Deep learning‐based ecological analysis of camera trap images is impacted by training data quality and quantity

Aim To test for control of vascular plant species richness in the riparian corridor by exploring three contrasting (although not mutually exclusive) hypotheses: (1) longitudinal patterns in riparian plant species richness are governed by local, river‐related processes independent of the regional species richness, (2) riparian plant species richness is controlled by dispersal along the river (longitudinal control), and (3) the variation in riparian plant species richness mirrors variation in regional richness (lateral control).Location The riparian zones of the free‐flowing Vindel River and its surrounding river valley, northern Sweden.Methods We used data from three surveys, undertaken at 10‐year intervals, of riparian reaches (200‐m stretches of riverbank) spanning the entire river. In addition, we surveyed species richness of vascular plants in the uplands adjacent to the river in 3.75‐km2 large plots along the same regional gradient. We explored the relationship between riparian and upland flora, and various environmental variables. We also evaluated temporal variation in downstream patterns of the riparian flora.Results Our results suggest that local species richness in boreal rivers is mainly a result of local, river‐related processes and dispersal along the corridor. The strongest correlation between species richness and the environment was a negative one between species number and soil pH, but pH varied within a narrow range. We did not find evidence for a correlation between species richness on regional and local scales. We found that the local patterns of species richness for naturally occurring vascular plants were temporally variable, probably in response to large‐scale disturbance caused by extreme floods. Most previous studies have found a unimodal pattern of species richness with peaks in the middle reaches of a river. In contrast, on two of three occasions corresponding to major flooding events, we found that the distribution of species richness of naturally occurring vascular plants resembled that of regional diversity: a monotonic decrease from headwater to coast. We also found high floristic similarity between the riparian corridor and the surrounding landscape.Main conclusions These results suggest that local processes control patterns of riparian species richness, but that species composition is also highly dependent on the regional species pool. We argue that inter‐annual variation in flood disturbance is probably the most important factor producing temporal variability of longitudinal species richness patterns.

Diversität und Biogeographie der Farne und Vögel Boliviens: Niche Modellierung GIS Applicationen

Knowledge on the spatial distribution of plant species richness and its associated factors is crucial for addressing several ecological problems. Climate factors have long been used to explain species richness patterns across space and time. In this study, we investigated how water and energy variables shape species richness patterns on the alpine Tibetan plateau. An empirical study involving 800 plots of size 1 m2 on 80 different sites was conducted across two transects, covering the arid and semi‐arid alpine region of the western part of the Tibetan plateau to explore plant species composition and richness in relation to climatic factors at a regional scale. We performed detrended correspondence analysis and generalized linear models to explore the effect of energy and water variables on species composition and species richness, respectively. We used a generalized linear mixed model to investigate the interactive effect of energy variables and water variables on species richness. We found that both energy variables and water variables significantly influenced species richness and composition on the alpine Tibetan plateau and supported the water–energy dynamics (WED) hypothesis. Specificall, plant species richness increased with greater liquid water availability. However, it exhibited a unimodal relationship with potential evapotranspiration. Our results also showed that precipitation played a more substantial role in the model than potential evapotranspiration, but their interaction strengthened the model. In conclusion, this study highlights the dominant role of precipitation in explaining species composition and richness patterns on the alpine Tibetan plateau and suggests conducting research on additional environmental factors affecting water availability in this region.

The regressive succession model hypothesizes tropical savanna‐woodlands to be a degraded stage of primary deciduous forests. Species diversity, richness and evenness of woody species in savanna‐woodlands, secondary deciduous forests and mature deciduous forests of central India were compared to test if the regressive succession explained pattern in species richness, diversity, functional diversity and basal area. At the plot scale (0.1 ha) secondary deciduous forests and savanna‐woodlands had similar species diversity, a pattern not consistent with the regressive model of deciduous forest succession, and mature deciduous forests had greater species diversity and richness (p<0.05). When examined at a larger scale or community scale by pooling all plots within a community type, the trend in diversity persisted even with greater effort allocated to sampling of secondary deciduous forests. Species richness at the community scale was greatest in secondary deciduous forest as expected from species area relationship. The communities shared 28 woody species but the species composition was significantly different between the communities. I suggest that conservation of tropical deciduous forests based on regressive succession model is problematic.

