This study was conducted in a dense evergreen rainforest located between 550 and 950 meters a.s.l. in the wilderness reserve of Mont Panié (20°37'30"S, 164°46'56"E, 5400 ha) in New Caledonia, South Pacific (Fig. 1). This area has been protected since 1950 on the basis of the cultural importance of Mont Panié and the high micro-endemism rates observed for most taxa (plants, insects, reptiles) (Tron et al. 2013). The climate is moist subtropical, with a hot season between December and mid-April and a cool season between mid-May and September. Mean annual precipitations in the Mont Panié region are around 3 000–4 000 mm, while mean temperatures range from about 18 °C to 25 °C throughout the year, and minimum temperatures can approach 0 °C (Tron et al. 2013). Two invasive rat species, Rattus rattus and R. exulans, are present in the study area. R. exulans was introduced in New Caledonia by the first Austronesian colonisers ca. 3 000 years ago, while R. rattus was introduced later by European settlers ca. 150 years ago (Beauvais et al. 2006). Both rat species are omnivorous, eating mainly plants, invertebrates and, to a lesser extent, Squamata, with R. rattus being more frugivorous and R. exulans more omnivorous (Duron et al. 2019).

Figure 1.

Location of the study area in the Pacific, northern New Caledonia. The study was conducted in Mont Panié montane rainforest on two adjacent 20 × 20 m trapping grids: a capture-mark-recapture (CMR) grid (96 live-traps) and a removal trapping grid (209 snap-traps).

The best SECR model had 29.81% relative support and included the effects of group of individuals and session on g₀, and group of individuals on σ (Table 2). The probability of capture g₀ was significantly lower for adult male R. rattus and higher for juvenile R. rattus than for R. exulans, whereas there was no significant difference in g₀ between adult female R. rattus and R. exulans (Table 3). Moreover, g₀ was significantly lower for CMR sessions 2 (December 2014) and 6 (September 2015) than for session 1 (September 2014; Table 3). Parameter σ was significantly lower for juvenile R. rattus than R. exulans, and higher for adult male R. rattus than R. exulans (Table 3, 4). Rat densities varied greatly across CMR sessions, from 9.48 ± 1.84 ind.ha^-1 in July 2015 to 33.61 ± 9.00 ind.ha_-1 in November–December 2014 (Fig. 2). R. rattus adult densities were higher for females than for males during the three first capture sessions and similar during the three last ones (see Suppl. material 1) R. rattus juvenile densities markedly increased from CMR session 1 (September 2014: 0.27 ± 0.27 ind.ha^-1) to sessions 2 and 3 (December 2014: 8.74 ± 2.74 ind.ha^-1; March 2015: 9.94 ± 1.68 ind.ha^-1) and dropped again to low levels (< 4 ind.ha^-1) from session 4 to 6 (May to September 2015) (see also Suppl. material 1). No R. exulans was captured during the first two CMR trapping sessions. R. exulans were first live-trapped during session 3 (March 2015) and their densities remained stable over the three following sessions (Fig. 2; see also Suppl. material 1). Home range sizes (HR₉₅ and HR₅₀) were on average three times as large for R. rattus adult males as for R. rattus adult females (Table 4).

Figure 2.

Density of rats (± SE) in the study area from October 2014 to November 2015. Densities were estimated from the best SECR model (g₀ ~ indgr + session, σ ~ indgr) in the CMR area and from the Zippin removal method in the removal area.

While nine trapping days were required to approach a near zero capture rate during removal session 1 (10 rats were captured at day 9 out of 209 traps), this rate was achieved after only two or three trapping days during the subsequent removal sessions. Rat density at the start of our removal experiment (D_before: May 2015) was estimated at 32.1 ind.ha^-1 (Fig. 2; see also Suppl. material 2). When comparing to the densities before each removal session (D_before: 32.11–7.78 inds.ha^-1), lethal trapping achieved a 80–93% reduction in rat density after each of the four subsequent sessions (D_after: 0.56–2.56 ind.ha^-1) and allowed maintaining low numbers throughout a trapping period of seven months (Fig. 2, see Suppl. material 2). Densities of the two rat species remained constant from July to November 2015 (Fig. 2). A higher R. exulans:R. rattus ratio was observed in the removal area as compared to the CMR area in May and September 2015 (Table 5).

