119 + 0.87 log altitude � 0.120 log dist

mill + 0.121 log ave dis mills qua � 1.43 log num mills buf + 0.190 log

ave dis mills buf � 0.30 log degraded area

79 0.266 4.40**

Autumn

log amphibian richness = �0.3037 + 0.154 log flora richness A + 0.087

log total cover

67 0.108 3.93*

log bird richness = 0.8068 � 0.239 log herbaceous cover � 0.61 log

degraded area

67 0.224 9.36***

log mammal richness = 0.7529 � 0.111 log dist mill � 0.37 log

degraded area

67 0.132 4.94*

log reptile richness = �1.161 + 0.53 log altitude � 0.254 log total

cover + 0.079 log herbaceous cover

67 0.247 6.98***

M. Santos et al. / Ecological Indicators 10 (2010) 192–205196

regimes (e.g. Calvo et al., 2005), and (c) the maximum number ofmills and average area affected by wind farms in this region (Strix,2007). The equations and source code of the model was availableon line as electronic Supplementary material (Appendix 2). Themodel is able to work with scenarios imposed by the user,although, if required, it can generate automatic scenarios based onrealistic situations. Actually, the user can choose between a modelwhere he controls all inputs (e.g. number of mills, time ofinstallation, interval between fires) by deactivating the Automaticvariable (choosing 0) or a model where all inputs are stochasticallygenerated, by activating this variable (choosing 1) (Figs. 3 and 4and Appendix 2—Constants).

In Fig. 2, the sub-model diagram attempts to foresee theresponse of the indicators to the changes that take place in eachquadrate. The independent variables were the logarithms of theenvironmental variables considered and the selected indicatorswere the logarithms of dependent variables (Tables 1 and 2). Theinitial values for these state variables were assumed to be zero(our initial situation in t0) (Appendix 2, Process equations). Later,for simulations representation, the initial value was discarded,since only in t1 (the first point of the simulation) it was possible totake into account the influences of the environmental variables onthe richness estimates. The processes that affect the statevariables are described by difference equations (Appendix 2,Difference equations). The inflows (ex.: spring amphibian gains)affecting the state variables (e.g. log amphibian richness A) werebased on positive constants and all positive partial coefficientsresulting from the previous multiple regression analysis (Table 2,

Fig. 2, Appendix 2—Difference and Process equations). On theother hand all the state variables were affected also by outflows(ex.: spring amphibian losses) related to the negative constantsand partial regression coefficients influences (Table 2, Fig. 2,Appendix 2—Difference and Process equations). Although theoutput for each vertebrate richness measure simulated iscomposed of a given value per time unit, the respective statevariable could have a cumulating behaviour over time in responseto changes in the environmental conditions. Thus, to avoid this, anadditional outflow adjustment was incorporated in each statevariable (ex.: amphibians adjust A). These outflow adjustmentsaimed to empty the state variables in each time step, by a ‘‘flushingcistern’’ mechanism, before beginning the next step with newenvironmental influences (Fig. 2 and Appendix 2—Difference andProcess equations). For process compatibilities and a morerealistic comprehension of the model simulations, some conver-sions were introduced, denominated associated variables (Fig. 2and Appendix 2—Associated variables). Regarding the selectedindicators, these conversions were obtained through an inversetransformation (anti-logarithmic), which transforms logarithmsinto the original measurement units (e.g. amphibian richness A).Other variables, resulting from simple mathematical operationsbetween the associated variables, such as the global vertebratespecies richness (total richness A, total richness B) (Fig. 2 andAppendix 2—Composed variables), were used to complete theoutput of the model and named composed variables. Theenvironmental variables were logarithm transformed for acompatible integration in the balances of the state variables

Fig. 2. Conceptual diagram of the sub-model used to predict vertebrate richness in response to environmental conditions. Rectangles represent state variables; other

variables, parameters or constants are small circles; sinks and sources are cloudlike symbols; flows are thick arrows; all the relations between state variables and other

variables are fine arrows. The specification of all variable codes is expressed in Table 1 and in Sections 3.1 and 3.2.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 197

(Figs. 2–4 and Appendix 2—Associated variables). This transfor-mation (e.g. log flora richness A) was incorporated because thedata required for the state variables balances should have thesame units used to obtain the partial regression coefficients,assumed as holistic ecological parameters (see Section 2).Therefore, only logarithms of the environmental variables areacceptable in the inflows and outflows of the richness indicators(Fig. 2 and Appendix 2—Difference equations and Processequations). Therefore, the model is prepared to accept andtransform real data from the environmental variables and toconvert logarithmic outputs from specific passerine estimationback into the original units.

The Fig. 3 shows the sub-model diagram that intends to predictthe vegetation recovery following fire in different ecologicalcircumstances. In fact, secondary successions are largely controlledby fire regimes, previous vegetation, soil characteristics, nutrientsupplies, climatic patterns and slope exposition (Buhk et al., 2007).Considering that the forecast of vegetation post-fire successionhighly depends on local characteristics (Capitanio and Carcaillet,2008), three secondary succession pathways were included, basedon previous studies in similar ecosystems (e.g. Calvo et al., 2005,

