Landscape and Urban Planning 120 (2013) 107–118 Contents lists available at ScienceDirect Landscape and Urban Planning journal homepage: www.elsevier.com/locate/landurbplan Research paper Using high-resolution remote sensing data for habitat suitability models of Bromeliaceae in the city of Mérida, Venezuela Caroline Judith a,b,∗ , Julio V. Schneider a,b,c , Marco Schmidt a,b,c , Rengifo Ortega d , Juan Gaviria e,f , Georg Zizka a,b,c a Senckenberg Research Institute and Natural History Museum Frankfurt, Department of Botany and Molecular Evolution, Senckenberganlage 25, 60325 Frankfurt am Main, Germany b Goethe-University, Institute of Ecology, Evolution and Diversity, Max-von-Laue-Str. 13, 60439 Frankfurt am Main, Germany c Biodiversity and Climate Research Centre (BiK-F), Senckenberganlage 25, 60325 Frankfurt am Main, Germany d Geomatikk AS, Oslo Section, Postbox 103, Økern, Norway e Instituto Jardín Botánico, Facultad de Ciencias, Núcleo La Hechicera, Universidad de los Andes, Apartado 52, 5212 Mérida, Venezuela f GeoBio-Center LMU, Ludwig-Maximilians-University München, Richard-Wagner-Str. 10, 80333 München, Germany h i g h l i g h t s • Twenty species and four genera of Bromeliaceae were found in the city of Mérida. • Bromeliaceae differ in their ecological preferences, with CAM species being adapted to sealed areas. • Combining satellite and elevation data is suitable for bromeliad distribution models. a r t i c l e i n f o Article history: Received 24 October 2012 Received in revised form 26 April 2013 Accepted 19 August 2013 Keywords: Andes Bromeliaceae Habitat suitability modelling Satellite imagery Tillandsia Urban biodiversity a b s t r a c t Little information is available concerning the effects of the increasing urbanization on biodiversity in tropical regions. Species distribution modelling based on interpolated climate data is a widely applied, time- and cost-effective tool to estimate the potential species richness in a target area. However, high fragmentation, strong environmental gradients on a small-scale, and lack of fine-scale environmental data in tropical urban areas require alternative approaches. In this study we combined a rapid species assessment approach with environmental niche modelling based on high-resolution ASTER satellite imagery to predict species distributions of Bromeliaceae in the city of Mérida, Venezuela. Twenty species of Bromeliaceae, e.g. 36% of the total bromeliad diversity of the state of Mérida, were observed in the city, including seven species with CAM physiology. CAM species showed significantly higher occurrence probabilities in zones with higher soil sealing, whereas in C3 species a trend across soilsealing zones was not observed. The remarkable urban species richness of Bromeliaceae is here attributed to the species’ different adaptive strategies, as well as to the strong elevation gradient of Mérida city. Our species modelling approach provides new possibilities for the identification of indicator species in different urban built-up areas. © 2013 Elsevier B.V. All rights reserved. 1. Introduction Urbanization continues to increase on a global scale and individual cities are attaining unprecedented sizes. Nowadays, more than 50% of the world’s population lives in urban areas ∗ Corresponding author at: Goethe-University, Institute of Ecology, Evolution and Diversity, Max-von-Laue-Str. 13, 60439 Frankfurt am Main, Germany. Tel.: +49 069 798 42226. E-mail addresses: caroline.judith@ymail.com (C. Judith), Julio.Schneider@senckenberg.de (J.V. Schneider), Marco.Schmidt@senckenberg.de (M. Schmidt), rengifo.ortega@geomatikk.no (R. Ortega), gaviria@ula.de (J. Gaviria), Georg.Zizka@senckenberg.de (G. Zizka). 0169-2046/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.landurbplan.2013.08.012 (United Nations, 2010). This trend imposes severe pressure upon the biodiversity in urban areas, making biodiversity research in populated areas a major concern. For historical reasons, most ecological and biodiversity assessments in urban areas have been conducted in industrialized countries of the northern hemisphere, whereas few studies are known from tropical or subtropical major cities (e.g. Franceschi, 1996; Sattler, Schmidt, & da Silva Alves, 2010), many of which are among the fastest growing ones worldwide (Cohen, 2006). Urbanization is accompanied by a multitude of effects on the local species composition, such as habitat loss, fragmentation, isolation, and alteration. Generally, urban areas are characterized by highly fragmented green areas with different levels of connectivity 108 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 and altered environmental conditions (Antrop, 2000; McKinney, 2002). Permanently changing conditions require specific adaptation of species for survival. It is therefore of particular interest for conservation and planning issues to understand the potential distribution of species in such an extreme habitat. In recent years, habitat suitability models (HSMs; or similar models like ecological niche models, environmental niche models, species distribution models) have become increasingly important for studying species distribution patterns. The general purpose of these modelling approaches is to estimate the potential geographic species distribution by considering the ecological requirements that are suitable for a species’ occurrence (Guisan & Zimmermann, 2000). This procedure diminishes misinterpretations made by traditional techniques, like point-occurrence maps (Hijmans & Graham, 2006). However, this technique is rather underrepresented in the field of urban biodiversity research. Developments in geographic information systems (GIS), terrain modelling, and climate data interpolation techniques have led to more precise environmental niche models (Elith et al., 2006; Franklin et al., 2013; Kriticos et al., 2012; Seguardo & Araújo, 2004). Until now, resource direct and indirect environmental gradients in terms of raster climate data sets and digital elevation models (DEM) were mainly used as input data during modelling. A major drawback of raster climate data and DEM is that fine-scale environmental data are usually not available for the species-rich tropical countries (Soria-Auza et al., 2010). In contrast, satellite data are widely available. Due to the direct measurement of the earth surface, these data provide spatially refined information of landscape and hydrological characteristics. Consequently, fine-scale habitat heterogeneity, as is observed in urban areas, is adequately represented and their incorporation in HSMs opens new opportunities. The successful application of satellite data has been shown in various ecological investigations of different taxonomic groups (Buermann et al., 2008; Morán-Ordóñez, Suárez-Seoane, Elith, Leonor, & de Estanislao, 2012; Schmidt, König, & Müller, 2008; Zimmermann, Edwards, Moisen, Frescino, & Blackard, 2007). However, to our knowledge no modelling of the habitat suitability of vascular plant species in urban habitats using satellite data has been conducted to date. In this study, the family Bromeliaceae is used as an example for modelling species distributions in a neotropical urban area. Bromeliaceae is one of the largest monocot families, comprising 58 genera and 3248 species (Luther, 2010), and is the second most important family of neotropical epiphytic phanerogams in terms of species numbers after the Orchidaceae (Benzing, 1990; Gentry & Dodson, 1987). They are exclusively neotropical with the West African Pitcairnia feliciana being the only exception. The family includes terrestrial species as well as extreme epiphytes. Due to numerous morphological and physiological adaptations, like leaf succulence, crassulacean acid metabolism (CAM), and specialized water-absorbing trichomes, bromeliads successfully colonize a broad range of natural habitats including rain forests, savannahs, deserts, and rocks. Tank bromeliads, species that impound water due to their tight leaf bases, provide aquatic habitats for a large number of