The Importance of Climate, Productivity, Habitat Heterogeneity, and Human Impact on the Distribution of Bird of Prey Species

Home , Productivity (ecology)

NORTH-WESTERN JOURNAL OF ZOOLOGY 13 (2): 285-296 ©NwjZ, Oradea, Romania, 2017 Article No.: e161608 http://biozoojournals.ro/nwjz/index.html

The importance of climate, productivity, habitat heterogeneity, and human impact on the distribution of bird of prey species richness across two continents latitudinal gradient (Europe and Africa)

Dragoș Mihail ȘTEFĂNESCU1*, Daniel RĂDUȚOIU1 and Emil MARINESCU2

1. University of Craiova, Faculty of Horticulture, Department of Biology and Environmental Engineering, Craiova, Romania. 2. University of Craiova, Faculty of Science, Department of Geography, Craiova, Romania. * Corresponding author, D.M. Ștefănescu, E-mail: [email protected]

Received: 14. April 2016 / Accepted: 16. August 2015 / Available online: 13. October 2016 / Printed: December 2017

Abstract. The objective of this paper was to identify the main factors that explain the greater part of variation in bird of prey species richness (BPSR) along latitudinal gradient, particularly from Northern Europe to South Africa. For this purpose we used 26 environmental variables grouped according to the main hypotheses proposed to explain the regional variability in species richness. Because the relationship between species richness and environmental variables vary with spatial scale, we used two type of grain size (100 km × 100 km and 200 km × 200 km). To link explanatory environmental variables to BPSR, for each grain size, we used two multiple regression methods: a global model, the Ordinary Least Squares (OLS) and a local model, the Geographically Weighted Regression (GWR). Further, we designed 11 different models based on combination of the environmental variables, both for OLS and GWR methods, in order to select the best one. For OLS method, variation in BPSR across the study area was best predicted by total model (all variables), both for

finer (R2 adjusted = 0.712; AICc = 22394.89) and coarser spatial scales (R2 adjusted = 0.804; AICc = 4888.98). For GWR method, the (climate + habitat heterogeneity) model was the best model regarding variation

explaination and model performances, both for finer (R2 adjusted = 0.953; AICc = 16890.49) and coarser (R2

adjusted = 0.958; AICc = 3909.18) resolutions. The GWR regression models performed better than OLS models in explaining variation in BPSR, the improvement of models performance was evident.In our study, both regression methods and analysis of variance partitioning lead to the conclusion that climate (particularly PC2) and habitat heterogeneity (particularly HABNB) were the most influental factors in determining BPSR at the spatial scales analyzed in this study.

Key words: bird of prey species richness, latitudinal gradient, climate, productivity, habitat heterogeneity, human influence.

Introduction known as the latitudinal species richness gradient (Willis & Whittaker 2002, Willing et al. 2003, Biological diversity or biodiversity is “the variety Stiling 2012). The number of species of most taxa of life”, encompassing all species on Earth, all (plants, invertebrates, amphibians, reptiles, birds forms, levels and combinations of natural varia- and mammals) exhibit highest species richness in tion (Gaston & Spicer 2004). If we are concern- the tropics, then richness declining as latitude in- edabout conserving biodiversity, we should un- crease (Blackburn & Gaston 1996, Cox & Moore derstand how the number of species in a particu- 2005,Lomolino et al. 2006, Townsend et al. 2008), lar place is determined and which factors are be- as for example the birds of North America where hind this (Townsend et al. 2008). The species are species richness increase from Arctic Canada to not evenly distributed around the globe. Physical Panama (Stiling 2012). Nevertheless, birds have features of the environment (e.g. temperature, been frequently used in studies regarding rich- precipitations) and biotic factors (e.g. predation) ness-environment relationships (Qian et al. 2009), limit the distribution patterns of species, according considering that they are highly mobile, easy to to the niche concept in the Grinnell tradition (So- observe and occupy a wide range of habitats beron 2007, Peterson et al. 2011, Stiling 2012). throughout the globe (Møller et al. 2010). Brown (1995) argued that the limits of species More than one hundred hypotheses were pro- geographic ranges are set mostly by physical con- posed to explain regional variability in species ditions. richness (Rahbeck & Graves 2001, Rahbeck et al. One of the most striking pattern in species di- 2007). Between them, many authors focused usu- versity is the gradient ranging from low at the ally on: i) ambient energy hypothesis and related poles to high at the equator, a phenomenon climatic hypotheses (H-Acevedo & Currie 2003, 286 D.M. Ștefănescu et al.