2008). The dynamics of herbaceous vegetation cover, shrublandcover and vegetation height were introduced in the model as tablefunctions (Fig. 3, Appendix 2—Table functions) and the respectivetrend was determined depending on the selected Successionvariable (Fig. 3, Appendix 2—Constants). This option may beintroduced by the user (imp success) (Fig. 3 and Appendix 2—Constants) either when the model is set as deterministic or whenthe succession is generated automatically by the model in astochastic mode (rand succes gen and success) (Fig. 3 andAppendix 2—State variables and Constants). In the deterministicmode the time of the fire occurrence is established by the user (impfire event, Fig. 3 and Appendix 2—Other variables). In contrast,when the model runs stochastically, the fire occurrence (fire event,Fig. 3 and Appendix 2—Other variables) is only determined by thevegetation characteristics (fire probability, Fig. 3 and Appendix 2—Other variables), a consequence of goat herdsmen vegetationmanagement, namely by induced fires in order to achieve suitablepastures for grazing (Vazquez and Moreno, 1998). The otherpertinent factors are randomly generated (probability of fire set upby goat herdsmen—random fire, Fig. 3 and Appendix 2—Othervariables). The fire occurrence (fire event) induces automatically a

Fig. 3. Conceptual diagram of the sub-model used to simulate vegetation dynamics. The specification of all variable codes is expressed in Table 1 and in Sections 3.1 and 3.2.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205198

new secondary succession and the pathway of the vegetationcharacteristics starts from the beginning (periodic succession,Fig. 3 and Appendix 2—Complementary state variables). The samebehaviour happens to the type of climatic conditions (and othereffects such as grazing pressure and/or pest’s occurrence) that willinfluence the dynamics of post-fire vegetation dynamics (climaconditions, Fig. 3 and Appendix 2—Complementary state vari-ables). This variable controls the speed of regeneration, althoughwithout influence in the path of succession (Bell, 2001; Buhk et al.,2007). The flora richness and flora diversity (Shannon, 1948) (florarichness A, flora richness B, flora diversity A, flora diversity B, Fig. 3and Appendix 2—Associated variables) were considered dependentof the vegetation characteristics and dynamics (Grytnes, 2000;Guo, 2001; Capitanio and Carcaillet, 2008). A multiple regressionanalysis was used to substantiate this relationship, based on thedata collected during the field works (for both sub-modelsconsidered, Table 3).

In Fig. 4 the sub-model diagram shows the components relatedto the wind farm installation, such as the number of wind millsinstalled by quadrate, the area affected by soil mobilizations andlocal characteristics such as altitude and inclination. The range ofvalues was chosen taking into account the gradients of the Maraolandscape where the changes were occurring (Strix, 2007). In thisperspective, the maximum number of wind mills by quadrate istwo (num mills qua, Fig. 4 and Appendix 2, Composed variables)(four by buffer; num mills buf, Fig. 4 and Appendix 2, Composed

variables) and the area affected by soil mobilizations never risesabove 6% (Bastos, 2007; Strix, 2007). The number of wind mills, thedistance of the wind mill from the centre of the quadrate and thetime of installation may be automatically generated (Stochasticrun; Automatic = 1, Fig. 3 and Appendix 2—Constants) ordetermined by the user (deterministic run; automatic = 0, Fig. 4and Appendix 2—Constants). When the model is set as automatic(stochastic run), in the first moment when the model startsrunning, it dictates if there is a wind mill installation (e.g. inst t

time gener 1 and inst time mill 1, Fig. 4 and Appendix 2,Complementary state variables process equations and Comple-mentary state variables) and when it will occur (e.g. mill 1 gen 1and auto mill 1, Fig. 4 and Appendix 2, Complementary statevariables process equations and Complementary state variables).After this course of action the next step is to install effectively themill (e.g. mill 1 and mil qua 1, Fig. 4 and Appendix 2,complementary state variables process equations and Comple-mentary state variables) at a random distance (until 150 m for thequadrate and 150–300 m for the buffer) (e.g. dist 1 and dist millqua 1, Fig. 4 and Appendix 2, complementary state variablesprocess equations and complementary state variables). On theother hand, if the model is running in the deterministic mode theoption to install wind mills is decided by the user (e.g. instal 1,Fig. 4 and Appendix 2, constants) as well as the respective time ofinstallation (e.g. time of inst 1, Fig. 4 and Appendix 2, constants)and the distance from the centre of the quadrate (e.g. imp dis mill

Fig. 4. Conceptual diagram of the sub-model used to generate the dynamics associated to wind mill installations and geomorphologic characteristics. The specification of all

variable codes is expressed in Table 1 and in Sections 3.1 and 3.2.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 199

1, Fig. 4 and Appendix 2, constants). Other variables that arecomposed by the previous ones are the distance to nearest mill,number of mills by quadrate, the number of mills by buffer, theaverage distance of the mills of the quadrate and average distanceof the mills of the buffer (dis mill, num mills qua, num mills buf, avedis mills qua, ave dis mills buf, Fig. 4 and Appendix 2, Composedvariables). The area transformed by the installation of the wind

mills and associated structures was designated as degraded area(degraded area, Fig. 4 and Appendix 2, Complementary statevariables) and was considered as dependent on the number of millsby quadrate (num mills qua, Fig. 4 and Appendix 2, Composedvariables). The deterministic and the stochastic pathway optionsfor the degraded area, altitude and slope (degraded area, altitudeand slope, Fig. 4 and Appendix 2, Complementary state variables)

Table 3The regression equations, degrees of freedom (DF), coefficient of determination (R2), F-value and their significance level (*P < 0.05, **P < 0.01 and ***P < 0.001) for all

combinations reported, as selected by stepwise multiple regression analysis.