animals, including arthropods and amphibians, thus contributing to the total biodiversity. Noteworthy is that bromeliads are either obligate C3 plants or facultative to obligate CAM plants (Martin, 1994), with the latter being characterized by a higher drought tolerance which in turn might be advantageous for successfully colonizing the generally drier urban areas. Thus, the diversity displayed by bromeliads and their ability to colonize different habitats makes them an excellent model plant family for the study of ecological niches in neotropical urban environments. Additionally, they may serve as surrogate taxa for urban species diversity because many species are widespread, occurring in large parts of the Neotropics. Numerous studies have been conducted regarding diversity and distribution patterns of bromeliads in their natural or semi-natural habitats (e.g. Hornung, 1998; Krömer, Kessler, & Gradstein, 2007; Schneider, Gaviria, & Zizka, 2003; Zizka et al., 2009). However, their distribution in urban habitats is largely unknown. The aims of this study are (1) to examine the diversity of Bromeliaceae in a tropical urban area and their distribution among urban habitats, (2) to test whether CAM bromeliads are more successful in colonizing urban areas than C3 bromeliads, (3) to evaluate the utility of high resolution satellite images for habitat suitability modelling in highly fragmented urban areas. 2. Materials and methods 2.1. Study site The study site encompasses the city of Mérida (Mérida state, Venezuela) as displayed by the official city boundary designated in the map “Plan de desarrollo urbano local de Mérida, marco de referencia” (Universidad de Los Andes, Facultad de Arquitectura y Arte, Unidad de Consultoría Externa y Proyectos). The city is located at a mean elevation of 1630 m a.s.l. (1000–2400 m) on a quaternary plateau in the Venezuelan Andes flanked by two mountain ranges, the Sierra Nevada to the east and the Sierra de la Culata to the northwest. Mérida has approximately 230,000 inhabitants and covers an area of 55 km2 . The climate is tropical mountainous. The precipitation has a bimodal annual distribution, with peaks in May–June and September–November, and the mean annual precipitation is 1737 mm (Stansell, Polissar, & Abbott, 2007). The mean annual temperature is between 16 and 20 ◦ C (Foghin-Pillin, 2002; Rojas & Alfaro, 2000) and is typical of low latitudes, showing strong diurnal oscillations but little seasonal variability. The main watercourses of Mérida are the Río Chama and the Río Albarregas. The city is located in an area of the lower montane rainforest (sensu Kappelle, 1996), but most of the original vegetation has been transformed and only small remnants of the forests currently persist within the limits of the city. The surrounding montane rainforests are characterized by tree species of genera such as Billia (Hippocastanaceae), Brunellia (Brunelliaceae), Weinmannia (Cunoniaceae), Clusia (Clusiaceae), and Decussocarpus (Podocarpaceae), accompanied by abundant epiphytes of the families Bromeliaceae, Orchidaceae, and Piperaceae (Engwald, 1999; Kelly, Tanner, Lughadha, & Kapos, 1994; Schneider, 2001; Schneider et al., 2003). The suburbs to the North, East, and West are built into the mountain slopes and are therefore situated at higher altitudes than the city center, which is located in the stream’s channel. These outskirts are characterized by a lower population density, and contain more semi-natural environments, such as house gardens and areas with typical ruderal components, as well as natural sites. Besides building areas, Mérida includes also sectors of used and abandoned agricultural land and forest fragments. All identified land cover classes are listed in Table 1. 2.2. Sampling design and botanical collections In order to assess the species diversity of bromeliads across the different urban habitat types, species occurrences were recorded within 79 representative plots (Fig. 1). These study plots were selected using an ASTER satellite image combined with a roadmap, and are equally distributed among the various built-up classes, vegetation types, and altitudinal zones of the city. Only study plots with at least one species occurrence were taken into consideration, in accordance with the requirements for the HSM technique (see below). Inaccessibility of the terrain restricted the number of plots between 1000 and 1200 m, and between 2000 and 2200 m. C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 109 Table 1 System of grouped land cover classes to urban zones of Mérida city according to the degree of soil sealing. Zone Soil sealing (%) Relative area in Mérida city (%) Land cover classes 1 2 75–100 50–74.9 22 36 3 4 25–49.9 5–24.9 6 17 5 0–4.9 19 Core area, city center; streets (paved); bare, highly impervious compacted soils; corrugated iron sheets; roofing tiles Suburbans with small gardens and groves; sport grounds; industrial used areas; intensively used agricultural fields; abandoned agricultural land Public parks; small plots of ruderal plant communities Open evergreen to semi-deciduous broad-leaved forests fragments; forest-bush transitions; plantations; water courses Cloud forests fragments; dense evergreen to semi-deciduous broad-leaved forests fragments; water courses root mean square equation (RMSE) (Eq. (1)), resulting in a RMSE of 10.18 m. RMSE = Fig. 1. Distribution of sample sites (triangles) within the study area (polygon). Additionally, the extremely steep flanks of the Río Chama and Albarregas valleys were not included in the survey. Each plot was 50 m2 in size (10 m × 5 m) and its geographic position and elevation were measured with a handhold GPS receiver (etrex Summit, Garmin, USA). Representative botanical specimens were collected for each species and deposited at the herbarium of the Universidad de Los Andes, Mérida, Venezuela (MERC). Epiphytes of the upper tree strata were collected using telescope pruning scissors or identified by means of binoculars. Species identification relied predominantly on Davidse, Sousa, and Chater (1994), Hornung (1998), Smith (1971), and Smith and Downs (1977), or on comparison with well-determined herbarium specimens of the herbarium MERC. 2.3. Remote sensing data and urban habitat classification An Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite image consisting of a L1A ASTER scene was used to classify the urban area of Mérida. It was collected by the ASTER/Terra sensor on February 1, 2004 and only 1% cloud cover was present at the time. The spatial resolution of the ASTER image differs between the three spectral groups of VNIR, SWIR and TIR. In this study we considered the first nine bands of the image, the VNIR spectral region (bands 1–3), which is characterized by a spatial resolution of 15 m, and the SWIR (bands 4–9) wavelength region with a spatial resolution of 30 m. The spectral region TIR with a spatial resolution of 90 m was not included. The image was obtained in a HDF EOS format with image center coordinates of 71◦ 11′ 06′′ W and 8◦ 39′ 49′′ N. Georeferencing was carried out with the image processing software TNT mips 6.5 (Microimages, Lincoln, USA) by using an affine georeferencing model. Therefore, a total of 7 Ground Control Points (GCP) were collected with the help of a handhold GPS receiver. We assessed the relative vertical spatial accuracy with the  n d2 i=1 i n (1) di2 = Zest − Zobs · Zest is the DEM value, Zobs the field-measured elevation value, n the number of collected GCPs. We performed a supervised maximum likelihood classification of the study region, using a soft classifier (De Smith, Goodchild, & Longley, 2009). The classification refers to 85 training sites based on field survey. The vegetation-impervious surface-soil (VI-S) model was taken into consideration to develop the training site assortment, as is recommended during classification of urban environments. The basic idea of the V-I-S model is to simplify the heterogeneous urban area and divide it into three main components: vegetation, impervious surface, and soil (Hung, 