During the last four trapping sessions (May to November 2015), contrasted patterns of rat juvenile proportions were observed between the CMR and removal areas (Table 5): in the CMR area, juveniles were recorded in highest numbers in May and July and then dropped to zero in September, while in the removal area, juveniles were recorded in lowest numbers in May, July and September before reaching highest numbers in November. Sex-ratios were balanced during May and July, but biased towards males both in September in the CMR area and in November in the removal area (Table 5). Heavier adult R. rattus and R. exulans were generally found in the removal area than in the CMR area (Table 5). The proportion of males with well-developed scrotal sacs increased earlier in the removal area (75% in July; Table 5).

In the removal area, average distances of trapped rats to the edge of the trapping grid (Dist_rats) fluctuated greatly during the four removal sessions. In July, six weeks after removal trapping had been initiated, most rats were captured near the edge of the removal trapping grid (Dist_rats ± SD = 11.98 ± 10.34 m), whereas in May, September and November, rat captures were distributed more evenly within the removal grid (Dist_rats = 19.98 ± 7.55 m; 31.81 ± 8.04 m; 22.79 ± 2.24 m, respectively; see also Suppl. material 3). A total of 27 individuals were captured in the CMR area at least once before and once after the beginning of the removal experiment. The barycenters of their home range center coordinates did not significantly shift towards the adjacent removal area in response to rat removal (X-axis barycenter coordinates, Z = 198, p = .84; Y-axis barycenter coordinates, Z = 257, p = .11) (Fig. 3). Furthermore, although rat control efforts appeared in the second best SECR model, this variable did not have any significant effect on σ (β = -0.21, SE = 0.16, 95%CI = [-0.52–0.10]) (Table 2).

Figure 3.

Estimated barycenters of home range centers (HRC) of rats recaptured in the CMR grid. Only HRC of rats recaptured at least once in the CMR grid (n = 27), before (HRC_before) and after (HRC_after) rat removal had been initiated are presented here. Grey and black dots represent individual rat HRC barycenters before and after removal, respectively, while green and red filled circles represent the average HRC barycenters across individuals before and after removal, respectively. Twenty-seven other individuals first captured in the CMR area were recaptured in snap-traps in the removal area (red lines and black stars).

A total of 27 rats equipped with PIT-tags in the CMR area were recaptured in the removal area over the course of the study (Fig. 3; see Suppl. material 4): 6 R. exulans (3 females: 1 adult and 2 juveniles, and 3 adult males) and 21 R. rattus (12 females: 8 adults and 4 juveniles, and 9 adult males). On average, adult male R. exulans and juvenile female R. rattus were the individual categories that travelled the longest distances, with respectively 138.30 ± 79.28 m and 145.01 ± 69.12 m travelled. Nineteen individuals travelled between 50 and 100 m and only six travelled > 100 m before being recaptured in the removal area. The longest distance was travelled by a R. exulans adult male (203.78 m in one day, from the CMR to the removal area).

To simulate scenarios of removal efficiency relative to trapping session duration and project expenditure, we used the average value of parameters g₀ (0.09) and σ (14.6 m) specifically estimated for our wire cage type of rodent live-trap, based on the best SECR model. With this simulation exercise and set of parameter values, we intend to demonstrate the usefulness of the proposed approach to compare the cost-effectiveness of plausible rodent management scenarios; however the approach could be easily adapted to other study systems, once key parameters are estimated for the relevant species and trapping devices.

If we were to ignore local social and economic constraints and assume that each trapping grid could be completely covered within one day, two grid layouts would allow achieving a cumulative rat capture probability ≥ .80 after 15 trapping days (Fig. 4A): layout 1 (15 × 15 m; 1.00) and layout 2 (25 × 25 m; .95), while three other grid layouts would achieve a cumulative capture probability ≥ .50: layout 3 (25 × 50 m; .75), layout 4 (25 × 75 m; .54) and layout 6 (50 × 50 m; .50). In our models, splitting the 200-ha grid into smaller adjacent management units allowed accomodating for these constraints, but in the meantime this strategy faced the risk of being too costly if trapping was carried out for a long period. Interestingly, however, when plotting the cumulative capture probability against cumulative project expenditure (in euros) over a theoretical scenario of 10 persons trapping over 30 days, it became apparent that the highest gain:cost ratio was achieved by layout 1 (.66: 24 900), layout 3 (.57: 22 700) and layout 2 (.63: 25 980) (Fig. 4B). Under this scenario, none of the ten different layouts reached a cumulative capture probability of .80. Given the local social constraints of 10 persons available for 15 days and 4 daily working hours per person, layouts 3 and 4 appeared to be the most cost-efficient, with a cumulative capture probability > .30 for a total cost < 17 200 euros (Fig. 4B). It is to be noted, however, that as layout 1 would require a total of 8 days to complete the initial baiting, this strategy could not be implemented if trappers were only available for 15 days.

