Effective sampling of pond-dwelling larval stages of the federally-listed flatwoods salamander (Ambystoma cingulatum) requires sufficient knowledge of when larvae are present and how best to sample them. The primary objective of this study was to maximize field sampling efficiency of flatwoods salamanders and minimize the uncertainty associated with declaring absence after repeated non-detects.

Technical Approach

The approach was to evaluate sampling method efficiency, the relationship between pond habitat and larvae residency, and the effect of sample design on sampling success. Salamander larvae and pond habitat were surveyed at Fort Stewart military installation for three years and at several locations in Florida on a limited basis using different methods. Statistical models were then used to predict larval occupancy of wetlands and to quantify the similarity among wetlands based upon the habitat information derived from pond-specific vegetation and landscape data.

Estimating the probability of detection is crucial to being able to calculate how many unsuccessful surveys are necessary to conclude that a site is unoccupied with a known level of confidence (e.g., 95%). Two methods were developed using the larval survey data to estimate probability of detection, which is a key component for calculating sampling effort needed to be confident that an undetected pond is actually unoccupied. The first method uses detection success rate for 5-min survey intervals and the second uses per-minute capture rate. Finally, a computer model was developed to simulate periodic sampling of ponds for flatwoods salamander larvae.



The studies on method effectiveness demonstrated that the most time efficient and cost effective sample method is dipnetting. Larvae were detected at one pond at Fort Stewart in 3 of 4 years of the study, but not at 59 other ponds that were sampled. These detections represent the only observations of successful breeding in a natural wetland in Georgia since 1999. The inability to find larvae in more than one pond at Fort Stewart confirms what appears, based on historic capture records, to be a significant decline in flatwoods salamander abundance at Fort Stewart. To better predict annual variation in pond residency, a model was developed that uses rainfall data and likely growth rates to predict hatching dates and period of pond residency. With two years of data that likely represent the extremes of pond residency in Georgia, it was found that date of hatching can occur as early as the first week of January and as late as the first week of March and, that in some years, larvae were found in the pond as late as the end of May. The most opportune time of year to sample flatwoods salamander larvae was during the second and third months after a pond fills to at least half, typically sometime within February to April. Using removal study data from temporary enclosures at St Marks National Wildlife Refuge (NWR), a relationship was derived between larval density and capture rate that enables future estimates of larval density based on observed capture rates. The proportion of total larvae present that were captured during a pass averaged about 40% (range of 22-65%) and was not dependent on larval density.


The presence of flatwoods salamanders was positively associated with a native iris species and the presence of facultative and obligate wetland plants. Increased canopy cover was strongly associated with unsuitable wetlands. Landscape structure was also an important predictor of habitat use. For example, including distance from streams improved prediction of habitat in regression analyses, and discriminant function analyses with landscape data had a higher rate of correctly classifying known breeding sites as being occupied. Models developed for Fort Stewart populations did a poor job of predicting presence for sites at St Marks NWR and Apalachicola NF. Results from hierarchical clustering indicate that this may be due to differences in wetland vegetation and landscape structure among the sites in Florida and Georgia. Developing habitat models for rare or declining species is particularly challenging because a species may decline for reasons that are not directly related to habitat structure. If populations become restricted in their distribution or shift habitat associations as a result of other ecological factors such as disease, predation, and competition, then this may cause habitat models to perform poorly. Amphibians are also challenging subjects for habitat modeling because their occupancy in ponds is highly variable through time. Because the habitat modeling did not identify the habitat of the historically occupied ponds at Fort Stewart as particularly unique, this may indicate that either (1) many ponds occur at Fort Stewart that potentially are capable of supporting flatwoods salamander breeding or (2) the features of the habitat that were used as input variables to the models do not include all the important features.


Using either developed method for estimating the probability of detection, the probability of detection can be related to capture rate and, based on the enclosure study results, also to larval density. These relationships enable basing calculations of the number of necessary surveys on an expected larval density instead of a detection probability, a concept that for most biologists has less meaning and utility than population density. Therefore, when an estimate of detection probability for a pond is unavailable or unattainable, biologists now have two options if they want to determine how many surveys are necessary. They can either (1) select a minimum density above which they want to be reasonably sure larvae do not exist and calculate the required number of non-detects necessary to reach that conclusion or (2) select a level of confidence with which they are comfortable and, knowing that their sampling effort is limited to a set number of surveys, they can calculate the density above which they are reasonably certain that larvae do not exist.

The computer model was used to determine the effectiveness of random versus adaptive sampling and the effect of stopping rules on survey success. Stopping rules for the purpose of this study refer to the number of times a pond is sampled without finding a larval salamander before sampling should be ceased because either the probability of the pond being occupied is extremely low or because distributing the effort to other ponds increases the probability of detecting larvae elsewhere. Neither sampling mode (random versus adaptive) was better than the other in all circumstances. In general, when occupied ponds were distributed in clusters, adaptive sampling based on pond proximity produced more detections than random sampling. In fact, if surveyors are able to identify any factor that is correlated with salamander presence, then adaptive sampling with that factor as a basis will produce the most efficient sampling design. When occupied ponds were randomly distributed across the pondscape, random sampling was slightly better. With regards to stopping rules, when adaptive sampling is justified, stopping rules of four or five produce more detections, but when surveyors have no basis for adaptive sampling, random sampling with a stopping rule of one or two trips is better or equal to longer stopping rules. The model was coded in the R, a programming language which is widely used and available at no cost.


The results (including model code) of this study will be distributed to biologists and resource managers that are involved in flatwoods salamander conservation in the southeastern U.S. A workshop was held at the conclusion of the study to brief regional biologists and resource managers on the findings of the study. Specific recommendations are presented to direct future monitoring of flatwoods salamanders at Fort Stewart with regard to how and when to sample, which ponds to sample, and how to determine when enough sampling has been completed after repeated non-detects. The methods developed here are also applicable to the detection of other rare species including pests and exotic invaders.

  • Aquatic ,

  • Habitat