Hawkins et al. 2003, Forsman &Mönkkönen dicators) (Sergio et al. 2008).Top predators may 2003,Moreno-Rueda & Pizzarro 2007, Qian et al. causally structure a whole community through 2009, Ortiz-Yusty 2013), which is based on the trophic cascade, such as when a predator limits its concept that environments at high latitudes have prey, which in turn limits its own prey, and thus mean conditions farther from species optima than play an important role in ecosystem functioning their low-latitude counterparts; ii) productivity (Sergio et al. 2008). hypothesis (van Rensburg et al. 2002, Hurlbert & Haskell 2003, Hawkins et al. 2003,Ding et al. 2006, Qian et al. 2009, Kennedy et al. 2014, Mouchet et Materials and methods al. 2015, Nieto et al. 2015), which emphasizes a Species data positive relationship between productivity (en- Range maps of 110 birds of prey species from two orders ergy) and richness; and iii) habitat heterogeneity (Accipitriformes and Falconiformes) which inhabit hypothesis (Kerr & Packer 1997, Fraser 1998, Mo- Europe and Africa where acquired from Bird species distri- reno-Rueda& Pizzarro 2009, Stein et al. 2014, Xu et bution maps of the world database al. 2015) according to which spatially heterogene- (http://www.biodiversityinfo.org), version 5.0 (BirdLife ous environments can be expected to contain more International and Nature Serve 2015) as polygon layers. Further, all these maps where combined in ArcGis 10.1 species because they provide a greater variety of and SAM software version 4.0 (Rangel et al. 2010) to ob- habitats. tain an overall birds of prey richness map. In this study, we analyze the relationship between different environmental variables and the Study area distribution of bird of prey species richness (BPSR) Studies regarding latitudinal gradient of species richness from northern Europe to southern Africa (latitu- should encompass a large area, that extends from the Arc- dinal gradient of richness), to identify the factors tic to theTropics, because partial coverage may lead to misleading patterns (Ding et al. 2006), and that was why that explain most of variation in BPSR across the we have chosen two continents, Europe and Africa, to study area. This vast area included a wide variety analyze variation in BPSR from higher to low latitudes. of climatic gradients and habitat types. Birds of We divided the study area into a numbers (n) of prey include 328 species (del Hoyo et al. 1994, Fer- equal-area quadrats (grid-cell format) using the World guson-Lees 2001, BirdLife International Taxo- Behrmann equal-area projection, with two grain sizes of nomic Checklist v8.02015) grouped in three main 100 km × 100 km (n = 3153) and 200 km × 200 km (n = orders, Cathartiformes (New World vultures) with 709), to take into account the effect of variation in spatial resolution on the relationship between BPSR and envi- 7 species, Accipitriformes (eagles, buzzards, harri- ronmental variables. We retained for analysis only quad- ers, hawks, kites) with 251 species and Falconi- rats with the same area (10,000 km2 for finer spatial extent formes (falcons) with 70 species, all thesespecies and 40,000 km2 for coarser resolution), avoiding the distributed around the globeand adapted to dif- quadrats where land area where < 50% of the quadrat ferent environments, from tundra to desert and size, such as most of coastal and island zones. Although the tropical forest. several authors (Rahbeck & Graves 2001, Rahbeck et al. Top vertebrate predators (including birds of 2007, Davies et al. 2007) have used latitude-longitude quadrats as units of analysis in their studies, this method prey species) have fascinated humans for millen- can create some problems of results interpretation, know- nia (Sergio et al. 2006). The charisma of these spe- ing that the area of latitude-longitude quadrats decreases cies is now used by conservationists in planning with increasing latitude (Ding et al. 2006). Preparation of protected areas, sites occupied by top predators the maps was done in ArcGis 10.1. having gnumbers and more diversity of avian species, vulnerable avian species, butterfly species Environmental variables and tree species (Sergio et al. 2005). Also, (Martin To test the effect of the environment on the BPSR distribution across the study area we used values of 26 envi- & Ferrer 2013) argued that birds of prey species ronmental variables (Table 1),including also four human may by an easy and cost-effective group to moni- (influences) variables, considering the current strong im- tor biodiversity compared with other vertebrate pact of human activities on biodiversity and the impor- groups. There are two main reason why conserva- tance of human activity in explaining species richness tion strategies based on top predators may lead to (Wilson et al. 2008, Filloy et al. 2015). These variables se- broader biodiversity benefits: i) the predators may lected as predictors of BPSR were rescaled and averaged directly cause high biodiversity (by resource facili- at the level of a quadratof 100 km × 100 km and 200 km × 200 km with ArcGis 10.1 and SAM software version tation and trophic cascades), or ii) they may be 4.0.As we mentioned above, the most frequently used hy- spatio-temporally associated with it (acting as in- potheses for testing environment-species richness Distribution of bird of prey species richness 287

Table 1. Environmental variables used in the present study.

Spatial resolu- Environmental variables Abbreviation Temporal extent/unit Source tion (original) Mean temperature for Dec/Jan (C) TDJ 2011-2012/degree Celsius 0.025º WorldGrid Mean temperature for Feb/Mar (C) TFM 2011-2012/degree Celsius 0.025º WorldGrid Mean temperature for Apr/May (C) TAM 2011-2012/degree Celsius 0.025º WorldGrid Mean temperature for Jun/Jul (C) TJJ 2011-2012/degree Celsius 0.025º WorldGrid Mean temperature for Aug/Sep (C) TAS 2011-2012/degree Celsius 0.025º WorldGrid Mean temperature for Oct/Nov (C) TON 2011-2012/degree Celsius 0.025º WorldGrid Mean annual temperature(C) MAT 2011-2012/degree Celsius 0.025º WorldGrid Maximum annual temperature(C) MAXT 2011-2012/degree Celsius 0.025º WorldGrid Minimum annual temperature(C) MINT 2011-2012/degree Celsius 0.025º WorldGrid Long-term precipitation for Nov/Dec/Jan(C) PNDJ 1950-2000/mm 0.025º WorldClim Long-term precipitation for Feb/Mar/Apr(C) PFMA 1950-2000/mm 0.025º WorldClim Long-term precipitation for May/Jun/Jul(C) PMJJ 1950-2000/mm 0.025º WorldClim Long-term precipitation for Aug/Sep/Oct(C) PASO 1950-2000/mm 0.025º WorldClim Mean annual precipitation(C) MAP 2003-2006/mm 0.025º WorldClim Net radiation (E) NETR 2015/W m-2 0.025º NEO Solar insolation (E) SOLAR 2015/W m-2 0.025º NEO Normalized Difference Vegetation Index (P) NDVI 2015/- 0.025º NEO Net Primary Productivity (P) NPP 2015/gC m-2 year 0.025º NEO Elevation (HH) ELEV 2000/meters 0.05º SRTM 30+ Elevation range (HH) ELEVR 2000/meters 0.05º SRTM 30+ Number of habitat types (HH) HABNB 2012/- 0.025º LP DAAC Simpson Habitat Diversity (HH) SDIV 2012/- 0.025º LP DAAC Long-term lights at night images* (H) LTLN 1992-2010/- 0.025º NOAA Average population density(H) POPDENS 1990-2015/no. km-2 0.2º GPW Artificial surfaces (urban areas >50%) (H) ARTS 2000-2005/% 0.025º WorldGrid (GCLC) Rain fed croplands (H) CROP 2000-2005/% 0.025º WorldGrid (GCLC)

WorldGrid (http://worldgrids.org) based on value of the 8-day MODIS (Moderate resolution imaging spectroradi- ometer) day-time LST time series data (https://lpdaac.usgs.gov) for temperature data; WorldClim (http://www.worldclim.org); NEO, Nasa Earth Observation (http://neo.sci.gsfc.nasa.gov); SRTM 30+ (http://topex.ucsd.edu); LP DAAC, Land Processes Distributed Active Archive Center, based on MODIS Land Cover data (https://lpdaac.usgs.gov); NOAA, National Oceanic and Atmospheric Administration (http://ngdc.noaa.gov); GPW, Gridded Population oh the World (http://sedac.ciesin.columbia.edu); GCLC, Globe Cover Land Cover developed by ESA, European Space Agency (http://maps.elie.ucl.ac.be/CCI). * Based on values of an index of reflection levels from 1 to 63. C – climate variables; E – energy variables; P – productivity variables; HH – habitat heterogeneity variables; H- human influences variables.

