To alleviate the burden of labeling data to train crowd counting models, we propose a prototype-based learning approach for semi-supervised crowd counting with an embeded understanding of perspective. Our key idea is that image patches with the same density of people are likely to exhibit coherent appearance changes under similar perspective distortion, but differ significantly under varying distortions. Motivated by this observation, we construct multiple prototypes for each density level to capture variations in perspective. For labeled data, the prototype-based learning assists the regression task by regularizing the feature space and modeling the relationships within and across different density levels. For unlabeled data, the learnt perspective-embedded prototypes enhance differentiation between samples of the same density levels, allowing for a more nuanced assessment of the predictions. By incorporating regression results, we categorize unlabeled samples as reliable or unreliable, applying tailored consistency learning strategies to enhance model accuracy and generalization. Since the perspective information is often unavailable, we propose a novel pseudo-label assigner based on perspective self-organization which requires no additional annotations and assigns image regions to distinct spatial density groups, which mainly reflect the differences in average density among regions. Extensive experiments on four crowd counting benchmarks demonstrate the effectiveness of our approach.

The stock assessment model for the Antarctic krill fishery is a population model operating on daily timesteps, which permits modeling within-year patterns of some population dynamics. We explored the effects of including within-year patterns in natural and fishing mortality on catch limits of krill, by incorporating temporal presence of key predator species and contemporary temporal trends of the fishing fleet. We found that inclusion of within-year variation in natural and fishing mortalities increased catch limits. Fishing mortality had a greater effect than natural mortality despite differences in top-down predation on krill, and potentially increased catch limits by 24% compared to the baseline model. Additionally, the stock assessment model allowed a higher catch limit when fishing was during peak summer months than autumn. Number of days with active fishing was negatively related to precautionary catch limits. Future stock assessments should incorporate contemporary spatiotemporal fishing trends and consider implementing additional ecosystem components into the model.

Knowledge of reproductive traits in exploited marine populations is crucial for their management and conservation. The maturity status in fish is usually assigned by traditional methods such as macroscopy and histology. Macroscopic analysis is the assessing of maturity stages by naked eye and usually introduces large amount of error. In contrast, histology is the most accurate method for maturity staging but is expensive and unavailable for many stocks worldwide. Here, we use the Random Forest (RF) machine learning method for classification of reproductive condition in fish, using the extensive data from Chilean hake (Merluccius gayi gayi). Gonads randomly collected from commercial industrial and acoustic surveys were classified as immature, mature-active and mature-inactive. A classifier for these three maturity classes was fitted using RFs, with the continuous covariates total length (TL), gonadosomatic index (GSI), condition factor (Krel), latitude, longitude, and depth, along with month as a factor variable. The RF model showed high accuracy (>82%) and high proportion of agreement (>71%) compared to histology, with an OOB error rate lower than 15%. GSI and TL were the most important variables for predicting the reproductive condition in Chilean hake, and to lesser extent, depth when using survey data. The application of the RF shows a promising tool for assigning maturity stages in fishes when covariates are available, and also to improve the accuracy of maturity classification when only macroscopic staging is available.

To alleviate the heavy annotation burden for training a reliable crowd counting model and thus make the model more practicable and accurate by being able to benefit from more data, this paper presents a new semi-supervised method based on the mean teacher framework. When there is a scarcity of labeled data available, the model is prone to overfit local patches. Within such contexts, the conventional approach of solely improving the accuracy of local patch predictions through unlabeled data proves inadequate. Consequently, we propose a more nuanced approach: fostering the model's intrinsic 'subitizing' capability. This ability allows the model to accurately estimate the count in regions by leveraging its understanding of the crowd scenes, mirroring the human cognitive process. To achieve this goal, we apply masking on unlabeled data, guiding the model to make predictions for these masked patches based on the holistic cues. Furthermore, to help with feature learning, herein we incorporate a fine-grained density classification task. Our method is general and applicable to most existing crowd counting methods as it doesn't have strict structural or loss constraints. In addition, we observe that the model trained with our framework exhibits a 'subitizing'-like behavior. It accurately predicts low-density regions with only a 'glance', while incorporating local details to predict high-density regions. Our method achieves the state-of-the-art performance, surpassing previous approaches by a large margin on challenging benchmarks such as ShanghaiTech A and UCF-QNRF. The code is available at: https://github.com/cha15yq/MRC-Crowd.