Evolutionary processes underlying spatial patterns in species richness remain largely unexplored, and correlative studies lack the theoretical basis to explain these patterns in evolutionary terms. In this study, we develop a spatially explicit simulation model to evaluate, under a pattern‐oriented modeling approach, whether evolutionary niche dynamics (the balance between niche conservatism and niche evolution processes) can provide a parsimonious explanation for patterns in species richness. We model the size, shape, and location of species’ geographical ranges in a multivariate heterogeneous environmental landscape by simulating an evolutionary process in which environmental fluctuations create geographic range fragmentation, which, in turn, regulates speciation and extinction. We applied the model to the South American domain, adjusting parameters to maximize the correspondence between observed and predicted patterns in richness of about 3,000 bird species. Predicted spatial patterns, which closely resemble observed ones ($$r^{2}=0.795$$), proved sensitive to niche dynamics processes. Our simulations allow evaluation of the roles of both evolutionary and ecological processes in explaining spatial patterns in species richness, revealing the enormous potential of the link between ecology and historical biogeography under integrated theoretical and methodological frameworks.

AimOwing to the broad use of camera traps, integration and standardization among camera trap studies has become key to maximizing their utility for local and global biodiversity conservation. Our goal was to introduce the use of a hierarchical modelling framework in the context of coordinated biodiversity monitoring to compare species richness and occupancy by integrating camera trap data from multiple study areas.LocationSouthwest China.MethodsWe used hierarchical occupancy models to integrate camera trap data for elusive mammal and pheasant communities from three study areas representing different habitat types: alpine and subalpine zones, dry‐hot valleys and subtropical montane forests. We evaluate the responses of species occurrence to human influence and habitat parameters based on a Bayesian approach.ResultsWe captured photographs of 23 mammal and 7 pheasant species over 10,095 trap nights. The model revealed that the alpine and subalpine zones supported the highest species richness of the target communities among the three habitat types. Surprisingly, dry‐hot valleys supported similar levels of species richness to subtropical montane forest. Species richness showed a similar bell‐shaped relationship with elevation, with the richness curve peaking at intermediate elevations at about 3500 m above sea level (asl). Posterior distributions for community‐level hyper‐parameters indicated the consistent and negative effects of human disturbance on species occupancy. The community model also revealed a strong quadratic relationship between elevation and occupancy, with the highest occupancy occurring at about 3700 m asl.Main conclusionUsing hierarchical occupancy models for integrating camera trap data from multiple study areas, we show that alpine/subalpine zone and dry‐hot valleys have the highest richness and should be given more priority for conservation of biodiversity in southwest China. We recommend broader application of the hierarchical occupancy modelling approach to camera trap data to obtain more comprehensive insights relevant to regional biodiversity conservation.