2002). Due to the fact that homogenous training sites generally improve the land cover classification, our training sites were homogenous whenever possible, leading to 19 land cover classes. The gradual transition between land use and land cover within urban areas has led to the consolidation of the two under the heading land cover. The qualitative evaluation of the land cover classification’s performance was carried out by a visual assessment of an aerial photograph obtained by the mission ISS005 of the International Space Station in October 2002 and an up-to-date roadmap. Additionally, GCPs were taken during two field excursions in May and two in September 2007. The 19 thematic land cover classes were grouped into five major categories (zones) based on the percentage of soil sealing using the software ArcGIS 10.0 (ESRI, Redlands, USA) (Table 1). Soil sealing is defined as soils that have been made impervious to water by human construction activities. The highest degree of soil sealing is found in the core area of the city where asphalted streets, buildings and further highly impervious compacted soils predominate. The degree of soil sealing is strongly linked to the observed land cover, and thus reflects the potential occurrence of trees that provide the main substrate for epiphytic bromeliads. Furthermore, the resolution of our satellite image did not allow for a clear differentiation between the vegetation-covered riverbanks and the water bodies during the process of supervised classification. For this reason water bodies were not classified and riverbanks were partly assigned to zones 4 and 5 considering the respective vegetation. To evaluate the quality of the supervised classification results of the five major zones a quantitative accuracy assessment was conducted in GRASS GIS, version 6.4.0 (GRASS Development Team, 2010). Two hundred random control points were allocated on the aerial photograph of 2002. Each control point was assigned a reference value. The reference values were compared to the classification value of the respective pixel. Producer’s, user’s, overall accuracy, and the kappa coefficient (see Congalton, 1991; Foody, 2002) were calculated in Microsoft Excel (2007). 110 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 2.4. Habitat suitability modelling (HSM) In this study, we follow the terminology of Bradley et al. (2012) and use the term HSM to describe models of potential species distribution based on environmental correlates. Predictive models of a species’ habitat were created using Maxent 3.3.3 (Phillips, Anderson, & Schapire, 2006; Phillips, Dudík, & Schapire, 2004), which has proven powerful for this purpose (Buermann et al., 2008; Elith et al., 2011; Lentz, Bye, & Sánchez-Cordero, 2008; PratesClark, Saatchi, & Agosti, 2008). Maxent utilizes presence-only point occurrences with high performance even on small datasets (Elith et al., 2006; Evangelista et al., 2008; Loiselle et al., 2008; Pearson, Raxworthy, Nakamura, & Peterson, 2007; Phillips et al., 2004; Wisz et al., 2008). As predictor variables we used the VNIR and the SWIR range of the ASTER image. Furthermore, we included a DEM taken by the Shuttle Radar Topography Mission (SRTM) with a resolution of 90 m. The DEM was obtained from the U.S. Geological Survey (USGS) and preprocessed using spline interpolation and a low-pass filter in GRASS GIS, version 6.4.0 (GRASS Development Team, 2010) to remove gaps and random errors (Li, Zhu, & Gold, 2005; Neteler & Mitasova, 2002). Thereafter, the DEM and the bands 1–3 of the ASTER image were resampled to 30 m resolution to obtain an equal spatial accuracy of the layers. The derivates slope and aspect were calculated using the DEM in ArcGIS 9.2 (ESRI, Redlands, USA). The number of point occurrences is a limiting factor for HSMs. Studies by Hernandez, Graham, Master, and Albert (2006), Pearson et al. (2007), and Wisz et al. (2008) unequivocally found that Maxent can still build meaningful models at a minimum of 5–10 occurrences. Therefore, we choose the mean of this range as a critical value and included only species with a minimum of eight occurrences in the analysis, thus excluding Guzmania mitis, Tillandsia longifolia, Tillandsia pruinosa, Tillandsia schiedeana, and Tillandsia schultzei. Species occurrence data usually are split into training and test data to provide statistical analyses of the HSMs. The available data set for the present study was limited in size. Consequently, training and test data sets would have become too small by splitting for an appropriate statistic testing (Pearson et al., 2007). As recommended by Phillips (2006), when working with small data sets, we ran our models with an 8-fold cross-validation and a testpercentage of 10% during each run. We evaluated the accuracy of the models, considering the average of the Test-AUC value (area under the curve) calculated for each single species, a commonly used but still controversial measure (Lobo, Jiménez-Valverde, & Real, 2008). We therefore tested the significance of the obtained AUC values of the species’ habitat suitability models using the nullmodel approach described in Raes and ter Steege (2007). To build the null distribution we ran 1000 random models with randomly drawn sample localities equal to the number of observations for each species and the same environmental predictors as described in the Maxent procedure. We used a 95% confidence interval to test if the HSM performs significantly better than expected by chance. For the further statistic analyses, bromeliad species were classified either as CAM or C3 species according to data from Martin (1994) and Crayn, Winter, and Smith (2004). Species were also classified according to their ability to impound water with their densely arranged rosette leaves, thus, species with tank habit were compared to species without or with a weakly developed tank habit. Our classification is based on morphological observations. Pearson correlations and a two-tailed t-test were used to test for trends of the point occurrence probabilities for each species and for CAM versus C3 species across urban soil sealing zones. For calculating the correlation coefficients we used 10,000 point occurrence probabilities randomly sampled from the total probability distribution per zone and species. Additionally, differences between mean occurrence probabilities (of the total point occurrence probabilities) per zone of CAM versus C3 species were tested using a two-tailed t-test. All statistical tests were run in R, version 2.15.1 (R Core Team, 2012). 3. Results 3.1. Identification, evaluation and spatial distribution of urban zones based on land cover classes The supervised satellite image classification of Mérida city resulted in a thematic map of 19 land cover classes. Grouping of these classes into five major urban zones based on soil sealing shows that the first two zones (50–100% sealed soils) make up ∼60% of the total urban area (Table 1 and Fig. 2). These highly sealed surfaces include the core area of Mérida (ca. 12%), housing developments with small gardens and groves, areas of intensive agricultural or industrial use, and abandoned agricultural land. Approximately 6% are covered by urban spaces, as well as small areas of ruderal plant communities, both being categorized as Zone 3. A total of about 10% are watercourses which correspond to zones 4 and 5. Zone 4 is characterized by relatively low soil-sealing (5–24.9%). It is covered by open evergreen to semi-deciduous broad-leaved forest fragments, plantations, and forest-bush transitions. About 19% of the total area is covered by closely forested areas, of which around 4% are cloud forest fragments (Zone 5). The accuracy assessment of the classification results of the five major urban zones in Mérida city is summarized in Table 2. The overall classification accuracy is 74.11% and the kappa coefficient is 0.45. 