Occupancy Estimation Research Articles

Occupancy estimation using IoT sensors and machine learning: Incorporating ventilation system operating state and preprocessed differential pressure data

Indoor occupancy should be measured to reduce energy consumption in buildings and predict infection risk. Various studies used machine learning to measure occupancy using carbon dioxide concentration, Wi-Fi probes, and camera image data. The carbon dioxide concentration depends on factors such as whether doors and windows are open and whether ventilation systems are running. Carbon dioxide concentrations fluctuate depending on variables, even with the same occupants in the room. We estimated the number of occupants using indoor environmental data for considering the changes in the carbon dioxide concentration caused by these variables. The differential pressure and operating state of a ventilation system were used as input variables for machine learning. Data were acquired from a living lab, which is used for seminars, lectures, etc. For up to 50 people. We used 85,366 data for training and 8418 data for validation without training. The accuracies of Artificial Neural Network (ANN), random forest, and support vector machine models were compared for different input variables. This study considered the carbon dioxide concentration fluctuation due to leakage as an input variable coupled with differential pressure sensor data. We compared whether the accuracy increases when differential pressure data are used. In the ANN model, the lowest root mean squared error is obtained when the carbon dioxide concentration, differential pressure data, and operating state of the ventilation system are used as input variables. In the future, we plan to increase the accuracy by utilizing various Internet of Things sensors and adopt diversify the input variables.

Semisupervised Learning-Based Occupancy Estimation for Real-Time Energy Management Using Ambient Data

Occupancy status information is essential for efficient energy management of deferrable and flexible electricity loads in both residential and commercial sectors. Motivated by the hardness to collect ground-truth occupancy data, we propose a new, semi-supervised learning based occupancy estimator, the Label-expanded Time-series Propensity Weighted Estimator (LTPWE), to improve the prediction accuracy using privacy-preserving ambient sensing data (in temperature, humidity, light, etc.), especially when the labelling frequency (the ratio of the labeled positive instances) is low. The proposed energy management scheme integrates the LTPWE occupancy estimator into a data-driven deep reinforcement learning algorithm, the Soft Actor Critic (SAC), to make real-time scheduling decisions without any prior knowledge on the dynamics of random renewable generation and electricity price. Simulation results on real-world datasets show that compared with multiple state-of-the-art baselines, the proposed energy management scheme can reduce the energy cost by 18.79-55.79% without sacrificing occupants’ thermal comfort.

Occupancy estimation with environmental sensors: The possibilities and limitations

Occupancy detection and estimation in buildings paves the way to improve the utilization of lighting and HVAC systems, induce energy savings and enhance the well-being of the occupants. This paper presents a comparative study of state-of-art machine learning techniques that solve two different occupancy monitoring problems using environmental sensor data. One is the regression problem that estimates the actual count of occupants while the other is the classification problem which estimates the level of occupancy (empty, sparse, full). The results of the best performing machine learning techniques that solve both problems for the open dataset from the University of Southern Denmark, Odense are presented to compare the accuracy of both approaches and the ease of implementation. The impact of CO2, temperature, and humidity features on the occupancy count/levels and detection accuracy (occupied versus unoccupied) are studied. Comprehensive analysis with different combinations of environmental features and other free features such as time-of-day along with different sampling techniques for training and testing are performed to understand how such models can be adapted for actual deployment. Our results indicate detection accuracy between 66% to 82% for different sampling schemes; with day-based sampling showing a better performance while random sampling generically showcasing lower accuracy (66.2%). The occupancy estimation (levels or counts) has accuracy in the range of 69% to 79% for random sampling and 71% to 80% for day-based sampling. Finally, results demonstrate that models based single environmental sensor data streams do not perform as well as the models with sensor fusion.

The successes and pitfalls: Deep-learning effectiveness in a Chernobyl field camera trap application.