Animal abundance estimation is increasingly based on drone or aerial survey photography. Manual postprocessing has been used extensively; however, volumes of such data are increasing, necessitating some level of automation, either for complete counting, or as a labour-saving tool. Any automated processing can be challenging when using such tools on species that nest in close formation such as Pygoscelis penguins. We present here a customized CNN-based density map estimation method for counting of penguins from low-resolution aerial photography. Our model, an indirect regression algorithm, performed significantly better in terms of counting accuracy than standard detection algorithm (Faster-RCNN) when counting small objects from low-resolution images and gave an error rate of only 0.8 percent. Density map estimation methods as demonstrated here can vastly improve our ability to count animals in tight aggregations and demonstrably improve monitoring efforts from aerial imagery.

The southern rays bream (Brama australis) is a highly migratory, epi-mesopelagic species supporting an important artisanal fishery off central-southern Chile. Despite its importance, several questions exist about this species's demography and migratory routes. The first step in understanding the migratory behavior of B. australis is to test the feasibility of a conventional tagging program, a standard mark-recapture method, to infer migration in fish. Between February 2020 and December 2021, conventional tagging was conducted during 21 fishing trips on board artisanal vessels off Lebu harbor (Biobío Region, Chile) using gillnets, longlines, and handlines. Three thousand nine hundred forty-six individuals of B. australis between 30 and 55 cm fork length were tagged using external T-anchor bar labels (commonly known as "spaghetti"). Approximately 100 and 200 fish were tagged per fishing trip using longlines and gillnets, respectively. The size distribution of the tagged individuals was consistent with those retained in the catch, with 90% of tagged fish being longer than the fork length at 50% maturity. Eight tags have been recovered off the coast of Lebu up to May 2022. With times at liberty between 50 and 537 days. These preliminary recaptures are also analyzed in the context of the conceptual model for demography and migration proposed for this species in Chile. The main conclusion of this research is that a conventional tagging program is feasible for B. australis in Chile.

1. Animal abundance estimation is increasingly based on drone or aerial survey photography. Manual post-processing has been used extensively, however volumes of such data are increasing, necessitating some level of automation, either for complete counting, or as a labour-saving tool. Any automated processing can be challenging when using the tools on species that nest in close formation such as Pygoscelid penguins. 2. We present here an adaptation of state-of-the-art crowd-counting methodologies for counting of penguins from aerial photography. 3. The crowd-counting model performed significantly better in terms of model performance and computational efficiency than standard Faster RCNN deep-learning approaches and gave an error rate of only 0.8 percent. 4. Crowd-counting techniques as demonstrated here have the ability to vastly improve our ability to count animals in tight aggregations, which will demonstrably improve monitoring efforts from aerial imagery.

Growth in fishes is usually modelled by a function encapsulating a common growth mechanism across ages. However, several theoretical works suggest growth may comprise two distinct mechanistic phases arising from changes in reproductive investment, diet, or habitat. These models are termed two-state or biphasic, where acceleration in growth typically changes around some transition age. Such biphasic models have already been successfully applied in elasmobranch species, where such transitions are detectable from length-at-age data alone, but where estimation has assumed normally distributed errors, which is inappropriate for such slow-growing and long-lived fishes. Using recent advances in growth parameter estimation, we implement a biphasic growth model with asymmetric and heavy-tailed errors. We use data from six datasets, encompassing four species of elasmobranchs, to compare the performance of the von Bertalanffy and biphasic models under normal, skew-normal, and Student-t error distributions. Conditional expectation maximization estimation proves both effective and efficient in this context. Most datasets analysed here supported asymmetric and heavy-tailed errors and biphasic growth, producing parameter estimates different from previous studies.

