Leveraging social media and deep learning to detect rare megafauna in video surveys

Deep learning has become a key tool for the automated monitoring of animal populations with video surveys. However, obtaining large numbers of images to train such models is a major challenge for rare and elusive species because field video surveys provide few sightings. We designed a method that takes advantage of videos accumulated on social media for training deep-learning models to detect rare megafauna species in the field. We trained convolutional neural networks (CNNs) with social media images and tested them on images collected from field surveys. We applied our method to aerial video surveys of dugongs (Dugong dugon) in New Caledonia (southwestern Pacific). CNNs trained with 1303 social media images yielded 25% false positives and 38% false negatives when tested on independent field video surveys. Incorporating a small number of images from New Caledonia (equivalent to 12% of social media images) in the training data set resulted in a nearly 50% decrease in false negatives. Our results highlight how and the extent to which images collected on social media can offer a solid basis for training deep-learning models for rare megafauna detection and that the incorporation of a few images from the study site further boosts detection accuracy. Our method provides a new generation of deep-learning models that can be used to rapidly and accurately process field video surveys for the monitoring of rare megafauna.     

Animal abundance estimation in independent observer line transect surveys

The theory of conventional line transect surveys is based on an essential assumption that 100% detection of animals right on the transect lines can be achieved. When this assumption fails, independent observer line transect surveys are used. This paper proposes a general approach, based on a conditional likelihood, which can be carried out either parametrically or nonparametrically, to estimate the abundance of non-clustered biological populations using data collected from independent observer line transect surveys. A nonparametric estimator is specifically proposed which combines the conditional likelihood and the kernel smoothing method. It has the advantage that it allows the data themselves to dictate the form of the detection function, free of any subjective choice. The bias and the variance of the nonparametric estimator are given. Its asymptotic normality is established which enables construction of confidence intervals. A simulation study shows that the proposed estimator has good empirical performance, and the confidence intervals have good coverage accuracy.     

On the choice of statistical models for estimating occurrence and extinction from animal surveys

In surveys of natural animal populations the number of animals that are present and available to be detected at a sample location is often low, resulting in few or no detections. Low detection frequencies are especially common in surveys of imperiled species; however, the choice of sampling method and protocol also may influence the size of the population that is vulnerable to detection. In these circumstances, probabilities of animal occurrence and extinction will generally be estimated more accurately if the models used in data analysis account for differences in abundance among sample locations and for the dependence between site-specific abundance and detection. Simulation experiments are used to illustrate conditions wherein these types of models can be expected to outperform alternative estimators of population site occupancy and extinction.     

Random denominators and the analysis of ratio data

Ratio data, observations in which one random value is divided by another random value, present unique analytical challenges. The best statistical technique varies depending on the unit on which the inference is based. We present three environmental case studies where ratios are used to compare two groups, and we provide three parametric models from which to simulate ratio data. The models describe situations in which (1) the numerator variance and mean are proportional to the denominator, (2) the numerator mean is proportional to the denominator but its variance is proportional to a quadratic function of the denominator and (3) the numerator and denominator are independent. We compared standard approaches for drawing inference about differences between two distributions of ratios: t-tests, t-tests with transformations, permutation tests, the Wilcoxon rank test, and ANCOVA-based tests. Comparisons between tests were based both on achieving the specified alpha-level and on statistical power. The tests performed comparably with a few notable exceptions. We developed simple guidelines for choosing a test based on the unit of inference and relationship between the numerator and denominator.     

A new satellite-based methodology for continental-scale disturbance detection.

The timing, location, and magnitude of major disturbance events are currently major uncertainties in the global carbon cycle. Accurate information on the location, spatial extent, and duration of disturbance at the continental scale is needed to evaluate the ecosystem impacts of land cover changes due to wildfire, insect epidemics, flooding, climate change, and human-triggered land use. This paper describes an algorithm developed to serve as an automated, economical, systematic disturbance detection index for global application using Moderate Resolution Imaging Spectroradiometer (MODIS)/Aqua Land Surface Temperature (LST) and Terra/MODIS Enhanced Vegetation Index (EVI) data from 2003 to 2004. The algorithm is based on the consistent radiometric relationship between LST and EVI computed on a pixel-by-pixel basis. We used annual maximum composite LST data to detect fundamental changes in land-surface energy partitioning, while avoiding the high natural variability associated with tracking LST at daily, weekly, or seasonal time frames. Verification of potential disturbance events from our algorithm was carried out by demonstration of close association with independently confirmed, well-documented historical wildfire events throughout the study domain. We also examined the response of the disturbance index to irrigation by comparing a heavily irrigated poplar tree farm to the adjacent semiarid vegetation. Anomalous disturbance results were further examined by association with precipitation variability across areas of the study domain known for large interannual vegetation variability. The results illustrate that our algorithm is capable of detecting the location and spatial extent of wildfire with precision, is sensitive to the incremental process of recovery of disturbed landscapes, and shows strong sensitivity to irrigation. Disturbance detection in areas with high interannual variability of precipitation will benefit from a multiyear data set to better separate natural variability from true disturbance.     