relationships are climate-ambient energy hypothesis, BPSR, for each grain size, we used two multiple regres- productivity hypothesisand habitat heterogeneity hy- sion methods: a global model, the Ordinary Least Squares pothesis, further grouping of variables for statistical (OLS) and a local model, the Geographically Weighted analysis was made based on thisstandpoint (see statistical Regression (GWR), (Fotheringham et al. 2002). The as- analysis). sumption of global regression method is that the relationship under study is stationary and therefore the estimated Statistical analysis parameters remain constant in space (for this reason, OLS All statistical analyses were made for both types of grain regression was used mainly for comparative purposes). In sizes. To gather a set of uncorrelated climatic variables, contrast, the GWR regression method is a technique that we conducted a principal component analysis (PCA) to expands standard regression for use with spatial data reduce collinearity and the number of climatic explana- (which are not often stationary), enabling local influences tory variables (the first 14 variables of Table 1). Only (Wang et al. 2005, Ortiz-Yusty 2013). principal component with eigenvalues > 1 were retained Since the GWR approach focuses on spatial non- for further analyses. stationary issues, considering spatial autocorrelation, we To link the explanatory environmentalvariables (a to- expect that the GWR regression method to perform better tal of 15 by reducing the number of climatic variables) to than classical OLS regression techniques (Legendre 1993, 288 D.M. Ștefănescu et al.

Diniz-Filho et al. 2003, Wang et al. 2005; Dormann et al. pooled), and spatially structured environment by parti- 2007). We estimated the spatial autocorrelation in the tioning of variance with respect to the variance in BPSR, data, including the response variable (BPSR), the explana- following the procedure described by Borcard et al. 1992, tory environmental variables, and regression residuals, by Legendre 1993, Moreno-Rueda & Pizzaro 2007). For this constructing spatial correlograms using Moran’s I coeffi- purpose, we used a OLS multiple partial regression cients at 10 distance classes (for BPSR and environmental analysis (Legendre & Legendre 2012), which estimates the variables) and 20 distance classes (for OLS and GWR re- percentage of variance explained by one variable after gression residuals), using SAM software version 4.0. We controlling for the effect of the other predictors, and ap- grouped environmental variables into 11 different models plied to refined best OLS model, for both grain sizes. and tested which of themwas the best predictor of variation in BPSR across study area, for both OLS and GWR methods, in order to select the best one: (1) Climate- Results Energy model: PC1 + PC2 + PC3 + SOLAR + NETR (see Table 1 for variable explanation); (2) Productivity model: For finer resolution (100 km× 100 km) mean BPSR NPP + NDVI; (3) Habitat heterogeneity model: ELEV + per quadrat was 28.22 (1SD: 15.67), with highest ELEVR + HABNB + SDIV; (4) Human impact model: species richness, up to 64 species per grid cell, de- POPDENS + LTLN + ARTS + CROP; (5) Climate-Energy + Productivity model; (6) Climate-Energy + Habitat het- tected in East Africa and the lowest, 4 species per erogeneity model; (7) Climate-Energy + Human impact grid cell, occurred in North Africa (Sahara region), model; (8) Productivity + Habitat heterogeneity model; Fig. 1.BPSR followed the same pattern of distribu- (9) Productivity + Human impact model; (10) Habitat het- tion and for coarser spatial scale (200km× 200 km), erogeneity + Human impact model; and (11) Total model: with mean BPSR equal to 31.38 (1SD: 16.78), up to Climate-Energy + Productivity + Habitat heterogeneity + 67 species per grid cell as highest species richness, Human impact. Comparisons of model quality (selection and 4 species per grid cell as lowest richness value procedure that allow the identification of the best model to describe patterns under study) for both OLS and GWR (Fig. 1). methods were made using Akaike’s information criterion

(AICc) corrected for small sample size (Quinn & Keough 2002,Fotheringham et al. 2002, Chatterjee & Simonoff 2013) and the maximum model fit (R2adj). The lower the

AICc the closer the approximation of model to reality, thus the best model is the one with the smallest AICc (Fotheringham et al. 2002; Wang et al. 2005). We also compare the relative performance of the GWR and OLS models to replicate the observed data set using an approximate likelihood ratio test based on F-test (Fother- ingham et al. 2002). This test is carried out by dividing the residual sum of squares for OLS model by that for the

GWR model with d1 (for OLS model) and d2 (for GWR model) degrees of freedom (DF), with effective degrees of freedom equal to the effective number of parameters. The best model (per above selection procedure) for OLS regression method and for each grain sizes was refined by conducting a backward stepwise regression (applied with XLSTAT 2014 Addinsoft) and a “Model Selection and Multi-Model inference” (implemented in SAM software version 4.0), to further elucidate the relationship between environmental variables and BPSR and to minimize VIF (variance inflation factor) of predictors. The relative importance of explanatory variables for refined OLS best model were evaluated with an approximate likelihood ratio test for nested models (Lichstein et al. 2002, Tognelli & Kelt 2004, Weisberg 2005). Different groups of factors that influence the distribution of BPSR may coincide or coun- teract one another, and both the environmental variables and BPSR can be spatially autocorrelated, consequently we estimated the relative importance of spatial structure Figure 1. Bird of prey species richness across the study (accounted by longitude, latitude, longitude2; latitude2 area per 100 km × 100km (top) and 200 km × 200km and longitude × latitude variables), environment (climate, (down) grid cell. productivity, habitat heterogeneity and human effect, Distribution of bird of prey species richness 289