Figure 4.

Cumulative rat capture probability over 15 trap-nights and their associated cumulative costs. A cumulative rat capture probability over 15 trap-nights for each of the 10 different grid layouts assessed in this study. Layout 1: 15 × 15 m; layout 2: 25 × 25 m; layout 3: 25 × 50 m; layout 4: 25 × 75 m; layout 5: 25 × 100 m; layout 6: 50 × 50 m; layout 7: 50 × 75 m; layout 8: 50 × 100 m; layout 9: 75 × 75 m; layout: 10 = 75 × 100 m (see full details of grid layouts in Table 1). Note that the asymptote for some of the sparser grids is < 1 (not shown) B cumulative rat capture probability against cumulative cost of the operations over a hypothetical maximum of 30 work-days per session. Dotted lines represent arbitrary capture probability thresholds (.50 and .80) to help read and interpret the figure. Dots mark the maximum effort available within the local constraints that apply in the wilderness reserve of Mont Panié, which is 10 persons working 4 hours a day for up to 15 consecutive days. Note that layout 1 (15 × 15 m) cannot be achieved in 15 days since a complete coverage of the grid with 10 people requires 8 splits and a total of 16 days (1 day for baiting plus 1 day to complete one trapping occasion, for each split).

Rat population biology and dynamics on tropical rainforest islands remain less well understood than in other systems. Our CMR study showed great seasonal variation in rat densities, with a marked peak of ca. 34 ind.ha^-1 during the austral summer that dropped three times lower in winter. In another study conducted in a New Caledonian lowland rainforest, rat densities were estimated at ca. 25 ind.ha^-1 in winter (Brescia 2011). Overall, rat densities in New Caledonia appear relatively high compared to average densities recorded in an Hawaiian mesic forest – R. rattus: 7.1 ind.ha^-1; R. exulans: .3 ind.ha^-1 (Shiels 2010); or in a low altitude New Zealand temperate forest – R. rattus: 6.2 ind.ha^-1 (Hooker and Innes 1995). By contrast, very high black rat densities (up to 65 ind.ha^-1 in summer) have been observed in the dry tropical forests of Europa Island in the Indian Ocean (Russell et al. 2011). In our study, black rat densities fluctuated seasonally, as observed elsewhere in the tropics, probably also in response to seasonal changes in food availability (Harris and Macdonald 2007; Russell et al. 2011). By contrast, densities of Pacific rats remained lower throughout our study, ranging from 0 to 4.3 ind.ha^-1, although a peak of ca. 17 ind.ha^-1 was detected in the first removal trapping session (May-June 2015). This pattern could be explained by negative interspecific interactions (e.g. Russell et al. 2004), in particular the competition release of Pacific rats from the removal of black rats over the course of the nine-day removal session, as indicated by the steady increase in Pacific rat capture rates as Black rat capture rate decreases. Some level of competition for traps may have occurred between species but given that trap saturation was relatively moderate in our CMR grid (average successful traps per session = 57.6%), it is unlikely that competition risk solely explains the observed differences in density between species.

In our study conducted in the Mont Panié wildnerness reserve, black rat home ranges varied between .2 and .7 ha according to age and sex. This is much smaller than what was found in a Hawaiian mesic forest (3.8 ha; Shiels 2010), but more similar to the dry forest habitats of Juan de Nova and Europa Islands in the Indian Ocean (from .3 to .8 ha) (Ringler et al. 2014). Rat home range size is expected to vary widely in space and time in relation to in situ rat densities and the intensity of competitive interactions, the underlying food resource availability, and the timing of rat breeding season (Russell et al. 2011; Harper and Bunbury 2015; Harper et al. 2015). In our study, home ranges did not vary significantly across seasons, despite large temporal variations in rat densities. However, it is possible that we could not detect any seasonal variation in home range sizes due to insufficient recapture data by session. Care should be taken, however, when comparing home range size estimated from spatially-explicit capture-recapture and telemetry location data due to methodological differences and/or sampling error (Nathan 2016). Although SECR-derived σ has been shown to underestimate telemetry-based home range radius, it still remains a useful estimate of space use as it concurrently incorporates movement and rat-trapping device interaction behaviours.

