The Selection, Testing, and Application of Ecological Bioindicator Birds: A Case Study of the Bale Mountains, Southeast Ethiopia

The interest in using ecological bioindicators species or higher taxa, whose presence/absence or abundance reflect the abiotic or biotic state of an ecosystem as cost-effective means of ecological monitoring has been globally increasing. The main aim of this study was to assess if such ecological bioindicator species could be identified within Afromontane grassland avifauna that would be used for monitoring the effects of livestock grazing on ecosystem in the Bale Mountains of Ethiopia. We collected data on birds and vegetation structure along 14 transects each in the light (protected) and heavy (unprotected) grazing sites in June 2014 (referred to as the first sampling period) and in December 2014 (the second sampling period). Then, we (i) initially identified potential bioindicator species for the light grazing site, based on data collected during the first sampling period; (ii) examined relationships between abundance of these bioindicators and level of grazing pressure; (iii) tested the consistency of those initially selected bioindicator species, based on independent data collected during the second sampling period. We used the Indicator Value (IndVal) Analysis method to identify bioindicator species for the light grazing site. Species with significant IndVal>60% were considered as potential bioindicator for the site compared to the heavy grazing site. Six species were initially identified as potential bioindicators from the first sampling period dataset, and four of these species were again consistently identified from the second sampling period. Furthermore, abundance of the bioindicators had significantly declined with increasing grazing pressure, but positively correlated with four habitat variables (i.e. heights of shrub, herb and grass, and cover of shrub). These findings suggest that those consistently identified four species represent suite of reliable bioindicators that can successfully be used for monitoring of changes in habitat structure in the site. Recommendations on how to apply these findings for ecological monitoring are provided.


INTRODUCTION
Bioindicators are species or group of species, or higher taxa, whose biological or ecological attributes (e.g. presence/absence, abundance, survival rate, reproductive success, etc) readily Niemi and McDonald, 2004). Ecological bioindicators are primarily used either to assess the condition of (e.g., as an early-warning system) or to predict trends in state of an ecosystem (Dale and Beyeler, 2001). The premise to use bioindicators for ecological monitoring has come from the fact that most ecosystems are biologically and ecologically highly diverse and complex, making difficult to undertake surveys on tall taxa during monitoring (Noss, 1990). Further, most of such diverse and complex ecosystems are found in developing tropical countries like Ethiopia where there is often lack of resources (funding and expertise), making ecological monitoring activities more difficult (Addisu Asefa et al., 2015a). Consequently, application of the bioindication concept in conservation initiatives has been advocated to be used as a simple and cost-effective means of ecological monitoring to assess the current and predict the future healthiness of ecosystems (Noss, 1990;McGeoch, 1998;Carignan and Villard, 2002;Niemi and McDona, 2004).
Application of the bioindication concept using various biological taxa in conservation programmes have been reported by several authors. For example, Andersen et al. (2002) have used ants as bioindicators in land management; Davis (2001) has used Dung beetles as indicators of change in the forests of northern Borneo; and Kitching et al. (2000) have used moth assemblages as indicators of environmental quality in remnants of upland Australian rain forest.
Birds have been also applied as indicators of environmental change (e.g., Morrison, 1986;Temple and Wiens, 1989). Similarly, Vilches et al. (2013) have used plant indicator species of broad-leaved oak forests in the eastern Iberian Peninsula. Despite the globally growing interest in studying and using of biological taxa as bioindicators to detect environmental changes and determine the causes and consequences of such changes on ecosystems (e.g. Kitching et al., 2000;Davis, 2001;Andersen et al., 2002;Vilches et al., 2013), inappropriate selection and application of bioindicators have put under question the utility of the bioindication concept as a conservation tool (Kremen, 1992;Landres et al., 1998;Carignan and Villard, 2002;Manne and Williams, 2003;Urban et al., 2012). Nonetheless, some authors (e.g. McGeoch, 1998) have provided a step-by-step procedure to be followed to select reliable bioindicator taxa that would be successfully applied for ecological monitoring. According to McGeoch (1998), the first step during bioindicator species identification is clearly defining the specific objectives-i.e. what is to be indicated by monitoring the bioindicators? Once the objectives are defined and potential bioindicator species are identified based on a priori suitability criteria (for detail on such criteria, consequently on birds (Yosef Mamo et al., 2014;Addisu Asefa et al., 2015b). Thus, monitoring the impact of livestock grazing, using bioindicators such as birds, on ecosystem of this grassland has been identified as key priority action by the BMNP management (OARDB, 2007). We chose birds in this study because they, among vertebrate groups of animals, have been a primary focus for most terrestrial applications of the bioindication concept (Mazerolle and Villard, 1999;Niemiand MacDonald, 2004). Overall, the reasons for choosing birds as bioindicators are: (a) relative ease of identification, (b) relative ease of measurement, (c) relatively large number of species with known responses to disturbance and (d) relatively low cost for monitoring (Morrison, 1986;Temple and Wiens, 1989;Mazerolle and Villard, 1999;Carignan and Villard, 2002;Niemi and MacDonald, 2004).
The specific objectives of this study were therefore to: (i) identify potential bioindicator bird species for the low grazing (protected) grassland site based on data collected during the first sampling period (in June 2014); (ii) test the responses of bird species that were initially identified as bioindicators of habitat change to both grazing pressure and grazing-induced changes in vegetation structure; (iii) test the consistency, and thus estimate the reliability, of initially selected potential bioindicator species using data collected in the area during the second sampling period (in November 2014); and, (iv) develop predictive models relating abundance (number of individuals of birds recorded along the sampling units) of bioindicators with habitat parameters which the bioindicators are supposed to be indicator for.