2.5. Modelled species richness and statistical analyses Our species richness maps were obtained by aggregating the HSMs of the single species for which the Test-AUC value of the HSM significantly differed from random, excluding the three species Tillandsia balbisiana, Tillandsia biflora, and Tillandsia myriantha. A Visual Basic routine (described in Schmidt, 2006) counted for each grid cell species with a probability of occurrence above the ‘maximum training sensitivity plus specificity threshold’ as recommended by Liu, Berry, Dawson, and Pearson (2005). In order to evaluate the similarity between the HSMs, a simple Pearson correlation analysis was carried out (Table 4). For the calculation of the correlation coefficients we used each pixel-value of the HSMs. Fig. 2. Map of land cover classes within Mérida city based on the supervised satellite image classification. C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 111 Table 2 Accuracy assessment of the five major urban zones of Mérida city. Classification data Data of geo-referenced aerial photograph Zone 1 Zone 2 Zone 4 Zone 5 Zone 1 Zone 2 Zone 3 Zone 4 Zone 5 23 1 6 2 0 5 46 0 1 3 3 2 6 3 1 2 9 0 32 3 0 4 0 6 39 33 62 12 44 46 Column total 32 55 15 46 49 197 User’s accuracy (%) Producer’s accuracy (%) 69.7 71.8 74.2 83.6 50 40 72.7 69.6 84.8 79.6 Overall accuracy (%) Kappa coefficient (k) Chance agreement Zone 3 Row total 69.7 74.1 0.45 0.23 3.2. Bromeliad diversity and species occurrences A total of 20 bromeliad species were recorded in the study area (Table 3), all belonging to the subfamily Tillandsioideae, and usually growing epiphytic (occasionally also saxicolous or terrestrial). Tillandsia (14 species) was the most diverse genus, followed by Racinaea (3), Guzmania (2), and Catopsis (1). Of these, eleven species are C3 plants, eight are CAM plants, and for one species (T. myriantha) data is ambiguous. Ten species exhibit a tank habit (impounding water), four possess a poorly developed tank habit, while six species are classified as lacking a tank habit (Table 3). The most frequent species was Tillandsia recurvata, which was found in 57 plots, whereas T. pruinosa was recorded in only two plots. Species richness was higher at elevations above 1400 m, with a maximum of 12 species per plot at Loma de la Virgen site (1869 m), a grazing area with many shade trees. Four plots contained only a single species (T. recurvata). The most widespread species are Catopsis nutans, Guzmania monostachia, Racinaea tenuispica, Tillandsia fendleri, Tillandsia juncea, T. recurvata, Tillandsia usneoides, and Tillandsia variabilis, occurring in all five urban zones. Racinaea spiculosa, T. biflora, Tillandsia complanata, T. myriantha, and T. schiedeana were reported from zones 2 through 5. The other species show an unequal distribution, especially G. mitis, which is confined to the two leastsealed urban zones. G. mitis, Racinaea tetrantha, T. balbisiana, T. longifolia, T. myriantha, T. pruinosa, T. schiedeana, and T. schultzei were comparatively rare with less than ten point occurrences. C. nutans was recorded above 1200 m and mostly in the urban center, less frequently in the semi-natural areas, and was absent from the northern cloud forest area. G. monostachia was recorded in areas above 1300 m and was widespread in central and marginal areas. It was especially frequent in gardens and public parks. R. tenuispica was among the most widespread species of the city, occurring in the center, urban pastureland, and forests. It was only absent from the north-western cloud forest and some highly sealed areas. T. usneoides and T. variabilis showed a similar distribution and were reported from areas above 1200 m from pastureland, parks, and highly sealed areas, but not from the north-western cloud forest. T. recurvata was observed most frequently in the highly sealed areas and was less frequent in pastureland and forest areas at the periphery. It was absent from the north-western cloud forest. T. balbisiana was a rare species that occurred only below 1600 m in green areas and parks, partially also in highly sealed areas. Tillandsia fasciculata was documented in semi-natural and highly sealed Table 3 List of Bromeliaceae found in the study site and their ecological and ecophysiological characteristics. Distribution and elevation according to Smith and Downs (1977). Photosynthetic pathway according to Martin (1994)a or Crayn et al. (2004)b . Tank habit refers to the rosulate, basally tight leaves that allow the plants to accumulate water and create phytotelmata. Some species (marked by parentheses) have more or less loosely arranged or narrow leaves and may impound only small volumes of water. Species names according to Luther (2010). Species Distribution Elevation range (m a.s.l.) Photosynthetic pathway Tank habit Catopsis nutans (Sw.) Griseb. Guzmania mitis L.B. Sm. Guzmania monostachia (L.) Rusby ex Mez in A. DC. & C. DC. Racinaea spiculosa (Griseb.) M.A.Spencer & L.B.Sm. Racinaea tenuispica (André) M.A.Spencer & L.B.Sm. Racinaea tetrantha (Ruiz & Pav.) M.A. Spencer & L.B. Sm. Tillandsia balbisiana Schult. f., in Roem. & Schult. Tillandsia biflora Ruiz & Pav. Tillandsia complanata Benth. Tillandsia fasciculata Sw. Tillandsia fendleri Griseb. Tillandsia juncea (Ruiz & Pav.) Poir. Tillandsia longifolia Baker Tillandsia myriantha Baker Tillandsia pruinosa Sw. Tillandsia recurvata (L.) L. Tillandsia schiedeana Steud. Tillandsia schultzei Harms Tillandsia usneoides (L.) L. Tillandsia variabilis Schltdl. Mexico to Ecuador Costa Rica to Ecuador USA to Bolivia and Brazil Costa Rica to Bolivia and Brazil Venezuela and Greater Antilles to Peru Venezuela to Bolivia USA to Colombia Costa Rica to Bolivia Costa Rica to Bolivia USA to Colombia Venezuela and Greater Antilles to Bolivia USA to Bolivia Costa Rica to Peru Venezuela and Colombia Mexico to Ecuador and Brazil USA to Argentina Mexico to Colombia and Venezuela Venezuela and Colombia USA to Chile Mexico to Venezuela 0–1700 1600–3300 900–1500 300–2800 300–2100 750–3300 30–1500 1300–3000 800–3500 50–1600 400–2900 150–1900 750–2800 1300–2200 0–1300 0–3000 50–1800 1200–2800 30–2400 100–2200 C3 ab C3 ab C3 -[CAM]a C3 b C3 ab C3 ab C3 ab CAMab [CAMa ] C3 b C3 ab CAMab C3 ab CAMab C3 ab CAMa C3 b CAMb CAMab CAMab C3 b CAMa CAMb Yes Yes Yes Yes Yes Yes No Yes Yes (Yes) Yes No (Yes) (Yes) No No No Yes No (Yes) 112 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 Fig. 3. (a) Occurrences (dots) and modelled distributions of Bromeliaceae species in the city of Mérida (polygon). (b) Modelled species diversity. C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 1 1 0.97 1 0.93 0.90 T. recurvata T. usneoides T. variabilis areas below 1600 m. T. juncea was found above 1200 m and was widespread across a variety of urban habitats ranging from green areas, parks and pastureland to forests. T. fendleri was recorded at areas above 1500 m and occurred especially in green areas, parks, pastureland and forests. T. biflora grew above 1750 m in forests and pastureland, but was absent from the central urban area. A similar distribution was observed for T. complanata. T. longifolia was a rare species occurring only in the north-western cloud forest above 2100 m. R. spiculosa occurred above 1700 m in forests, pastureland and cropland, but was absent from highly sealed areas of the city center. T. myriantha was observed above 1700 m in pastureland, gardens and groves. 