Research Article |
Corresponding author: Emma B. Hanslowe ( ebhanslowe@gmail.com ) Corresponding author: Francesca T. Erickson ( fte0001@auburn.edu ) Academic editor: Sandro Bertolino
© 2022 Emma B. Hanslowe, Amy A. Yackel Adams, Melia G. Nafus, Douglas A. Page, Danielle R. Bradke, Francesca T. Erickson, Larissa L. Bailey.
This is an open access article distributed under the terms of the CC0 Public Domain Dedication.
Citation:
Hanslowe EB, Yackel Adams AA, Nafus MG, Page DA, Bradke DR, Erickson FT, Bailey LL (2022) Chew-cards can accurately index invasive rat densities in Mariana Island forests. NeoBiota 74: 29-56. https://doi.org/10.3897/neobiota.74.80242
|
Rats (Rattus spp.) are likely established on 80–90% of the world’s islands and represent one of the most damaging and expensive biological invaders. Effective rat control tools exist but require accurate population density estimates or indices to inform treatment timing and effort and to assess treatment efficacy. Capture-mark-recapture data are frequently used to produce robust density estimates, but collecting these data can be expensive, time-consuming, and labor-intensive. We tested a potentially cheaper and easier alternative, chew-cards, as a count-based (quantitative) index of invasive rat densities in tropical forests in the Mariana Islands, an archipelago in the western North Pacific Ocean. We trialed chew-cards in nine forest grids on two Mariana Islands by comparing the proportion of cards chewed to capture-mark-recapture density estimates and manipulated rat densities to test whether the relationship was retained. Chew-card counts were positively correlated with rat capture-mark-recapture density estimates across a range of rat densities found in the region. Additionally, the correlation between the two sampling methods increased with the number of days chew-cards were deployed. Specifically, when chew-cards were deployed for five nights, a 10% increase in the proportion of cards chewed equated to an estimated increase in rat density of approximately 2.4 individuals per ha (R2 = 0.74). Chew-cards can provide a valid index of rat densities in Mariana Island forests and are a cheaper alternative to capture-mark-recapture sampling when relative differences in density are of primary interest. New cost-effective monitoring tools can enhance our understanding and management of invaded islands while stretching limited resources further than some conventional approaches, thus improving invasive species management on islands.
Abundance estimation, capture-mark-recapture, Guam, invasive predator control, Rattus, Rota, spatial capture-recapture, tropical ecology
Invasive species jeopardize worldwide biodiversity (
Islands are often the focus of invasive species research and control efforts (
Rats (Rattus spp.) are difficult to detect (
Trapping (
Count-based indices are commonly used as relative measures of abundance or density (
Chew-track-cards, a tool for indexing rodents, are baited pieces of plastic that retain animal tooth impressions and footprints. Seminal work conducted in Australia and New Zealand determined that chew-track-cards are a cost-effective means of accurately indexing small mammal abundances across multiple species, including rats (
The Marianas are a chain of 15 volcanic islands in the western North Pacific Ocean (Fig.
Nine forest grids sampled via chew-cards and live-trapping for rats (Rattus spp.) during June 2018–August 2019 on Guam (G1–4) and Rota (R1–5) in the Mariana Islands. In the inset map, the red rectangle indicates the location of the two islands in the western North Pacific Ocean. In the main map, the blue circles indicate 11 × 11 grids with 12.5-m intervals between each station (grid area = 1.56 ha), and orange circles indicate 10 × 10 grids with 10-m intervals between each station (grid area = 0.81 ha).
Guam and Rota are the southernmost and larger (Guam = 550 km2; Rota = 85 km2) of the Mariana Islands (Fig.
Pacific rats (Rattus exulans), brown rats (Rattus norvegicus), and black rats (Rattus rattus) have been established in the Mariana Islands—where bats are the only native mammals—for centuries (
The brown treesnake was accidentally introduced from its native range in the South Pacific (
Our work on Guam occurred during 2018 within a 55-ha plot of homogenous disturbed limestone forest located on Andersen Air Force Base, termed the Habitat Management Unit. An extensive, interagency restoration plan including removal of non-native animals, constructing barriers, native plant recovery, and bird reintroductions exists for the Habitat Management Unit (
We sampled nine forest grids on Guam (n = 4 grids) and Rota (n = 5 grids; Fig.
