Rapid human population growth in Ethiopia has resulted in the degradation of vast areas of wildlife habitats due to agricultural expansion, infrastructure development, and urbanization. The Omo Valley in the southwestern part of Ethiopia has been particularly affected by land use changes, but despite its ecological importance, few relevant studies have been conducted there in the last two decades. Our aim is to provide updated and scientifically verifiable information for medium and large terrestrial mammal species richness and community structure in four Protected Areas in the Omo Valley. We used bycatch camera trap data from a large carnivore survey and nonparametric incidence-based estimators for data analysis. A total of 52 mammals from nine orders and eighteen families were recorded, of which approximately 29.4% are listed as globally threatened and one as an endemic subspecies. We present the current species lists and compare them with historical records and observed the highest species number in Omo National Park, even though nine species were no longer recorded there. We applied the Morisita-Horn similarity index to reveal a high degree of overlap in mammalian species among adjacent Protected Areas, but less overlap between Protected Areas far from each other, indicating distance decay of similarity. The distribution of feeding guilds was significantly different across Protected Areas, and carnivore detection frequency was relatively low in Tama Community Conservation Area compared to our other study sites. This study confirmed the conservation importance of the area in terms of mammalian diversity, albeit with low detection levels, especially of large carnivores, underscoring the importance of promoting landscape connectivity to maintain population viability across the Omo Valley. From our experience, the use of camera trap bycatch data proved to be effective in surveying large- and medium-sized mammalian species, but less so in capturing the rarer species in the area.

Camera trap surveys for the purpose of the large carnivore study took place from 2020 to 2022 in CCNP, MNP, TCCA and ONP for a two-month period each.  Due to the large carnivore target the grid size was informed by the smallest home range sizes for carnivores in the regions (Asfaw et al., in review). We used a mixture of Bushnell Trophy Trail Camera 119717cw (Bushnell, California, USA), Minox DTC 550 (Minox, Wetzlar, Germany), Dorr Snapshot Trail Camera (DÖRR, Hamburg, Germany) and Rollei WK cameras (Rollei, Hamburg, Germany). The camera trap survey lasted for 10301 camera days (2799 in ONP; 3089 in TCCA; 717 in MNP and 3696 in CCNP)

We used Quantum GIS (QGIS) 3.24 Vector Grid Research Tool (Quantum GIS Development Team, 2021) to divide each protected area into 25 km² grid squares. The center point of each grid cell was then determined using the QGIS Polygon Centroid Geometry Tool (Quantum GIS Development Team, 2018) from which the geographic coordinates were extracted and uploaded onto a handheld GPS device for use in the field.

Single un-baited movement activated cameras were then placed within a 100 - 800m radius of the center of each of these cells, in a location that optimized detection such as close to water or wildlife spoor, scat or trail and away from anthropogenic activity. In total 137 cameras (from 19 to 46 in each site, based on the area coverage of the study site) were placed for a minimum of 60 continuous days and effectively retrieved. On placement we trimmed vegetation and grass, without altering the immediate habitat, within the camera's detection zone to reduce false trigger events.

All camera trap images were then identified to the species level using the Field Guide to African Mammals as a reference (Kingdon, 2015); all pictures other than large to medium sized mammals were avoided from this study. DigiKam (https://bugs.kde.org/) and exiftoolr (https://exiftool.org) programs were used for image tagging and metadata extraction, respectively. Datasets were organized and managed for statistical analysis in R studio using camtrapR package (Niedballa et al., 2016). The images were filtered according to species using 5-minute intervals to ensure independence and a presence/absence data frame was created for all detected species.