113 1 0.85 0.94 0.93 1 0.58 0.45 0.59 0.64 1 0 0.62 0.77 0.66 0.60 1 −0.06 0.49 0.16 −0.02 0.08 0.12 1 0.88 −0.04 0.36 0.11 −0.05 0.02 0.06 1 0.01 0.07 0.59 0.64 0.92 0.92 0.97 0.97 1 0.11 0.73 0.93 0.01 0.47 0.22 0.01 0.14 0.16 1 0.14 0.87 0.08 0.12 0.53 0.59 0.82 0.80 0.87 0.92 T. fendleri T. fasciculata T. complanata R. tetrantha R. tenuispica R. spiculosa G. monostachia C. nutans 1 0.84 0.06 0.97 −0.02 0.03 0.61 0.63 0.90 0.94 0.97 0.96 C. nutans G. monostachia R. spiculosa R. tenuispica R. tetrantha T. complanata T. fasciculata T. fendleri T. juncea T. recurvata T. usneoides T. variabilis Table 4 Values of the Pearson correlation coefficients of the modelled habitat suitability maps. Species HSMs were generated for 15 species, with 12 of the modelled distribution patterns differing significantly from a random one (Appendix A). C. nutans, G. monostachia, R. tenuispica, T. juncea, T. recurvata, T. usneoides, and T. variabilis show similar distribution ranges according to the HSMs (Fig. 3a and Table 4), each species prevailing in the city center and, thus, in the area of maximum anthropogenic modification of the landscape. A contrasting pattern is observed in R. spiculosa, R. tetrantha, and T. complanata, which show higher occurrence probabilities in areas along the two rivers and outside the urban sector. It is also remarkable that the percental contribution of the environmental variable slope is relatively high (40.3%, 50.4%, and 49.3%) in these models. Thus, they predominantly occur in shaded, forested depressions along the western mountainside (>50% occurrence probability). However, R. tetrantha shows generally the lowest probability to occur whereas T. myriantha shows the highest. T. fendleri was predicted to be absent from the southern part of Mérida, while the north-eastern part of Mérida’s center shows the highest cumulative probability (>60%). Also the program Maxent considered the contribution of the variables DEM (25%) and slope (42.6%) relatively high in this case. In contrast, T. fasciculata has the highest predicted distribution for the south-western part of the study area (>60% probability). The percent contribution of the environmental variable DEM to the model of T. fasciculata was highest (59%). According to the modelled habitat suitability, the species show different probabilities for their occurrence in the five urban zones (Fig. 4). There is a highly significant trend (p < 0.001) to occur in areas with higher soil sealing for C. nutans (r = 0.25, t = 58.3), G. monostachia (r = 0.13, t = 29.5), R. tenuispica (r = 0.13, t = 30.3), T. fasciculata (r = 0.08, t = 17.6), T. fendleri (r = 0.04, t = 9.9), T. recurvata (r = 0.25, t = 58.9), T. usneoides (r = 0.11, t = 23.8), and T. variabilis (r = 0.12, t = 27.3). An inverse trend significant at p < 0.001 is observed for R. spiculosa (r = −0.38, t = −91.5), R. tetrantha (r = −0.29, t = −67.9), and T. complanata (r = −0.35, t = −85.0). The distributions of CAM versus C3 species were different (Fig. 5), but a trend toward higher occurrence probabilities in zones with higher soil sealing is only significant for CAM species (r = 0.47, t = 2.582, df = 23, p < 0.017; for C3 species: r = −0.064, t = −0.3671, df = 33, p = 0.716). Because the species with a well-developed tank habit correspond exactly with those with a C3 physiology, they do not show any significant trend across the land cover zones. In contrast, the non- or poorly impounding species have higher occurrence probabilities in zones with higher soil sealing. Our modelled map of species richness predicts the highest species diversity in the core area of the city. A relatively high probability of species diversity can be also observed along the two watercourses, the Río Chama and Río Albarregas. The lowest species diversity is displayed in the more natural to semi-natural habitats. T. juncea 3.3. Modelled habitat suitability and species richness 114 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 Fig. 4. Boxplots of point occurrence probabilities per urban zone (means of the percentage of soil sealing) for each species. Only species occurring ≥8 times per study plot were considered. Analysis based on 10,000 randomly sampled probabilities per zone. Fig. 5. Scatterplots of mean occurrence probabilities for each species per urban zone (means of the percentage of soil sealing). (left) CAM species and (right) C3 species. C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 Among the latter, the highest diversity has been modelled for the humid mountain depressions. 4. Discussion and conclusions 4.1. Urban diversity of Bromeliaceae and the role of CAM versus C3 species CAM plants as compared to C3 plants are generally more drought-tolerant due to their peculiar physiology. Drought tolerance, in turn, is favourable for successfully colonizing the generally drier urban habitats. Therefore, it is not surprising that the CAM species among the bromeliads investigated here show higher occurrence probabilities in urban zones with high soil sealing. Comparing our data with two floristic inventories of successional and mature upper montane rainforests between 2300 and 3300 m a.s.l. in the immediate vicinity of Mérida city (La Montaña site, 1.5 ha, Kelly et al., 1994; La Caña site, ∼1.2 ha, Schneider, 2001) reveals that all their species lacking in our study area but known to occur also at altitudes covered by Mérida city are C3 plants (Mezobromelia capituligera [=Tillandsia capituligera], Tillandsia compacta, Tillandsia denudata, Tillandsia tovarensis, Vriesea robusta, Vriesea tequendamae, Werauhia cowellii [=Vriesea cowellii]). Thus, CAM bromeliads are not only well-adapted to live in tropical urban habitats, but are also suitable indicators for the degree of urbanization. The absence of the C3 species of the La Montaña and La Caña sites mentioned above from our study area may have various reasons. They might have been unable to migrate into the city, gone extinct therein, or the number of study plots may have been insufficient to discover all bromeliad species actually present. Especially critical for the successful colonization of isolated urban phorophytes by epiphytes is the distance that the diaspores have to travel from the source populations. Although the bromeliads observed in this study all belong to the pogonochorous (plumed, wind-dispersed) subfamily Tillandsioideae, isolation by distance is certainly an important reason for a species-poor epiphytic flora (Hietz-Seifert, Hietz, & Guevara, 1996; Wolf, 2005). Local extinction may also play a critical role as it has been well-established that already after a decade of isolation the epiphytic flora of phorophytes (i.e., the plants carrying the epiphytes) substantially impoverishes regarding species richness, mainly due to the altered microclimatic conditions on isolated trees (Nöske et al., 2008; Werner & Gradstein, 2008; Werner, Homeier, & Gradstein, 2005). As isolation of phorophytes or alternative substrates and the therewith associated altered microclimate is a longstanding condition in most of the urban habitats, these factors are supposed to be the main triggers of species decline. On the other hand, the city’s flora is still comparatively rich, considering that it contains 36% of the total bromeliad diversity of the state of Mérida (56 spp., see Hornung, 1998; Hornung & Gaviria, 1999) which covers an area of 11,300 km2 . These numbers illustrate the resilience of many bromeliad species to the ecological changes caused by urbanization. All bromeliad species encountered are epiphytes, which in turn depend on their phorophytes. Although the microclimate differs between the urban and natural non-urban sites, the epiphytic substrate is similar throughout. The tiny T. recurvata is probably the only species with the ability to colonize new habitats in urban areas, since it often grows on telephone wires or similar substrates. Therefore, resilience is also due to competitive advantage in the epiphytic niche over other urban angiosperm species. The relatively high modelled species diversity along the watercourses (Fig. 3b) may be due to the fact that these areas are mainly covered by old phorophytes, are relatively humid, and are less influenced by human impact. To explain the relatively high diversity of bromeliads in the core area of Mérida city two factors are 115 important. First, the upper montane rainforests above 2300 m in the vicinity of Mérida do not harbour any CAM species (Kelly et al., 1994; Schneider, 2001), which means that this species group apparently has a lower potential for colonizing the more natural areas in the marginal sectors of the city. Second, the C3 species colonize successfully the natural areas, but also occur in the core areas with higher soil sealing. Thus, the core area is habitat to both groups, whereas the marginal sectors are less favourable for CAM species. 