Using indicator species to detect high quality habitats in an East African forest biodiversity hotspot

East African dry coastal forest consists of three forest types: Cynometra forest (further subdivided into Cynometra thicket and Cynometra woodland), Brachystegia forest, and Mixed forest (formerly named Afzelia forest or Hymenaea-Manilkara forest; Muriithi and Kenyon 2002). The Arabuko Sokoke forest in southern Kenya is the largest remaining dry coastal forest block in East Africa, with a total extent of 41,600 ha (Wass 1995; Bennun et al. 1996). Considered an Important Bird Area (IBA), the Arabuko Sokoke forest is listed as the second most important site for bird conservation in Africa due to the high number of bird species, including many forest endemics (Collar and Stuart 1988; Bennun and Njoroge 1999).

We performed radio-tracking of O. ireneae during the transitions between the Wet and Dry Seasons, in July and August 2016–2017. Ten individual owls (all adults, gender could not be determined) were caught with mist-nets set around their roosting trees. We banded each owl and measured their head, wing, tarsus, and body weight. We then attached a radio transmitter (Pip Ag376; Biotrack Ltd, Wareham, UK, with a weight of 1.5 g) as a back-pack with a "wing-loop" harness made of natural rubber to each of the owls. The weight of the radio transmitter was less than 2% of the total body weight (mean 50 ± 1.7 g), well below the commonly used threshold (4–5% of body weight; see Kenward 2001).

We conducted radio-tracking with hand-held receivers (AR8200-MK3, AOR Inc. USA) and four-element Yagi antennas. Two observers simultaneously took bearings of tagged owls every 20 min from 6 pm until 6 am using a compass, recording their own position with a hand-held GPS device (Garmin GPSmap 64x). The two observers maintained positions of at least 300 m distant from each other. Data collection was split into two 6 h periods (6 pm till midnight; midnight till 6 am). During the early morning, the tagged owls were followed in order to locate their roosting trees. The exact position of each owl was calculated by triangulating the bearings of both observers. Positions calculated for the first day after tagging were excluded due the potential for atypical movement behaviour. Positions with distances > 1000 m were excluded from the data set as the transmitters had a maximum range of 1000 m. The roosting trees of each individual were identified, and subsequently geographically determined.

Home range sizes of O. ireneae individuals were estimated using Minimum Convex Polygons (MCP) and Autocorrelated Kernel Density Estimation (AKDE) for 95% (transitional area) and 50% levels (core area). The MCP home range sizes enabled us to compare our data with the results of previous studies (e.g. Virani 1994). The AKDE method allowed for the more precise estimate of particularly small home range sizes, compared with conventional Kernel density estimators, as it corrects estimates for temporal autocorrelation (here, 20 min intervals; Noonan et al. 2019). The AKDE home range size were calculated by fitting continuous-time movement models (ctmm), which were visually inspected in order to assess the autocorrelation structure of the movement data and selected using the Akaike Information Criterion, corrected for small sample size (AICc). From the best model, the AKDE utilisation distribution and home range sizes were calculated (Fleming and Calabrese 2017). Incremental area analysis was then conducted in order to determine whether home range size reached an asymptote, thus indicating that the number of locations was sufficient. In this study, home range estimates reached an asymptote after 180 to 200 relocations. The analyses were performed using the “Animove-Triangulation” package in QGIS v. 2.0.1; MCP home range estimators were calculated with the R package “adehabitatHR” (Calenge 2006); and AKDE estimators were generated with the R package “ctmm” (Calabrese et al. 2016).

SDMs were created on the basis of calculated positions (fixes) obtained from radio-tracking and the following environmental variables: tree cover, tree height, Normalized Difference Vegetation Index (NDVI), spectral information from blue, Mid-Infrared (MIR), Near-Infrared (NIR) bands, and vegetation type. These environmental variables were selected following observations of the species´ habitat characteristics and behaviour (see Cuadros-Casanova et al. 2018) and are based on land cover data derived from satellite imagery (sources detailed in Table 1). These were resampled to a resolution of 30 m using a nearest neighbour approach (package “raster”, Cran R, Hijmans 2016). Maximum Entropy modelling (MaxEnt v. 3.4.1) was employed to perform SDM (Phillips et al. 2006) using a Jackknife test in order to measure variable importance in the model development. Model performance was evaluated by examining the area under the receiver operating characteristic curve (AUC), where values of 0.5–0.7 were considered low and thus represented poor model performance, whereas values of 0.7–0.9 were considered moderate, and values > 0.9 represented excellent model performance (Phillips et al. 2006). Based on the assumptions that the home ranges of individuals, i.e. pairs, do not overlap, and that O. ireneae primarily occurs as pairs (i.e. two individuals per home range), the following total pair numbers were estimated by combining the home range estimate per individual (E/I) with the potentially suitable habitat size (SHS). Following a bootstrap approach, 20% of the fixes were randomly selected as test records, while the remaining 80% were used to train the model. This procedure was repeated ten times and averaged over all repetitions. The lowest ten percentile training omission threshold accounting for potential spatial triangulation errors was selected as the presence-absence threshold. All computations were performed in QGIS v. 2.0.1, and MaxEnt v. 3.4.0 (Phillips et al. 2006). Finally, the carrying capacity for O. ireneae was estimated based on the spatial expansion of potentially suitable habitats obtained from SDMs, and the mean home range size of the AKDE calculations.