All Guam grids and Rota grids R4 and R5 consisted of 11 × 11 trap stations with 12.5-m intervals between each station (grid area = 1.56 ha). The remaining three grids on Rota (R1–3) were part of a concurrent study (
We conducted capture-mark-recapture trapping of rats ≤ 2 days before (G1, G4, G2.2, G3.2, R1, R2, and R3) or after (G2.1, G3.1, and R5) a five-day card deployment so the cards would reflect the same rat densities estimated with capture-mark-recapture methods. We did not deploy live-traps and cards simultaneously to avoid competing baits on the landscape. We set baited, fixed-open traps two days prior to the start of live-trapping to allow the rats to acclimate to their presence (
We constructed rat indexing cards by cutting 4-mm thick, twin-walled polypropylene sheeting into 90 × 180-mm rectangles and aligned the flutes parallel to the short sides of the cards (Fig.
A chew-card and B chew-track-card designs used to index rat (Rattus spp.) density in forest habitats on Guam and Rota in the Mariana Islands during June 2018–August 2019. Designs were patterned after
At each station, we stapled the cards to trees approximately one meter off the ground with the baited half up (Fig.
To confirm or refute rat-chew identification, we deployed a RECONYX PC900 HyperFire Professional Covert Camera Trap (RECONYX, Holmen, WI, USA) at six randomly selected cards from each grid, except R2 and R3, for the duration of the five-day card deployment. We initially programmed the cameras to trigger upon motion detection (for G1, G2.1, G3.1, G4, G3.2) but switched to a time-lapse setting after December 2018 (for grids G2.2, R1, R4.1, R4.2, and R5) to better capture species interactions with the cards. We reviewed all camera-trap photos and cross-referenced our field assessments of rat chews with the photos from the corresponding camera-trap night. We measured daily rainfall via rain gauges at all grids except R2 and R3.
We calculated individual body condition indices by dividing mass by head-body length (
We used spatially explicit capture-recapture models (
Capture probabilities can vary by time, behavior, and individual heterogeneity (
We analyzed data from each grid separately. At grids with sufficient data (R4–5), we used a two-step approach to model capture probabilities from which we derived density estimates. First (Step 1), we accounted for all available hypothesized sources of individual heterogeneity in capture probability by including sex, age, and body condition index as predictors. We fit models with additive combinations of temporal covariates, including a two-night neophobic response (neophobia2), a time trend (Time), daily rainfall amount (rain; when available), a behavioral response (behavior), and no temporal variation (.). We did not include neophobia2 with either rain or Time in the same model. We retained the best-supported temporal variation structure(s) to test all possible additive combinations of individual covariates, including sex, age, body condition index, and no individual heterogeneity (Step 2). We failed to collect individual covariate and rain data for Rota grids R1–3, and thus did not have sufficient data for the two-step approach. For these grids, we simply fit all other possible additive combinations of the remaining temporal covariates. We held the spatial parameter (σ) constant (i.e., null) in all models.
Data from grids on Guam were too sparse (< 10 total captures per grid) to use spatially explicit models, so we used simpler closed-capture conditional likelihood models (
We did not analyze tracking ink data because we deemed our tracking ink methods ineffective in this system and instead treated all cards as ‘chew-cards’ and limited our analysis to teeth impressions. We summed the cumulative number of cards with rat chews for each deployment day (1–5 days) for each grid and calculated the daily proportion of cards with rat chews. We used linear regression models and Pearson’s product-moment correlations, implemented in base R, to assess the relationship between card indices and capture-mark-recapture density estimates. We conducted these analyses five times, where the predictor variable in each regression analysis was the proportion of cards that detected rats after one, two, three, four, and five deployment nights, respectively, for each grid.
We captured 233 individual rats a total of 444 times in 10,090 corrected trap nights over the course of our study, where one corrected trap night equaled one active trap night corrected for sprung (via target and non-target captures and false trips) and non-functioning/missing traps by considering them to represent half of a night of trapping effort and no trapping effort, respectively (Table
Corrected trap nights†, number of individual rats (Rattus spp.) captured (# indiv. rats), total number of rat captures (including recaptures; total rat caps.), sex (M = male; F = female; U = undetermined sex), age (A = adult; J = juvenile; U = undetermined age), density estimate plus/minus standard error (D̂ ± SE), and proportion of chew-cards with rat chews after nights 1–5 for each sampling grid in forest habitats on Guam and Rota in the Mariana Islands during June 2018–August 2019.