South Asian river dolphins (Platanista gangetica) are among the most endangered of the world's cetaceans. The two subspecies in the family Platanistidae, Indus, and Ganges river dolphins (P. g. minor and P. g. gangetica), are both threatened by dams and barrages, declining river flows, fisheries bycatch, and pollution. We examine differences in external and skull morphology between dolphins in each river system to clarify their taxonomic status. Skulls from each river system could easily be differentiated using diagnostic differences in the shape of the frontal bones behind the nasals. This feature was present in all individuals irrespective of size, age, and sex. Ganges river dolphins are sexually dimorphic with females larger than males, but there was no evidence of dimorphism in the small sample of Indus river dolphins. There were no mitochondrial DNA haplotypes shared between the two river systems, and five fixed differences suggested a long-term (approximately 0.55 million years) absence of gene flow. Diagnosable differences in morphological and genetic characteristics indicate long-term reproductive as well as geographic isolation of Indus and Ganges river dolphins. We conclude that Indus and Ganges river dolphins should each be recognized as distinct species, and elevate the Indus subspecies, Platanista gangetica minor, to species level, Platanista minor Owen, 1853. Formal redescriptions are provided for both species.

Abstract Intense underwater sounds caused by military sonar, seismic surveys, and pile driving can harm acoustically sensitive marine mammals. Many jurisdictions require such activities to undergo marine mammal impact assessments to guide mitigation. However, the ability to assess impacts in a rigorous, quantitative way is hindered by large knowledge gaps concerning hearing ability, sensitivity, and behavioral responses to noise exposure. We describe a simulation‐based framework, called SAFESIMM (Statistical Algorithms For Estimating the Sonar Influence on Marine Megafauna), that can be used to calculate the numbers of agents (animals) likely to be affected by intense underwater sounds. We illustrate the simulation framework using two species that are likely to be affected by marine renewable energy developments in UK waters: gray seal (Halichoerus grypus) and harbor porpoise (Phocoena phocoena). We investigate three sources of uncertainty: How sound energy is perceived by agents with differing hearing abilities; how agents move in response to noise (i.e., the strength and directionality of their evasive movements); and the way in which these responses may interact with longer term constraints on agent movement. The estimate of received sound exposure level (SEL) is influenced most strongly by the weighting function used to account for the specie's presumed hearing ability. Strongly directional movement away from the sound source can cause modest reductions (~5 dB) in SEL over the short term (periods of less than 10 days). Beyond 10 days, the way in which agents respond to noise exposure has little or no effect on SEL, unless their movements are constrained by natural boundaries. Most experimental studies of noise impacts have been short‐term. However, data are needed on long‐term effects because uncertainty about predicted SELs accumulates over time. Synthesis and applications. Simulation frameworks offer a powerful way to explore, understand, and estimate effects of cumulative sound exposure on marine mammals and to quantify associated levels of uncertainty. However, they can often require subjective decisions that have important consequences for management recommendations, and the basis for these decisions must be clearly described.

Loud anthropogenic underwater noise, such as that associated with sonar operations, pile driving, or seismic surveys, can cause behavioral and physiological disturbance to many animals that may affect their survival or ability to breed. However, no formal framework for assessing the population-level consequences of this disturbance is currently available. We describe an interim version of a framework developed by a working group on the population consequences of disturbance, funded by the US Office of Naval Research through the University of California, that can be used to assess the effects of offshore renewable energy developments on marine mammal populations.

The development of guidelines for mitigating noise impacts on marine fauna requires data about the biological relevance of noise effects and about the practicality of mitigation options. Recent expansion of scientific knowledge has shown that noise effects vary among animals with different behavioral ecophysiology. Beaked whales exemplify that some sensitive species may experience negative effects of sound at kilometers from the source, due to behavioral reactions leading to indirect physiological damage. Moored hydrophone arrays have contributed substantially to our understanding of naval sonar effects on beaked whales and have been used to refine techniques for passive acoustic detection of cetaceans. Similarly, broadband Ocean Bottom Cables/Nodes could facilitate learning about effects of seismic sounds and cetaceans' distribution offshore. This information is essential to improve spatial mitigation in the planning-phase of activities. Also, passive acoustics can help real time mitigation, which requires early detection of vulnerable species and practical mitigation protocols triggered by detection. Detection could be aided by large-scale portable acoustic arrays, which are now technologically feasible. Pilot studies of technological applications for mitigation and cost-benefit modelling of potential mitigation scenarios will help to inform effective mitigation design. Mitigation reduces social conflict regarding noise effects, a win-win for all stakeholders.