Controlling rats for 5–9 consecutive nights every five to seven weeks over a 5.5-ha area (on a 20 × 20 m trapping grid) allowed reaching an initial 16-fold density decrease (from 32.1 to 2.6 rats.ha^-1), followed by a five to fourteen-fold decrease (down to 0.6–1.6 rats.ha^-1) after the following sessions, with densities going back to pre-removal levels (ca. 8 rats.ha^-1) in-between sessions. This indicates that trapping every five to seven weeks is not sufficient to reach near zero rat densities in the Mont Panié area. In a mesic forest in Hawaii, rat removal efforts deployed over a 26-ha area (25 × 50 m grid with trap spacing of 12.5 m; traps checked daily for two weeks, then every two weeks) allowed maintaining rat numbers ca. 3 times lower (ca. 1 rat/100 trap.night) than the initial pre-removal state, and was shown to enhance the reproduction of an endangered endemic plant Cyanea superba subsp. superba released from rat predation pressure (Pender et al. 2013). In Australia, a rat removal experiment conducted with live traps (first intensively for two months; then with three-days-long capture sessions every 4–8 weeks) maintained black rat densities 3.3 times lower (3.7 ind.ha^-1) than untreated sites, and was associated with a substantial increase in the populations of native rat R. fuscipes (Stokes et al. 2009). These examples, with many others, show that rat control operations can be highly beneficial for native biodiversity (Duron et al. 2017), although the optimal level of management required to ensure the long-term persistence of vulnerable prey is often not known. Monitoring native biodiversity while supressing rat numbers would help identify rat density thresholds below which positive effects on biodiversity can be detected, and test whether maintaining rat densities around 8 inds.ha^-1 in the Mont Panié area would be enough to protect the most vulnerable native species. In addition to improving the knowledge base, monitoring how native systems respond to the removal of alien predators is crucial in order to detect unexpected outcomes (Ruscoe et al. 2011; Walsh et al. 2012) and may help convince both funding bodies and participating local communities of the relevance of continuing the operation (Ferraro and Pattanayak 2006).

A reduction in rodent abundance may be followed by a rapid reinvasion, induced by enhanced immigration and/or increased breeding and survival of remaining adults and juveniles (Innes et al. 1995; Gundersen et al. 2001; King et al. 2011; Hansen et al. 2020). In New Zealand, following the eradication of eight forest fragments averaging 5.3 ha in size, black rats were shown to travel up to 600 m in a few days and reinvade cleared areas very quickly despite large areas of intervening non-preferred habitat (King et al. 2011). In a Vietnamese farming system, the populations of Rattus argentiventer and R. losea were shown to respond to pest control mainly through an increase in recruitment, immigration and survival of residents, but not through enhanced breeding productivity (Brown and Tuan 2005). A recent study conducted on black rats in Australia showed that reinvasion of controlled areas was slow and occurred both through immigration and settlement by different categories of individuals: juveniles, adults or sub-adults without permanent home ranges, and individuals occupying neighboring areas expanding their home ranges (Hansen et al. 2020).

In our study, the level of control efforts applied (5–9 trapping days every 5–7 weeks) appeared to have prevented rapid, complete reinvasion of a 5.5-ha forest area. As our removal trapping only covered the austral winter, it is possible that reinvasion rates would have been higher during the subsequent summer period. We found, however, some evidence of rats travelling from the CMR area to the adjacent removal area already on the first removal trapping day, and then throughout the trapping period. Moreover, six weeks after the initial knock-down of the rat population, captures were mainly recorded along the edge of the removal area. Nevertheless, it remains unclear whether these scattered movements and spatio-temporal patterns of captures were driven by immigration, were a consequence of the expansion of existing home ranges in the adjacent CMR area, or simply reflect the higher capture probability of neighboring individuals dwelling in the near vicinity of the removal area. Our analyses of rat home range barrycenters and size (sigma) did not reveal any marked shift in space use nor any enlargement of home ranges within the trappable rat population of the CMR area in response to ongoing control efforts. Our data also suggest that some reestablishement may have been driven by density-dependent enhancement of breeding: a larger proportion of reproductively active males as well as young juveniles (R. rattus < 52 g) were detected at an earlier stage in the removal area (July 2015) compared to the CMR area. It is unlikely that these sex and age distribution patterns would have been explained by seasonal differences in trappabilities within individual rat groups since sex-and age-specific capture probabilities remained relatively constant over the course of the removal trapping experiment (see Suppl. material 2). Given the low numbers of rats captured in each of the three removal trapping sessions following the initial knock down, if enhanced breeding had occurred, it probably did not contribute to a substantial or sustained recovery of the controlled population.