Study Area
The Bale Mountains region is located in the south-eastern highlands of Ethiopia (Fig 1). It is part of the Eastern Afromontane Hotspot Biodiversity area designated by Conservation International (Williams et al., 2004). At the heart of these mountains is the Bale Mountains National Park (BMNP), which is located at about 400 Km southeast of the capital, Addis Ababa (OARDB, 2007). The national park covers an area of 2200 km 2 and ranges in altitude from 1500 -4377m a.s.l. (OARDB, 2007). To date about 78 species of mammals and 278 bird species have been recorded from the Bale Mountains area; of which 17 mammals and 6 bird species are endemic to Ethiopia (Addisu Asefa, 2007Asefa, , 2011. The Bale Mountains area is characterized by eight months (March-October) of rainy season and four months (November-February) of dry season (OARDB, 2007). The present study was carried out in the northern montane grassland area which occurs as a central broad flat valley (between altitudes of 3000 -3150 m a.s.l.) between two mountainous ranges (Fig 1). This grassland has an area of c. 37 km 2 , of which ~15 km 2 falls inside the BMNP boundary (hereafter referred to as light grazing site). This site is relatively well-protected from illegal livestock grazing. The remaining area of the grassland falls outside the park boundary and is being used as a communal livestock grazing land by the surrounding local community (hereafter referred to as heavy grazing site) (Fig 1; see also OABRD, 2007). On the average (mean ± S.D.) 1528 ± 86 heads of livestock (cattle and horses) has been reported to use this heavy grazing site every day (Yosef Mamo et al., 2014). Artemesia afra and Helichrysum splendidum) (OARDB, 2007). The extent of the open grasslands is ~4 and 12 km 2 and of marsh grassland is 5 and 5 km 2 , respectively, in the light and heavy grazing sites. Shrublands (~5 km 2 ) are completely destroyed in the heavy grazing site and currently occur only in the light grazing site (Hillman, 1986;Yosef Mamo et al., 2015).