For grid cell with grain size of 100 km ×100 km, principal component (Table 2). At the grain size of three principal components (PC) - out of fourteen - 200 ×200 km three principal components were also produced by climate data (temperature and pre- retained, based on their eigenvalues, with similar cipitation) had eigenvalues higher than 1.0. First pattern of variable correlations and contributions. principal component (PC1) explained 54.4% of the BPSR and environmental variables are spa- total variance in the climate data, the second one tially autocorrelated, with positive autocorrelation (PC2) 23.2% and the third one 14.4%, together all over short distance and negative autocorrelation at these variables accounted for 92% of overall cli- large distance, Moran’s I values being significantly mate variance (Table 2). Table 2 showed also load- larger than the expected values (P<0.005) for all ings (correlations) between climatic variables and correlograms (Fig. 2). The spatial correlogram for the three principal components.PC1 scores are BPSR showed that this variable is positively auto- positively correlated with all temperature data correlated up to c. 3600 km. The same pattern of and negatively correlated with precipitation vari- autocorrelation was observed for both cell grid ables, MAT, MAXT, TAM, TAS and TONvariables dimensions. having high contribution to PC1 variable (Table 2). The OLS and GWR regression models pro- Note the inverse relationship between tempera- duced significant relationships (P<0.001 in all ture and precipitation data loadings, indicating cases) between BPSR and environmental variables that high values of temperature variables are asso- with similar pattern of overall performances for ciated with low values of precipitation data. PC2 both types of resolution (Table 3). scores are positively correlated with MAP, PASO, TDJ, PMJJ and MINT (variables with highest con- OLS regression tribution) and negatively correlated with three Performances of each global OLS regression model temperature variables (MAXT, TAS and TJJ), with are shown in Table 3.Variation in BPSR across low contribution. The second component is mainly study area was best predicted, per R2 and AICc, by a gradient in mean annual precipitation (MAP), total model (all variables), both for finer (R2 ad- long term precipitation for Aug/Sep/Oct (PASO) justed = 0.712; AICc = 22394.89) and coarser spatial and mean temperature for Dec/Jan (TDJ). The scales (R2 adjusted = 0.804; AICc = 4888.98), Table third principal component variables (PC3) are cor- 3. The model performance (AICc) of refined OLS related positively with PNDJ and PFMA variables best model (Table 4) was slightly superior to the and negatively with PMJJ and PASO, precipitation OLS best model (Table 3), with AICc = 22392.38 for variables had the greatest contribution to this finer resolution and AICc=4881.39 for coarser resolution. Parameter estimates of environmental vari- Table 2. Principal components analysis of climate data ables for the refined OLS best model, for finer and withPC1, PC2 and PC3 loadings, for finer spatial resolu- coarser resolutions are indicated in Table 4, mostly tion 100 km ×100 km. Full name of variables are indi- considered variables contributing significantly to cated in Table 1. accounting for the variation in BPSR. Per likeli- Climate variable PC1 PC2 PC3 hood ratio test, the most important environmental Eigenvalues 7.611 3.251 2.021 variables in the refined OLS best model was Proportion 0.544 0.232 0.144 NETR, followed by PC2, HABNB, NDVI and SDIV MAP -0.591 0.754 0.048 (all these variables indicated a positive association MAT 0.960 0.263 0.023 MAXT 0.942 -0.028 0.025 with BPSR) for grid cells with finer resolution, and MINT 0.752 0.591 0.179 HABNB, followed by PC2, SDIV, NDVI, and PASO -0.435 0.692 -0.532 SRTM (with the same direction of association) for PFMA -0.575 0.426 0.653 coarser grid (Fig. 3). CROP, SOLAR and PC1 were PNDJ -0.523 0.259 0.772 the environmental variables that contributed least PMJJ -0.453 0.593 -0.635 to BPSR variation. Therefore, principal component TAM 0.899 0.065 -0.274 loadings (particularly PC2) and regression coeffi- TAS 0.857 -0.303 0.239 cients from the refined OLS best model indicate TDJ 0.661 0.680 0.134 TFM 0.790 0.571 -0.101 that BPSR is positively correlated with tempera- TJJ 0.779 -0.381 -0.144 ture and precipitation (based on PC2), habitat het- TON 0.822 0.393 0.308 erogeneity and productivity (NDVI), for both types of spatial scales.

290 D.M. Ștefănescu et al.

Figure 2. Spatial correlograms for (a) BPSR, (b) PC2, (c) NETR, (d) NDVI, (e) HABNB and (f) CROP, for finer spatial resolution 100 km × 100 km (all correlograms are significant at P< 0.005).Full names of variables are indicated in Table 1.

Table 3. Performance of the global ordinary least square (OLS) and geographically weighted regression (GWR) models explaining bird of prey species richness (BPSR) distribution across the study area at the grain sizes of 100 km× 100 km and 200 km× 200 km.

MODEL (100km ×100 km) Methods AICc R2(adjusted) DF F Climate-Energy OLS 24254.93 0.479 6 7.10*** GWR 18195.75 0.928 77.23 Habitat heterogeneity OLS 24715.91 0.396 5 5.69*** GWR 19375.24 0.892 77.92 Human impact OLS 25786.76 0.152 5 1.80NS GWR 23944.31 0.539 17.62 Productivity OLS 25529.15 0.218 3 6.12** GWR 19790.44 0.879 39.88 Climate + Productivity OLS 24217.14 0.485 8 8.89*** GWR 17504.62 0.942 102.35 Climate + Human impact OLS 23923.28 0.531 10 2.07NS GWR 21658.60 0.775 32.00 Climate + Habitat heterogeneity OLS 22590.58 0.693 10 6.59*** GWR 16890.49 0.953 137.27 Productivity + Habitat heterogeneity OLS 24267.07 0.477 7 6.49*** GWR 18520.99 0.919 100.64 Productivity + Human impact OLS 24840.92 0.372 7 1.89NS GWR 22852.13 0.671 24.10 Habitat heterogeneity + Human impact OLS 24103.87 0.503 9 1.93NS GWR 22066.56 0.742 32.52 Total model (all variables) OLS 22394.89 0.712 16 2.01NS GWR 20274.48 0.855 53.14

MODEL (200km ×200 km) Methods AICc R2(adjusted) DF F Climate-Energy OLS 5523.05 0.504 6 7.27*** GWR 4236.32 0.929 61.72 Habitat heterogeneity OLS 5464.18 0.543 5 5.85*** GWR 4337.16 0.915 62.38 Human impact OLS 5846.31 0.217 5 1.85NS GWR 5833.22 0.585 14.50 Distribution of bird of prey species richness 291

Table 3. (Continued).

MODEL (200km ×200 km) Methods AICc R2(adjusted) DF F Productivity OLS 5833.22 0.229 3 6.45*** GWR 4523.53 0.885 32.84 Climate + Productivity OLS 5522.54 0.506 8 9.06*** GWR 4123.28 0.941 80.28 Climate + Human impact OLS 5402.67 0.584 10 2.09NS GWR 4913.17 0.798 26.9 Climate + Habitat heterogeneity OLS 4920.52 0.789 10 5.74*** GWR 3909.18 0.958 105.91 Productivity + Habitat heterogeneity OLS 5405.49 0.581 7 7.13*** GWR 4166.82 0.936 79.05 Productivity + Human impact OLS 5625.03 0.428 7 1.89NS GWR 5198.18 0.698 19.79 Habitat heterogeneity + Human impact OLS 5231.22 0.673 9 1.77NS GWR 4860.67 0.811 26.95 Total model (all variables) OLS 4888.98 0.804 16 1.78NS GWR 4540.01 0.884 44.35

NS = not significant; **p<0.01; ***p<0.001.