Grid | Live-trap dates | Corrected trap nights† | # indiv. rats | Total rat caps. | Sex | Age | D� ± SE | Chew-card proportions | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
M | F | U | A | J | U | 1 | 2 | 3 | 4 | 5 | ||||||
Guam | ||||||||||||||||
G1 | 11–20 Jun 2018 | 1,296.5 | 1 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 0.26 ± 0.78 | .00 | .00 | .00 | .01 | .01 |
G2.1 | 19–28 Jul 2018 | 1,153.0 | 4 | 5 | 2 | 2 | 0 | 4 | 0 | 0 | 1.37 ± 0.14 | .00 | .00 | .02 | .05 | .10 |
G3.1 | 19–28 Jul 2018 | 879.0 | 3 | 3 | 2 | 1 | 0 | 3 | 0 | 0 | 0.79 ± 2.35 | .00 | .01 | .03 | .05 | .11 |
G4 | 04–13 Aug 2018 | 1,009.5 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.00 ± 0.00 | .00 | .02 | .04 | .05 | .10 |
G3.2 | 29 Nov–08 Dec 2018 | 1,155.5 | 6 | 11 | 4 | 2 | 0 | 5 | 1 | 0 | 1.01 ± 9.65 | .02 | .05 | .06 | .07 | .10 |
G2.2 | 02–11 Feb 2019 | 1,243.5 | 6 | 7 | 5 | 1 | 0 | 5 | 1 | 0 | 1.93 ± 0.18 | .03 | .09 | .14 | .18 | .29 |
Guam total | 6,737.0 | 20 | 27 | 14 | 6 | 0 | 18 | 2 | 0 | |||||||
Rota | ||||||||||||||||
R1 | 04–07 Jun 2019 | 286.0 | 20 | 35 | 9 | 5 | 6 | 5 | 9 | 6 | 7.09 ± 1.71 | .28 | .47 | .61 | .67 | .73 |
R2 | 11–14 Jun 2019 | 334.0 | 12 | 14 | 4 | 4 | 4 | 2 | 6 | 4 | 20.83 ± 12.65 | .02 | .07 | .14 | .25 | .39 |
R3 | 11–14 Jun 2019 | 311.0 | 17 | 27 | 3 | 5 | 9 | 7 | 1 | 9 | 9.37 ± 1.79 | .27 | .76 | .84 | .87 | .87 |
R4.1 | 28 Jun–07 Jul 2019 | 1,285.5 | 92 | 196 | 48 | 44 | 0 | 51 | 41 | 0 | 34.73 ± 4.52 | .12 | .59 | .83 | .91 | .94 |
R4.2‡ | 19.99 ± 4.52 | .69 | .89 | .96 | .96 | .96 | ||||||||||
R5 | 27 Jul–05 Aug 2019 | 1,136.5 | 72 | 145 | 40 | 32 | 0 | 56 | 16 | 0 | 21.86 ± 3.21 | .02 | .10 | .21 | .36 | .48 |
Rota total | 3,353.0 | 213 | 417 | 104 | 90 | 19 | 121 | 73 | 19 | |||||||
Total | 10,090.0 | 233 | 444 | 118 | 96 | 19 | 139 | 75 | 19 |
Boxplots depicting the medians (bold lines), interquartile ranges (IQRs; 25th–75th percentiles; rectangles), minimums (first quartile-1.5*IQR) and maximums (third quartile+1.5*IQR; dashed lines), and any outliers (black dots) for A mass, B head-body length, and C body condition index for live-trapped rats (Rattus spp.) in forest habitats on Guam (n = 19 rats) and Rota (n = 163 rats) in the Mariana Islands during June 2018–August 2019. Statistics shown in the bottom-left corners are for Wilcoxon rank-sum tests (α = 0.05). Rats were A heavier and had C higher body condition indices on Guam compared to Rota, but there was no difference in B head-body lengths between the two islands.
We found that rat capture probability on both islands exhibited a behavioral effect (Fig.
Capture (p̂) and recapture (ĉˆ) probability estimates from closed-capture conditional likelihood models for rats (Rattus spp.) in forest habitats on Guam (G1–4) and Rota (R1–5) in the Mariana Islands during June 2018–August 2019.