Data Collection
We used our previously published data on abundance and occurrence data of birds in the study area (Addisu Asefa et al., 2015b) both for initial identification and subsequent testing of bioindicator species. Bird data were first collected in June 2014 during the wet season (hereafter referred to as the first sampling period) along systematically established 28 transects (14 each in the light grazing and heavy grazing sites and at a minimum distance of 300m apart) of each 1-km long. The start and end geographical coordinates of each transect were saved in Garmin GPS unit to ensure same transects were repeated during the dry season, which was undertakenin November 2014 (hereafter referred to the second sampling period). We undertook the second sampling work to test whether the species identified as bioindicators from the first sampling period would indeed be consistently appeared to fulfil the selection criteria. According to the recommendations of Weaver (1995), Majer andNichols (1998), andMcGeoch et al. (2002), the dataset to be used for such consistency testing should come from samples taken at different environmental conditions (e.g., sampling the same area during different seasons) compared to the samples taken for the initial bioindicator species identification. Thus, we collected the two datasets during different seasons based on this recommendation, as it would enable us to finally retain only subset of species, among initially identified potential ecological bioindicators, that would be effectively applied for the intended ecological monitoring (McGeoch et al., 2002).
During both sampling periods, birds were counted within 50 m width on both sides of each transect. Transects were surveyed randomly and only one transect was surveyed per day. Bird surveys were undertaken early in the morning (between 07:00-10:00) when birds are thought to be more active, while slowly walking at speed of ~2 km hr -1 . Aerial feeders (raptors, swallows, and swifts) and wetland birds were not recorded as the primary objective of the study was on terrestrial birds. For list of species recoded in each site during each sampling period, see table 1 in Addisu Asefa et al. (2015b).
Data on six habitat parameters (heights and percentage covers of three plant functional forms [shrub, herb, and grass]) were also recorded within four 10 m × 10 m quadrates established along each transects at 200 m distance intervals (Addisu Asefa et al., 2015b).To determine heights of each plant functional form, four different measurements were taken at each quadrate (totalling to 16 measurements per transect) using a labelled measuring stick and cover was visually estimated (Newton, 2007).

Data Analysis
A given bird species was considered to be potential bioindicator of habitat condition and used for long-term monitoring of the impact of livestock grazing on vegetation structure in the light grazing site of our study area if it: (i) fulfils a priori suitability selection criteria, (ii) show clear response to disturbance, (iii) consistently fulfils again the a priori suitability selection criteria based on independent dataset collected from same sites during the second sampling period (in different season) and (iv) shows strong positive correlations with the habitat variables which it was supposed to be indicative(see also Kremen, 1992;McGeoch, 1998;Hilty and Merelender, 2000). We tested each of these assumptions in a step-by-step fashion as follows.

Initial Identification of Bioindicators
We initially identified potential bioindicator bird species for the light grazing sites, based on data collected during the first sampling period (i.e. wet season data), in two steps process. First, we For the purpose of this study, we regarded those species with maximum significant IndVals >60% in a given site as potential bioindicator species for that site.
Then, we used additional a priori suitability criteria to refine the selection processbecause the IndVal analysis approach provides information only on some aspects of bioindicator species' niche (e.g. habitat specialty and fidelity) (Dufrene and Legendre, 1997; McGeoch and Chown, 1998), but there are other additional properties -i.e., traits/characteristics which a given potential bioindicator taxa should possess if it is to be considered as reliable that such species should also fulfill to be regarded as a reliable bioindicators (Kremen, 1992;Hiltyand Merelender, 2000;Manne and Williams, 2003). Among such species-specific properties which we used as additional a priori suitability selection criteria were whether the potential bioindicator species: i) has a clear taxonomic status, i) is a non-migrant, with wide distribution (national, regional or global distribution), iii) is easy to find and measure (i.e. high abundance) (McGeoch, 1998;Hiltyand Merelender, 2000). Species identified as potential bioindicators based on the IndVal analysis were therefore refined using these a priori suitability criteria based on species-specific information obtained from Redman et al. (2009) and Bird Life International (2015).