Camera traps have become in situ sensors for collecting information on animal abundance and occupancy estimates. When deployed over a large landscape, camera traps have become ideal for measuring the health of ecosystems, particularly in unstable habitats where it can be dangerous or even impossible to observe using conventional methods. However, manual processing of imagery is extremely time and labor intensive. Because of the associated expense, many studies have started to employ machine-learning tools, such as convolutional neural networks (CNNs). One drawback for the majority of networks is that a large number of images (millions) are necessary to devise an effective identification or classification model. This study examines specific factors pertinent to camera trap placement in the field that may influence the accuracy metrics of a deep-learning model that has been trained with a small set of images. False negatives and false positives may occur due to a variety of environmental factors that make it difficult for even a human observer to classify, including local weather patterns and daylight. We transfer-trained a CNN to detect 16 different object classes (14 animal species, humans, and fires) across 9576 images taken from camera traps placed in the Chernobyl Exclusion Zone. After analyzing wind speed, cloud cover, temperature, image contrast, and precipitation, there was not a significant correlation between CNN success and ambient conditions. However, a possible positive relationship between temperature and CNN success was noted. Furthermore, we found that the model was more successful when images were taken during the day as well as when precipitation was not present. This study suggests that while qualitative site-specific factors may confuse quantitative classification algorithms such as CNNs, training with a dynamic training set can account for ambient conditions so that they do not have a significant impact on CNN success.

Ultra-Wideband-Based Time Occupancy Analysis for Safety Studies.

This study investigates the use of ultra-wideband (UWB) tags in traffic conflict techniques (TCT) for the estimation of time occupancy in a real-world setting. This study describes UWB technology and its application in the TCT framework. Many experiments were conducted to evaluate the accuracy of the occupancy time measurement using a UWB-based tag. The UWB performance was measured using data from UWB tags as well as a video camera system by subtracting the time occupancy within a conflict zone. The results show that the UWB-based system can be utilized to estimate occupancy time with a mean absolute error difference from ground truth measurements of 0.43 s in the case of using two tags and 0.06 s in the case of using one tag in an 8 m × 8 m study area with double-sided two-way communication. This study also highlights the advantages and limitations of using UWB technology in TCT and discusses potential applications and future research directions. The findings of this study suggest that the UWB-based localization of multiple tags needs further improvements to enable consistent multi-tag tracking. In future work, this technology could be utilized to estimate post-encroachment time (PET) in various traffic scenarios, which could improve road safety and reduce the risk of collisions.

Estimating population size for California spotted owls and barred owls across the Sierra Nevada ecosystem with bioacoustics

Monitoring population size at ecosystem scales is difficult for most species of conservation concern. While assessing site occupancy at broad scales has proven feasible, rigorous tracking of changes in population size over time has not – even though it can provide a stronger basis for assessing population status and conservation-decision making. Therefore, we demonstrate how relatively low-intensity, ecosystem-scale passive acoustic monitoring (PAM) can be linked to local-density monitoring to estimate the population size of native California spotted owls (Strix occidentalis occidentalis) and invasive barred owls (S. varia) across the western Sierra Nevada, California. Based on a PAM sampling grid with 400 ha cells (the approximate home range size of these species), we estimated site occupancy to be between 0.42 (SE = 0.02) and 0.30 (SE = 0.02) for California spotted owls using relatively liberal and strict criteria, respectively, for considering a cell occupied. PAM-based site occupancy estimates within local-scale density monitoring study areas (range = 0.41–0.78 and 0.28–0.76 for liberal and strict criteria, respectively) were strongly and positively correlated with local density (range = 0.08–0.31 owl/km2) for this species. In contrast, ecosystem-wide site occupancy of barred owls was very low based on PAM (0.034, SE < 0.01), as were densities within local monitoring studies (range = 0–0.005 owls/km2). By scaling ecosystem-wide site occupancy estimates to densities estimated with local monitoring studies, we estimated that, depending on occupancy criteria, 2,218 (SE = 278) or 2,328 (SE = 489) California spotted owls occurred in the Sierra Nevada ecosystem in 2021. Thus, while California spotted owls are a rare subspecies, they were well-distributed across the Sierra Nevada. Because there were so few barred owl detections, we could not estimate ecosystem-scale abundance, which reflects the success of prior experimental removals in the region. In conclusion, our study provides a generalizable framework for estimating population size for territorial species with PAM at ecosystem scales when local-scale estimates of density are available. Thus, we demonstrate that this approach can provide novel and valuable insights into monitoring populations to aid species conservation.