Equations DF R2 F

Annual (B equations)

log flora richness B = 0.703 + 0.117 log vegetation height � 0.0008 log shrubland

cover + 0.078 log herbaceous cover

240 0.082 6.58***

log flora diversity B = 0.096 + 0.317 log flora richness B 240 0.37 141.5***

Seasonal (A equations)

Winter

log flora richness A = 0.254 + 0.147 log vegetation height + 0.285 log herbaceous cover 40 0.37 11.18***

log flora diversity A = 0.004 + 0.345 log flora richness A + 0.0384 log vegetation height 40 0.57 24.88***

Spring

log flora richness A = 0.900 + 0.156 log herbaceous cover 51 0.09 4.85*

log flora diversity A = 0.103 + 0.297 log flora richness A 51 0.29 20.06***

Summer

log flora richness A = 0.953 79 ***

log flora diversity A = 0.189 + 0.357 log flora richness A � 0.121 log total cover + 0.0621 log shrubland cover 79 0.46 21.55***

Autumn

log flora richness A = 1.11 + 0.117 log vegetation height � 0.178 log total cover 67 0.194 7.82**

log flora diversity A = 0.181 + 0.452 log flora richness A � 0.085 log vegetation height + 0.3 log total

cover � 0.0721 log shrubland cover � 0.108 log herbaceous cover

67 0.41 8.51***

M. Santos et al. / Ecological Indicators 10 (2010) 192–205200

are selected by a toggle (deterministic run, automatic = 0;stochastic run, automatic = 1, Fig. 4 and Appendix 2—Constants).

3.3. StDM performance and simulations

The temporal unit chosen was the week, because it wasassumed acceptable to monitor the changes that occur in this typeof systems (Bastos, 2007). The confrontation between simulatedvalues (based on average environmental conditions) and real ones

Fig. 5. Graphical comparisons, using a fitted line plot, between simulated values produced

axis) for the vertebrate indicators by season: (1) summer, (2) autumn, (3) winter and (4)

richness; (c) open triangle, bird richness; (d) solid triangle, flora diversity; (e) open cir

richness.

are illustrated in the Figs. 5, 6 (model B) and 7 (model A). Theindependent data allowed us to compare values for vertebraterichness (amphibians’ richness, reptile richness, bird richness,mammal richness, and total richness), flora richness and floradiversity (Table 4). The models predict with success the values ofthose indicators, although the Base model (model B), as a rule,operates in more accurate manner (Figs. 5–7 and Table 4).

For illustrative purposes, and taking into account that: (1) bothsub-models can reasonably simulate the indicators response

by model B (simulation results, y-axis) and observed values (averaged real values, x-

spring. Symbols explanation: (a) solid quadrate, flora richness; (b) solid circle, total

cle, mammal richness; (f) cross, reptiles richness; (g) open quadrate, amphibians

Fig. 6. Graphical comparisons, using a fitted line plot, between simulated values

produced by model B (simulation results, y-axis) and observed values (averaged

real values, x-axis) for the vertebrate indicators throughout the year as a whole.

Symbols explanation: (a) solid quadrate, flora richness; (b) solid circle, total

richness; (c) open triangle, bird richness; (d) solid triangle, flora diversity; (e) open

circle, mammal richness; (f) cross, reptiles richness; (g) open quadrate, amphibians

richness.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 201

(Figs. 5–7 and Table 4) and (2) the model’s simulations should beshowed and discussed in a integrative manner, only the globalvertebrate species richness (total richness A, total richness B—Fig. 4and Appendix 2) was selected to simulate holistic trends facingnew dynamic scenarios. In fact species richness is considered alevel-headed indicator (Fleishman et al., 2006) with the advan-

Fig. 7. Graphical comparisons, using a fitted line plot, between simulated values produced

axis) for the vertebrate indicators by season: (1) summer, (2) autumn, (3) winter and (4)

richness; (c) open triangle, bird richness; (d) solid triangle, flora diversity; (e) open cir

richness.

tages demonstrated by other modelling applications (Santos andCabral, 2004). The selection of the option 0 in the ‘‘Automatic’’toggle (Figs. 2 and 3 and Appendix 2—Constants) determined thevariation in the environmental conditions taking into account apossible path of landscape succession (a fire in time = 0 activatesthe initial plant succession) (Strix, 2007). The scenarios considered(Figs. 8 and 9) were based on possible temporal drifts that couldoccur in a quadrate of the studied area. The Fig. 8a (the referencecondition) shows a scenario with no fire occurrence and no windmills installation: (1) the vegetation cover is expected to increaseuntil the shrubs became dominant and then its likely to stabilize(around 130%), although some variance occurs each year, manlyrelated to annual herbs that are only present during spring/summer seasons; (2) vegetation height has the same pattern of thevegetation cover, although when the shrubs are above the 80 cmthe influence of the annual herbs in vegetation height is considerednegligible (maximum height 130 cm). In response to this scenariothe simulations of vertebrate richness based on the Base model(total richness B) and on the Alternative model (total richness A)are showed in Fig. 8b. According to these simulations in the earlyphases of the succession there seems to be a higher richness (5–7species and 7–11 species for models B and A respectively),sequenced by a step reduction and an evident stabilization (4species and 5–7 species for models B and A respectively). TheFig. 8c illustrates a scenario with the incidence of periodic fires incycles of 5 years. Each fire sets the succession in the early phaseobserved in the beginning of all simulations (equivalent to t = 1)and the same pattern of vegetation evolution is considered(reference conditions). The response of the total vertebrate

by model A (simulation results, y-axis) and observed values (averaged real values, x-

spring. Symbols explanation: (a) solid quadrate, flora richness; (b) solid circle, total

cle, mammal richness; (f) cross, reptiles richness; (g) open quadrate, amphibians

Table 4The regression equations, degrees of freedom (DF), coefficient of determination (R2),

F-value and their significance level (*P < 0.05 and ***P < 0.001) for all the observed

versus simulated values on the variables considered (using the annual model and

the seasonal model, model B and model A respectively).