The population of O. ireneae in the Arabuko Sokoke forest occurs at much higher densities than the only other population for which density measures exist—the Usambara Mountains where there were 1.5–4.0 pairs/km² over an area of 9700 ha (Evans 1997). Nevertheless, the average home range size we obtained for tracked O. ireneae based on AKDE95 (18.9 ± 2.2 ha), which is thought to be particularly accurate (Noonan et al. 2019), continue a long-standing increasing trend: Home range size estimates have gone from 12.5 to 14.3 ha/pair in 1973–1979 (Britton and Zimmerman 1979) and 1984 (Kelsey and Langton 1984) to 14.9 ha/pair in 1994 (Virani 1994), 16.1 ha/pair in 2003 and 17.5 ha/pair in 2008 (Virani et al. 2010). The increase in home range size is reflected in the estimate of the population size: 1025 pairs in 2000 (Virani 2000), 800 pairs in 2008 (Virani et al. 2010) and 387–544 pairs in our study. While such differences in population densities and sizes could be a result of differences in data collection techniques, data quality or in the algorithms applied to estimate home ranges, they could also be caused by a decline in habitat quality.

Increasing home range sizes may be a response to decreasing habitat quality with reduced resources (e.g. food). Wiktander et al. (2001) showed that home range sizes decrease when resources become scarce (e.g. during the dry season). Such environmental changes may also be due to anthropogenic habitat deterioration, which influences habitat structures, species communities, and the availability of resources. This, subsequently impacts individual behaviour (Villard et al. 1999; Banks et al. 2012, 2017; Ndang'ang'a et al. 2013). Previous studies have shown that birds living in degraded environments need to cover larger areas to find suitable food sources when compared with birds living in high-quality habitats, i.e. in non-fragmented environments (Carey et al. 1990; Hansbauer et al. 2008). This might also be the case in the Arabuko Sokoke forest, where habitat quality is decreasing due to selective hardwood logging, charcoal production, and the removal of dead wood (Glenday 2005; Virani et al. 2010; Matiku et al. 2013).

In addition to anthropogenic activities, changes in the density of mega-herbivores such as elephants can also impact ecosystem structure (Villard et al. 1999; Ndang'ang'a et al. 2013; Banks et al. 2012, 2017), species composition and populations of single taxa (Fanshawe 1995; Clergeau and Burel 1997; Barros and Cintra 2009). Previous studies have shown that elephants increase the accumulation of dead wood and the availability of elephant dung, leading to a rise in insect activity and changes to arthropod community composition which then affects higher trophic levels, such as birds (Otieno et al. 2014). In addition, elephants may destroy large trees, which provide important breeding sites to O. ireneae.

Our SDMs indicated that tree height (availability of large old trees), NDVI (biomass), and vegetation type (Cynometra woodland) are the strongest explanatory variables for predicting O. ireneae occurrence within the Arabuko Sokoke forest. The SDM demonstrated the restriction of O. ireneae to the Cynometra woodland forest type, a finding in line with the study by Virani (1994) on O. ireneae habitat demands in the Arabuko Sokoke forest. Our data further indicated several roosting places per owl, always located in the centre of the AKDE50 area. Owls return to these roosting places after nocturnal foraging, using different roosting places in an alternating manner. All roosting places were characterised by high Cynometra trees and were covered by dense lianas and vines (Camilo Zamora & Ivon Cuadros-Casanova, personal observations). This again underlines the importance of undisturbed Cynometra woodland to this sensitive bird species.

Our SDMs reveal that O. ireneae is neither homogeneously distributed across Arabuko Sokoke forest, nor homogeneously distributed across Cynometra woodland (a fraction of Arabuko Sokoke forest; Fig. 2). The only suitable habitat in the Arabuko Sokoke forest is Cynometra woodland with dense vegetation and large, old trees and a high proportion of deadwood. Decreasing habitat quality may reduce the carrying capacity of O. ireneae, and may also negatively impact other forest specialist species. With a mean body weight of 50 ± 1.7 g, O. ireneae is the smallest owl of East Africa, and its colourful plumage makes it a very charismatic species. Along with being an indicator of intact Cynometra woodland forest, makes this owl an excellent flagship species to promote and conserve this endangered biodiversity hotspot.