Conservation planning in national parks may benefit from site occupancy and detection estimates of native animal species

Protected areas are the best strategy to conserve native biodiversity around the world. In Southeast Asia, most of the pristine tropical forest habitats are limited to terrestrial protected areas such as national parks and wildlife reserves. Endau Rompin National Park (ERNP) is one of undisturbed biodiversity-hotspots in the southern region of Peninsular Malaysia but surrounded by industrial oil palm plantations and threatened by habitat fragmentation. To date, little is known about the distribution and habitat requirement of wildlife, including those conservation priority species in the ERNP. This study aims to investigate the mammal distribution in the ERNP by estimating the probability of occurrence of a species among sampled sites. At 24 camera-trapping sites, we used passive infrared camera traps to detect the forest wildlife. We estimated wildlife occupancy by using single-season occupancy analysis while considering of imperfect detection and relating it to stand- and landscape-level variables. We detected 26 species of native wild animals in the study area including several conservation priority species such as Asian elephant, Asian tapir, and Malayan tiger. The site occupancy of conservation-priority species attests to ERNP's status as an important regional biodiversity hotspot. The detection probabilities of wildlife species were highly sensitive to specific ecological variables at stand- and landscape-level. Our data indicated that tree canopy cover, forest edges, adjacent human settlements, and riparian habitats (e.g., stream and river) are crucial for the persistence of the native wildlife species. Our results suggest that pristine forest habitat such as those typical to the ERNP are critical for wildlife conservation. We suggest that the stand- and landscape-level ecological variables are considered in the conservation planning to protect the forest biodiversity and ecological integrity of the national park and other terrestrial protected areas in the tropics.

Considerations for fitting occupancy models to data from eBird and similar volunteer-collected data

Abstract An occupancy model makes use of data that are structured as sets of repeated visits to each of many sites, in order to estimate the actual probability of occupancy (i.e. proportion of occupied sites) after correcting for imperfect detection using the information contained in the sets of repeated observations. We explore the conditions under which preexisting, volunteer-collected data from the citizen science project eBird can be used for fitting occupancy models. Because the majority of eBird’s data are not collected in the form of repeated observations at individual locations, we explore 2 ways in which the single-visit records could be used in occupancy models. First, we assess the potential for space-for-time substitution: aggregating single-visit records from different locations within a region into pseudo-repeat visits. On average, eBird’s observers did not make their observations at locations that were representative of the habitat in the surrounding area, which would lead to biased estimates of occupancy probabilities when using space-for-time substitution. Thus, the use of space-for-time substitution is not always appropriate. Second, we explored the utility of including data from single-visit records to supplement sets of repeated-visit data. In a simulation study we found that inclusion of single-visit records increased the precision of occupancy estimates, but only when detection probabilities are high. When detection probability was low, the addition of single-visit records exacerbated biases in estimates of occupancy probability. We conclude that subsets of data from eBird, and likely from similar projects, can be used for occupancy modeling either using space-for-time substitution or supplementing repeated-visit data with data from single-visit records. The appropriateness of either alternative will depend on the goals of a study and on the probabilities of detection and occupancy of the species of interest.

Unsupervised domain adaptation with and without access to source data for estimating occupancy and recognizing activities in smart buildings

Energy-efficient buildings have gained increasing interest in the last decades as they provide optimal energy management. With the emergence of smart homes, many smart tools have been developed to optimize energy efficiency such as activities recognition (AR) and occupancy estimation (OE). The creation of these smart building tools may be disrupted by the scarcity of labeled data. Indeed, data labeling is a tedious and time-consuming task and, it can be very expensive for building objectives. However, labeled data scarcity can be solved by sharing knowledge from different domains using unsupervised domain adaptation techniques. Also, privacy issues can emerge and prevent the use of certain types of data that share residents’ attitudes. However, data privacy can be preserved by considering approaches that do not have direct access to original data. In this research, we provide a comparative analysis between unsupervised domain adaptation (UDA) methods, applied to the tasks of AR and OE, with and without direct access to source domain data. We have 6 adapted approaches for UDA with access to labeled source data: domain separation networks (DSN), cluster alignment with a teacher (CAT), CAT+ gradient reversal (RevGrad), CAT + robust RevGrad (rRevGrad), Auxiliary Target Domain-Oriented Classifier (ATDOC) with nearest centroid classifier (NC), and ATDOC with neighborhood aggregation (NA). Also, we have 6 adapted methods for UDA without labeled source data: confidence score weighting adaptation using joint model data structure (CoWA-JMDS), CoWA-JMDS without weights mixup, divide and contrast (DaC), attracting and dispersing (AaD), source hypothesis transfer with information maximization (SHOT-IM), and source hypothesis transfer with self-supervised pseudo-labeling (SHOT-Pseudo-labeling). All the considered methods have been tested on AR and OE datasets collected using ambient sensors. The comparative analysis made in this work has shown us impressive findings and has given great ideas about the type of approaches (with access or without access to the source data) that we should consider for real-world smart building applications.