Relationships between Grazing Level, Habitat Structure and Bioindicators
We tested the responses of both habitat variables and bioindicators (identified for the light grazing site from the first sampling period) to grazing pressure. We used the summed abundances of all species identified as potential bioindicators, rather than individual species' abundance, as an input for these analyses, following De Cáceres et al. (2010). Using the summed abundance is advantageous to minimize dependence on individual species and to improve confidence by basing conclusions on a wider array of responses than on response of individual species (Hilty and Merenlander, 2000;McGeoch et al., 2002). These analyses were undertaken using Generalized Linear Mixed Models (GLMMs) with normal distribution and identity link function in SPSS version 20 (IBM Corporation, 2001). In the models, vegetation attributes (height and cover), and abundance of the bioindicators were entered as dependent variables, while grazing level (light vs heavy grazing) as fixed factor and site (light vs heavy grazing sites) identity as a random factor to account for potential independence of transects within a site (Quinn and Keough, 2002). We also examined the relationship of habitat variables with abundance of the bioindicators within the light grazing site using a linear regression model. As most habitat variables had showed co-linearity between each other, we undertook PCA and used the first two component axes that explained 82% of the variation in the dataset for the regression modelling (for detail on the correlation between each pair of the variables and between them and the PCA components, see Appendix A and B).

Testing Consistency of the Bioindicators
Using independent data collected during the second sampling period (in November 2014) from same transects along which the first dataset was collected, we tested the consistency of species initially selected as potential bioindicators from the first sampling period. The IndVal analysis method and the other additional suitability criteria used for the initial selection (see Bioindicator selection above) were followed to identify bioindicator species from this independent dataset collected during the second sampling period. Those species that were initially selected from the first sampling period data as potential bioindicators were considered to be reliable bioindicators if they attained again IndVals of >60% based on the dataset of the second sampling period. Thus, the selection process was refined whereby only a subset of species that showed consistency across the two sampling periods were finally retained as robust bioindicators.

Application of Bioindicators for Ecological Monitoring
To assess the potential application of species-those species that showed consistency across sampling periods and thus were finally selected as reliable bioindicators-in ecological monitoring, we tested the predictive power of the bioindicators for each of the six habitat variables in the light grazing site. We then developed predictive models relating each habitat variable that showed strong positive correlation with abundance of the bioindicators. These analyses were undertaken using simple linear regression models, where each habitat variable was treated separately as response variables, while average (from the two sampling periods) of the summed abundance of those four bird species as predictor. We assumed that strong and significant correlation between a given habitat variable and abundance of the bioindicators in the light grazing site (protected area) implies that the bioindicators will be used confidently for longterm monitoring of that habitat variable in the site.

RESULTS
Overall, 33 species (24 and 25 species form the light and heavy gazing sites, respectively) were recorded in the study sites across the two sampling periods. Number of species recorded during the two sampling periods was almost similar between sites, but was 27% fewer during the second sampling period than during the first sampling period in the light grazing site (Table 1).
Nonetheless, 32% fewer individuals were recorded across sites during the second sampling period compared to the first sampling period. Significant difference (P<0.05) within sites in number of individuals between the two sampling periods was revealed only for the heavy grazing site; bird individuals were 3% more in the light grazing site but 47% fewer in the heavy grazing in the second sampling period (Table 1).

Initial Identification of Potential Bioindicators
Based on the IndVal criteria (i.e. IndVal >60%), six species were identified as potential bioindicators from the dataset collected during the first sampling period for the light grazing site (  (Table 2b). All these 11 species had also fulfilled the additional a priori suitability selection criteria, thus were considered as potential bioindicators for their respective site. Note: * = species discarded from the final selection.