A phenomenological model without dispersal kernel to model species migration

Phenomenological approaches to model species migration are usually based on kernel-based methods. These methods require a good knowledge of the dispersal agent behaviour for a given species. They also calculate the location of individuals independently to each other (except the mother plant) and then suppress some of them according to additional interactions such as competition, facilitation and recruitment. In this paper, we propose to use a new phenomenological method, the Gibbs method, to model tree species migration at large scale. The Gibbs method handles the location of adult individuals in terms of pairwise interactions described by a potential function. This function summarizes the set of known and unknown factors determining the spatial distribution of the individuals (or cohorts). The principle of the Gibbs method is to minimize the sum of all pairwise interactions, also called the cost function, in order to optimize the spatial point pattern according to the chosen potential function.     

Modeling Abundance Index Data from Anuran Calling Surveys

Abstract:  Evaluation of anuran populations is commonly based on calling surveys that report categorical abundance index data. I present a statistical model for abundance index data that are observations representing ordered abundance classes (e.g., none, some, many). The proposed model provides a formal treatment of detection probability, factors that affect detection, and variation in abundance. The model can be viewed as a generalization of that proposed by MacKenzie et al. (2002) for estimating site-occupancy rates in that it allows for more than two abundance classes. Because the abundance distribution is characterized by multiple abundance classes, it may be more sensitive to subtle changes in the underlying abundance that may go undetected with simple occupancy estimates under which sites are characterized merely as occupied or not. The method is most immediately applicable to surveys of anurans in which index data related to the intensity of calling activity are collected. I applied the proposed method to calling index data from the green frog (  Rana clamitans ) collected as part of the North American Amphibian Monitoring Program. The best model indicated considerable variation in detectability over time and in response to temperature. The resulting adjusted (for detectability) abundance-state distribution demonstrates the negative bias in abundance state obtained from simplistic summaries of calling index data that disregard these sources of variation in detectability.     

Experimental and environmental factors affect spurious detection of ecological thresholds

Threshold detection methods are increasingly popular for assessing nonlinear responses to environmental change, but their statistical performance remains poorly understood. We simulated linear change in stream benthic macroinvertebrate communities and evaluated the performance of commonly used threshold detection methods based on model fitting (piecewise quantile regression [PQR]), data partitioning (nonparametric change point analysis [NCPA]), and a hybrid approach (significant zero crossings [SiZer]). We demonstrated that false detection of ecological thresholds (type I errors) and inferences on threshold locations are influenced by sample size, rate of linear change, and frequency of observations across the environmental gradient (i.e., sample-environment distribution, SED). However, the relative importance of these factors varied among statistical methods and between inference types. False detection rates were influenced primarily by user-selected parameters for PQR (tau) and SiZer (bandwidth) and secondarily by sample size (for PQR) and SED (for SiZer). In contrast, the location of reported thresholds was influenced primarily by SED. Bootstrapped confidence intervals for NCPA threshold locations revealed strong correspondence to SED. We conclude that the choice of statistical methods for threshold detection should be matched to experimental and environmental constraints to minimize false detection rates and avoid spurious inferences regarding threshold location.     

Spatial statistical models that use flow and stream distance

We develop spatial statistical models for stream networks that can estimate relationships between a response variable and other covariates, make predictions at unsampled locations, and predict an average or total for a stream or a stream segment. There have been very few attempts to develop valid spatial covariance models that incorporate flow, stream distance, or both. The application of typical spatial autocovariance functions based on Euclidean distance, such as the spherical covariance model, are not valid when using stream distance. In this paper we develop a large class of valid models that incorporate flow and stream distance by using spatial moving averages. These methods integrate a moving average function, or kernel, against a white noise process. By running the moving average function upstream from a location, we develop models that use flow, and by construction they are valid models based on stream distance. We show that with proper weighting, many of the usual spatial models based on Euclidean distance have a counterpart for stream networks. Using sulfate concentrations from an example data set, the Maryland Biological Stream Survey (MBSS), we show that models using flow may be more appropriate than models that only use stream distance. For the MBSS data set, we use restricted maximum likelihood to fit a valid covariance matrix that uses flow and stream distance, and then we use this covariance matrix to estimate fixed effects and make kriging and block kriging predictions. Received: July 2005 / Revised: March 2006     

