Introduction

Urban forest remnants are an isolated patches of woody/tree vegetation which grows within city or town such as parks, wooded buffers, greenways etc. (Forman 2014). These remnants had formed due to the destruction of forests as a result of intensive urbanization. These areas may vary in shape, size and vegetation species diversity (McKinney 2002; Petr et al. 2004) and provide several goods (bye products) and services to the society (Asan et al. 2007; Sander et al. 2010). Now a day, protection of biological diversity, recreation and esthetic services are more important in urban forest remnants (Tyrväinen et al. 2005; Kark and Dearborn 2009). The urban forests have a rich diversity of plant species since they are not managed (Bagnall 1979; McKinney 2002; Petr et al. 2004). Furthermore, they contribute to the protection of the biological diversity by accommodating food, shelter, concealment and nest for a great number of animal species (Mansfield et al. 2005; Tyrväinen et al. 2005). Urban forest remnants are significant spaces where people learn about the nature within the heart of cities. People may enjoy aesthetic beauty, an ideal place for recreation, exercising, and trekking at the free of cost and due to easy access (Konijnendijk 2005; Tyrväinen et al. 2005; Kong et al. 2007).

Even though urban forest remnants are highly important, but unfortunately facing various threats such as air pollution, intensive utilization, excessive use of recreational activities, and illegal logging. These factors directly or indirectly altered the vegetation structure and species composition and also productivity. Therefore, efforts in recent years are increasing to protect the biological diversity in urban forest remnants (Alvey 2006; Kark and Dearborn 2009). It is very challenging to make such an evaluation by taking account of all components of biological diversity. Species diversity is the main component in the evaluation of biological diversity (Rocchini et al. 2004). The most accurate way of collecting the data needed to evaluate the biological diversity is taking the inventory based on intensive terrestrial measures. However, this is a labor-intensive process, expensive and time consuming (Gillespie et al. 2008; Mutowo and Murwira 2012). In practice, the inventory can be taken only for a small portion of the planning area due to the lack of resources and time (Luoto et al. 2002). Therefore, it is critically important to evaluate rapidly and monitor the spatial distribution and temporal variation in biological diversity in order to decide on the appropriate protection strategies (Nagendra 2001; Moser et al. 2002; Singh et al. 2010; Varghese, et al. 2010).

Satellite images have a great potential as a source of information to monitor the biological diversity at local, regional and national scales at species level (Luoto et al. 2002; Rocchini et al. 2007; Gillespie et al. 2008; Singh et al. 2010; Adam et al. 2012). There have been efforts to determine the species diversity through remotely sensed data (Bawa et al. 2002; Fairbank and McGwire 2004; Rocchini et al. 2004; Cayuela et al. 2006; Rocchini 2007; Levin et al. 2007; Oldeland et al. 2010). These studies usually focused on the relationships between the spectral values and the species diversity in the plots (α diversity). Besides, the species rarefaction function relying on dissimilarities in species between the plots (β diversity) was adapted by some researchers to the spectral rarefaction function relying on spectral difference (Rocchini et al. 2005; Rocchini et al. 2007; Rocchini et al. 2009; Rocchini et al. 2011). These studies explored the ways of estimating β diversity with the help of the spectral rarefaction curve derived from the presence/absence matrix created based on presence/absence of each Digital Numbers (DNs) in the plots. They were conducted mainly in natural forests, whereas they did not explore adequately the possibility of using satellite images to evaluate the species diversity in the urban forest remnants.