4.2. The utility of remote sensing data for habitat suitability modelling in urban areas Generally our HSMs were in line with field experience, and high AUC values indicate a good model fit. Only for three species with comparatively few (T. balbisiana, T. biflora) or clustered (T. myriantha) occurrence points, test AUCs did not significantly differ from those of the null-models. Previous studies showed that plant distributions in cities are strongly influenced by human impact and that each component of the urban environment, that is, the different types of built-up areas, has its own specific set of species (Godefroid & Koedam, 2007; McKinney, 2006). Such a pattern was also confirmed by our data on bromeliads with C3 and CAM species showing different success in colonizing the urban zones. The resolution of environmental variables is of particular importance for HSMs. Especially in heterogeneous fragmented landscapes, species distributions cannot be reflected by coarse scale environmental data like interpolated temperature or precipitation maps (König, Schmidt, & Müller, 2009). In contrast, high resolution satellite images can reflect the heterogeneity of surfaces on a very fine scale, resulting in an excellent environmental predictor in urban environmental studies. Several studies (e.g. Buermann et al., 2008; Cord & Rödder, 2011; Zimmermann et al., 2007) have already shown the predictive power of remote sensing data when incorporated into HSM approaches. Cord and Rödder (2011) found that for most species HSMs on the basis of remote sensing data combined with climate data yield the best results. However, the resolution of interpolated climate data for Mérida city is not high enough to be applicable for our study. Additionally, inferences concerning the improvement in model success for remote sensing data are limited by the relatively large contribution of the elevation and the slope to most of the HSMs. One major drawback concerning the applicability of remote sensing data in urban areas is the phenomenon of mixed pixels (Wang, 1990). The urban ground heterogeneity results in pixels that correspond to more than one land cover type and it is thus difficult to assign them unambiguously. Nonetheless, in our study the use of a maximum likelihood classification with a soft classifier in combination with a homogenous selection of training sites led to a satisfying result regarding the scope of our investigation. The accuracy assessment of the supervised classification of the five major urban zones shows a fair overall accuracy of 74.1% (Congalton, 1991; Table 2). Misclassification results from the complex spectral signatures of urban landscapes with mixed pixels and the small fractions of zone types within the city. However, when comparing k and the chance agreement the classification accuracy is 45% better than it would have been expected only by chance. In general the calculated producer’s and user’s accuracies are sufficiently high to distinguish between the five urban zones (Table 2). Only Zone 3 has a relatively low producer’s and user’s accuracy. This can be explained by the high fragmentation of this zone, which is composed of public parks and small plots of ruderal plant communities (Table 1). It is the smallest zone of all and covers only 6% of the study area. High fragmentation of urban environments often results in misclassified pixel-values, as a consequence of the C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 116 insufficient image resolution and the combination of the urban landscapes element size in relation to the sensor cell size. This leads to mixed pixels, which are a well-known phenomenon in urban landscapes (Lillesand & Kiefer, 2008; Lu & Weng, 2004). The classification accuracy can be improved by using higher resolution remote sensing data and advanced classification technics (object oriented classification). This, however, is beyond the scope of this study. 4.3. Conclusions The uninterrupted trend of rural-to-urban migration in tropical countries will result in a higher pressure on urban biodiversity, rendering effective conservation strategies for urban species of vital importance. Here we have shown that the incorporation of remote sensing data into habitat suitability modelling is a powerful, costeffective method to visualize fine-scale biodiversity patterns in complex tropical urban habitats. Understanding the potential distribution of the species, in turn, is essential for identifying more precisely the specific needs for maintaining diversity in urban areas, thus enhancing conservation planning. Bromeliaceae is a plant group resilient to urbanization due to their multiple morphological and physiological adaptations. It is thus a group suitable to work with for studies of urban environments related to biodiversity patterns. Especially the relative contribution of C3 versus CAM species to the total richness of bromeliad species may prove a valuable indicator, with the extreme predominance of the latter indicating not only a trend toward aridification and increased soil sealing of the local landscape but also a potential loss of the total diversity (due to the joint loss of more mesic species). However, this needs to be tested more thoroughly in future studies. Since bromeliads provide secondary aquatic habitats for insects or amphibians in the water-impounding tank species (phytotelmata), they contribute to the overall urban biodiversity. On the other hand, they participate in the spreading of insect-borne epidemic diseases such as dengue fever (Frank, 1983; Lounibos, O’Meara, Nishimura, & Escher, 2003). Thus, a better knowledge of their potential distribution in urban areas will be helpful to monitor and control such diseases. Acknowledgements We like to thank M. Alberto, L. Anzola for help during fieldwork in Venezuela, K. König and G. Bocksberger for helpful advices regarding the modelling approach and statistical analyses, S. Connery and R. Byer for language advices, and three anonymous reviewers for fruitful comments on an earlier version of the manuscript. The first author is also grateful to the German Academic Exchange Service (DAAD) for financial support. We also acknowledge funding by the Biodiversity and Climate Research Center (BiK-F), Frankfurt. Appendix A. Test-AUC values of the HSMs and the null model. Species for which the Test-AUC value of the HSM does not differ significantly from random are highlighted by an asterisk. Additionally, the percent contributions of the environmental variables to the Maxent models are list, when they a higher than 10%. Species Number of species occurrences Average of the test AUC AUC of the null-model Band 2 Catopsis nutans Gutzmania monostachia Racinaea spiculosa Racinaea tenuispica Racinaea tetrantha Tillandsia balbisiana Tillandsia biflora Tillandsia complanata Tillandsia fasciculata Tillandsia fendleri Tillandsia juncea Tillandsia myriantha Tillandsia recurvata Tillandsia usneoides Tillandsia variabilis 46 24 14 55 11 8 15 11 17 40 34 8 57 54 50 0.83 0.87 0.79 0.82 0.95 0.77* 0.8* 0.87 0.92 0.83 0.84 0.70* 0.82 0.81 0.85 0.74 0.8 0.77 0.73 0.79 0.80 0.85 0.79 0.83 0.76 0.77 0.80 0.72 0.73 0.73 – – 32.5 – 20.9 – 13.4 17.8 14.4 – 14.9 21.6 – – 10 Band 3 – – 14.4 – – – – – – – – 11.3 – – – Band 5 38.8 12.7 – 33.7 – – – – – 14.9 26 – 29.6 35 26.9 Band 6 – – – – – 22 – – – – – – – – – DEM Slope – 27.1 – 12.6 21.2 54 16 23 59 25.6 – – 14 11.5 11.3 33.7 39.2 40.3 30.1 50.4 11.7 60 49.3 15.7 42.6 38.8 57.1 38.6 33.4 39.9 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 References Antrop, M. (2000). Changing patterns in the urbanized countryside of Western Europe. Landscape Ecology, 15(3), 257–270. http://dx.doi.org/10. 