Equations DF R2 F

Year

Simulation results = �0.0998048 + 0.978879

Real values (B model)

6 0.993 734.99****

Autumn

Simulation results = �0.338615 + 1.10242

Real values (B model)

6 0.962 125.04***

Simulation results = �0.417062 + 1.06838

Real values (A model)

6 0.940 78.17***

Winter

Simulation results = �0.290106 + 1.61174

Real values (B model)

6 0.977 216.84***

Simulation results = �0.519858 + 1.74493

(A model)

6 0.963 131.73***

Spring

Simulation results = 0.168026 + 0.74030

Real values (B model)

6 0.995 1097.90***

Simulation results = �0.0445303 + 1.11485

Real values (A model)

6 0.993 730.38***

Summer

Simulation results = �0.116271 + 0.990382

Real values (B model)

6 0.976 199.37***

Simulation results = 0.886441 + 0.739346

Real values (A model)

6 0.748 14.87*

M. Santos et al. / Ecological Indicators 10 (2010) 192–205202

richness (Fig. 8d) to this scenario shows similar trends to thoseobserved in the early phases of Fig. 8b, reflecting vertebrateresilience to fire, largely compelled by the vegetation dynamics.The scenario showed in Fig. 9a reflects the installation of two windmills per quadrat (line 3) at time = 400 which degrade an averageof 1% of the quadrate original vegetation (line 4). The response ofthe total vertebrate richness (Fig. 9b) to these alterations(time = 400) is a decrease in approximately 20% when comparedwith reference response (Fig. 8b). The last scenario considered(Fig. 9c) aims at simulating a quadrate where wind mills are

Fig. 8. Computer simulations for the total richness response (b and d) to different sce

occurrence and no wind mills installation (a) and with fire occurrence and no wind mills

cover in percentage) and (2) vegetation height (vegetation height in cm). (b and d) lines

response of the total vertebrate richness produced by model A. The arrows represent t

installed and fire occurs (the timing of these events is the same ofthe previous Figs. 8c and 9a). The response of the total vertebraterichness to this scenario (Fig. 9d) reveals two different influences.In fact, while the total vertebrate richness response to fire isresilient (as observed in Fig. 8d), the construction of wind millsreduces the total vertebrate richness in a more permanent way bydecreasing the reference values (as observed in Fig. 9b). Compara-tively, the Base model (model B) responds with a stylised patternand lower average values of total species richness. On the otherhand, only the simulations of the Alternative model (model A)captures the seasonal variations in species richness.

4. Discussion

Most studies concerning the ecological impacts of wind farms(and associated structures) on wildlife have been focused onmortality of birds and bats, although some reported alsobehavioural changes in this scope (Hoover and Morrison, 2005;Rabin et al., 2006). The effects of the modifications in the habitatstructure (directly and indirectly related to the installation of windfarms) on biodiversity in general and on terrestrial animal groupsin particular received almost no attention (Larsen and Madsen,2000; Rabin et al., 2006). In this study, the aim was to assess otherecological attributes usually not included in reports about windfarms environmental impacts, such as the holistic changes at thelevel of the communities and ecosystems. As fire plays animportant factor in the dynamics of Mediterranean ecosystems(Montenegro et al., 2004; Herrando et al., 2005; Pausas, 2006), itwas considered a fundamental implicit factor to the comprehen-sion of the wind farm effects in animal and plant communities.

Our results suggest that wind farm disturbance and firedisturbance are significant issues influencing the species richnessof vertebrates. The obtained results are in conformity with severalstudies that tried to identify the disturbance effects on the localcommunities by using indicator taxa approaches (Anand et al.,2005; Ficetola et al., 2007). In fact, the response of vertebrates tofire is considered inconsistent, depending on the type of fire,

narios of land use in a representative quadrate of Marao Mountain: with no fire

installation (c). (a and c) lines explanation: (1) percentage cover of all stratum (total

explanation: (1) response of the total vertebrate richness produced by model B; (2)

he occurrence of fire. A simulated period of 780 weeks (15 years) was considered.

Fig. 9. Computer simulations for the vertebrate total richness response (b and d) to different scenarios of land use in a representative quadrate of Marao Mountain: with no fire

occurrence and with wind mills installation (a) and with fire occurrence and wind mills installation (c). (a and c) lines explanation: (1) percentage cover of all stratum (total

cover in percentage); (2) vegetation height (vegetation height in cm); (3) number of mills installed in the buffer (num mills buf); (4) percentage of area degraded by the

installation of the wind farm (degraded area). (b and d) lines explanation: (1) response of the total vertebrate richness produced by model B; (2) response of the total

vertebrate richness produced by model A. The arrows represent the occurrence of fire. A simulated period of 780 weeks (15 years) was considered.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 203

ecosystem and group studied (Smith, 2000). Anyway, the apparentresilience of the flora and fauna to fire events, a recurrentphenomenon in the study region, is in agreement to resultsobtained in other studies in similar conditions (for revisions seeSmith, 2000; Buhk et al., 2007).