In recent years, there has been a great interest in the texture properties of the image (spatial orders of pixels) for natural resource applications of image analysis (Ozdemir et al. 2008; Kim et al. 2011; Ozdemir and Karnieli, 2011; Ardila et al. 2012; Ozdemir and Donoghue 2013; Dian et al. 2015). These studies reported that there are relations between textural properties and structural heterogeneity of the forest. However, the relationship between the species diversity and the texture properties of images has not been evaluated extensively yet. In studies related to species diversity, the alignment between the size of the object and the spatial resolution of the satellite image is very important. The use of a satellite image with a coarser resolution compared to the size of the object may conceal the actual spatial heterogeneity and lead to the underestimation of species diversity (Rocchini et al. 2011). Mixed pixels constitute the most important problem encountered in using the satellite images for urban forest remnants as they contain rich species diversity (Hung 2002). For that reason, it is important to compare the satellite images of different spatial resolution in order to decide on the optimal spatial resolution in evaluating the species diversity in urban forest remnants.

This study aimed at exploring the possibilities to evaluate the woody species diversity in urban forest remnants by using satellite images of different spatial resolution. To this end, the relationships between the α diversity calculated by using terrestrial measures and observations and the spectral & texture properties measured on satellite images were evaluated. Then, it was tested whether or not β diversity can be estimated in urban forest remnants by using spectral rarefaction function.

Materials and Methods

Study Area

The study area covers the urban forest remnants in Istanbul that has the most crowded city in Turkey with a population of around 15 million people. The urban forest remnants both in the European side and the Asian side of Istanbul are located at altitudes starting from sea level up to 250 m (Fig. 1). Formerly being natural forests, these forest remnants have been preserved and survived until today while the species diversity has been enriched through planting of new plant species (Asan et al. 2007). Residents of Istanbul use these areas for recreation, sports, and education purpose.

Fig. 1
figure 1

Geographical location of the study site

Full size image

Field Data

The study site was divided into square grids of 90 × 90 m. The focus of the study was comprised of the woody species diversity. Therefore, 35 square grids that did not contain man-made objects were selected as sampling units. Since, the urban forest remnants were rich in woody species diversity, the sampling units were further divided into sub-plots. The circular plots of a fixed size (400 m2) were positioned on four corners and one in the center of the sampling units. In this way, the sampling patter consisted of 35 sampling units and 175 sub-plots. The center of the plots was founded on the land with the help of GPS and applied to the land with a laser. The woody species diversity in the plots of 400 m2 and the number of species were recorded in the inventory. The field study was performed between August 2014 and November 2014.

Satellite Data

A RapidEye image with 6.5 m spatial resolution (5 m-orthorectified pixel size), ASTER image with 15 m spatial resolution and Landsat-8 OLI image with 30 m spatial resolution were used in the study. The radiometric resolution of the RapidEye image is 12 bits and has 5 bands (Blue 440–510 nm; Green 520–590 nm; Red 630–685 nm; Red Edge 690–730 nm; Near Infrared 760–850 nm). ASTER image has a radiometric resolution of 8 bits and 3 bands in the Visible, Near Infrared (VNIR) (Green 0.52–0.60 μm, Red 0.63–0.69 μm, Near Infrared 0.76–0.86 μm). The Landsat 8 image that acquires images in the range of visible, near infrared, short wave infrared and thermal infrared has 5 bands in the Visible, Near Infrared (Coastal 0.433–0.453 μm, Blue 0.450–0.515 μm, Green 0.525–0.600 μm, Red 0.630–0.680, Near Infrared 0.845–0.885 μm). It has 12 bits radiometric resolution. RapidEye and ASTER images were obtained on 02 July 2014, while Landsat-8 image was obtained 30 June 2013.

In order to provide more detailed data regarding the areas that were covered with vegetation in multi-channel images, vegetation indexes were developed by using the original bands (Jensen and Lulla 1996). Four vegetation indexes (Atmospherically Resistant Vegetation Index-ARVI; Difference Vegetation Index-DVI; Infrared Percentage Vegetation Index-IPVI; Normalized Difference Vegetation Index-NDVI) were produced in the study in order to evaluate their potential for the evaluation of woody species diversity (Table 1).

