Landscape structure influences avian species diversity in tropical urban mosaics

In this study, we tested whether urban landscape structure influences avian species diversity using data for Harare, Zimbabwe. Initially, we quantified landscape structure using fragmentation indices derived from a 5m resolution SPOT 5 imagery. We collected bird species data through field-based observations of birds at 35 locations occurring in five land use/land cover types. We quantified avian species diversity using Barger-Parker, Menhinick and Simpson’s Indices. Regression analysis was used to determine the nature and strength of the relationships between avian species diversity and fragmentation indices. Results indicated that woodland specialist avian species are negatively associated with landscape fragmentation, while grassland specialist and generalist avian species positively responded to patch edge density, habitat patch size and shape complexity. Overall, our results suggest that changes in landscape structure due to expansion of built-up areas in tropical urban areas may influence avian species diversity.


Introduction
Understanding the factors that influence biodiversity within urban landscapes is fundamental to 31 the planning and development of biodiversity tolerant cities. In the 21 st Century, increasing 32 landscape fragmentation resulting from urban development and transportation infrastructure is 33 considered a predominant driver of biodiversity loss in tropical ecosystems [1]. Urban 34 development has a marked impact on the environment [2] as it replaces wildlife habitat with 35 artificial surfaces that are unsuitable as wildlife habitat e.g., asphalt surfaces [3]. Although urban 36 areas occupy <3% of the Earth's land surface area [4], their ecological impacts span over large 37 spatial extents and sometimes beyond the urban boundaries [5]. Thus, understanding biological 38 diversity-landscape structure (spatial configuration of a given land cover class) relationships is 39 increasingly becoming critical in urban planning [6]. In urban areas, the expansion of built-up 40 areas as well as its configuration is hypothesised to have differential but significant impacts on 41 biodiversity patterns [3], thereby making objective methods for quantifying this phenomena 42 critical.

44
The quantification of landscape structure in urban landscapes is an important step towards 45 developing urban growth management plans that promote biological diversity. Thus, the 46 development of methods for understanding the impact of urban development on biological 47 diversity in the tropics is critical for biodiversity conservation and enhancement of wildlife 48 persistence in these ecosystems. Such methods may need to focus on improving the estimates of 49 landscape structure-biodiversity relationships. Although field measurements are regarded as the 50 most accurate method of quantifying landscape structure-biodiversity relationships, these 51 measurements are costly and labour intensive and can only be feasible over smaller scales [7,8]. 52 In this regard the development of methods that supplement field measurements is important.

54
Developments in Geographic Information Systems (GIS) and satellite remote sensing have made 55 it possible to quantify landscape structure rapidly [2,3]. In the past, several studies have 56 demonstrated the utility of landscape indices derived from satellite remotely sensed GIS data in 57 estimating landscape-biodiversity relationships across various spatiotemporal scales in temperate 58 landscapes [9][10][11]. For example, in a study by Coops et al. [12] satellite-derived landscape 59 metrics were used to predict bird species richness in Ontario, Canada using the Moderate-60 resolution Imaging Spectroradiometer (MODIS) and explained variance ranging between 47 to 61 75%. Similarly, Guo et al. [10]  observed that air photo (R 2 = 0.54) and Landsat TM satellite image (R 2 = 0.52) were better 66 predictors of avian species density than field-measured vegetation structure (R 2 = 0.32). In urban 67 landscapes, relatively higher resolution imagery could be of use in modelling the relationship 68 between landscape structure and biodiversity.

70
The availability of high spatial resolution sensors such as SPOT 5 has provided data that could 71 be used to improve the quantification and mapping of landscape structure indices in urban 72 landscapes that in turn may allow for improved understanding of landscape structure-biodiversity 73 relationships. To date, studies that assess the utility of high spatial resolution multispectral 74 imagery such as SPOT 5 in estimating landscape structure-biodiversity relationships in tropical 75 urban ecosystems remains rudimentary.

77
In this study, we tested whether and in what way landscape structure indices derived from 78 remotely sensed land cover relate with avian species diversity patterns in Harare, Zimbabwe.

79
Specifically, we tested whether and to what extent avian species diversity respond to constraints 80 including habitat patch size, habitat shape complexity, and habitat inter-patch distance. We 81 derived bird species data from field surveys and landscape structure data from high spatial 82 resolution sensors, i.e. SPOT 5 for Harare, Zimbabwe. We expect differential responses of avian 83 species diversity to habitat constraints. For example, woodland and grassland specialist avian 84 species may be negatively related to decrease in habitat patch size, increased shape complexity 85 and habitat isolation distance. While generalist species will respond positively to changes in 86 habitat conditions. 87 88 89

90
Study area 91 The study was carried out in the Harare Metropolitan province of Zimbabwe ( Figure 1). The

92
Harare metropolitan area is approximately 892km 2 in spatial extent and has a human population 93 of approximately 2.5 million [13]. The center of the study area, is located at Longitude 31º7ꞌE 94 and Latitude 17º55ꞌS with an altitude range of 1400-1500m above sea level. The city experience 95 two distinct seasons i.e., hot wet summers (October -April) and cool dry winters (May -96 September). The mean annual rainfall ranges between 800-1000mm, while mean annual 97 temperature ranges between 25 -27 ºC [14].  kirkiana [15]. The bare ground cover type consists of exposed surfaces and area under active 106 urban development. The water cover type includes impoundments and rivers. The urban built-up 107 area is made up of impervious surface covering including road networks, industrial areas, high 108 and low density residential areas. The study site was selected because it represents an ideal 109 location to study landscape structure-biodiversity relationships in the context of regional and 110 urban planning. The area is currently undergoing a rapid increase in human population associated 111 with unguided urban development patterns whose impacts have not been quantified.