In contrast and independently of the model approach used, ourresults show that vertebrate species responds with a decline inrichness when the wind farm installation was simulated. Since therates of vertebrate mortality caused by wind farm activities areusually low in the study region (Strix, 2007), similar to otherstudies in wind farms located on mountain areas (Barrios andRodriguez, 2004), the identified decrease in richness may thereforebe produced by other factors associated to the wind millsinstallation, such as direct disturbance, structural habitat changesand induced behavioural segregation (Rabin et al., 2006; Fergun-son et al., 2008).

One of the central requirements of StDM is that the data setrecorded includes true gradients of changes (Cabral et al., 2008).In this way, the factors of time and space were implicit in therespective treatment. Such a procedure allows more realism, asthe respective parameters are being considered with regard totheir embedding in time and space. This is of particularimportance when it comes to the comprehension of theindicator’s response. Most of the published papers with StDMapplications were based on average annual databases forpredicting the generic trends of pertinent ecological indicators(Santos and Cabral, 2004; Cabecinha et al., 2004, 2007; Silva-Santos et al., 2006; Cabral et al., 2007; Santos et al., 2007, 2009).However, if the main objective of the modelling effort is, forinstance, the demonstration of the applicability of StDM in thescope of the ecological monitoring and management carried outin more realistic scenarios (Cabecinha et al., 2009; Silva-Santoset al., 2008), this annual approach becomes insufficient. Sincemost of the components of an ecosystem depend on weatherconditions or phenological aspects, different complementaryequations should be introduced into the state variable balance.

Consequently, the simulation performance of a given statevariable results from the calculations of alternative equationsautomatically selected in response to the seasonal influence.Thus, with reference to the modelling goals, the two StDM modelapproaches described in the present paper, the Base model(model B) and the Alternative model (model A), represent atrade-off between realism, complexity and accuracy. The modelA introduces the seasonal variation in the simulations of thevertebrate richness trends, a pattern that is much more likely torepresent what is happening in the field (Fergunson et al., 2008).Therefore, when the studies are based on inventories collected indifferent times of the year and/or regarding specific population,guilds or community dynamics, the seasonal structured modelsshould be the better option (Fergunson et al., 2008). On the otherhand, the model B captures, in a one-dimensional but intuitiveway, the generic trend for the 15 years simulated, highlightingthe effects of wind farms on the structure of the studycommunity. This performance seemed to be satisfactory whendealing with the aims of the most Environmental ImpactAssessments, usually supported by incomplete data (ex.: oneseason data, data from poor inventories).

The StDM models developed in this study seem to represent auseful contribution for detecting key changes in mountaincommunities affected by wind farm installations, namely byquantifying species richness in different ecological circumstances.In this scope, since the richness of a community is holisticallydetermined by the habitat structure and wind farm characteristics(Rabin et al., 2006; Fergunson et al., 2008), our StDM simulationsallowed a better perception of the ecological consequences whenwind farms are installed. Therefore, our proposed methodologyshould be considered as a complementary tool for assessing theintegrity of the communities, by using the global vertebratespecies richness as an indicator within the ‘‘data space’’ of theenvironmental gradients monitored in particular ecosystems, suchas our case-study in mountains dominated by heathlands of NWPortugal.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205204

The species richness is considered the most discriminatingindicator among the commonly used diversity indicators andindices (Magurran, 1998). Nevertheless the species richness shouldbe considered with caution, and other indicators must be used for adeeper analysis of the trends detected (Fleishman et al., 2006).However, since the integrity of these systems can be only partlyassessed by species richness, this approach should be furthercomplemented with the introduction of other indicators, interac-tions and interferences with precise applicability conditions(Magurran, 1998; Fleishman et al., 2006).

5. Conclusion

Complex spatio-temporal environmental data sets are becom-ing common in ecology because of the increasing use of simulationmodels and automated data collection devices. The spatial andtemporal dimensions represent real challenges for the interpreta-tion of these data. A particularly complex problem is thesimulation of the relationships among variables that can divergedramatically in response to the same environmental changes(Steele et al., 2005).

The main objective of the StDM approach proposed is amechanistic understanding of the holistic ecological functioning inthe scope of the need for rapid, standardized and cost-savingassessment methodologies (Santos and Cabral, 2004). The obtainedsimulation results are encouraging since they seem to demonstratethe StDM reliability in capturing the dynamics of the studiedecosystems by predicting the behavioural pattern for the keycomponents selected under very complex and variable environ-mental scenarios. These simulations showed that the ecologicalindicators selected, as state variables, were not indifferent tochanges in the ecological conditions, namely when conditionsrelatively unaffected by human activities were changed by man-induced disturbances. The relevant ecological drifts simulated arein agreement with real observations and other studies thatinvestigated the biological consequences of ecosystem changesby particular anthropogenic impacts, such as the wind farminstallation on threatened ecosystems.

Another goal when developing methods for assessing speciesoccurrence is the feasibility of application and extent to whichthe results can be useful in other areas (Andreasen et al., 2001).Despite the limitations inherent to this demonstration, themethodology proposed is adaptable and was easily applicable todifferent environmental problems in diverse ecosystems. Whencompared to other modelling methodologies, such as ArtificialIntelligence and neural networks (Dzeroski et al., 1997; Kuoet al., 2006), our methodology is more intuitive, namely inmathematical terms, providing easy explanations for theunderlying relations between independent and dependentvariables and because is based on conventional linear methodsthat allowed a more direct development of testable hypotheses.Dzeroski et al. (1997) referred that models produced in the formof rules, based on machine learning approaches, are transparentand can be easily understood by experts. The integrativeapplication of the StDM exhibits these structural qualities butprovides also simple, suitable and intuitive outputs, easilyinterpreted by non-experts (ranging from resource users tosenior policy makers).