1023/A:1008151109252 Benzing, D. H. (1990). Vascular epiphytes: General biology and related biota. Cambridge: Cambridge University Press. Bradley, B. A., Olsson, A. D., Wang, O., Dickson, B. G., Pelech, L., Sesnie, S. E., et al. (2012). Species detection vs. habitat suitability: Are we biasing habitat suitability models with remotely sensing data? Ecological Modelling, 244, 57–64. http://dx.doi.org/10.1016/j.ecolmodel.2012.06.019 Buermann, W., Saatchi, S., Smith, T. B., Zutta, B. R., Chaves, J. A., Milá, B., et al. (2008). Predicting species distributions across the Amazonian and Andean regions using remote sensing data. Journal of Biogeography, 35(7), 116–1176. http://dx.doi.org/10.1111/j.1365-2699.2007.01858.x Cohen, B. (2006). Urbanization in developing countries: Current trends, future projections, and key challenges for sustainability. Technology in Society, 28(1–2), 63–80. http://dx.doi.org/10.1016/j.techsoc.2005.10.005 Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37(1), 35–46. http://dx.doi.org/10.1016/0034-4257(91)90048-B Cord, A., & Rödder, D. (2011). Inclusion of habitat availability in species distribution models through multi-temporal remote-sensing data? Ecological Applications, 21(8), 3285–3298. http://dx.doi.org/10.1890/11-0114.1 Crayn, D. M., Winter, K., & Smith, J. A. C. (2004). Multiple origins of crassulacean acid metabolism and the epiphytic habitat in the Neotropical family Bromeliaceae. Proceedings of the National Academy of Sciences of the United States of America, 101(10), 3703–3708. http://dx.doi.org/10.1073/pnas.0400366101 Davidse, G., Sousa, M., & Chater, A. O. (Eds.). (1994). Flora Mesoamericana (flora of Mesoamerica), Vol. 6, Alismataceae a Cyperaceae. St. Louis: Missouri Botanical Garden. De Smith, M. J., Goodchild, M. F., & Longley, P. A. (2009). Geospatial analysis: A comprehensive guide to principles, techniques and software tools (3rd ed.). Leicester: The Winchelsea Press. Elith, J., Graham, C. H., Anderson, R. P., Dudík, M., Ferrier, S., Guisan, A., et al. (2006). Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29(2), 129–151. http://dx.doi.org/10.1111/j. 2006.0906-7590.04596.x Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee En, Y., & Yates, C. J. (2011). A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17, 1–15. Engwald, S. (1999). Diversität und Ökologie der Epiphyten eines Berg- und eines Tieflandregenwaldes in Venezuela (diversity and ecology of the epiphytes of a mountain and a lowland rainforest in Venezuela). Norderstedt: Universität Bonn, Libri-BoD [Dissertation]. Evangelista, P. H., Kumar, S., Stohlgren, T. J., Jarnevich, C. S., Crall, A. W., Normann, J. B., III, et al. (2008). Modelling invasion for a habitat generalist and a specialist plant species. Diversity and Distributions, 14(5), 808–817. Foghin-Pillin, S. (Ed.). (2002). Tiempo y clima en Venezuela: Aproximación a una geografía climática del territorio venezolano (weather and climate of Venezuela: An aproximation of a climatic geography of a Venezuelian territory). Caracas: Instituto Pedagógico José Manuel Siso Martínez. Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote Sensing of Environment, 80(1), 185–201. http://dx.doi.org/10. 1016/S0034-4257(01)00295-4 Franceschi, E. A. (1996). The ruderal vegetation of Rosario City, Argentina. Landscape and Urban Planning, 34(1), 11–18. http://dx.doi.org/10.1016/ 0169-2046(95)00203-0 Frank, J. H. (1983). Bromeliad phytotelmata and their biota, especially mosquitoes. In J. H. Frank, & L. P. Lounibos (Eds.), Phytotelmata: Terrestrial plants as hosts for aquatic insect communities (pp. 101–128). Medford: Plexus Publishing, Inc. Franklin, J., Davis, F. W., Ikegami, M., Syphard, A. D., Flint, L. E., Flint, A. L., et al. (2013). Modelling plant species distributions under future climates: How fine scale do climate projections need to be? Global Change Biology, 19(2), 473–483. http://dx.doi.org/10.1111/gcb.12051 Gentry, A. H., & Dodson, C. H. (1987). Contribution of nontrees to species richness of a tropical rain forest. Biotropica, 19(2), 149–156. Godefroid, S., & Koedam, N. (2007). Urban plant species patterns are highly driven by density and function of built-up areas. Landscape Ecology, 22(8), 1227–1239. http://dx.doi.org/10.1007/s10980-007-9102-x GRASS Development Team. (2010). Geographic Resources Analysis Support System (GRASS) Software, Version 6.4. Open Source Geospatial Foundation. http://grass.osgeo.org Guisan, A., & Zimmermann, N. E. (2000). Predictive habitat distribution models in ecology. Ecological Modelling, 135(2–3), 147–186. http://dx.doi.org/10. 1016/S0304-3800(00)00354-9 Hernandez, P. A., Graham, C. H., Master, L. L., & Albert, D. L. (2006). The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography, 29(5), 773–785. http://dx.doi.org/10.1111/j.0906-7590.2006.04700.x Hietz-Seifert, U., Hietz, P., & Guevara, S. (1996). Epiphyte vegetation and diversity on remnant trees after forest clearance in southern Veracruz, Mexico. Biological Conservation, 75(2), 103–111. http://dx.doi.org/10. 1016/0006-3207(95)00071-2 Hijmans, R. J., & Graham, C. H. (2006). The ability of climate envelope models to predict the effect of climate change on species distributions. Global Change Biology, 12(12), 2272–2281. http://dx.doi.org/10.1111/j.1365-2486.2006.01256.x 117 Hornung, C. (1998). Flora de las Bromeliáceas del Estado Mérida (Flora of the Bromeliaceae of the state Mérida). Venezuela: Universidad de los Andes, Mérida [Thesis]. Hornung, C., & Gaviria, J. (1999). Novedades para la flora del estado Mérida: 1. Nuevos registros de Bromeliaceae (news about the flora of the state of Mérida: 1. New records of Bromeliaceae). PlantULA, 2(1–2), 87–101. Hung, M. (2002). Urban land cover analysis from satellite images. In Proceedings of Pecora 15-Land Satellite Information IV/ISPRS Commission I/FIEOS 2002 Conference 10–15 November, Denver, Colorado. Kappelle, M. (1996). Los Bosques de Roble (Quercus) de la Cordillera de Talamanca, Costa Rica: Biodiversidad, Ecología, Conservación y Desarrollo (The Oak (Quercus) forests of the Talamancan Mountain Range. Costa Rica: Biodiversity, conservation and development). Universiteit Amsterdam (UvA) & Instituto Nacional de Biodiversidad (INBio), Amsterdam. Kelly, D. L., Tanner, E. V. J., Lughadha, E. M. N., & Kapos, V. (1994). Floristics and biogeography of a rain forest in the Venezuelan Andes. Journal of Biogeography, 21(4), 421–440. König, K., Schmidt, M., & Müller, J. V. (2009). Modelling species distributions with high resolution remote sensing data to delineate patterns of plant diversity in the Sahel zone of Burkina Faso. In A. Röder, & J. Hill (Eds.), Recent advances in remote sensing and geinformation processing for land degradation assessment (pp. 199–209). Leiden: CRC Press/Balkema. Kriticos, D. J., Webber, B. L., Leriche, A., Ota, N., Macadam, I., Bathols, J., et al. (2012). CliMond: Global high-resolution historical and future scenario climate surfaces for bioclimatic modelling. Methods in Ecology and Evolution, 3(1), 53–64. http://dx.doi.org/10.1111/j.2041-210X.2011.00134.x Krömer, T., Kessler, M., & Gradstein, S. R. (2007). Vertical stratification of vascular epiphytes in submontane and montane forest of the Bolivian Andes: The importance of the understory. Plant Ecology, 189(2), 261–278. http://dx.doi.org/10.1007/s11258-006-9182-8 Lentz, D. L., Bye, R., & Sánchez-Cordero, V. (2008). Ecological niche modeling and distribution of wild sunflower (Helianthus annuus L.) in Mexico. International Journal of Plant Sciences, 169(4), 541–549. Li, Z., Zhu, Q., & Gold, C. (Eds.). (2005). Digital terrain modeling: Principles and methodology. Boca Raton: CRC Press. Lillesand, T., & Kiefer, R. W. (2008). Remote sensing and image interpretation (6th ed.). United Kingdom: Wiley. Liu, C., Berry, P. M., Dawson, T. P., & Pearson, R. G. (2005). Selecting thresholds of occurrence in the prediction of species. Ecography, 28(3), 385–393. http://dx.doi.org/10.1111/j.0906-7590.2005.03957.x Lobo, J. M., Jiménez-Valverde, A., & Real, R. (2008). AUC: A misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography, 17(2), 145–151. http://dx.doi.org/10.1111/j.1466-8238.2007.00358.x Loiselle, B. A., Jørgensen, P. M., Consiglio, T., Jiménez, I., Blake, J. G., Lohmann, L. G., et al. (2008). Predicting species distributions from herbarium collections: Does climate bias in collection