Table 1 Vegetation indexes obtained from the satellite images
Full size table

Calculation of Spectral & Textural Properties

In the study, the texture properties were determined in addition to the spectral properties by using satellite images (Table 2). The Brightness of the pixels in each image window corresponding to the sampling units were taken into account as the spectral properties. Texture properties were calculated according to Grey Level Co-occurrence Matrix (GLCM). A large number of texture properties can be determined through GLCM. They are categorized under three main groups (Haralick et al. 1973). Those texture properties that are in the same group have high correlation with one another; therefore, one property from each group may suffice. Accordingly; “Contrast” from the contrast group; “Entropy” from the regularity group, and “Correlation” from the statistics group were used in the study. Spectral & texture properties were identified separately for the original bands of the satellite images or each set of data based on the original bands.

Table 2 Spectral & textural properties used in the study
Full size table

Alpha (α) Diversity

In the study, α diversity was represented by Species Richness, Shannon-Wiener Diversity Index (Shannon and Weaver 1949), Simpson’s (Simpson 1949) Diversity Index and Species Mingling Index (Gadow and Hui 2002) (Table 3). Species Richness refers to the total number of the sampled species in the area and it is acknowledged as the simplest indicator of biological diversity (Battles et al. 2001; Rocchini et al. 2005). Shannon-Wiener Index combines both number of species and their evenness into a single measure. Simpson’s Index is a dominance index that differently assigns weight to common or dominant species. The main principle of the Species Mingling Index is based on the neighborhood of the four closest trees to the reference tree.

Table 3 Indicators of alpha diversity used in the study
Full size table

For each sampling unit, the correlation analysis was applied to test whether, there is a significant association between the α diversity or not by taking the average of the species diversity measured in the sub-plots and the spectral & texture properties obtained from the satellite images. The Pearson correlation coefficient (PCC) was used to evaluate the association. PCC can be used when the data had a normal or near-normal distribution. Shapiro-Wilk method was used to test whether or not all variables had a normal distribution. When P > 0,05, the variable in question was considered to have a normal distribution (Kalaycı 2006; Kayitakire et al. 2006). The work flow used to identify the alpha diversity and associate it with the image properties is presented in Fig. 2.

Fig. 2
figure 2

Flow chart of the methodology applied

Full size image

Beta (β) Diversity

β diversity was calculated by using the rarefaction function. To this end, first of all rarefaction analysis was applied to obtain the values of the actual species diversity from the plots and the rarefaction curve was modeled depending on the presence and absence of the species in the plots. Taking into account of the approach proposed by Rocchini et al. (2009); the minimum accumulated number of species was taken as average alpha (ᾱ) diversity while the maximum accumulated number of species was taken as gamma (ɣ) diversity based on the rarefaction curve. The actual β diversity (βreal) was calculated as the difference between ɣ and ᾱ. Then, the same procedure was applied to the brightness obtained from the satellite images and the spectral rarefaction curves were created.

In order to estimate the β diversity with the help of the image brightness values, the relationship between the species rarefaction values and the spectral rarefaction values was tested through employing the multiple regression models. The stepwise selection procedure was applied to develop a model. With the help of the most compatible models, the species rarefaction curves were remodeled, based on which the estimated β diversity (βest) values were found. Βest values and βreal values were compared on the basis of three satellite images. There is a need for data from one band in order to model the rarefaction curves. Previous studies preferred PCA or NDVI (Rocchini et al., 2009, 2011). As we used satellite images with different number of bands in our study, it was decided to use NDVI image in order to ensure a standard between hem. Rarefaction functions were calculated based on 500 iterations through EstimateS (Version 9.1.0) software. The work flow, followed in the analysis of rarefaction is presented in Fig. 2.