Relationships of Grazing Level with Bioindicators and Vegetation Parameters
Of the six vegetation variables considered in this study, only three (shrub height and cover and grass height) showed significant differences between the light and heavy grazing sites (ANOVA, in all cases, F 1,26 = 6.516-32.415, P <0.01). All these variables were in greater values in the light grazing site compared to the heavy grazing site (Fig 2). Similarly, abundance of the bioindicators -i.e. summed abundance of the set of species identified from the first sampling period-was

Consistency of the Bioindicators
From the independent dataset collected during the second sampling period (i.e. dry season data), four species were identified as potential bioindicators in each site, based on the IndVal analysis approach (Table 2a and  were among those species that were initially identified from the first sampling period (Table 2a and b). These results indicate that three-fourth of species initially identified for the light grazing site, but only two-fifth of the species identified for the heavy grazing site, were found to be consistent across the two sampling periods. These consistently selected species were thus regarded as robust bioindicators of habitat condition in the site for which they were supposed to be bioindicators.

Application of the Bioindicators for Ecological Monitoring
Although we identified bioindicators and tested their consistency for both the light and heavy grazing sites, we focused analysis of the application of the bioindicators for ecological monitoring only on those bioindicators selected for the light grazing site. We decided this because our goal was to propose the use of birds as bioindicators for long-term monitoring of habitat change in light grazing site (protected site  Table 3). The coefficients of determination (R) of regression models relating each of these four vegetation parameters against the abundance of the bioindicators in this site were significantly high and ranged between 0.632-0.921 (Table 3).
These suggest that 40-85% of the variations in these vegetation variables in the site were explained by variations in abundances of the bioindicators. Estimated parameters (regression slopes) for these four vegetation variables were significant and indicted positive relationships between habitat variables and abundance of the bioindicators (Table 4). These results, therefore, suggest that the four species finally retained as reliable bioindicators for the light grazing site can readily and successfully be used for long-term monitoring of trends in vegetation structural composition in the site. Table 4. Predictive regression models relating the values of four vegetation variables with abundance of the bioindicators (BI abun ) for the light grazing site in the northern BMNP. Abundance was modelled as average number of individuals of the four species finally selected as reliable bioindicators from the two sampling periods, and vegetation height was expressed in cm and cover expressed in percentage.