Table 4. Model results and variable parameters for refined OLS best model, for both types of grid resolutions.Full names of variables are indicated in Table 1.

100 km × R2 = 0.713, R2(adjusted) = 0.712, 100 km F13, 3139 = 599.86, P< 0.0001, AICc = 22392.38 Variable Coeffi- t -value Standard b - SD b + SD cient error Constant -8.153 -2.76** 2.956 -11.109 -5.197 NDVI 0.019 9.01*** 0.002 0.017 0.021 SDIV 0.071 8.31*** 0.008 0.063 0.079 HABNB 1.840 24.29*** 0.075 1.765 1.915 NPP -0.002 -8.47**** 0.000 -0.002 -0.002 SOLAR 0.044 4.31*** 0.010 0.034 0.054 NETR 0.089 6.34*** 0.014 0.075 0.103 SRTM 0.006 12.76*** 0.000 0.006 0.006 CROP 0.156 8.04*** 0.019 0.137 0.175 POPDENS 0.022 6.63*** 0.003 0.019 0.025 LTLN -0.859 -9.16*** 0.093 -0.952 -0.766 PC1 0.948 4.64*** 0.204 0.744 1.152 PC2 1.433 5.38*** 0.266 1.167 1.699 PC3 0.566 3.75*** 0.151 0.415 0.717 200 km × R2 = 0.804, R2(adjusted) = 0.801,

200 km F11, 697 = 259.98, P< 0.0001, AICc = 4881.39 Figure 3. Likelihood ratio estimates of environmental Variable Coeffi- t -value Standard b - SD b + SD variables of refined OLS total model for both grid cell cient error resolutions.Full names of variables are indicated in Ta- *** Constant -26.859 -4.55 5.901 -32.760 -20.958 ble 1. *** NDVI 0.028 6.63 0.004 0.024 0.032 SDIV 0.129 7.41*** 0.017 0.112 0.146

HABNB 1.878 14.59*** 0.128 1.750 2.006 NPP -0.000 -1.67NS 0.000 0.000 0.000 GWR regression SOLAR 0.086 4.18*** 0.020 0.066 0.106 Variation in BPSR distribution was better pre- SRTM 0.007 6.98*** 0.001 0.006 0.008 dicted with the local GWR regression (Table 3), CROP 0.115 2.54* 0.045 0.070 0.160 compared with global OLS regression, per the se- * POPDENS 0.017 2.04 0.008 0.009 0.025 lected evaluation criteria (R2, AICc and F-test re- LTLN -1.077 -4.47*** 0.241 -1.318 -0.836 sults). The GWR (climate + habitat heterogeneity) PC1 1.620 4.07*** 0.397 1.223 2.017 model was determinedas the best model,both for PC2 1.884 6.02*** 0.312 1.572 2.196 2 finer (R adjusted = 0.953; AICc = 16890.49) and NS = not significant; *p <0.05; **p<0.01; ***p<0.001; coarser (R2 adjusted = 0.958; AICc = 3909.18) reso- ****p<0.0001 292 D.M. Ștefănescu et al. lutions. Descriptive statistics of the parameter es- rameter estimates from refined OLS best model for timates for the best GWR model (climate + habi- both resolutions. For example, interquartile range tat), for both spatial scales, are presented in Table of GWR PC2 coefficient (-4.181 to 5.816) was far 5, indicating a wide variation across the study outside of the corresponding ± 1 standard devia- area, and a spatially complex relationship between tion of the OLS PC2 coefficient (1.167 to 1.699), for BPSR and environmental predictors. Per the me- 100 km × 100 km grain size. dian value obtained for each predictor of (climate Partial regression results further indicated that + habitat) model, HABNB, PC2 and PC3 were the most of the variation in BPSR is explained by envi- variables that most contributed (positively) to this ronment (climate, productivity, habitat heteroge- model, and presented the greatest range of varia- neity and human factors) and spatially structured tion, for both kinds of spatial resolutions, indicat- environment, from 71.3% for finer spatial scale to ing that the habitat heterogeneity and the climate 80.4% for coarser scale (Table 6), whereas the spa- (temperature and precipitations) are factors most tial structure alone explained only between 1.5% linked with BPSR variation across study area. for coarser scale to 2.9% for finer spatial scale. When we consider only the environmental com- Table 5. Descriptive statistics of the parameter estimates ponent, non-human factors explained 70% of vari- for GWR best model (climate + habitat heterogeneity) ance in BPSR for finer spatial scale and 79.7% for for both types of grid resolutions. coarser scale. Among non-human factors, for both 100 km × 100 km spatial scales, climate and habitat heterogeneity Variable Statistics explained much of the variation in roughly equal Minimum Median Maximum 25% 75% percentage (Table 6), while productivity explained quartile quartile Constant -73.919 21.291 328.158 -1.744 48.258 only between 0.1% (finer resolution) and 1.2% PC1 -5.844 0.121 8.311 -1.751 2.848 (coarser resolution) of variance. PC2 -27.573 0.307 14.670 -4.181 5.816 PC3 -22.317 0.466 15.153 -1.889 4.676 Table 6. Variation partitioning (%) of bird of prey species SOLAR -1.045 0.016 0.387 -0.103 0.110 richness data for both grain sizes.