Results

Relationships Between Alpha Diversity and Satellite-Based Variables

The relationship between alpha diversity and the spectral & texture properties of the image was observed to be linear or near-linear; therefore, the relationship between the image properties and the indicators of α diversity was evaluated through PCC. Statistically significant relationships were found between diversity indicators calculated on the basis of terrestrial measures and the spectral & texture properties obtained from the satellite images. To sum up, the Brightness derived from the red band of RapidEye image had a statistically significant relationship with S (Table 4). The Brightness derived from the near infrared band showed a significant relationship with S and MI. Moreover, the Brightness derived from NDVI and ARVI were associated with all the diversity indicators. ARVI Brightness had the strongest relationship with S and H′ (r = 0.80 and 0.75 p < 0.01). Apart from the spectral values, the texture properties obtained from RapidEye image were observed to have significant relationships with the diversity indicators. Among the texture properties, GLCM Contrast had a significant relationship with SR in the red band and vegetation indexes. In contrast, the relationship between the GLCM Contrast and MI, S and H′ was, in general, weak on all bands. GLCM Correlation had a significant relationship, in general, with SR; whereas the relationship between MI, S and H′ and GLCM Correlation was observed to be weak on all bands. Besides, GLCM Entropy also had a statistically significant relationship with SR on the red band. Among the texture properties, the strongest relationship was observed to be between the GLCM Contrast derived from IPVI and SR (r = 0.57, p < 0.01).

Table 4 Pearson correlation coefficients between spectral & texture features and indicators of α diversity (RapidEye)
Full size table

The strongest relationship between the image properties derived from ASTER image and indicators of α diversity was observed between the Brightness derived from NDVI and S (r = 0.55, p < 0.01) (Table 5). It was understood that the Brightness derived from the vegetation indexes other than NDVI had, in general, significant relationships with S, H′ and MI. Furthermore, significant relationships were found between the Brightness and S in the red band, and between the Brightness and S and MI on the near infrared band. The strongest relationship between the texture properties and the diversity indicators was found between the GLCM Contrast derived from the red band and SR (r = 0.56, p < 0.01). Furthermore, the GLCM Contrast on green and red bands highlight a significant relationships with H′, S, MI and SR. The original bands of ASTER image and GLCM Entropy obtained from the vegetation indexes were, in general, associated with all diversity indicators. The relationship between the GLCM Entropy derived from DVI and SR was calculated as r = 0.56 (p < 0.01). As another texture property, GLCM Correlation in all vegetation indexes had, in general, significant relationships with all diversity indicators.

Table 5 Pearson correlation coefficients between spectral & texture features and indicators of α diversity (ASTER)
Full size table

The Brightness derived from the coastal and red bands of Landsat-8 with a spatial resolution of 30 × 30 m had significant relationships with SR and S (Table 6). The Brightness derived from the vegetation indexes, however, were observed to have significant relationships with S and MI. Only the GLCM Correlation out of the texture properties that were derived from the red band had a significant relationship with SR. Apart from this, the texture properties derived from the original bands of Landsat-8 satellite image and vegetation indexes had very weak relationships with H′, S, MI and SR. The strongest relation between the image properties of Landsat-8 and indicators of α diversity was found between the Brightness derived from NDVI and S (r = 0.50, p < 0.01).

Table 6 Pearson correlation coefficients between spectral & texture features and indicators of α diversity (Landsat-8)
Full size table

Modeling Species Rarefaction Curves based on Spectral Rarefaction Curves

The species and spectral rarefaction curves modeled as a result of the rarefaction analysis in the study are presented in Fig. 4. According to the species rarefaction curve; the average number of species of the sampling units (ᾱreal) was 20, while the actual total number of species in the sampling units (ɣreal) was 50 and the diversity between the sampling units (βreal) was 30 (Fig. 3). The species rarefaction curve rose quickly until the seventh sampling unit, after which the rising trend of the curve slowed down. This trend of the rarefaction curve shows that there is a low species difference between the sampling units. The trends in the spectral rarefaction curves were similar to those of the species rarefaction curve. It can be concluded based on spectral rarefaction curves that the spectral difference between the image windows corresponding to the sampling units was low on each of the three satellite images.