Our grids represented a range of rat density estimates (D̂ range = 0.00–34.73 rats/ha) to test card indices. Rat densities on Rota (D̂ range = 7.09–34.73 rats/ha) were higher than those on Guam (D̂ range = 0.00–1.93 rats/ha). At the two grids we re-sampled after lethal snake treatments on Guam, G3.2 and G2.2, rat density increased by 28% and 41%, respectively, but remained comparatively low even three months after snake control was applied (D̂ = 1.01; SE [D̂] = 9.65 and D̂ = 1.93; SÊ [D̂] = 0.18, respectively; Table
We deployed 1,389 chew-cards during 60 days of sampling on Guam (n = 6 deployments) and Rota (n = 6 deployments). The mean proportion of cards chewed after five days was 0.12 (SD = 0.09) on Guam and 0.73 (SD = 0.24) on Rota. On average, the proportion of cards with chews increased by 0.03 (SD = 0.03) a day on Guam and 0.10 (SD = 0.10) a day on Rota.
The proportion of cards chewed by rats was correlated with density estimates when cards were left in the field for at least three nights (Fig.
Linear regressions and Pearson’s product-moment correlations to assess the relationship between the cumulative proportion of cards with rat (Rattus spp.) chews after one, two, three, four, and five nights (x-axis) and capture-mark-recapture density estimates plus/minus standard error (D̂ ± SE; y-axis) in forest habitats on Guam and Rota in the Mariana Islands during June 2018–August 2019.
Note that an intercept (B0) was not included in this equation because it rounded to zero.
We deployed cameras on 60 cards and processed > 24,000 photos with animals on the cards. Twenty-eight of these cards had field recordings of rat chews, and we confirmed rat identification via photos at 27 of 28 (96%) of the card/camera nights (e.g., Fig.
Trail-camera photo of a rat (Rattus spp.) leaving visible chews on a chew-card. We used trail cameras to confirm or refute rat-chew identification at randomly selected cards from each grid. Emma B. Hanslowe photograph captured by an automated camera trap on 10 July 2019 in forest habitat on Rota in the Mariana Islands.
Our study demonstrated a positive, monotonic relationship between chew-card counts and rat density estimates across a range of densities in Guam and Rota forests, and we thus conclude that chew-cards provided a valid index of rat densities and may be effective on similar tropical islands. Specifically, counts from chew-cards deployed for 3–5 nights correlated with rat capture-mark-recapture density estimates. This relationship was retained across rat density estimates ranging from 0–35 rats/ha and after management. The correlation between the proportion of cards with rat chews and capture-mark-recapture density estimates increased daily and was highest after five nights, when nearly three quarters of the variance in capture-mark-recapture density estimates was predicted by variation in chew-card proportions (R2 = 0.74). Accordingly, chew-cards should be deployed for a minimum of three nights, but five nights is optimal as this duration provided the smallest standard error around the regression line. Evaluating longer chew-card deployment periods (≥ 6 nights) may be advantageous, as additional nights might have even stronger correlations with rat density. However, the proportion of cards chewed will eventually stabilize or become 1.0 when all the cards are chewed, and this may occur more quickly at high rat densities (
While chew-cards have been widely used to assess invasive small mammal populations (
Rat chews were easily distinguished from non-target chews (e.g., feral cats [Felis catus] and crab [Coenobita brevimanus; Birgus latro] pinches) and correctly identified in our study, as confirmed by our camera-trap data. Specifically, rats were photographed chewing cards at nearly all cards positive for rat chews (27 of 28 [96%] card/camera nights). The single unconfirmed chew was likely not misidentified but was more likely not captured because the camera’s motion detection did not trigger. We switched camera settings from motion detection to time-lapse after this occurrence to improve rat detection on cameras, and all rat chews corresponding to a camera-trap night were photographed thereafter. Our study was the first to confirm chew-card species identification with cameras, as recommended by
We encountered significant issues with tracking ink during our study. First, the Marianas’ tropical climate caused the ink to run and fade. Second, a multitude of non-target species (e.g., geckos, skinks, crabs, snails/slugs, ants, worms) left unidentifiable tracks that made distinguishing any rat tracks difficult, time-consuming, and erroneous. Similar to other studies, we found that tracking ink provided little additional information relative to chew marks alone (P. J. Sweetapple, Manaaki Whenua Landcare Research, written comm, 08 Sep 2018), and recent studies have discontinued its use in New Zealand (
