Aquatic turtles and crocodilians have a long and complex history of ecological coexistence. For example, many species of crocodilians routinely prey on turtles (Pope 1949; Cott 1961; Platt et al. 2006; Heithaus et al. 2008), and turtles can constitute the dominant prey remains in American alligator (Alligator mississippiensis) stomachs (Delany and Abercrombie 1986; Janes and Gutzke 2002). Crocodilians also exhibit positive allometry of bite force throughout development (Erickson et al. 2003), and this may make them particularly potent predators of many turtle species. Even so, some turtle species appear to benefit from the presence of crocodilians; a comprehensive analysis in Florida found that 26.6% of alligator nests were used by a variety of turtle species as nesting sites and suggested that female alligators may provide incidental protection to conesting turtles (Enge et al. 2000). These varied, sometimes opposing ecological relationships imply that predator avoidance mechanisms in turtles may be complex, species specific, and dependent on local conditions.

Although it is generally recognized that the chelonian shell has evolved as a protective antipredator defense mechanism, morphometric analyses have shown that different shell geometries confer more or less protection from large predators that crush their prey, including crocodilians. A recent Finite Element Analysis demonstrated this point among 3 species of North American emydine turtles that varied in shell shape. Relatively higher-domed and thicker-shelled Glyptemys muhlenbergii were more resistant to crushing than more dorsoventrally flattened Emys (Actinemys) marmorata, which in turn were more crush resistant than Clemmys guttata, the least domed and thinnest shelled of the 3 species (Stayton 2009). Similar reasoning, but with less quantitative support, has been applied to other turtles with respect to specific predators. For example, the relatively high-domed thick shell of Pseudemys concinna in the coastal southeastern United States (below the “Fall Line”) has been postulated to have evolved as a response to American alligator predation (Aresco and Dobie 2000). Alternatively, turtles with low-domed, easily crushed thinner shells in the same areas may instead utilize behavioral antipredator strategies, including hiding in refuges, crypsis, or increased speed and agility (Greene 1988). Although we know that this pertains in other taxa (Caro 2005), to our knowledge this trade-off between morphological and behavioral antipredator tactics has never been formally explored in wild turtles.

We used a comparative field sampling approach to examine the relationship between turtle abundance and crocodile presence/absence during a recent survey of aquatic turtle populations in several localities in Tanzania, East Africa, an area with an ancient history of co-occurrence between turtles and crocodilians (O'Connor et al. 2006). We trapped 2 species of pelomedusid turtles across a range of aquatic habitat types and conditions, including those that routinely have large populations of the Nile crocodile (Crocodylus niloticus), a large predator that preys on turtles (Cott 1961, Wallace and Leslie 2008). Pelusios subniger is a small, dorsoventrally flattened turtle that is abundant across much of East Africa (Spawls et al. 2002). Although individuals may reach 20 cm in carapace length (Ernst and Barbour 1989; Spawls et al. 2002), most adults average around 15 cm (Spawls et al. 2002; HBS, unpubl. data, 2009). Cott (1961) found that P. subniger was common in the stomachs of Nile crocodiles in Uganda and Northern Rhodesia (now Zambia); his extensive study showed that 15 crocodiles from 5 of 8 sites had remains of 16 individual specimens in their stomachs, making P. subniger one of the most common tetrapod prey species. Given its low-domed, modest shell proportions, we predicted that P. subniger would favor pools and watercourses not frequented by Nile crocodiles. We further predicted that its congener P. sinuatus, a large (up to 55-cm carapace length; Ernst and Barbour 1989; Spawls et al. 2002), high-domed species that appears to have a relatively crush-resistant shell, would occur more frequently in water bodies that contain Nile crocodiles.

Methods

Research was carried out primarily in Rukwa Region of western Tanzania in and around Katavi National Park but also between Kipili and Karando near the eastern shore of Lake Tanganyika during the wet season in February 2009. We sampled a total of 78 sites during 4 weeks of field work, where a site is defined as an individual trap or seining location. Based on TC's extensive knowledge of the Katavi landscape, we trapped for several days at all of the permanent and seasonal aquatic habitats in the national park and surrounding areas that were holding water.

We set turtle traps in isolated vernal pools, pools temporarily connected to flowing water, and in bends, eddies, and backwaters of flowing rivers where water movement was sluggish. Our turtle traps were modified, collapsible crab pots (87 cm long, 55 cm wide, 26 cm tall, enclosed with 3-cm nylon mesh netting) that were baited with local dried catfish, set, and left without disturbance for 28–72 hours. In water that was less than 25 cm deep, traps were placed directly on the bottom near logs, snags, and emergent vegetation. In deeper water we used modified traps that also have a 2-m-long, internally baffled, 1-cm mesh “chimney”. For these, the trap was placed on the bottom adjacent to snags or potential basking sites and the chimney was hung from emergent vegetation. The chimney allows trapped turtles access to air, and the baffles prevent escape of captured turtles. Although several traps were often set along a single stretch of waterway (a river, for example), they were generally set 100–300 m, and frequently much further, apart. We also used a 5-m-long, 1.5-m-tall, 3-mm mesh seine net to sample 13 shallow, permanent and temporary pools and ponds. Two individuals would drag the net repeatedly across the pool until either 50% of the surface of the pool had been sampled or at least 10 turtles had been captured. Crocodile presence at each location was determined by observing one or more crocodiles during at least 15 minutes of direct observation by 3–4 observers prior to HBS and an assistant entering the water. Crocodile absence was determined by observing no crocodiles at a site during the observation period and by additionally noting that crocodiles had never been observed by TC in these locations during the previous 15 years' field work. Although crocodiles may have occasionally gone undetected, Katavi National Park fully protects its crocodiles, and they tend to be both bold and visible throughout the park.