Fig. 3
figure 3

Species and spectral rarefaction curves

Full size image

According to the results of the multiple regression analyses between the species rarefaction values and spectral rarefaction values, very strong and positive relationships were found between the species and spectral rarefaction values (Fig. 4). The relationships between the spectral rarefaction values derived from RapidEye NDVI, ASTER NDVI and Landsat-8 NDVI images and species rarefaction values (respectively R2 = 0.99, 0.99, 0.98) were found to be similar. The species rarefaction curves that were remodeled according to those relationships are presented in Figure 5. According to these curves; ᾱest was estimated to be 20, ɣest to be 47 and βest to be 27 based on the spectral properties of Landsat-8 NDVI image. Ᾱest, ɣest and βest were estimated to be 20, 50, and 30 respectively, according to the ASTER NDVI image and to be 19, 50 and 31, respectively according to the RapidEye NDVI image. According to these findings; β diversity between the sampling units could be estimated accurately by using the brightness in the satellite images.

Fig. 4
figure 4

Modeling of the relationship between the species and spectral rarefaction values

Full size image
Fig. 5
figure 5

Estimation of the species rarefaction curves (a-Landsat-8 b-ASTER c-RapidEye) and β diversity based on the spectral rarefaction curves

Full size image

Discussion

The findings of the study demonstrate that satellite images can be useful tools to evaluate the α diversity in the urban forest remnants. The results of the correlation analysis revealed that there were statistically significant relationships between the spectral & texture properties of the image and α diversity. The strongest relationship between the indicators of α diversity and spectral values was found between the Brightness derived from ARVI of RapidEye satellite image and S. The Brightness derived from ARVI and other indicators of α diversity that were SR, H′ and MI also had strong relationships. On ASTER satellite image, the strong relationship between the image properties and indicators of α diversity was observed to be between the Brightness derived from NDVI and S. The coastal, red bands of Landsat-8 satellite image as well as some vegetation indexes were found to have statistically significant relationships with the diversity indicators. However, the level of such relationships was rather weaker compared to the other two satellite images because of mixed pixels. In the light of these findings, it can be suggested that it could be possible to evaluate the species diversity of a specific area by using the spectral values of an image.

These findings are also consistent with the results of similar studies that were conducted previously. For instance; Rocchini et al. (2004) verified the “spectral variation hypothesis” by using QuickBird image, according to which the areas with the highest spectral variation are expected to have the highest species richness. In another study conducted by Rocchini et al. (2007), the effect of the bands in evaluating the local species richness using the QuickBird satellite image was tested, while it concluded that near infrared band could be used for such evaluation. Bawa et al. (2002) studied the potential use of satellite images in order to identify the areas that had low and high species richness in tropical forests. This study in which NDVI derived from IRS 1C LISS III was used revealed that there was a high and positive relationship between NDVI and species richness (R2 = 0.63 in broad-leaved forests, 0.42 in macquis shrub lands, and 0.66 in all vegetation types). Levin et al. (2007) conducted a study in which they tested the relationship between the NDVI images derived from Landsat, Aster and QuickBird satellite images and plant species richness. They concluded that there was a positively significant relationship between all of the NDVI images and plant richness (with R2 as high as 0.87 between NDVI and total plant richness and 0.89 for annual plant richness).