Local multiplicity adjustments for spatial cluster detection

The spatial scan statistic is a widely applied tool for cluster detection. The spatial scan statistic evaluates the significance of a series of potential circular clusters using Monte Carlo simulation to account for the multiplicity of comparisons. In most settings, the extent of the multiplicity problem varies across the study region. For example, urban areas typically have many overlapping clusters, while rural areas have few. The spatial scan statistic does not account for these local variations in the multiplicity problem. We propose two new spatially-varying multiplicity adjustments for spatial cluster detection, one based on a nested Bonferroni adjustment and one based on local averaging. Geographic variations in power for the spatial scan statistic and the two new statistics are explored through simulation studies, and the methods are applied to both the well-known New York leukemia data and data from a case–control study of breast cancer in Wisconsin.     

Residential radon concentration in the Abruzzo region (Italy): a different perspective for identifying radon prone areas

Indoor radon is an important risk factor for human health. Indeed radon inhalation is considered the second cause of lung cancer after smoking. During the last decades, in many countries huge efforts have been made in order to measuring, mapping and predicting radon levels in dwellings. Various researches have been devoted to identify those areas within the country where high radon concentrations are more likely to be found. Data collected through indoor radon surveys have been analysed adopting various statistical approaches, among which hierarchical Bayesian models and geostatistical tools are worth noting. The essential goal of this paper regards the identification of high radon concentration areas (the so-called radon prone areas) in the Abruzzo Region (Italy). In order to accurately pinpoint zones deserving attention for mitigation purpose, we adopt spatial cluster detection techniques, traditionally employed in epidemiology. As a first step, we assume that indoor radon measurements do not arise from a continuous spatial process; thus the geographic locations of dwellings where the radon measurements have been taken can be viewed as a realization of a spatial point process. Following this perspective, we adopt and compare recent cluster detection techniques: the simulated annealing scan statistic, the case event approach based on distance regression on the selection order and the elliptic spatial scan statistic. The analysis includes data collected during surveys carried out by the Regional Agency for the Environment Protection of Abruzzo (ARTA) in 1,861 random sampled dwellings across 277 municipalities of the Abruzzo region. The radon prone areas detected by the selected approaches are provided along with the summary statistics of the methods. Finally, the methodologies considered in this paper are tested on simulated data in order to evaluate their power and the precision of cluster location detection.     

Human health risk from Pb in urban street dust in northern UK cities

The presence of Pb in the environment can cause significant health problems. These issues are exasperated when the lead is in a more amenable form for potential ingestion. This study investigates the potential human health risk from Pb in urban street dusts. The lead levels in urban street dust in major city centres in northern UK have been compared to levels determined in 35 cities around the world. With a few exceptions, it was noted that the mean Pb levels in this study exceeded those found in other cities worldwide. Samples (n = 15) of urban street dust were collected across five city centres, and specifically in areas in which pedestrians are likely to concentrate, as well as near historical buildings. Typical total lead concentrations across all sampling sites ranged from 306 to 558 mg/kg. The human health risk was assessed using oral bioaccessibility testing of the urban street dust. The mean oral bioaccessibility data, irrespective of site and sample location, were in the range 43 ± 9 %. The total and bioaccessible concentrations of lead were compared to the estimated tolerable daily intake (TDI_oral) values based on unintentional soil/dust consumption. It is noted, in all cases, that the maximum estimated lead daily intake exceeded the TDI_oral. An alternative approach for assessing the daily intake is proposed based on the actual measured air quality in selected cities on the same day as the sampled urban dust.     

应用超敏感酵母HUG1-yEGFP生物传感器筛选基因毒性化合物

为了扩展本实验室构建的响应于DNA损伤信号的超敏感酵母HUG1-yEGFP生物传感器的检测范围并积累参考数据,为实际应用打下实验基础,本研究对26种与环境及人类生活息息相关的化合物的基因毒性进行了检测,筛选出8种具有基因毒性的化合物,并将其结果与现有微生物检测系统的检测结果进行了对比。结果显示,超敏感酵母HUG1-yEGFP生物传感器的检测结果与以细菌为宿主细胞的Ames试验或SOS显色法的检测结果的吻合度高于50%,与以酵母为宿主细胞的GSA检测系统的检测结果的吻合度高于85%。以真核生物酵母为宿主的检测系统与以原核生物细胞为宿主的检测系统的结果的差异反映了它们抗胁迫机制的不同。本研究应用超敏感酵母HUG1-yEGFP生物传感器特异性检测出3种化合物为基因毒性阳性,这3种化合物是代森锰锌、孔雀石绿及丙烯酰胺,均被报道具有潜在致癌性。本研究结果表明,超敏感酵母HUG1-yEGFP生物传感器可以有效地对环境污染有关的基因毒性化合物进行检测,可以作为其他现有检测系统的补充,应用于环境化学污染的健康风险评价。     