In analyzing our data, we felt that the spatial separation of samples was a key aspect of the biological and statistical independence of each observation. We therefore treated each pool that was seined and each individual trap as a single sampling unit. Each site along a river or at a pool was only trapped once, ensuring that the same turtles could not be retrapped. Given this spatial separation, we feel fairly confident that each trap was providing an independent estimate of the turtles in a local patch of aquatic habitat. Traps set in the presence of crocodiles were placed along 3 major watercourses > 15 km apart. Given the size and mobility of crocodiles, it is possible that the same crocodiles could have influenced traps that were several hundred meters apart within a site, but the 3 watercourses were completely independent sampling units. The vast majority of traps were set for 2 consecutive days and nights although a few were set for 3 days and nights.

Results and Discussion

We caught a total of 50 P. subniger and only 1 P. sinuatus in Rukwa Region, and we limit our quantitative analysis to the former species. We caught 1 or more P. subniger in a greater proportion of sites that contained no crocodiles (10 out of 33 sites) than in sites where crocodiles were present (2 out of 45) (χ²  =  5.889, df  =  1, p < 0.01). The same trend was found when sites with some river flow (which may be less optimal habitat for P. subniger in East Africa; Spawls et al. 2002) were excluded from analyses (no crocodiles, 9 out of 27; crocodiles, 1 out of 9) although this result failed to reach statistical significance (Fisher exact test, p  =  0.166). Furthermore, the number of P. subniger captures per trap night (defined as a 24-hour trapping period) was approximately 8 times higher in areas without crocodiles (0.48 turtles per trap night, n  =  40 trap nights) than in areas with crocodiles (0.06 turtles per trap night, n  =  81 trap nights). Likewise, seining success in pools with no crocodiles (3.71 turtles per seined pool, n  =  7 pools) was greater than pools containing crocodiles (0 in 6 pools).

Katavi National Park is famous for its large population of Nile crocodiles (Katavi National Park General Management Plan 2002) although these are patchily distributed throughout the park. This allowed us to test the prediction that low-domed, thin-shelled P. subniger would be found in microhabitats that are not occupied by crocodiles. Although not conclusive, our small dataset supports this prediction. Given that our data are purely observational, we cannot yet distinguish between active behavioral avoidance of crocodile-occupied habitat by P. subniger and successful predation by crocodiles in the habitats where they co-occur. We also cannot exclude confounding ecological variables that may affect both crocodiles and P. subniger, such as alternative prey abundance, water quality, or frequency of disturbance. However, at least 2 factors suggest that behavioral avoidance, rather than elimination by crocodiles, may be important. First, we captured 8 times as many P. subniger in ponds that were free of crocodiles but that were frequently adjacent to crocodile-inhabited watercourses. Given that these small ponds are within the home range of crocodiles, but were not occupied by them when we were trapping, suggests that P. subniger may be opportunistically choosing sites that are crocodile-free, or at least infrequently occupied. Second, we captured 2 P. subniger in habitats occupied by crocodiles, suggesting that Nile crocodile predation does not entirely eliminate turtles from a given site.

Additionally, we caught the congeneric P. sinuatus, a turtle with a high-domed armored carapace, in both Lake Tanganyika and the Wami River in Eastern Tanzania where Nile crocodiles are also found. At the Wami River site, we trapped 6 P. sinuatus in 14 trap-nights but caught neither P. castanoides nor Pelomedusa subrufa, the 2 smaller, low-domed species that also occur in this area of Tanzania. Moreover, in Lake Tanganyika, local fishermen caught one adult P. sinuatus and confirmed that Nile crocodiles are common at this site. When we seined small waterways immediately adjacent to Lake Tanganyika where crocodiles are absent, we found P. subniger, but not P. sinuatus. Cott (1961) also found much lower predation by Nile crocodiles on P. sinuatus compared to P. subniger, with only a single P. sinuatus found in the stomach of 1 crocodile. Again, these results are purely observational, and confounding effects of habitat preference and other environmental conditions may be driving these associations (note also that Broadley and Boycott 2009 stated that P. sinuatus can be heavily depredated by Nile crocodiles). However, these observations are consistent with the interpretation that congeneric turtle species may be differentially susceptible to crocodile predation as a result of their morphologies, and that the micro-distribution of crocodiles may partially explain both the local distribution and behavioral defenses of different Pelusios species.