Equation-based modeling and optimization-based parameter estimation in multimodal virtual sensing platforms for smart buildings

This paper considers a unified approach for virtual sensing in smart buildings that utilizes equation-based modeling and optimization-based parameter estimation. A modeling method is introduced to describe relationships among existing sensory information and application-oriented parameters in a reusable manner. Measurement errors in existing sensors are also described in this equation-based model. When parameters to be estimated and existing sensors used for virtual sensing are identified, relevant equations are selected and combined as simultaneous equations. The parameter estimation method takes an optimization-based approach to cope with uncertainty in sensory information. Simultaneous equations corresponding to target sensing are used as a constraint in the optimization problem. In cases where many different sensors are utilized to increase accuracy, we may need to consider compromising contradictions among sensory information due to sensing errors. If the target building does not have enough sensors for the intended parameter estimation, we may need certain assumptions to determine the value. Such over- or under-defined situations are automatically detected and considered in our parameter estimation mechanism. For this purpose, we introduce a method, based on an idea inspired by randomized algorithms, for repeatedly solving an optimization problem with different weight factors in the objective function and for analyzing any fluctuations in estimated values. This is computationally inexpensive and can be applied to big real-world problems. We applied this method to building occupancy estimations for efficient air-conditioning control and to customer attribute analysis for an office building cafeteria during lunch time.

Efficacy of machine learning image classification for automated occupancy‐based monitoring

AbstractRemote cameras have become a widespread data‐collection tool for terrestrial mammals, but classifying images can be labor intensive and limit the usefulness of cameras for broad‐scale population monitoring. Machine learning algorithms for automated image classification can expedite data processing, but image misclassifications may influence inferences. Here, we used camera data for three sympatric species with disparate body sizes and life histories – black‐tailed jackrabbits (Lepus californicus), kit foxes (Vulpes macrotis), and pronghorns (Antilocapra americana) – as a model system to evaluate the influence of competing image classification approaches on estimates of occupancy and inferences about space use. We classified images with: (i) single review (manual), (ii) double review (manual by two observers), (iii) an automated‐manual review (machine learning to cull empty images and single review of remaining images), (iv) a pretrained machine‐learning algorithm that classifies images to species (base model), (v) the base model accepting only classifications with ≥95% confidence, (vi) the base model trained with regional images (trained model), and (vii) the trained model accepting only classifications with ≥95% confidence. We compared species‐specific results from alternative approaches to results from double review, which reduces the potential for misclassifications and was assumed to be the best approximation of truth. Despite high classification success, species‐level misclassification rates for the base and trained models were sufficiently high to produce erroneous occupancy estimates and inferences related to space use across species. Increasing the confidence thresholds for image classification to 95% did not consistently improve performance. Classifying images as empty (or not) offered a reasonable approach to reduce effort (by 97.7%) and facilitated a semi‐automated workflow that produced reliable estimates and inferences. Thus, camera‐based monitoring combined with machine learning algorithms for image classification could facilitate monitoring with limited manual image classification.

The Value of Better Vehicle Occupancy Estimates for Project Prioritization

Transportation projects are increasingly evaluated based on how they improve person throughput, which requires knowing the passenger car occupancy. Yet obtaining occupancies is labor intensive, leading agencies to rely on statewide values. How useful are locality-specific vehicle occupancies? A case study of 38 candidate Virginia highway investments showed an occupancy change of .15 randomly applied to one-half of the projects affects 24% to 42% of project rankings if locality-specific occupancies are used. Occupancy is not the only data element affecting project prioritization, but it is an underlooked one. The study showed that an occupancy uncertainty of .15 is equivalent to an uncertainty of 3.90% in the discount rate or 13.2% in assigned traffic volume. As these investments had a mean value of $17 M, this study demonstrated that when project prioritization metrics are based on person throughput, city-or county-specific occupancies are an integral component of a data-driven prioritization process.