113
Quantifying landscape structure 114 We derived landscape structure data from a 5-m spatial resolution SPOT 5 image of Harare.

130
Sampling design 131 In a GIS, we processed the study area into a LULC categories layer representing three LULC 132 types i.e., low urbanization grasslands, low urbanization forested area and built-up areas (Table   133 1). Subcategories were defined for each category to account for variations each context 134 presented. Altogether we had seven LULC subcategories and representing three LULC types and 135 35 transect sampling sites (Table 1).

137
#Insert Table 1 138 139 Using the LULC categories base map and the Random Sampling Tool in Quantum GIS 2.6.1 140 (QGIS Development Team, Switzerland) we stratified the study area (excluding private and 141 security areas e.g., military and airport land) into five sampling sites for each LULC subcategory 142 (total 35 sites) ( Table 1). We deemed the sample of 35 sites representative for statistical purposes flying or foraging within a 50m distance from the 600m transect line (see SI 1). We identified the 157 birds to species level based on expert knowledge and a field guide book i.e., Roberts Birds of 158 Southern Africa [29]. We also categorized avian species into three ecological guilds (generalists, 159 woodland specialists as well as grassland specialists) because we investigated landscape 160 influence on the birds at guild level.

161
Avian species were selected as the model species, because they are highly mobile and can 162 respond to landscape change quickly than ground dwelling mammals or other rarely seen species 163 [9] which makes birds useful indicators of species responses to urban development induced 164 environmental change. The study focused on overall avian species than select target species, 165 common in many studies [30]. The advantage of focusing on overall avian species is that it allows the study to account for avian species with different life histories and behaviors [18,30,167 31].

169
Quantifying avian species diversity 170 We used the Menhinick, Berger-Parker and Simpson's indices to quantify avian species diversity 171 [9, 32, 33] ( Table 2) 185 Prior to regression analysis we tested the avian species data for normality using the  Smirnov test to test [37] for conformity to the simple regression assumption for randomness and

194
Avian species diversity-landscape structure relationships 195 We surveyed 6081 birds representing 69 species in 35, 600m transects. Thirty percent of the 196 surveyed birds were observed in low urbanization grassland habitat, 46% in built-up areas and 197 24% in low urbanization forested land. We also observed that bird species abundance, richness 198 and diversity (i.e., Menhinick's, Berger-Parker and Simpson's Indices) varies across the three 199 LULC classes (Table 2).

201
#Insert Table 2 202 203 Woodland specialist avian species -landscape structure relationships 204 Simple regression showed that woodland specialist avian species were negativity associated with 205 patch metrics derived from low urbanization forested cover type, specifically shape complexity 206 (R 2 = 0.635), shape size (R 2 = 0.616) and isolation distance (R 2 = 0.778) (Figure 2). Simple regression showed that grassland specialist avian species had a strong positive 212 polynomial relationship with patch edge derived from low urbanization grassland cover type (R 2 213 = 473, Figure 3) and not significant (p>0.05) association with patch size and isolation distance.

232
Results also indicated that avian species diversity of woodland specialists negatively correlated 233 with edge density of the low urbanization forested cover type, suggesting that for these specialist

252
The observation that generalist avian species diversity positively correlates with landscape 253 fragmentation also suggest that generalist bird species benefit from forest loss and fragmentation. Overall, this study provides evidence that high resolution satellite imagery offer improved 259 opportunities for estimating the effect of urban development on biodiversity in particular avian 260 species diversity. The best model explained 79% variation in avian species diversity. This 261 coefficient of determination is higher than obtained by Coops et al. [12] and Guo et al. [10]  planned, thus making high spatial resolution satellite imagery an excellent alternative to 287 delineating spatial variability habitat fragmentation. However, it will be useful to test the 288 applicability of these models in independent study sites to observe whether the form of remotely 289 sensed models of landscape metrics are consistent and can be improved further. Nevertheless, we 290 make a claim that this finding provides an opportunity to quantifying the impact of urban 291 landscape pattern on biodiversity in tropical urban landscapes of sub-Saharan Africa.

294
The main objective of this study was to test whether and to what extent avian species respond to 295 constraints including habitat fragment size, shape complexity and isolation distance in urbanizing 296 tropical ecosystems. From the results of this study, we conclude that the: 297 1. size, shape and isolation distance of habitat fragments matter to woodland specialist avian 298 species; 299 2. shape of habitat fragments matter to grassland specialist species, than isolation and size 300 of grassland fragments; and 301 3. the increasing complexity of habitat fragment shape and size increases the diversity of 302 generalist species than isolation. 303 We therefore conclude that urban planning can improve biodiversity in urban landscapes by 304 managing the size, shape and isolation distance of habitat fragments. Such approaches to urban 8 305 development can create conditions suitable for avian species persistence in urban landscapes.

306
Large, regular shaped and interconnected habitat fragments are also fundamental to the 307 conservation of avian species in urban landscapes. Future urban development strategies should 308 therefore consider habitat conditions necessary for species persistence, by managing the size, 309 shape and isolation distance of undeveloped grassland and forested areas in urban ecosystems. 310 We suggest further studies that aim to assess the variation of avian species diversity in relation to 311 land use, primary productivity, climatic and topographic variables to assess the pattern of the 312 distribution and assess whether or not further improvements for estimating biodiversity impacts 313 of urban development can be achieved.

315
Acknowledgements 316 We are grateful to Anna Zivumbwa for support in the field.    Berger-Parker's Indexes) and landscape structure (a = size, b= shape, c = isolation distance)