Acknowledgments

The authors are indebted to all the students from the Universityof Tras-os-Montes e Alto Douro (UTAD) who assisted in field work.A special thanks is addressed to Dr. Antonio Crespı, from theHerbarium-Botanical Garden of the UTAD, for the supervision offlora and vegetation identification. This study was supported by

Energiekontor Lda. and Gesfinu/EnergiaVerde Lda. in the scope of‘‘Plano Geral de Monitorizacao da Serra do Marao’’ project.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in

the online version, at doi:10.1016/j.ecolind.2009.04.014.

References

Anand, M., Laurence, S., Rayfield, B., 2005. Diversity relationships among sixdifferent groups in recovering and restored forest. Conserv. Biol. 19, 955–962.

Andreasen, J.K., O’Neill, R.V., Noss, R., Slosser, N.C., 2001. Considerations for thedevelopment of a terrestrial index of ecological integrity. Ecol. Indicators 1, 21–35.

Araujo, M.B., Nogues-Bravo, D., Reginster, I., Rounsevell, M., Whittaker, R.J., 2008.Exposure of European biodiversity to changes in human-induced pressures.Environ. Sci. Policy 11, 38–45.

Barrios, L., Rodriguez, A., 2004. Behavioral and environmental correlates of soaring-bird mortality at on-shore wind turbines. J. Appl. Ecol. 41, 72–81.

Bastos, R., 2007. Seleccao e resposta de indicadores ecologicos a cenarios deperturbacao em ecossistemas de montanha. Relatorio Final de Estagio (Licen-ciatura em Biologia Ramo Cientıfico) Universidade de Tras-os-Montes e AltoDouro, Vila Real, Portugal, 60 pp.

Bell, D.T., 2001. Ecological response syndromes in the flora of southwesternAustralia: fire resprouters versus reseeders. Bot. Rev. 67, 417–440.

Bibby, C.J., Burgess, N.D., Hill, D.A., Mustoe, S.H., 2000. Bird Count Techniques,second ed. Academic Press, London, 302 pp.

Bond, W.J., Keeley, J.E., 2005. Fire as a global ‘‘herbivore’’: the ecology and evolutionof flammable ecosystems. Trends Ecol. Evol. 20, 387–394.

BP (British Petroleum), 2007. Cumulative installed wind turbine capacity. http://www.bp.com/sectiongenericarticle.do?categoryId=9017928&conten-tId=7033483 Cited 10 October 2007.

Buhk, C., Meyn, A., Jentsch, A., 2007. The challenge of plant regeneration after fire inthe Mediterranean Basin; scientific gaps in our knowledge on plant strategiesand evolution of traits. Plant Ecol. 192, 1–19.

Cabecinha, E., Silva-Santos, P., Cortes, R., Cabral, J.A., 2007. Applying a stochastic-dynamic methodology (StDM) to facilitate ecological monitoring of runningwaters, using selected trophic and taxonomic metrics as state variables. Ecol.Model. 207, 109–127.

Cabecinha, E., Cortes, R., Cabral, J.A., 2004. Performance of a stochastic-dynamicmodelling methodology for running waters ecological assessment. Ecol. Model.175, 303–317.

Cabecinha, E., Pardal, M.A., Cortes, R., Cabral, J.A., 2009. A Stochastic DynamicMethodology (StDM) for reservoir’s water quality management: validation ofa multi-scale approach in a south European basin (Douro, Portugal). Ecol.Indicators 9, 329–345.

Cabral, J.A., Rocha, A., Santos, M., Crespı, A.L., 2007. A stochastic dynamic metho-dology (SDM) to facilitate handling simple bird indicators in the scope of theagri-environmental measures problematics. Ecol. Indicators 7, 34–47.

Cabral, J.A., Cabecinha, E., Santos, M., Travassos, P., Silva-Santos, P. 2008. Simulatingthe ecological status of changed ecosystems by holistic applications of a newStochastic Dynamic Methodology (StDM). In: Alonso, M.S., Rubio, I.M. (Eds.),Ecological Management, New Research. Nova Science Publishers, New York.

Calvo, L., Santalla, S., Valbuena, L., Marcos, E., Tarrega, R., Luis-Calabuig, E., 2008.Post-fire natural regeneration of a Pinus pinaster forest in NW Spain. Plant Ecol.197, 81–90.

Calvo, L., Tarrega, R., Luis, E., Valbuena, L., Marcos, E., 2005. Recovery after experi-mental cutting and burning in three shrub communities with different domi-nant species. Plant Ecol. 180, 175–185.

Capitanio, R., Carcaillet, C., 2008. Post-fire Mediterranean vegetation dynamics anddiversity: a discussion of succession models. For. Ecol. Manag. 3–4, 431–439.

Corn, P.S., Bury, R.B., 1990. Sampling Methods for Terrestrial Amphibians andReptiles. Gen. Tech. Rep. PNW-GTR-256. U.S. Department of Agriculture, ForestService, Pacific Northwest Research Station, Portland, OR, 34 pp.