Dependent variable Equation
Shrub

DISCUSSION
In this study potential bioindicator bird species were initially identified from a first sampling period based on a priori established selection suitability criteria, and the robustness (consistency) of these initially identified species were tested on independent dataset collected from same sites during the second sampling period. Sets of species were found to be robust bioindicators, i.e. had consistently fulfilled all the bioindicator selection criteria during the two sampling periods in a site. Whereas, certain initially identified potential bioindicator species showed a wide variation in their indicator values (IndVals); these species were therefore considered to be unreliable and thus were discarded from the final suite of species. This testing process has led to refine the selection process by retaining only subset of species that were found to be robust bioindicators and improved the confidence with which the final suite of species may be regarded as reliable bioindicators (McGeoch et al., 2002). Overall, similar to reports of several authors around the globe (e.g. Kitching et al., 2000;Andersen et al., 2002;McGeoch et al., 2002;Vilches et al., 2013), our results provide additional insights into the potential application of the bioindication concept in biodiversity conservation programmes. Our findings, in particular, support previous works of many authors (e.g. Morrison, 1986;Temple and Wiens, 1989) , 1997;McGeoch and Chown, 1998;McGeoch et al., 2002;De Cáceres et al., 2010). However, in views of some authors (e.g. Hilty and Merenlander, 2000) the IndVal approach itself, despite several variants have been developed to overcome such issues, still suffers from some shortcomings, especially if the selection criteria relies solely on the degree of species' IndVals and the objective of the selection is to consider species for bioindicator-based ecological monitoring. Although assessment of species' local abundance and frequency of occurrence (using the IndVal method) is the foremost priority step in ecological bioindicator selection, other species-specific traits that influence their utility as reliable bioindicator should also be accounted for. For example, relying on a migratory species selected, by virtue of its being attaining high degree of IndVal, as bioindicator-based ecological monitoring may lead to erroneous conclusions and wrong management decisions to be made (Hilty and Merelender, 2000;Manne and Williams, 2003). Further, the choice of the minimum threshold IndVals should be achieved by a species to be considered as bioindicator is usually arbitrarily defined based on authors' judgement (De Cáceres et al., 2010). Therefore, in addition to the IndVal criteria, species-specific life-history traits, ecological specializations (diet/habitat) and other relevant traits should be used as secondary selection criteria to refine the selection process. Following such two-step process is essential to retain only suite of reliable species (those species that show consistency in indication power across sampling periods) for effective application of the bioindication concept, thus is advocated for all studies concerned with bioindicator identification and application.
In bioindication studies aimed for ecological monitoring, most methods used for bioindicators selection relies on quantitative (e.g. abundance) and/or qualitative (frequency of occurrence) data collected on species of assemblages, but data collected on such attributes (especially for mobile animals like birds) in a given site could be a result of chance event or sampling errors. Thus, before developing recommendations for their application in ecological monitoring programs, whether attributes of the bioindicators are strongly and significantly related to environmental stressors (e.g. grazing disturbances) and/or with the ecological variables which the bioindicators are supposed to be indicatives should be confirmed (McGeoch, 1998).
As was true in the present study, in addition to its importance to make predictions of future states in some vegetation structures (see discussion below), this testing process improves more the confidence with which the selected potential bioindicator species would be further considered to be appropriate for ecological monitoring (Majer and Nichols, 1998;McGeoch, 1998). However, some authors (e.g. McGeoch, 1998;McGeoch et al., 2002) still suggest that-because of speciesspecific differential responses to variations in spatio-temporal environmental conditionsachieving significant relationships between attributes of potential bioindicators and ecosystem in a given site alone may not be sufficient enough to reliably apply such species for monitoring purposes. Therefore, set of initially identified potential species should only be considered as reliable bioindicator if they show consistency-i.e. selected again as bioindicator species-when tested based on data independent from those used for initial identification, for example by resampling under different temporal or spatial conditions (Weaver, 1995;Majer and Nichols;1998, McGeoch, 1998McGeoch et al., 2002). In this study, such consistency testing has enabled to finally propose subset of species, among initially identified potential ecological bioindicators,  Afework et al., 2009). Consequently, one of the main management objectives of the BMNP has been to reduce the impact of livestock grazing on the extent and structure of shrub and grass vegetation in the grassland ecosystem, and birds were proposed as bioindicators to monitor of habitat change in the area (OARDB, 2007).
Our study therefore directly fits to the conservation management objective of the BMNP, and as such the information obtained and the conclusions drawn from this study will contribute to informed management of the light grazing site of the montane grassland in the BMNP. Applying bioindicator-based long-term ecological monitoring in the site will also compliment information obtained from the on-going ecological and threats monitoring programme in the BMNP (Kinahan, 2010).
In order to practically use the proposed final suite of bioindicators for ecological monitoring in future, we recommend that these species should be counted following same procedures and time of the year-either in June or November, or in both months-used in the grazing) and decide on what management actions should be taken to mitigate the stressor and its impact on the ecosystem.

CONCLUSION
From the study, it is demonstrated that how to: (i) select reliable ecological bioindicator bird species; (ii) use birds for scientifically rigorous bioindicator-based ecological monitoring programme in a given area; and (iii) transform information derived from such bioindicator-based ecological monitoring programs into direct practical application (i.e., for making informed conservation management decision). In addition, the results provide valuable information for effective management of the MBNP. The methods followed in this study can serve as a showcase which can be adopted by researches interested in the study and application of bioindicator-based ecological monitoring systems in protected areas.

ACKNOWLEDGMENTS
We thank Daniel Tilaye and Shubbisa Godana for their assistance during the field work. This study was conducted with a financial support provided by Wondo Genet College of Forestry and Natural Resources/Hawassa University, Ethiopia. We also thank the Ethiopian Wildlife Conservation Authority and the Bale Mountains National Park for the research permission provided.