Miguel A. Acevedo

Automated classification of bird and amphibian calls using machine learning: A comparison of methods

Abstract We compared the ability of three machine learning algorithms (linear discriminant analysis, decision tree, and support vector machines) to automate the classification of calls of nine frogs and three bird species. In addition, we tested two ways of characterizing each call to train/test the system. Calls were characterized with four standard call variables (minimum and maximum frequencies, call duration and maximum power) or eleven variables that included three standard call variables (minimum and maximum frequencies, call duration) and a coarse representation of call structure (frequency of maximum power in eight segments of the call). A total of 10,061 isolated calls were used to train/test the system. The average true positive rates for the three methods were: 94.95% for support vector machine (0.94% average false positive rate), 89.20% for decision tree (1.25% average false positive rate) and 71.45% for linear discriminant analysis (1.98% average false positive rate). There was no statistical difference in classification accuracy based on 4 or 11 call variables, but this efficient data reduction technique in conjunction with the high classification accuracy of the SVM is a promising combination for automated species identification by sound. By combining automated digital recording systems with our automated classification technique, we can greatly increase the temporal and spatial coverage of biodiversity data collection.

Using Automated Digital Recording Systems as Effective Tools for the Monitoring of Birds and Amphibians

Abstract There is a need to improve the quantity and quality of data in biodiversity monitoring projects. We compared an automated digital recording system (ADRS) with traditional methods (point-counts and transects) for the assessment of birds and amphibians. The ADRS proved to produce better quantity and quality of data. This new method has 3 additional advantages: permanent record of a census, 24 h/d data collection and the possibility of automated species identification.

Social network models predict movement and connectivity in ecological landscapes

Network analysis is on the rise across scientific disciplines because of its ability to reveal complex, and often emergent, patterns and dynamics. Nonetheless, a growing concern in network analysis is the use of limited data for constructing networks. This concern is strikingly relevant to ecology and conservation biology, where network analysis is used to infer connectivity across landscapes. In this context, movement among patches is the crucial parameter for interpreting connectivity but because of the difficulty of collecting reliable movement data, most network analysis proceeds with only indirect information on movement across landscapes rather than using observed movement to construct networks. Statistical models developed for social networks provide promising alternatives for landscape network construction because they can leverage limited movement information to predict linkages. Using two mark-recapture datasets on individual movement and connectivity across landscapes, we test whether commonly used network constructions for interpreting connectivity can predict actual linkages and network structure, and we contrast these approaches to social network models. We find that currently applied network constructions for assessing connectivity consistently, and substantially, overpredict actual connectivity, resulting in considerable overestimation of metapopulation lifetime. Furthermore, social network models provide accurate predictions of network structure, and can do so with remarkably limited data on movement. Social network models offer a flexible and powerful way for not only understanding the factors influencing connectivity but also for providing more reliable estimates of connectivity and metapopulation persistence in the face of limited data.

Bird Community Dynamics and Habitat Associations in Karst, Mangrove and Pterocarpus Forest Fragments in an Urban Zone in Puerto Rico

Abstract. Puerto Ricos forest cover decreased to less than 10% in the early 1900s leaving few forest patches available for migrant and resident birds. In this process of deforestation karst hills and coastal wetlands have been some of the most severely modified forest types; however, we know little about their bird community dynamics and their relation with habitat variables. To address this issue we studied bird species composition and habitat characteristics in karst forest and two coastal forested wetlands (mangrove and Pterocarpus forest) in the Caribbean island of Puerto Rico. In each forest type, we conducted 10 point counts monthly for two years and characterized habitat variables. We performed a non-metric multidimensional scaling (NMS) ordination and a multi-response permutation procedure (MRPP) for each year to determine the similarity of bird species composition among monthly censuses and the relation between species composition and habitat variables. This ordination technique grouped censuses into three groups: karst forest, Pterocarpus and mangrove in the migratory period, and Pterocarpus and mangrove in the non-migratory period. The high tree species richness in the karst forest, and the presence of standing water in coastal wetlands were the most important habitat variables associated with the different bird communities. Our results demonstrate that the karst and coastal wetlands forests, even if they are small patches surrounded by a mixed matrix of pasture and urban settlements may be important habitat for both residents and migrants, and suggest that the protection and restoration of these habitats should be high management and conservation priorities.