Maximum likelihood method for parameter estimation in non-linear models with below detection data

The maximum likelihood (ML) method for regression analyzes of censored data (below detection limit) for nonlinear models is presented. The proposed ML method has been translated into an equivalent least squares method (ML-LS). A two stage iterative algorithm is proposed to estimate statistical parameters from the derived least squares translation. The developed algorithm is applied to a nonlinear model for prediction of ambient air CO concentration in terms of concentrations of respirable particulate matter (RSPM) and NO₂. It has been shown that if censored data are ignored or estimated through simplifications such as (i) censored data are equal to detection limit, (ii) censored data are half of the difference between detection limit and lower limit (e.g., zero or background level) or (iii) censored data are equal to lower limit, this can cause significant bias in estimated parameters. The developed ML-LS method provided better estimates of parameters than any of the simplifications in censored data.     

Inference about density and temporary emigration in unmarked populations

Few species are distributed uniformly in space, and populations of mobile organisms are rarely closed with respect to movement, yet many models of density rely upon these assumptions. We present a hierarchical model allowing inference about the density of unmarked populations subject to temporary emigration and imperfect detection. The model can be fit to data collected using a variety of standard survey methods such as repeated point counts in which removal sampling, double-observer sampling, or distance sampling is used during each count. Simulation studies demonstrated that parameter estimators are unbiased when temporary emigration is either "completely random" or is determined by the size and location of home ranges relative to survey points. We also applied the model to repeated removal sampling data collected on Chestnut-sided Warblers (Dendroica pensylvancia) in the White Mountain National Forest, U.S.A. The density estimate from our model, 1.09 birds/ha, was similar to an estimate of 1.11 birds/ha produced by an intensive spot-mapping effort. Our model is also applicable when processes other than temporary emigration affect the probability of being available for detection, such as in studies using cue counts. Functions to implement the model have been added to the R package unmarked.     

荧光素酶表达基因法(CALUX)用于二噁英检测的研究进展

二噁英是当前环境中存在的毒性最大的物质,具有难分解性和高生物蓄积性,对人类造成了极大危害,其检测及控制至关重要。二噁英的检测方法主要包括高分辨气相色谱-高分辨质谱联机方法(HRGC-HRMS)、基于抗原-抗体的酶联免疫吸附法(ELISA)以及基于总毒性当量(TEQ)的生物检测法等,荧光素酶表达基因法(chemical-luciferase gene expression,CALUX)因其检测灵敏度和准确性高、检测范围广、操作简便快速、分析费用低等优点成为二噁英筛选检测的首选方法。本文对CALUX技术用于二噁英的检测研究进展进行了系统评述,并提出二噁英CALUX检测方法体系未来研究方向。     

Tree growth inference and prediction from diameter censuses and ring widths.

Estimation of tree growth is based on sparse observations of tree diameter, ring widths, or increments read from a dendrometer. From annual measurements on a few trees (e.g., increment cores) or sporadic measurements from many trees (e.g., diameter censuses on mapped plots), relationships with resources, tree size, and climate are extrapolated to whole stands. There has been no way to formally integrate different types of data and problems of estimation that result from (1) multiple sources of observation error, which frequently result in impossible estimates of negative growth, (2) the fact that data are typically sparse (a few trees or a few years), whereas inference is needed broadly (many trees over many years), (3) the fact that some unknown fraction of the variance is shared across the population, and (4) the fact that growth rates of trees within competing stands are not independent. We develop a hierarchical Bayes state space model for tree growth that addresses all of these challenges, allowing for formal inference that is consistent with the available data and the assumption that growth is nonnegative. Prediction follows directly, incorporating the full uncertainty from inference with scenarios for "filling the gaps" for past growth rates and for future conditions affecting growth. An example involving multiple species and multiple stands with tree-ring data and up to 14 years of tree census data illustrates how different levels of information at the tree and stand level contribute to inference and prediction.     