Integrating community science and agency‐collected monitoring data to expand monitoring capacity at large spatial scales

AbstractMonitoring species to better understand their status, ecology, and management needs is a major expense for agencies tasked with biodiversity conservation. Community science data have the potential to improve monitoring for minimal cost, given appropriate analytical frameworks. We describe a framework for integrating data from the eBird community science platform with agency‐collected monitoring data using a multistate occupancy model. Our model accounts for the structural differences across datasets and allows for estimation of both occupancy and breeding probabilities. The framework was applied to Common Loons (Gavia immer) in Washington State. A total of 766 sites had observation effort, of which 713 sites had only eBird effort, 26 sites had only Washington Department of Fish and Wildlife (WDFW) effort, and 27 sites had both. We predicted that the probability of occupancy was only 0.07 (95% Bayesian credible interval, BCI = 0.02–0.51) at the 2324 sites in our sampling frame, though the probability that Common Loons were breeding at occupied sites was 0.95 (95% BCI = 0.71–1.00). We found that probability of occupancy was positively related to waterbody size (probability of a positive effect = 0.88) and negatively related to an index of human influence (probability of a negative effect = 0.94). We found that probability of breeding at occupied sites was positively related to tree canopy cover (0.86), negatively related to elevation (0.99), and negatively related to barren, scrub/shrub, and herbaceous land cover (0.98). We found that state agency biologists were 16 times more likely to detect breeding Common Loons at a site than were eBird users (0.94, 95% BCI = 0.78–0.99 for agency biologists vs. 0.08, 95% BCI = 0.06–0.10 for eBird users). However, the amount of effort expended by eBird users meant that they confirmed Common Loons at 94 sites while agency biologists confirmed them at just 24 sites, although evidence of reproduction was only contributed by agency biologists. Our results provide a better understanding of the distribution of Common Loons in Washington, while further demonstrating that community science data can be a valuable complement to agency‐collected data, if appropriate frameworks are developed to integrate these data sources.

Density and abundance estimate of Antillean manatees (Trichechus manatus) within a landlocked lake in Southeastern México

Manatees are threatened along their range. In México, this species is listed as endangered. Manatee conservation strategies require density or occupancy estimates, especially in areas where species face survival risks. On turbid waters, like those found in rivers and lakes of the Southern Gulf of México coast, visual methods used to detect and count manatees underestimate actual numbers. Our goal was to estimate the density and abundance (N) of Antillean manatees in a small State Natural Protected Area, where a manatee population of unknown size inhabits. We performed line transects using a side scan sonar to detect animals and mark-recapture in the isolated population of manatees within Laguna de las Ilusiones, México, a landlocked lake that excludes transit. Using distance sampling from 14 boat trips, estimates of density and abundance were 15.5 manatees km-2 and 27 ± 5 manatees in the lake (CV ≈ 16.6 %). With MARK, from six capture events and 19 individual encounter records, the abundance estimated was 24 manatees (CV ≈ 16 to 24 %). Previous number of manatees were based only on visual surveys, which reported at least seven manatees. Density is lower than other similar studies along narrow waterways in important areas in México and other Central and South America countries. Studying this endangered subspecies is limited by cryptic habits, turbid waters, poor funding, and low densities, making density or abundance estimates difficult. However, within particular areas and established monitoring areas, these methods could be useful to generate baselines for conservation strategies.

Job exposure matrices for occupational respiratory disease: a narrative review.

Job exposure matrices (JEMs) are epidemiological tools used to provide estimations of occupational exposures when it is not feasible to complete detailed individual occupational histories. To identify and summarize the characteristics of published general population JEMs (GPJEM) of inhalable occupational exposures applied in studies of respiratory disease. MEDLINE and EMBASE databases were searched using pre-defined search terms, with screening performed by two independent reviewers to identify studies reporting the use of a GPJEM. JEM creation papers were subsequently identified and reviewed for each individual GPJEM, noting its characteristics in terms of occupational classification system and exposure estimates. From 728 studies identified in initial searches, 33 GPJEMs of inhalable occupational exposures were identified. Versions of the International Standards Classification of Occupations were the most used occupational classification system. Binary, probability and intensity-based exposure estimates were most frequently reported in GPJEMs. Selection of a GPJEM to apply in epidemiological research should be based on the exposure(s) of interest, time period of occupations under review, geographical region for intended use, occupation classification system used and the exposure estimate outcome.