Cuesta, D., Taboada, A., Calvo, L., Salgado, J.M., 2006. Short-term effects of fire onarthropods in Calluna-heathlands. For. Ecol. Manag. 234S, 186.

Dale, V.H., Beyeler, S.C., 2001. Challenges in the development and use of ecologicalindicators. Ecol. Indicators 1, 3–10.

Dale, V.H., Peacock, A.D., Garten Jr., C.T., Sobek, E., Wolfe, A.K., 2008. Selectingindicators of soil, microbial, and plant conditions to understand ecologicalchanges in Georgia pine forests. Ecol. Indicators 8, 818–827.

Dzeroski, S., Grbovic, J., Walley, W.J., Kompare, B., 1997. Using machine learningtechniques in the construction of models: II data analysis with rule introduc-tion. Ecol. Model. 95, 95–111.

Elzinga, C.L., Salzer, D.W., Willoughby, J.W., Gibbs, J.P., 2001. Monitoring Plant andAnimal Populations. Blackwell Science, 300 pp.

Everaert, J., Stienen, E.W., 2007. Impact of wind turbines on birds in Zeebrugee(Belgium): significant effect on breeding stern colony due to collisions. Bio-divers. Conserv. 16, 3345–3349.

Fergunson, C.A., Carvalho, L., Scott, E.M., Bowman, A.W., Kirika, A., 2008. Assessingecological responses to environmental change using statistical models. J. Appl.Ecol. 45, 193–203.

M. Santos et al. / Ecological Indicators 10 (2010) 192–205 205

Ficetola, G.F., Sacchi, R., Scali, S., Gentilli, A., De Bernardi, F., Galeotti, P., 2007.Vertebrates respond different to human disturbance: implications for the use ofthe focal species approach. Acta Oecol. 31, 109–118.

Fielding, A.H., Whitfield, D.P., McLeod, D.R.A., 2006. Spatial association as anindicator of the potential for future interactions between wind energy devel-opments and golden eagles Aquila chrysaetos in Scotland. Biol. Conserv. 131,359–369.

Fleishman, E., Noss, R.F., Noon, B.R., 2006. Utility and limitations of species richnessmetrics for conservation planning. Ecol. Indicators 6, 543–553.

Flowerdew, J.R., Shore, R.F., Poulton, S.M.C., Sparks, T.H., 2004. Live trapping tomonitor small mammals in Britain. Mammal. Rev. 34, 31–50.

Gamboa, G., Munda, G., 2007. The problem of windfarm location: a social multi-criteria evaluation framework. Energy Policy 35, 1564–1583.

Gitzen, R.A., West, S.D., Maguire, C.C., Manning, T., Halpern, C.B., 2007. Response ofterrestrial small mammals to varying amounts and patterns of green-treeretention in Pacific Northwest forests. For. Ecol. Manag. 251, 142–155.

Grytnes, J.A., 2000. Fine-scale vascular plant species richness in different alpinevegetation types: relationship with biomass and cover. J. Veg. Sci. 11, 87–92.

Guo, Q., 2001. Early post-fire succession in California Chaparral: changes in diver-sity, density, cover and biomass. Ecol. Res. 16, 471–485.

Herrando, S., Brotons, L., Llacuna, S., 2005. Post-fire dynamics in Mediterraneanshrublands: are bird communities structure by arthropod availability? Rev. Cat.Ornit. 21, 17–28.

Hoover, S.L., Morrison, M.L., 2005. Behavior of red-tailed hawks in a wind turbinedevelopment. J. Wildl. Manag. 69 (1), 150–159.

Kuo, P.D., Banzhaf, W., Leier, A., 2006. Network topology and the evolution ofdynamics in an artificial genetic regulatory network model created by wholegenome duplication and divergence. Biosystems 85, 177–200.

Larsen, J.K., Madsen, J., 2000. Effects of wind turbines and other physical elementson field utilization by pink-footed geese (Anser brachyrhynchus): a landscapeperspective. Landsc. Ecol. 15, 755–764.

Magurran, A.E. (Ed.), 1998. Ecological Diversity and its Management. PrincetonUniversity Press, Princeton, New Jersey.

Molina, R.T., Tellez, T.R., Alcaraz, J.A.D., 1992. Aportacion a la bioclimatologia dePortugal. Anales Jardin Botanico Madrid 49, 245–264.

Montenegro, G., Ginoccho, R., Segura, A., Keel, J.E., Gomez, M., 2004. Fire regimes andvegetation responses in two Mediterranean-climate regions. Ver. Chil. Hist. Nat.77, 455–464.

Moreira, F., Ferreira, P.G., Rego, F.C., Bunting, S., 2001. Landscape changes andbreeding bird assemblages in northwestern Portugal: the role of fire. Landsc.Ecol. 16, 175–187.

Niemeijer, D., Groot, R.S., 2008. A conceptual framework for selecting environ-mental indicator sets. Ecol. Indicators 8, 14–25.

Noest, V., Van der Maarel, E., Van der Muelen, F., Van der Laan, D., 1989. Optimum-transformation of plant-species cover-abundance values. Vegetatio 83, 167–178.

O’Connell, T.J., Bishop, J.A., Brooks, R.P., 2007. Sub-sampling data from the NorthAmerican Breeding Bird Survey for application to the Bird Community Index, anindicator of ecological condition. Ecol. Indicators 7, 679–691.

Oberdorf, T., Pont, D., Hugheny, B., Chessel, D., 2001. A probabilistic model char-acterizing fish assemblages of French rivers: a framework for environmentalassessment. Freshwater Biol. 46, 399–415.