The matrix alters the role of path redundancy on patch colonization rates

Landscape connectivity is central to many problems in ecology and conservation. Recently, the role of path redundancies on movement of organisms has been emphasized for understanding connectivity, because increasing the number of potential paths (i.e., redundancy) is predicted to increase movement rates, which can alter predictions for foraging theory and population dynamics. Nonetheless, experiments that test for the effects of path redundancies on connectivity remain scarce. We tested for the role of path redundancies on the movements of a habitat specialist, Chelinidea vittiger, using experimental arenas that altered path redundancy by varying the amount and configuration of stepping stones across a gradient of matrix resistance. We found that stepping-stone redundancies increased colonization rates to target patches, but the effects differed depending on the configuration of redundancy and the structure of the matrix. In addition, matrix effects were better explained through the use of effective distance measures that incorporate redundancy in the matrix than those that ignore redundancy. Our results provide experimental evidence that measures that ignore redundancies may be inadequate for capturing functional connectivity, illustrate the ways in which redundancies alter colonization rates, and emphasize how habitat configuration and matrix structure can interact to guide movement of individuals across landscapes.

Spatial Heterogeneity, Host Movement and Mosquito-Borne Disease Transmission

Mosquito-borne diseases are a global health priority disproportionately affecting low-income populations in tropical and sub-tropical countries. These pathogens live in mosquitoes and hosts that interact in spatially heterogeneous environments where hosts move between regions of varying transmission intensity. Although there is increasing interest in the implications of spatial processes for mosquito-borne disease dynamics, most of our understanding derives from models that assume spatially homogeneous transmission. Spatial variation in contact rates can influence transmission and the risk of epidemics, yet the interaction between spatial heterogeneity and movement of hosts remains relatively unexplored. Here we explore, analytically and through numerical simulations, how human mobility connects spatially heterogeneous mosquito populations, thereby influencing disease persistence (determined by the basic reproduction number R 0), prevalence and their relationship. We show that, when local transmission rates are highly heterogeneous, R 0 declines asymptotically as human mobility increases, but infection prevalence peaks at low to intermediate rates of movement and decreases asymptotically after this peak. Movement can reduce heterogeneity in exposure to mosquito biting. As a result, if biting intensity is high but uneven, infection prevalence increases with mobility despite reductions in R 0. This increase in prevalence decreases with further increase in mobility because individuals do not spend enough time in high transmission patches, hence decreasing the number of new infections and overall prevalence. These results provide a better basis for understanding the interplay between spatial transmission heterogeneity and human mobility, and their combined influence on prevalence and R 0.

Place-based attributes predict community membership in a mobile phone communication network.

Social networks can be organized into communities of closely connected nodes, a property known as modularity. Because diseases, information, and behaviors spread faster within communities than between communities, understanding modularity has broad implications for public policy, epidemiology and the social sciences. Explanations for community formation in social networks often incorporate the attributes of individual people, such as gender, ethnicity or shared activities. High modularity is also a property of large-scale social networks, where each node represents a population of individuals at a location, such as call flow between mobile phone towers. However, whether or not place-based attributes, including land cover and economic activity, can predict community membership for network nodes in large-scale networks remains unknown. We describe the pattern of modularity in a mobile phone communication network in the Dominican Republic, and use a linear discriminant analysis (LDA) to determine whether geographic context can explain community membership. Our results demonstrate that place-based attributes, including sugar cane production, urbanization, distance to the nearest airport, and wealth, correctly predicted community membership for over 70% of mobile phone towers. We observed a strongly positive correlation (r = 0.97) between the modularity score and the predictive ability of the LDA, suggesting that place-based attributes can accurately represent the processes driving modularity. In the absence of social network data, the methods we present can be used to predict community membership over large scales using solely place-based attributes.