Passive acoustic monitoring indicates Barred Owls are established in northern coastal California and management intervention is warranted

Abstract Barred Owls (Strix varia) have recently expanded westward from eastern North America, contributing to substantial declines in Northern Spotted Owls (Strix occidentalis caurina). Passive acoustic monitoring (PAM) represents a potentially powerful approach for tracking range expansions like the Barred Owl’s, but further methods development is needed to ensure that PAM-informed occupancy models meaningfully reflect population processes. Focusing on the leading edge of the Barred Owl range expansion in coastal California, we used a combination of PAM data, GPS-tagging, and active surveys to (1) estimate breeding home range size, (2) identify patterns of vocal activity that reflect resident occupancy, and (3) estimate resident occupancy rates. Mean breeding season home range size (452 ha) was reasonably consistent with the size of cells (400 ha) sampled with autonomous recording units (ARUs). Nevertheless, false-positive acoustic detections of Barred Owls frequently occurred within cells not containing an activity center such that site occupancy estimates derived using all detected vocalizations (0.61) were unlikely to be representative of resident occupancy. However, the proportion of survey nights with confirmed vocalizations (VN) and the number of ARUs within a sampling cell with confirmed vocalizations (VU) were indicative of Barred Owl residency. Moreover, the false positive error rate could be reduced for occupancy analyses by establishing thresholds of VN and VU to define detections, although doing so increased false negative error rates in some cases. Using different thresholds of VN and VU, we estimated resident occupancy to be 0.29–0.44, which indicates that Barred Owls have become established in the region but also that timely lethal removals could still help prevent the extirpation of Northern Spotted Owls. Our findings provide a scalable framework for monitoring Barred Owl populations throughout their expanded range and, more broadly, a basis for converting site occupancy to resident occupancy in PAM programs.

Observers are a key source of detection heterogeneity and biased occupancy estimates in species monitoring

Reliable assessments of population status and trends underpin conservation management efforts but are complicated by the fact that imperfect detection is ubiquitous in monitoring data. We explore the most commonly considered variables believed to influence detection probabilities, quantifying how they influence detectability and assessing how occupancy rates are impacted when a variable is ignored. To do so, we used data from two multi-species amphibian monitoring programmes, collected by volunteers and professional surveyors.Our results suggest that although detection rates varied substantially in relation to commonly considered factors such as seasonal and annual effects, ignoring these factors in the analysis of monitoring data had negligible effect on estimated occupancy rates. Variation among surveyors in detection probabilities turned out to be most important. It was high and failing to account for it led to occupancy being underestimated. Importantly, we identified that heterogeneity among observers was as high for professional surveyors as for volunteers, highlighting that this issue is not restricted to citizen-science monitoring.Occupancy modelling has greatly improved the reliability of inference from species monitoring data, yet capturing the relevant sources of variation remains a challenge. Our results highlight that variation among surveyors is a key source of heterogeneity, and that this issue is just as pertinent to data collected by experts as by volunteers. Detection heterogeneity should be accounted for when analysing monitoring data. Furthermore, efforts to increase training of field crews and collecting data to quantify differences between observer abilities are important to avoid biased inference resulting from unmodelled observer differences.

Increasing sampling efficiency of Bombus communities with rare and endangered species by optimizing detection probabilities: A multi-species occupancy modelling approach using roadsides as a case study