Pausas, J.G., 2006. Simulating Mediterranean landscape pattern and vegetationdynamics under different fire regimes. Plant Ecol. 187, 249–259.

Pearmam, P.B., Weber, D., 2007. Common species determine richness patterns inbiodiversity indicator taxa. Biol. Consev. 138, 109–119.

Pimm, S.L., Brow, J.H., 2004. Domains of diversity. Science 304, 831–833.Portal do Governo, 2007. Assinatura de Contratos de Incentivos a Energia Eolica.

http://www.portugal.gov.pt/Cited 7 October 2007.

Rabin, L.A., Coss, R.G., Owings, D.H., 2006. The effects of wind turbines on anti-predator behavior in California ground squirrels (Spermophilus beecheyi). Biol.Conserv. 131, 410–420.

Ralph, C.J., Sauer, S.R., Droege, S. Monitoring bird populations by point countsGeneral technical report PSW 149. Forest Service, United States Department ofAgriculture, 1995, 187 pp.

Rivas Martınez, S., Gandullo, J.M., Allue, J.L., Montero, J.L., Gonzalez, J.L., 1987.Memoria del mapa de Series de Vegetacion de Espana. ICONA, Madrid.

Ryan, T.J., Philippi, T., Leiden, Y.A., Dorcas, M.E., Wigley, T.B., Gibbons, J.W., 2002.Monitoring herpetofauna in a managed forest landscape: effects of habitattypes and census techniques. For. Ecol. Manag. 167, 83–90.

Santos, M., Cabral, J.A., 2004. Development of a stochastic dynamic model forecological indicators prediction in changed Mediterranean agroecosytems ofnorth-eastern Portugal. Ecol. Indicators 3, 285–303.

Santos, M., Travassos, P., Repas, M., Cabral, J.A., 2009. Modelling the performance ofbird surveys in non-standard weather conditions: general applications withspecial reference to mountain ecosystems. Ecol. Indicators 9, 41–51.

Santos, M., Vaz, C., Travassos, P., Cabral, J.A., 2007. Simulating the impact of socio-economic trends on threatened Iberian wolf populations Canis lupus signatus innorth-eastern Portugal. Ecol. Indicators 7, 649–664.

Shannon, C.E., 1948. The mathematical theory of communication. Bell Syst. Tech. J.27, 379–423 623–656.

Silva-Santos, P., Pardal, M.A., Lopes, R.J., Murias, T., Cabral, J.A., 2006. A stochasticdynamic methodology (SDM) to the modelling of trophic interactions, with afocus on estuarine eutrophication scenarios. Ecol. Indicators 6, 394–408.

Silva-Santos, P., Pardal, M.A., Lopes, R.J., Murias, T., Cabral, J.A., 2008. Testing theStochastic Dynamic Methodology (StDM) as a management tool in a shallowtemperate estuary of south Europe (Mondego, Portugal). Ecol. Model. 210, 377–402.

Smith, J.K. (Ed.), 2000. Wildland Fire in Ecosystems: Effects of Fire on Fauna. RMRS-GTR-42-vol. 1 Ogden. U.T. Rocky Mountain Research Station, 83 pp.

Sokal, R.R., Rohlf, F.J., 1995. Biometry, third ed. W.H. Freeman and Company, NewYork.

Steele, B.M., Reedy, S.K., Nemani, R.R., 2005. A regression strategy for analyzingenvironmental data generated by spatio-temporal processes. Ecol. Model. 181,93–108.

Strix, 2007. Plano Geral de Monitorizacao—Parques eolicos de Penedo Ruivo,Mafomodes, Seixinhos e Teixeiro. Strix Environment and Innovation, PortoPortugal.

Tognelli, M.F., 2005. Assessing the utility of indicator groups for the conservation ofSouth American terrestrial mammals. Biol. Conserv. 121, 409–417.

Torre, I., Diaz, M., Martınez-Padilla, J., Bonal, R., Vinuela, J., Fargallo, J.A., 2007. Cattlegrazing, raptor abundance and small mammal communities in Mediterraneangrasslands. Basic Appl. Ecol. 8, 565–575.

Ukmar, E., Corrado, B., Luiselli, L., Bologna, M.A., 2007. The effect of fire on com-munities, guilds and species of breeding birds in burnt and control pinewoods incentral Italy. Biodivers. Conserv. 16, 3287–3300.

Van der Maarel, E., 1979. Transformation of cover-abundance values in phytoso-ciology and its effects on community similarity. Vegetatio 39, 97–114.

Vazquez, A., Moreno, J.M., 1998. Patterns of lightning, and people caused fires inPeninsular Spain. Int. J. Wildland Fire 8, 103–115.

Wittenberg, L., Malkinson, D., Ofer, B., Halutzy, A., Tesler, N., 2007. Spatial andtemporal patterns of vegetation recovery following sequences of forest fires in aMediterranean landscape, Mt. Carmel Israel. Catena 71, 76–83.

Yasuda, M., 2004. Monitoring diversity and abundance of mammals with cameratraps: a case study in Mount Tsukuba, central Japan. Mamm. Study 29, 37–46.

Yost, A.C., 2008. Probabilistic modeling and mapping of plant indicator in a North-east Oregon industrial forest, USA. Ecol. Indicators 8, 45–56.

Zar, J.H., 1996. Biostatistical Analysis. Prentice-Hall, Englewood Cliffs, NJ.