NETR -0.286 0.080 1.337 -0.060 0.260 Grain size HABNB -1.041 0.522 1.771 0.151 0.852 100 × 100 200 × 200 SDIV -0.097 0.031 0.136 0.002 0.065 km km SRTM -0.020 - 0.001 0.024 -0.004 0.007 Total explained variance 74.2 81.9 SRTMR -0.007 0.001 0.008 -0.000 0.002 Environment 38.0 39.4 200 km × 200 km Spatial structure 2.9 1.5 Variable Statistics Environment + spatial structure* 33.3 41.0 Minimum Median Maximum 25 % 75 % Total environmental variance 71.3 80.4 quartile quartile Human factors only 1.3 0.7 Constant -232.166 13.996 171.475 -7.578 40.079 Human factors + non-human factors* 13.9 20.8 PC1 -7.160 -0.222 8.798 -2.014 2.362 Non-human factors only 56.1 58.9 PC2 -47.850 1.380 14.940 -4.430 6.462 Total non-human factors 70.0 79.7 PC3 -29.719 0.692 48.861 -1.670 4.820 Climate only 22.5 21.6 SOLAR -0.495 0.040 0.948 -0.080 0.128 Productivity only 0.8 1.2 NETR -0.483 0.055 1.234 -0.059 0.234 Habitat heterogeneity 21.4 34.6 HABNB -0.394 0.527 1.425 0.126 0.894 Climate + Productivity* 9.1 3.9 SDIV -0.135 0.064 0.229 0.001 0.119 Climate + Habitat heterogeneity* 4.1 0.6 SRTM -0.049 0.000 0.025 -0.005 0.006 Productivity + Habitat heterogeneity* 0.0 2.3 SRTMR -0.006 0.001 0.019 -0.000 0.003 Climate + Productivity + 12.1 15.5 Habitat heterogeneity*

Relationships between BPSR and environ- *shared variance mental variables were spatially non-stationary, interquartile range of the GWR best model estimates exceeded the ± 1 standard deviation of the equiva- Discussion lent refined OLS best model parameters (Tables 4 and 5). These tables show that all the interquartile Climate (energy), productivity and habitat hetero- range of parameter estimates from GWR (climate geneity were the primary factors used for explain- + habitat) model were positioned outside the ing bird richness distribution at large spatial range of ± 1 standard deviation of equivalent pa- scales. In our study, both regression methods and Distribution of bird of prey species richness 293 analysis of variance partitioning lead to the con- ing species richness in dry zones (Hawkins et al. clusion that climate (particularly PC2) and habitat 2003; Rueda & Pizarro 2007). Several studies re- heterogeneity (particularly HABNB) were the ported the influence of weather fluctuations, espe- most influential factors in determining BPSR at the cially precipitation, on demographic parameters spatial scales analyzed in this study. The GWR on raptors (Hustler & Howells 1985, Hustler & (climate + habitat) best model, which accounts for Howells 1987, Ontiveros & Pleguezuelos 2003, spatial autocorrelation, performed better than re- Wichmann et al. 2003). Hustler & Howells 1985 fined OLS best model (all variables) in explaining noticed that the amount of rainfall received from variation in BPSR;the improvement of models per- January to the end of May appeared to influence formance was evident. Residuals from refined egg-laying times of Tawny Eagle (Aquila rapax) in OLS best model showed positive spatial autocor- Africa;pairs laid early if less than a third of the relation up to c. 1300 km, for both grain size, rain quantity fell in March and April, and laid late whereas residuals from the GWR (climate + habi- if the total rainfall for this period was 300 mm or tat) model showed very low spatial autocorrela- less. The same pattern of GWR parameter for PC2 tion, indicating that the last model better repre- was observed also for coarser resolution. Per PC2 sents the distribution of BPSR across the two con- loadings (Table 2), BPSR is also correlated with tinents (Fig. 4). Also, it was observed in a series of temperatures for winter and spring months, studies that broad scale environment-richness re- knowing that temperature is associated with the lationships become more evident when grain size breeding success of birds, either positively or increases (Rahbeck & Graves 2001; Xu et al. 2015). negatively (Newton 2003). Calderon et al. 1987 Our study confirmed an increase in variance ex- showed that in the case of Spanish Imperial Eagle plained by OLS and GWR models from finer to (Aquila adalberti) successful eggs hatching was coarser spatial scale. positively correlated with the average minimum The relation between BPSR and environmental temperature in March, in the coldest years hatch- variables is better understood by analyzing the ing success being as low as 30%. Also, Ontiveros & GWR parameter variation. At the finer scale, for Pleguezuelos (2003) highlight the fact that average example, the GWR parameter for PC2 (mean an- annual temperature is the main climatic variable nual precipitation, long term precipitation for explaining the breeding success of Bonelli’s eagle Aug/Sep/Oct and mean temperature for (Hieraaetus fasciatus) throughout its latitudinal Dec/Janhaving the greatest positive loading range in the Western Mediterranean, annual tem- scores on this principal component, Table 1) had a perature being a deterministic parameter during median of 0.307 with a range of - 27.573 to 14.670 certain periods of the raptor’s life history. (Table 5), with positive values for North and East Although there is a positive relationship be- Africa (zones with the lowest annual precipitation tween HABNB and BPSR, both for finer (R2 = values), and negative value (relationship) in the 0.264; P< 0.001) and coarser (R2 = 0.437; P< 0.001) rest of the study area. This underlines the fact that spatial scales, local GWR regressions revealed a precipitation would have a major role in determin- more complex spatial pattern of this relationship.

Figure 4. Spatial correlograms for the residuals of OLS and GWR best models, for both spatial scales.