Dispersal is commonly observed to be male-biased and principally realized by juveniles in most rodent species, including Rattus spp. (Brown and Tuan 2005; King et al. 2011; Krebs et al. 2011). Had immigration occurred in our removal area, we did not find any evidence for male rats being involved in the recolonization process more than females: the sex ratio observed in the removal area remained fairly balanced and constant as control efforts progressed, except during the last removal trapping session, six months after the initial knock down, when males represented 60% of captures. In addition, we did not find any support for juvenile-driven dispersal: juvenile numbers were very low in the removal area during each of the four removal trapping sessions. Without further evidence on the movements and origins of the rats captured in controlled areas (inferred from e.g. genetic analyses, rhodamine-B marking experiments), it remains unclear as to what processes typically govern rat population reestablishment in such contexts.

Our modelling exercise indicated that it is, in principle, realistic to control invasive rats over moderate-sized areas in challenging environments, using lethal trapping. For example, in the absence of local and economic constraints, a one-shot reduction in rat density of 93% was achievable over 200 ha on a 15 × 15 m grid in a total of 1585 work hours or 5 trapping days (and a 100% reduction in 3169 work hours or 10 trapping days) (see Fig. 4A). This represented 385 (if 93%) or 792 (if 100%) man-work-days for an approximate operational budget of 22.000–38.000 euros per trapping session, assuming four working hours per trapper per day. However, within the 150 man-work-days limit imposed by local workforces, none of the ten trapping grid layouts evaluated in this study could be practically implemented at the scale predefined by local stakeholders (200-ha pilot area). Treating the entire management area as smaller adjacent units that could be trapped in successive time stages was essential for achieving complete coverage. This strategy has been successfully implemented in the Scottish Outer Hebrides for the control of non-native American mink Mustela vison populations over large spatial scales (Macleod et al. 2019). While our simulations showed that splitting the entire management area would in theory enhance project feasibility, the limited workforce available in the Mont Panié wilderness reserve will inevitably hinder trapping effectiveness. Indeed, with 10 persons available for 15 days at a time, none of the grid layouts would achieve a cumulative rat capture probability higher than .30, which would probably not be sufficient to produce tangible benefits to native forest biodiversity. Alternative options to maximise removal efficiency may be to either reconsider the total size of the management area (e.g. 100 ha instead of 200 ha), if solely reliant on local workforces, or involve a few commited professional trappers to maintain a constant high trapping effort, although the latter would require additional financial resources over a protracted period of time. Other options include the use of self-resetting traps, which could be particularly useful in remote areas. The creation of “mainland islands” should also be considered as an alternative option to permanent rat removal, if and when resources to set up such an ambitious program are available. Ultimately, in areas where local workforce is limited and the use of aerial baiting is not accepted by the local communities, ensuring that the scale of management is fit for purpose is key to success. This decision should be based on the fine balance between maximising the use of available resources and achieving a cumulative capture probability that is meaningful for biodiversity targets.

With this study, we demonstrate that scenario modelling, informed by trapping experiments, is a flexible tool for informing the design of cost-effective control programs of island-invasive rodent populations, under idiosyncratic local circumstances. Due to rats’ productivity and reinvasion rates, a one shot reduction in density is clearly not enough to produce tangible benefits to native biodiversity. Given the prospect of assessing optimal strategies for a sustainable rat control program, acquiring rat demographic rates (productivity, survival, dispersal movements) will be helpful to explore finer components of trapping design (e.g. number of trapping sessions per season, year and habitat types) and inform on the most cost-effective trapping regime (how often to trap, for how long and where) in the long-term. While our modelling approach was developed for rats in the Mont Panié reserve, it could be easily adapted to other systems and invasive pest species that could be controlled by lethal trapping.