Conservation under uncertainty: optimal network protection strategies for worst-case disturbance events

Summary 1. Conservation goals are ideally set after a thorough understanding of potential threats; however, predicting future spatial patterns of threats, such as disturbance, remains challenging. Here, we develop a novel extension of network fortification‐interdiction models (NFIM) that deals with uncertainty in future spatial patterns of disturbance by optimally selecting sites that will best mitigate a worst-case scenario for a given magnitude of disturbance. 2. This approach uses information on between-patch movement probabilities and patchspecific survival, which can be estimated from mark–recapture data, to optimize life expectancy. Optimization occurs in three interrelated stages: protection, followed by disturbance and then assessment. 3. We applied the modelling approach to two mark–recapture data sets: roseate terns Sterna dougallii in the north-eastern United States and the Everglade snail kite Rostrhamus sociabilis plumbeus in Florida. We contrasted the results to a more conventional approach of protecting sites that maximize connectivity (by minimizing the distances among protected sites) and a biobjective model that maximizes connectivity and the number of individuals under protection. 4. Protecting sites that best mitigate future worst-case disturbance scenarios consistently resulted in higher predicted life expectancies than protecting patches that minimize dispersal distance. Predicted life expectancy was similar between NFIM and the bi-objective model for the small roseate tern network, yet the NFIM predicted higher life expectancy than any of the scenarios in the bi-objective model in the snail kite network. 5. Synthesis and applications. This application of interdiction models prescribed a combination of patches for protection that results in the least possible decrease in life expectancy. Our analyses of the snail kite and roseate tern networks suggest that managing to protect these prescribed patches by the network fortification -interdiction models (i.e. protecting against the worst-case disturbance scenario) is more beneficial than managing patches that minimize dispersal distance or maximize the number of individuals under protection if the conservation goal is to ensure the long-term persistence of a species.

Spatial asymmetries in connectivity influence colonization-extinction dynamics.

Movement has broad implications for many areas of biology, including evolution, community and population ecology. Movement is crucial in metapopulation ecology because it facilitates colonization and reduces the likelihood of local extinction via rescue effects. Most metapopulation modeling approaches describe connectivity using pair-wise Euclidean distances resulting in the simplifying assumption of a symmetric connectivity pattern. Yet, assuming symmetric connectivity when populations show net asymmetric movement patterns may result in biased estimates of colonization and extinction, and may alter interpretations of the dynamics and conclusions regarding the viability of metapopulations. Here, we use a 10-year time series on a wind-dispersed orchid Lepanthes rupestris that anchors its roots in patches of moss growing on trees or boulders along streams, to test for the role of connectivity asymmetries in explaining the colonization−extinction dynamics of this orchid in a network of 975 patches. We expected that wind direction could highly alter dispersal direction in this orchid. To account for this potential asymmetry, we modified the connectivity measure traditionally used in metapopulation models to allow for asymmetric effective distances between patches and subsequently estimated colonization and extinction probabilities using a dynamic occupancy modeling approach. Asymmetric movement was prevalent in the L. rupestris metapopulation and incorporating potential dispersal asymmetries resulted in higher colonization estimates in larger patches and more accurate models. Accounting for dispersal asymmetries may reveal connectivity effects where they were previously assumed to be negligible and may provide more reliable conclusions regarding the role of connectivity in patch dynamics.

The proximate causes of asymmetric movement across heterogeneous landscapes

ContextAsymmetric movements, in which the probability of moving from patch i to patch j is not necessarily the same as moving in the opposite direction, may be the rule more than the exception in nature where organisms move through spatially heterogeneous environments. Empirical tests of dispersal asymmetries are rare with even fewer tests of the mechanisms driving such patterns.ObjectivesWe tested for the mechanisms of asymmetric movement in the cactus-feeding insect, Chelinidea vittiger, using a combination of observational and experimental approaches.MethodsIn the observational approach, we analyzed movements from mark-recapture data in a large plot for over 4–5 generations and tested for the role of differences in patch area and wind direction driving broad-scale asymmetric movements. In the field experiment, we translocated individuals to experimental arenas where we tested for the roles of patch area, wind, presence of conspecifics, and matrix height driving directed movements at fine spatio-temporal scales.ResultsWe found that population-level patterns of movements in C. vittiger were generally asymmetric. At broad scales, observational data suggested that these asymmetries were related to variations in patch size, with movements being directed from small to large patches. At fine scales, experiments showed that movement was also directed from small to large patches, but this effect was mediated by the structure of the surrounding matrix.ConclusionsOur results illustrate how and why movement asymmetries can occur across landscapes. Accounting for such asymmetries may improve our understanding and prediction of spatially structured population dynamics and landscape connectivity.