294 D.M. Ștefănescu et al.

For finer resolution, we observed negative rela- dominant type of land cover (Mouchet et al. 2015). tionships between HABNB and BPSR in Eastern Climate and habitat heterogeneity were the Europe, and positive relationships for the rest of most influential factors in determining BPSR at the the study area. Also, we found negative relation- spatial scales analyzed in this study. The GWR ships in Eastern Europe and Central Africa for (climate + habitat) model, which accounts for spa- coarser resolution. According to Stein et al. (2014) tial autocorrelation, was the best predictor of environmental heterogeneity is one of the most variation in BPSRacross the study area. Also, local important factors influencing species richness GWR regressions help us to reveal a more com- gradients, particularly vegetation and topographic plex spatial pattern of this relationship. heterogeneity showing strong associations with species richness. In fact, the relative importance of the climate and habitat heterogeneity in determining species References richness varies with spatial scale (Rueda and Piz- Anderson, D.L. (2001): Landscape heterogeneity and diurnal raptor zaro 2009). In a multiscale assessment of avian diversity in Honduras: the role of indigenous shifting richness patterns, Rahbek and Graves (2001) cultivation. Biotropica 33: 511-519. Bellocq, M.I., Gomez-Insausti, R. (2005): Raptorial birds and found that climatic variables explained 58-72% of environmental gradients in the southern Neotropics: a test of the variance in species richness, with precipitation species-richness hypotheses. Austral Ecology 30: 892-898. being the most influential factor at finer spatial BirdLife International and Nature Serve (2015): Bird species distribution maps of the world. BirdLife International, scales, whereas topography (as a surrogate for Cambridge UK and NatureServe, Arlington, USA. habitat heterogeneity) was the dominant factor at Blackburn, T.M., Gaston, K.J. (1996): Spatial patterns in the species coarser spatial scales. Nevertheless, this fact was richness of birds in the New World. Ecography 19: 369-376. Borcard, D., Legendre, P., Drapeau, P. (1992): Partialling out the observed by us, with habitat heterogeneity being spatial component of ecological variation. Ecology 73: 1045-1055. relatively more important (34.6% of explained Brown, J.H. (1995): Macroecology. The University of Chicago Press, variance) than climate (21.6% explained variance) Chicago, USA. Calderon, J., Castroviejo, J., Garcia, L., Ferrer, M. (1987): El Aguila at coarser spatial scale (Table 6). Also, in a study Imperial Aquila adalberti en Donana: algunos aspectos de su that followed the description of spatial patterns in reproducción. Alytes 5: 47-72. bird of prey species richness of South America, Chatterjee, S., Simonoff, J.F. (2013): Handbook of regression analysis. John Wiley & Sons, New Jersey, USA. Diniz-Filho et al. (2002) showed that habitat type Cox, C.B., Moore, P.D. (2005): Biogeography. An ecological and and heterogeneity affect species richness at differ- evolutionary approach. 7th Edition. Blackwell Publishing, ent spatial scales. Anderson (2001) concluded that Oxford, UK. Davies, R.G., Orme, C.D.L., Storch, D., Olson, V.A., Thomas, G.H., landscape heterogeneity was more important in Ross, S.G., Ding, T-S, Rasmussen, P.C., Bennett P.M., Owens explaining differences in raptor species diversity I.P.F., Blackburn T.M., Gaston, K.J. (2007): Topography, energy than any single habitat or combination of habitats and the global distribution of bird species richness. Proceedings of the Royal Society B-Biological Science 274: 1189-1197. in Honduras. Also, Xu et al. (2015) pointed out del Hoyo, J., Elliott, A., Sargatal, J. (1994): Handbook of the birds of that elevation variability, as a measure of habitat the world, vol. 2, New World vultures to guineafowl. Lynx heterogeneity, was the most important predictor Edicions, Barcelona, Spain. Ding, T.S., Yuan, H.W., Geng, S., Koh, C.N., Lee, P.F. (2006): Macro- of mammals and resident birds in China. scale bird species richness patterns of the East Asian mainland Species richness may be affected also indi- and islands: energy, area and isolation. Journal of Biogeography rectly by climate through its effect on productivity 33: 683-693. Diniz-Filho, J.A.F, de Sant’ Ana, C.E.R., de Souza, M.C., Rangel, (Rueda and Pizzaro 2009). In our study productiv- T.F. (2002): Null models and spatial patterns of species richness ity (NPP and NDVI) had a minor influence in de- in South American birds of prey. Ecology Letters 5: 47-55. termining BPSR spatial distribution. Bellocq and Diniz-Filho, J.A.F., Bini, L.M., Hawkins, B.A. (2003): Spatial autocorrelation and red herrings in geographical ecology. Gomez-Insausti (2005) found that the mean annual Global Ecology and Biogeography 12: 53-64. temperature was the strongest environmental Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, predictor of raptor species richness, with 82% ex- J., Carl, G., Davies, R.G., Hirzel, A., Jetz, W., Kissling, W.D., Kuhn, I., Ohlemuller, R., Peres-Neto, P.R., Reineking, B., plained variance, followed by actual evapotran- Schroder, B., Schurr, F.M., Wilson, R. (2007): Methods to account spiration (AET) and vegetation structure which for spatial autocorrelation in the analysis of species explained 77% of spatial variance in raptors rich- distributional data: a review. Ecography 30: 609-628. Ferguson-Lees, J., Christie, D.A. (2001): Raptors of the World. ness. In a study regarding the importance of envi- Christopher Helm, London, UK. ronmental variables on European terrestrial verte- Filloy, J., Grosso, S., Bellocq, M.I. (2015): Urbanization altered brate species distribution, richness was best pre- latitudinal patterns of bird diversity-environment relationships in the southern Neotropics. Urban Ecosystems 18: 777-791. dicted by actual evapotranspiration (AET) and Distribution of bird of prey species richness 295