sampling influence model outcomes? Journal of Biogeography, 35(1), 105–116. http://dx.doi.org/10.1111/j.1365-2699.2007.01779.x Lounibos, L. P., O’Meara, G. F., Nishimura, N., & Escher, R. L. (2003). Interactions with native mosquito larvae regulate the production of Aedes albopictus from bromeliads in Florida. Ecological Entomology, 28(5), 551–558. http://dx.doi.org/10.1046/j.1365-2311.2003.00543.x Lu, D., & Weng. (2004). Spectral mixture analysis of the urban landscape in Indianapolis with Landsat ETM + Imagery. Photogrammetric Engineering and Remote Sensing, 70(9), 1053–1063. Luther, H. E. (Ed.). (2010). An alphabetical list of bromeliad binomials. Sarasota, Florida: Sarasota Bromeliad Society and Marie Selby Botanical Gardens. Martin, C. E. (1994). Physiological ecology of the Bromeliaceae. The Botanical Review, 60(1), 1–82. http://dx.doi.org/10.1007/BF02856593 McKinney, M. L. (2002). Urbanization, biodiversity, and conservation. BioScience, 52(10), 883–890. http://dx.doi.org/10.1641/0006-3568(2002) 052[0883:UBAC2.0.CO;2] McKinney, M. L. (2006). Urbanization as a major cause of biotic homogenization. Biological Conservation, 127(3), 247–260. http://dx.doi.org/10.1016/j. biocon.2005.09.005 Morán-Ordóñez, A., Suárez-Seoane, S., Elith, J., Leonor, C., & de Estanislao, L. (2012). Satellite surface reflectance improves habitat distribution mapping: A case study on heath and shrub formations in the Cantabrian Mountains (NW Spain). Diversity and Distributions, 18(6), 588–602. http://dx.doi.org/10.1111/j.1472-4642.2011.00855.x Neteler, M., & Mitasova, H. (2002). Open source GIS: A GRASS GIS approach. Berlin, Hamburg: Kluwer Academic Publishers. Nöske, N. M., Hilt, N., Werner, F. A., Brehm, G., Fiedler, K., Sipman, H. J. M., et al. (2008). Disturbance effects on epiphytes and moths in a Montane forest in Ecuador. Basic and Applied Ecology, 9(1), 4–12. http://dx.doi.org/10.1016/j.baae.2007.06.014 Pearson, R. G., Raxworthy, C. J., Nakamura, M., & Peterson, A. T. (2007). Predicting species distributions from small numbers of occurrence records: A test case using cryptic geckos in Madagascar. Journal of Biogeography, 34(1), 102–117. http://dx.doi.org/10.1111/j.1365-2699.2006.01594.x Phillips, S. (2006). A brief tutorial on Maxent, AT & T Research, Retrieved March 10, 2012 from http://www.cs.princeton.edu/∼schapire/maxent/tutorial/ tutorial.doc Phillips, S. J., Anderson, R. P., & Schapire, R. E. (2006). Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3–4), 231–259. http://dx.doi.org/10.1016/j.ecolmodel.2005.03.026 Phillips, S. J., Dudík, K. M., & Schapire, R. E. (2004). A maximum entropy approach to species distribution modelling. In C. E. Brodley (Ed.), Machine Learning, 118 C. Judith et al. / Landscape and Urban Planning 120 (2013) 107–118 Proceedings of the Twenty-first International Conference (ICML 2004), Banff, Alberta, Canada, July 4–8, 2004 (12th ed., pp. 472–486). New York: ACM Press. http://dx.doi.org/10.1145/1015330.1015412 Prates-Clark, C. D. C, Saatchi, S. S., & Agosti, D. (2008). Predicting geographical distribution models of high-value timber trees in the Amazon Basin using remotely sensed data. Ecological Modelling, 211(3–4), 309–323. http://dx.doi.org/10.1016/j.ecolmodel.2007.09.024 R Core Team. (2012). R: A language and environment for statistical computing. Vienna, Austria: R foundation for Statistical Computing. Raes, N., & ter Steege, H. (2007). A null-model for significance testing of presence-only species distribution models. Ecography, 30(5), 727–736. http://dx.doi.org/10.1111/j.2007.0906-7590.05041.x Rojas, M. I., & Alfaro, E. J. (2000). Influencia del océano Atlántico tropical sobre el comportamiento de la primera parte de la estación lluviosa en Venezuela (influence of the tropical Atlantic Ocean to the first part of the rain station in Venezuela). Tópicos Meteorológicos y Oceanográficas, 7(2), 88–92. Sattler, D., Schmidt, S., & da Silva Alves, M. V. (2010). Analysis of the planted and spontaneous vegetation at selected open spaces in Apicuous district of Recife. In N. Müller, P. Werner, & J. G. Kelcey (Eds.), Urban biodiversity and design (pp. 273–290). Chichester, Brazil: Wiley-Blackwell. http://dx.doi.org/10.1002/9781444318654.ch14 Schmidt, M. (2006). Pflanzenvielfalt in Burkina Faso – Analyse, modellierung und dokumentation (plant diversity of Burkina Faso – Analysis, modelling and documentation). Frankfurt a.M: Goethe-Universität [Dissertation]. Schmidt, M., König, K., & Müller, J. V. (2008). Modelling species richness and life form composition in Sahelian Burkina Faso with remote sensing data. Journal of Arid Environments, 72(8), 1506–1517. http://dx.doi.org/10.1016/j.jaridenv.2008.02.015 Schneider, J. V. (2001). Diversity, structure, and biogeography of a successional and mature upper montane rain forest of the Venezuelan Andes (La Caña, Valle de San Javier, Mérida State). Frankfurt a.M: Goethe-Universität [Dissertation]. Schneider, J. V., Gaviria, J., & Zizka, G. (2003). Inventario florístico de un bosque altimontano húmedo en el Valle de San Javier, Edo. Mérida, Venezuela (floristic inventory of an altimontane humid forest in the San Javier Valley, State of Mérida, Venezuela). PlantULA, 3(2), 65–81. Seguardo, P., & Araújo, M. B. (2004). An evaluation of methods for modelling species distributions. Journal of Biogeography, 31(10), 1555–1568. http://dx.doi.org/10.1111/j.1365-2699.2004.01076.x View publication stats Smith, L. B. (Ed.). (1971). Flora de Venezuela (Flora of Venezuela), Parte 1: Bromeliaceae, Carácas, Vol. 12. Instituto Botanico. Smith, L. B., & Downs, R. J. (Eds.). (1977). Tillandsioideae. Flora Neotropica Monographs, 14(2), 663–1492. Soria-Auza, R. W., Kessler, M., Bach, K., Barajas-Barbosa, P. M., Lehnert, M., Herzog, S. K., et al. (2010). Impact of the quality of climate models for modelling species occurrences in countries with poor climatic documentation: A case study from Bolivia. Ecological Modelling, 221(8), 1221–1229. http://dx.doi.org/10.1016/j.ecolmodel.2010.01.004 Stansell, N. D., Polissar, P. J., & Abbott, M. B. (2007). Last glacial maximum equilibrium-line altitude and paleo-temperature reconstructions for the Cordillera de Mérida, Venezuelan Andes. Quaternary Research, 67(1), 115–127. http://dx.doi.org/10.1016/j.yqres.2006.07.005 United Nations. (2010). World urbanization prospects: The 2009 revision. New York: United Nations, Department of Economic and Social Affairs, Population Division. Wang, F. (1990). Fuzzy supervised classification of remote sensing images. IEEE Transactions on Geoscience and Remote Sensing, 28(2), 194–201. http://dx.doi.org/10.1109/36.46698 Werner, F. A., & Gradstein, S. R. (2008). Seedling establishment of vascular epiphytes on isolated and enclosed forest trees in an Andean landscape, Ecuador. Biodiversity and Conservation, 17(13), 3195–3207. http://dx.doi.org/10.1007/s10531-008-9421-5 Werner, F. A., Homeier, J., & Gradstein, S. R. (2005). Diversity of vascular epiphytes on isolated remnant trees in the Montane forest belt of southern Ecuador. Ecotropica, 11, 21–40. Wisz, M. S., Hijmans, R. J., Li, J., Peterson, A. T., Graham, C. H., Guisan, A., et al. (2008). Effects of sample size on the performance of species distribution models. Diversity and Distributions, 14(5), 763–773. http://dx.doi.org/10.1111/j.1472-4642.2008.00482.x Wolf, J. H. D. (2005). The response of epiphytes to anthropogenic disturbance of pineoak forests in the highlands of Chiapas, Mexico. Forest Ecology and Management, 212, 376–393. http://dx.doi.org/10.1016/j.foreco.2005.03.027 Zimmermann, N. E., Edwards, J. R., Moisen, T. C., Frescino, G. G., & Blackard, T. S. J. A. (2007). Remote sensing-based predictors improve distribution models of rare, early successional and broadleaf tree species in Utah. Journal of Applied Ecology, 44(5), 1057–1067. http://dx.doi.org/10.1111/j.1365-2664.2007.01348.x Zizka, G., Schmidt, M., Schulte, K., Novoa, P., Pinto, R., & König, K. (2009). Chilean Bromeliaceae: Diversity, distribution and evaluation of conservation status. Biodiversity and Conservation, 18(9), 2449–2471.