The results obtained demonstrate that the texture properties of an image can be used to evaluate α diversity in addition to the Brightness. GLCM Contrast, GLCM Entropy and GLCM Correlation were observed to have statistically significant relationships with SR on RapidEye image. The strongest relationship was found between the GLCM Contrast values and SR. On ASTER image; some significant relationships were observed between the texture properties and diversity indicators. The strongest relationship was found to be between the GLCM Contrast and GLCM Entropy values and SR. On Landsat-8 image, however, only the GLCM Correlation and SR had a significant relationship. To the best of our knowledge, there is no other study that evaluated the relationship between the texture properties of the satellite image and species diversity, the findings of this study could not be compared directly with the relevant studies in the literature. However, it was reported that some texture properties obtained from satellite images could be used both at stand level (Ozdemir and Karnieli 2011) and landscape level (Ozdemir et al. 2012) to determine the heterogeneity of the forest structure. Moreover, in the study conducted by Ozkan (2014) it concluded that the texture properties of the image were high due to the increased heterogeneity in the areas with greater species diversity. Therefore, it can be suggested that the findings of this study are partially consistent with the relevant literature.

The results of the study showed that spectral rarefaction curves could be used as an important method to estimate β diversity in urban forest remnants. According to the multiple regression analysis, very high positive relationships were found between the species and spectral rarefaction values. Based on the species rarefaction curves that were remodeled according to those relationships, β diversity could be estimated with high precision. These results are, in general, consistent with the findings of the similar studies performed previously. In a study conducted by Rocchini et al. (2009) who used Landsat ETM+ satellite image, the species diversity in different regions was compared based on the spectral rarefaction curves and they emphasized that it was a strong tool to identify the differences in biological diversity between the regions. In another study conducted by Rocchini et al. (2011) using Landsat ETRM+ satellite image, the relationship between the species rarefaction and spectral rarefaction was evaluated on the basis of the growth rates in the rarefaction curves. They found high and positive relationships between the species rarefaction and spectral rarefaction (r = 0.88;0.95 p < 0.001). Based on those results, they concluded that spectral rarefaction curves could be used as a quick method to estimate the differences in vegetation diversity at regional scale.

In our study, three satellite images with different spatial resolution were tested in the evaluation of woody species diversity in urban forest remnants. For α diversity, RapidEye and ASTER satellite images with a spatial resolution of 5 m and 15 m had similar results, while the results of Landsat-8 satellite image with a spatial resolution of 30 m were lower than the others. For the estimation of β diversity, none of the satellite images were superior to another. Therefore, low-cost medium resolution satellite images (Landsat) can be used to estimate β diversity in urban forest remnants based on rarefaction.

This study demonstrated that satellite images could be used to evaluate the woody species diversity in urban forest remnants and its temporal variations. The species diversity could not be precisely identified on the satellite images due to the inability to evaluate the species in the substrata because of crown closure or due to specific challenges such as complicated pixels (Rocchini et al. 2011). However, it would be possible to identify the areas that are rich or poor in species diversity. Therefore, it could be possible to decide where to concentrate the sampling to collect the data needed for future planning of these areas. This can result in more sound data regarding the species diversity in a specific area. Moreover, the temporal variation in the species diversity can be quickly and practically evaluated through the analysis of satellite images recorded on different dates. In this way, the hot points where the protection strategies can be focused could be identified easily during planning. Hence, it would be useful to continue evaluating the possibilities of using next-generation remotely sensed data with different spatial and spectral resolution for this purpose.

Conclusions

The results revealed that the woody species diversity at α level and the spectral & texture properties. The strongest relationship was found between the Brightness derived from Atmospherically Resistant Vegetation Index (ARVI) of RapidEye image and Simpson’s Diversity Index (S). A very high correlation was found between the spectral rarefaction values and species rarefaction values. It was revealed that β diversity could be estimated with high precision through employing species rarefaction curves modeled on the bass of the spectral rarefaction curves. The woody species diversity in urban forest remnants can be evaluated at α and β levels by employing the spectral & texture properties derived from the satellite images. Based on findings, it is concluded that the satellite images can be used to obtain reliable data more quickly as compared to other field survey methods due to having a significant potential. However, the satellite image with a resolution of 5 m would be more appropriate to estimate the woody species diversity of urban forest remnants. Hence, we recommended that the satellite image with a resolution of 5 m would be used in future for planning and conservation of urban forest remnants in a sustainable manner.