Forsman, J.T., Mönkkönen, M. (2003): The role of climate in limiting Nieto, S., Flombaum, P., Garbulsky, M.F. (2015): Can temporal and European resident bird populations. Journal of Biogeography spatial NDVI predict regional bird-species richness? Global 30: 55-70. Ecology and Conservation 3: 729-735. Fotheringham, A.S., Brunsdon, C., Charlton, M. (2002): Ontiveros, D., Pleguezuelos, J.M. (2003): Influence of climate on Geographically Weighted Regression. The analysis of spatially Bonelli’s eagle’s (Hieraaetus fasciatus V. 1822) breeding success varying relationships. John Wiley & Sons, West Sussex, through the Western Mediterranean. Journal of Biogeography England. 30: 755-760. Fraser, R.H. (1998): Vertebrate species richness at the mesoscale: Ortiz-Yusti, C.E. (2013): Temperature and precipitations as relative roles of energy and heterogeneity. Global Ecology and predictors of species richness in northern Andean amphibians Biogeography Letters 7: 215-220. from Colombia. Caldasia 35: 65-80. Gaston, K.J., Spicer, J.Y. (2004): Biodiversity: an introduction. Peterson, A.T., Soberon, J., Pearson, R.G., Anderson, R.P., Martinez- Second Edition. Blackwell Publishing, Malden, USA. Meyer, E., Nakamura, M., Araujo, M.B. (2011): Ecological niches H-Acevedo, D., Currie, D.J. (2003): Does climate determine broad- and geographic distributions. Princeton University Press, scale patterns of species richness? A test of the causal link by Princeton, USA. natural experiments. Global Ecology and Biogeography 12: 461- Qian, H., Wang, S., Li, Y., Wang, X. (2009): Breeding bird diversity 473. in relation to environmental gradients in China. Acta Oecologica Hawkins, B.A., Field, R, Cornell, H.V., Currie, D.J., Guegan, J.F., 35: 819-823. Kaufman, D.M., Kerr, J.T., Mittelbach, G.G., Oberdorff, T., Quinn, G.P., Keough, M.J. (2002): Experimental design and data O’Brien, E.M., Porter, E.E., Turner, R.G. (2003): Energy, water, analysis for biologists. Cambridge University Press, Cambridge, and broad-scale geographic patterns of species richness. Ecology UK. 84: 3105-3117. Rahbek, C., Gotelli, NJ., Colwell, R.K., Entsminger, G.L., Rangel, Hawkins, B.A., Porter, E.E., Diniz-Filho, J.A.F. (2003): Productivity T.F.L.V.B., Graves, G.R. (2007): Predicting continental-scale and history as predictors of the latitudinal diversity gradient of patterns of bird species richness with spatially explicit models. terrestrial birds. Ecology 84: 1608-1623. Proceedings of the Royal Society B-Biological Science 274: 165- Hurlbert, A.H., Haskell, J.P. (2003): The effect of energy and 174. seasonality on avian species richness and community Rahbek, C., Graves, G.R. (2001): Multiscale assessment of patterns composition. The American Naturalist 161: 83-97. of avian species richness. Proceedings of the National Academy Hustler, K., Howells, W.W. (1985): A population study of Tawny of Science 98: 4534-4539. Eagles in the Hwange National Park, Zimbabwe. Ostrich 57: Rangel, T.F., Diniz-Filho, J.A.F., Bini, L.M. (2010): SAM: a 101-106. comprehensive application for Spatial Analysis in Hustler, K., Howells, W.W. (1987): Habitat preference, breeding Macroecology. Ecography 33: 46-50. success and the effect of primary productivity on Tawny Eagles Sergio, F., Caro, T., Brown, D., Clucas, B., Hunter, J., Ketchum, J., Aquila rapax in the tropics. Ibis 131: 33-40. McHugh, K., Hiraldo, F. (2008): Top predators as conservation Kennedy, J.D., Wang, Z., Weir, J.T., Rahbek, C., Fjeldså, J., Price, tools: ecological rationale, assumptions, and efficacy. Annual T.D. (2014): Into and out of the tropics: the generation of the Review of Ecology, Evolution, and Systematics 39: 1-19. latitudinal gradient among New World passerine birds. Journal Sergio, F., Newton, I., Marchesi, L. (2005): Top predators and of Biogeography 41: 1746-1757. biodiversity. Nature 436: 192. Kerr, J.T., Packer, L. (1997): Habitat heterogeneity as a determinant Sergio, F., Newton, I., Marchesi, L. (2006): Ecologically justified of mammal species richness in high-energy regions. Nature 385: charisma: preservation of top predators delivers biodiversity 252-254. conservation. Journal of Applied Ecology 43: 1049-1055. Legendre, P. (1993): Spatial autocorrelation: trouble or new Soberon, J. (2007): Grinnellian and Eltonian niches and geographic paradigm? Ecology 74: 1659-1673. distributions of species. Ecology Letters 10: 1115-1123. Legendre, P., Legendre, L. (2012): Numerical ecology. 3rd Edition. Stein, A., Gerstner, K., Kreft, H. (2014): Environmental Elsevier, Amsterdam, The Netherlands. heterogeneity as a universal driver of species richness across Lichstein, J.W., Simons, T.R., Shriner, S.A., Franzreb, K.E. (2002): taxa, biomes and spatial scales. Ecology Letters 17: 866-880. Spatial autocorrelation and autoregressive models in ecology. Stiling, P. (2012): Ecology. Global insights & investigations. Ecological Monographs 72: 445-463. McGraw Hill, New York, USA. Lomolino, M.V., Riddle, B.T., Brown, J.H. (2006): Biogeography. 3ed Tognelli, M.F., Kelt, D.A. (2004): Analysis of determinants of Edition. Sinauer Associates, Inc, MA, USA. mammalian species richness in South America using spatial Martin, B., Ferrer, M. (2013): Assessing biodiversity distribution autoregressive models. Ecography 27: 427-436. using diurnal raptors in Andalusia, Southern Spain. Ardeola 60: Townsend, C.R., Begon, M., Harper, J.L. (2008): Essentials of 15-28. ecology. 3rd Edition. Blackwell Publishing, MA, USA. Møller, A.P., Fiedler, W., Berthold, P. (2010): Effects of climate van Rensburg, B.J., Chown, S.L., Gaston, K.J. (2002): Species change on birds. Oxford University Press, New York, USA. richness, environmental correlates, and spatial scale: a testing Moreno-Rueda, G., Pizarro, M. (2007): The relative influence of using South African birds. The American Naturalist 159: 566- climate, environmental heterogeneity, and human population 577. on the distribution of vertebrate species richness in south- Wang, Q., Ni, J., Tenhunen, J. (2005): Application of a eastern Spain. Acta Oecologica 32: 50-58. geographically-weighted regression analysis to estimate net Moreno-Rueda, G., Pizarro, M. (2009): Relative influence of habitat primary production of Chinese forest ecosystems. Global heterogeneity, climate, human disturbance, and spatial structure Ecology and Biogeography 14: 379-393. on vertebrate species richness in Spain. Ecological Research 24: Weisberg, S. (2005): Applied linear regression. 3rd Edition. John 335-344. Wiley & Sons, New Jersey, USA. Mouchet, M., Levers, C., Zupan, L., Kuemmerle, T., Plutzar, C., Erb, Wichmann, M.C., Jeltsch, F., Dean, W.R.J., Moloney, K.A., Wissel, K., Lavorel, S., Thuiller, W., Haberl, H. (2015): Testing the C. (2003): Implication of climate change for the persistence of effectiveness of environmental variables to explain European raptors in arid savanna. Oikos 102: 186-202. terrestrial vertebrate species richness across biogeographical Willing, M.R., Kaufman, D.M., Stevens, R.D. (2003): Latitudinal scales. PloS ONE 10(7): e0131924. gradients of biodiversity: pattern, process, scale, and synthesis. Newton, I. (2003): Population limitation in birds. Elsevier Academic Annual Review of Ecology, Evolution, and Systematics 34: 273- Press Limited, San Diego, USA. 309.

296 D.M. Ștefănescu et al.

Willis, K.J., Whittaker, R.J. (2002): Species diversity-scale matters. bird species richness in China based and habitat groups. PloS Science 295: 1245-1248. ONE 10(12): e0143996. Wilson, J.W., van Rensburg, B.J., Ferguson, J.W.H., Keith, M. (2008): The relative importance of environment, human activity and space in explaining species richness of South African bird orders. Journal of Biogeography 35: 342-352. Xu, H., Cao, M., Wu, J., Cai, L., Ding, H., Lei, J., Wu, Y., Cui, P., Chen, L., Le, Z., Cao, Y. (2015): Determinants of mammal and