There are many developments for offshore renewable energy around the United Kingdom whose installation typically produces large amounts of far-reaching noise, potentially disturbing many marine mammals. The potential to affect the favorable conservation status of many species means extensive environmental impact assessment requirements for the licensing of such installation activities. Quantification of such complex risk problems is difficult and much of the key information is not readily available. Expert elicitation methods can be employed in such pressing cases. We describe the methodology used in an expert elicitation study conducted in the United Kingdom for combining expert opinions based on statistical distributions and copula-like methods.

•Fast and robust decomposition of a matrix representing a spatial grid through time.•Rapid approximation for robust principal component analysis.•Competitive performance in terms of recall and precision for motion detection.•GPU accelerated implementation allows faster computation.

Appropriate management of the effects of human activities on animal populations requires quantification of the rate at which animals encounter stressors. Such activities are heterogeneously distributed in space, as are the individual animals in a population. This will result in a heterogeneous exposure rate, which is also likely to vary over time. A spatially explicit analysis of individual exposure is therefore required. We applied Bayesian spatially explicit capture–recapture models to photo‐identification data to estimate the home range of well‐marked individuals in a protected coastal population of bottlenose dolphins. Model results were combined with the estimated distribution of boat traffic to quantify how exposure to this disturbance varied in time and space. Variability in exposure between individuals was also investigated using a mixed‐effects model. The cumulative individual exposure to boat traffic varied between summers, depending both on the overall area usage and the degree of individual movement around the activity centres. Despite this variability, regions of higher risk could be identified. There were marked inter‐individual differences in the predicted amount of time dolphins spent in the presence of boats, and individuals tended to be consistently over‐ or underexposed across summers. Our study offers a framework to describe the temporal, spatial and individual variation in exposure to anthropogenic stressors when individuals can be repeatedly identified over time. It provides opportunities to map exposure risk and understand how this evolves in time at both individual and population levels. The outcome of such modelling can be used as a robust evidence base to support management decisions.

SummaryChanges in natural patterns of animal behaviour and physiology resulting from anthropogenic disturbance may alter the conservation status of a population if they affect the ability of individuals to survive, breed or grow. However, information to forecast population‐level consequences of such changes is often lacking.We developed an interim framework to assess the population consequences of disturbance when empirical information is sparse. We show how daily effects of disturbance, which are often straightforward to estimate, can be scaled to the disturbance duration and to multiple sources of disturbance.We used expert elicitation to estimate parameters that define how changes in individual behaviour or physiology affect vital rates and incorporated them into a stochastic population model. Model outputs can be used to evaluate cumulative impacts of disturbance over space and time. As an example, we forecast the potential effects of disturbance from offshore wind farm construction on theNorthSea harbour porpoise (Phocoena phocoena) population.Synthesis and applications. The interim framework can be used to forecast the effects of disturbances from human activities on animal populations, to assess the effectiveness of mitigation measures and to identify priority areas for research that reduces uncertainty in population forecasts. The last two applications are likely to be important in situations where there is a risk of unacceptable change in a species' conservation status. The framework should, however, be augmented with empirical data as soon as these are available.

Human activities that impact wildlife do not necessarily remove individuals from populations. They may also change individual behaviour in ways that have sublethal effects. This has driven interest in developing analytical tools that predict the population consequences of short-term behavioural responses. In this study, we incorporate empirical information on the ecology of a population of bottlenose dolphins into an individual-based model that predicts how individuals' behavioural dynamics arise from their underlying motivational states, as well as their interaction with boat traffic and dredging activities. We simulate the potential effects of proposed coastal developments on this population and predict that the operational phase may affect animals' motivational states. For such results to be relevant for management, the effects on individuals' vital rates also need to be quantified. We investigate whether the relationship between an individual's exposure and the survival of its calves can be directly estimated using a Bayesian multi-stage model for calf survival. The results suggest that any effect on calf survival is probably small and that a significant relationship could only be detected in large, closely studied populations. Our work can be used to guide management decisions, accelerate the consenting process for coastal and offshore developments and design targeted monitoring.