Nearly one-quarter of bumble bee species assessed by the International Union for the Conservation of Nature are considered to be in decline, necessitating monitoring to support recovery planning. A key element of robust wildlife monitoring is accounting for imperfect detection in the observation process and efforts to understand sources of detection uncertainty in bumble bee surveys began recently in response to the listing of the rusty patched bumble bee, Bombus affinis, as federally endangered in the U.S. and Canada. We investigated within-community variation in detection probabilities among eight species that vary in occupancy, including B. affinis. We sampled bumble bees along roadsides in the greater metropolitan area of Minneapolis-St. Paul, Minnesota in the United States in 2018. We generated estimates of occupancy, detection probability, and cumulative detection probability from a multispecies occupancy model using a Bayesian framework that included covariates on occupancy and detection. Covariates selected a priori included floral cover and impervious surface effects on occupancy, and date, time of day, and observer effects on detection. We found that Bombus species with low occurrence are not necessarily less detectable than common species with higher occurrence, and that species reach their peak detection probability at different dates. Detection was higher in the afternoon and varied among observers. Well-designed Bombus monitoring programs can reduce sampling effort and optimize detection by sampling on dates near species detection peaks, sampling in the afternoon, and using cumulative detection probabilities to determine optimum sampling effort.

The recording behaviour of field-based citizen scientists and its impact on biodiversity trend analysis

Opportunistic species sightings submitted by citizen science volunteers are a valuable source of species data for trends analysis, as used in biodiversity indicators. However, projects collecting these data give people flexibility where and when to make records, and the recording behaviour of participants varies between individuals. Here we tested the effect of recorder behaviour on outputs of the analysis of temporal biodiversity trends. Using a large (c. 3 million records), 20 year unstructured citizen science dataset of butterfly records in Great Britain, we manipulated recorder behaviour by constructing biased 50% subsamples of the dataset by preferentially including different types of recorders (based on high and low values of four metrics independently describing the temporal, spatial and taxonomic attributes of recorder behaviour). We found that, in general, the three outputs (namely: occupancy trend, precision of the trend, and the estimate of occupancy) showed relatively little deviation from random expectation across most of the different types of recorder behaviour. Occupancy trends showed least deviation, while estimates of occupancy itself showed greatest deviation from the random expectation. Regarding the recorder behaviours, the outputs were most sensitive to variation in ‘recorder potential’, which describes the difference between ‘thorough’ and ‘incidental’ recorders. Importantly, by demonstrating the robustness of occupancy trends to differences in recorder behaviour, this study provides support for the appropriate use of occupancy trend modelling for unstructured citizen science. However, we did not consider change in recorder behaviour over time, so further research is required to assess the impact of this on trend modelling. This study highlights the value of developing solutions to further increase the robustness of biodiversity trend analysis. These solutions should include both analytical developments and enhancements in project design to engage participants.

Integrating multiple sign types to improve occupancy estimation for inconspicuous species.

Standard occupancy models enable unbiased estimation of occupancy by accounting for observation errors such as missed detections (false negatives) and, less commonly, incorrect detections (false positives). Occupancy models are fitted to data from repeated site visits in which surveyors record evidence of species presence. Use of indirect sign (e.g., scat, tracks) as evidence of presence can vastly improve survey efficiency for inconspicuous species but can also introduce additional sources of error. We developed a "multi-sign" occupancy approach to model the detection process separately for unique sign types and used this method to improve estimates of occupancy dynamics for an inconspicuous species, the American pika (Ochotona princeps). We investigated how estimates of pika occupancy and environmental drivers differed under four increasingly realistic representations of the observation process: (1) perfect detection (commonly assumed for modeling pika occupancy), (2) standard occupancy model (single observation process without possibility of false detection), (3) multi-sign with no false detections (non-false positive model), and (4) multi-sign with false detections (full model). For the multi-sign occupancy models, we modeled the detection of each sign type (fresh scat, fresh haypiles, pika calls, and pika sightings) separately as a function of climatic and environmental covariates. Estimates of occupancy processes and inferences about environmental drivers were sensitive to different detection models. Simplified representations of the detection processes generally resulted in higher occupancy estimates and higher turnover rates than the full multi-sign model. Environmental drivers also varied in their influence on occupancy models, where (e.g.) forb cover was estimated to more strongly influence occupancy in the full multi-sign model than the simpler models. As has been reported previously in other contexts, unmodeled heterogeneity in the observation process can lead to biases in occupancy processes and uncertainty in the relationships between occupancy and environmental covariates. Overall, our multi-sign approach to dynamic occupancy modeling, which accounts for spatio-temporal variation in reliability among sign types, has strong potential to generate more realistic estimates of occupancy dynamics for inconspicuous species.