Non-targets may further hinder chew-card efficacy in Mariana Island forests via bait consumption and interspecific interference. We observed bait consumption by ants in the field, and reduced bait availability likely reduces chew-card attraction/effectiveness. In forests with abundant ants, chew-cards may be ineffective (pers. obs.). Use by non-targets may also affect rat chew-card detection (i.e., interspecific interference); for example, two studies in New Zealand found that individuals of one species were less likely to chew cards if they had already been chewed by another species (
Our study results suggest that chew-cards can be appropriate for monitoring changes in rat distribution or relative density over space or time in association with invasive predator (e.g., brown treesnake) occurrence or suppression efforts in Mariana Island forests. Chew-cards have several advantages over capture-mark-recapture density estimation, at the forefront of which is cost. Extrapolating from cost analyses conducted by
Controlling invasive species on islands is a global conservation priority (
We thank S. Amburgey, A. Bristol, A. Bryant, A. Cummings-Krueger, T. Hinkle, K. Kabat, X. Lazaro, A. Leach, S. Lundy, M. Mendiola, Z. Quiogue, A. Reyes, C. Reza, N. Sablan, C. Schmokel, N. Van Ee, P. Xiong, and D. Young for their assistance in the field and H. Barbé, C. Campbell, C. Castagnet, T. Kelly, M. MacPhail, G. Nickerson, L. Roberts, P. Trifiletti, and S. Valencia for their assistance with data processing, entry, and quality control. L. Bonewell, A. Collins, G. Engler, J. Guilbert, K. Samsel, and M. Viernes provided administrative and logistical support. M. Mazurek and A. Wiewel facilitated project development. We thank R. Reed for intellectual contribution and administrative leadership. A. Dillon, H. McCaslin, and Z. Weller provided guidance on statistical analyses, code, and modeling. R. Boone gave feedback on the original project proposal and study design. A. Feuka and B. Hardy reviewed an earlier draft of this manuscript. We thank Andersen Air Force Base for providing access to their installation and E. Mori, C. King, K. Wilson and one anonymous reviewer for their thoughtful comments that improved this paper. This work was conducted in accordance with the Commonwealth of the Northern Mariana Islands Department of Lands and Natural Resources scientific research permits # 04121-19 and # 08835-20. The present research involved capturing, handling, marking, and field euthanasia of free-ranging rats. We followed guidelines approved by the American Society of Mammalogists and the Institutional Animal Care and Use Committee of Colorado State University (protocol # 18-7896A and amendments). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Funding was provided by the U.S. Department of the Interior Office of Insular Affairs, the U.S. Fish and Wildlife Service Wildlife Restoration Program, and U.S. Navy (MIPR No. N61128-18-MP-002GS; N41557-20-MP-002GS).
Appendix 1, Tables S1–S4
Data type: pdf file
Explanation note: Model selection results. TableS1. Guam: Huggins’ closed-capture conditional likelihood model selection results for combined Guam grids sampled during June 2018–February 2019. Table S2. R1–3: Model selection results for spatially explicit models fit to data collected during June 2019 from grids for which we did not collect individual covariates. Results from the temporal models only (Step 1) are provided by grid. Table S3. R4: Spatially explicit model selection for rats sampled during June–July 2019. Step 1 models include all hypothesized sources of individual variation in capture probability (sex + age + BCI + temporal structures) listed below. We retained the best-supported temporal structure (behavior) when testing all possible additive combinations of individual covariates in Step 2 (sex, BCI, age, and no individual heterogeneity). Table S4. R5: Spatially explicit model selection for rats sampled during July–August 2019. Step 1 models include all hypothesized sources of individual variation in capture probability (sex + age + BCI + temporal structures) listed below. We retained the best-supported temporal structure (behavior + neophobia2) when testing all possible additive combinations of individual covariates in Step 2 (sex, BCI, age, and no individual heterogeneity).
Figure S1
Data type: pdf file
Explanation note: Density estimator comparison. Fig. S1. Comparison of three density estimation approaches for rats (Rattus spp.) using capture-mark-recapture data from Guam (G1–4) and Rota (R1–5) forest grids during June 2018–August 2019. Black and dark gray bars represent density estimates (D̂s) calculated from model-averaged abundance estimates (N̂s) divided by effective trapping areas (ETAs) calculated by adding boundary strips equaling half of the mean maximum distances moved by rats captured more than once (0.5MMDM) and the full MMDM, respectively. Light gray bars represent D̂s from spatially explicit capture-recapture (SECR) models for sites on Rota only.