Terrestrial laserscanning of tidal flats—a case study in Jiangsu Province, China

Terrestrial laserscanning (TLS), also called ground-based LiDAR (Light Detection And Ranging) is a relatively new method which revolutionised geomorphological research in many domains. However, detailed studies of tidal flats by TLS have not been described in the literature yet. This study aims to fill this methodological gap by the application of TLS at two different locations on the coast of Jiangsu Province, Eastern China, and an assessment of the usability of this method for geomorphological research in such environments. The acquired point clouds are first processed to remove erroneous and noisy points. Subsequently, point clouds are computed to produce polygonal meshes and grid-based digital terrain model (DTM) more commonly used by the scientific community. The accuracy of the measurements is assessed by an analysis of elevation deviations for flat and horizontal concrete blocks. High quality point clouds with point densities of up to 4,000 points/m² were acquired for a distance of up to 200 m. The data allowed for the detection of small landforms such as tidal channels, creeks and ripples in centimetre and decimetre scale. The point clouds had an average error of approximately 3 mm, however for some few points errors of up to 1.8 cm were detected. Based on the results it can be concluded that TLS can be a useful additional method for geomorphological research on tidal flats due to its ability to describe the landforms from high density point clouds. Repeated scanning could therefore provide data to quantitatively and qualitatively describe geomorphological changes over wider areas and thereby improve the understanding of sedimentation and erosion on tidal flats.     

Geolocation of free-ranging fish on the European continental shelf as determined from environmental variables II. Reconstruction of plaice ground tracks

Between December 1993 and February 1997, 302 electronic data storage tags (DSTs), programmed to record depth at 10-min intervals and temperature daily, were attached to mature female plaice, Pleuronectes platessa, and released in the southern North Sea. Fifty tags were returned, 38 of which functioned fully and recorded 2,955 days of data. Twenty-seven tags recorded data over the full period at liberty, and 34 geographical ground tracks were reconstructed. Reconstruction was performed using a two-dimensional tidal stream simulation model that translated vertical movement of fish, recorded by DSTs, into horizontal movement assuming an initial down-tide swimming speed of 0.6 body lengths s^–1. Geographical accuracy of reconstructed tracks was assessed based on closeness of fit between (1) reconstruction endpoint and reported recapture position; (2) reconstructed locations and corresponding locations based on tidal data recorded by DSTs using the tidal location method (TLM; location of areas with similar tidal range and time of high water); and (3) DST temperature records and corresponding averaged sea surface temperature data records for corresponding locations. The results demonstrate that the assumptions of the tidal stream simulation model were sufficient to reconstruct geographically accurate representations of the migrations of individual plaice, which have in turn provided new information on the extent, duration, and directionality of movement. Our study demonstrates how DSTs can provide fishery-independent data with direct management applications in behaviourally driven, individual-based predictive models of fish migration.Communicated by J.P. Thorpe, Port Erin     

Hunter Reporting of Catch per Unit Effort as a Monitoring Tool in a Bushmeat‐Harvesting System

Abstract: Growing threats to biodiversity in the tropics mean there is an increasing need for effective monitoring that balances scientific rigor with practical feasibility. Alternatives to professional techniques are emerging that are based on the involvement of local people. Such locally based monitoring methods may be more sustainable over time, allow greater spatial coverage and quicker management decisions, lead to increased compliance, and help encourage attitude shifts toward more environmentally sustainable practices. Nevertheless, few studies have yet compared the findings or cost‐effectiveness of locally based methods with professional techniques or investigated the power of locally based methods to detect trends. We gathered data on bushmeat‐hunting catch and effort using a professional technique (accompanying hunters on hunting trips) and two locally based methods in which data were collected by hunters (hunting camp diaries and weekly hunter interviews) in a 15‐month study in Equatorial Guinea. Catch and effort results from locally based methods were strongly correlated with those of the professional technique and the spatial locations of hunting trips reported in the locally based methods accurately reflected those recorded with the professional technique. We used power simulations of catch and effort data to show that locally based methods can reliably detect meaningful levels of change (20% change with 80% power at significance level [α]= 0.05) in multispecies catch per unit effort. Locally based methods were the most cost‐effective for monitoring. Hunter interviews collected catch and effort data on 240% more hunts per person hour and 94% more hunts per unit cost, spent on monitoring, than the professional technique. Our results suggest that locally based monitoring can offer an accurate, cost‐effective, and sufficiently powerful method to monitor the status of natural resources. To establish such a system in Equatorial Guinea, the current lack of national and local capacity for monitoring and management must be addressed.