Machine learning is rapidly developing as a tool for gathering data from imagery and may be useful in identifying (classifying) visible specimens in large numbers of seabed photographs. Application of an automated classification workflow requires manually identified specimens to be supplied for training and validating the model. These training and validation datasets are generally generated by partitioning the available manual identified specimens; typical ratios of training to validation dataset sizes are 75:25 or 80:20. However, this approach does not facilitate the desired scalability, which would require models to successfully classify specimens in hundreds of thousands to millions of images after training on a relatively small subset of manually identified specimens. A second problem is related to the ‘class imbalance’, where natural community structure means that fewer specimens of rare morphotypes are available for model training. We investigated the impact of independent variation of the training and validation dataset sizes on the performance of a convolutional neural network classifier on benthic invertebrates visible in a very large set of seabed photographs captured by an autonomous underwater vehicle at the Porcupine Abyssal Plain Sustained Observatory. We tested the impact of increasing training dataset size on specimen classification in a single validation dataset, and then tested the impact of increasing validation set size, evaluating ecological metrics in addition to computer vision metrics. Computer vision metrics (recall, precision, F1-score) indicated that classification improved with increasing training dataset size. In terms of ecological metrics, the number of morphotypes recorded increased, while diversity decreased with increasing training dataset size. Variation and bias in diversity metrics decreased with increasing training dataset size. Multivariate dispersion in apparent community composition was reduced, and bias from expert-derived data declined with increasing training dataset size. In contrast, classification success and resulting ecological metrics did not differ significantly with varying validation dataset sizes. Thus, the selection of an appropriate training dataset size is key to ensuring robust automated classifications of benthic invertebrates in seabed photographs, in terms of ecological results, and validation may be conducted on a comparatively small dataset with confidence that similar results will be obtained in a larger production dataset. In addition, our results suggest that automated classification of less common morphotypes may be feasible, providing that the overall training dataset size is sufficiently large. Thus, tactics for reducing class imbalance in the training dataset may produce improvements in the resulting ecological metrics.

Several studies have recently reported that common species are more important for species richness patterns than rare species. However, most such studies have been based on broad‐scale atlas data. We studied the contribution of different species occupancy, i.e. number of plots occupied, to species richness patterns emerging from species data in 50 by 50 m plots within six 140–200 ha forests in Norway. The study included vascular plants, lichens, bryophytes, and polypore fungi. We addressed the following questions: 1) are common species more correlated with species richness than rare species? 2) How do occupancy classes combine at various levels of species richness? 3) Which occupancy class is best in identifying the overall most species‐rich sites (hotspots) by sampling? The results showed that rare species were better correlated with species richness than common species when the information content was accounted for, that high species richness was associated with a higher proportion of less frequent species, and that the best occupancy class for local hotspot identification was species present in 10–30% of the plots within a forest. We argue that the observed correlations between overall richness and sub‐assembly richness are primarily structured by the combination of the distributions of species richness and species occupancy. Although these distributions result from general ecological processes, they may also be strongly affected by idiosyncratic elements of the individual datasets caused by the specific environmental composition of a study area. Hence, different datasets collected in different areas may lead to different results regarding the relative importance of common versus rare species, and such effects should be expected on both broad and fine spatial scales. Despite these effects, we suggest that infrequent species will tend to be more strongly correlated to species richness at local scales than at broader scales as a result of more right‐skewed species‐occupancy distributions.

Aim To examine patterns in anuran species richness along an elevation gradient and identify factors that govern anuran species richness on a tropical elevational gradient.Location Sampling for anurans was carried out in Kalakad Mundanthurai Tiger Reserve (KMTR) in the southern Western Ghats, India.Methods Night‐time sampling for anuran species richness was carried out from 20 November 2004 to 20 April 2005, during the north‐east monsoon and dry seasons, using transects (50 × 2 m) and visual encounter surveys along the streams. The entire gradient was classified into thirteen 100‐m elevation zones. Sampling at the alpha (single drainage basin) level was carried out in the Chinnapul River drainage basin (40–1260 m a.s.l.) and at the gamma (landscape) level in four drainage basins. Additionally, published records were used to arrive at an empirical species richness (S) for the entire landscape. Mid‐Domain Null software was used to test for the possible influence of geometric constraints on anuran species at both the alpha and gamma levels. The influence of area under each elevation zone on empirical S was tested. The pattern in anuran species richness along the elevational gradient was investigated using: (1) species boundaries in each elevation zone and their habitat correlates, (2) abiotic factors as predictor variables, (3) mean snout vent lengths of anurans, and (4) correlation between the matrices of distance in the elevation zones based on microhabitat parameters and species composition. Cluster analysis on species presence–absence in the elevation zones was used to categorize the entire gradient into high, middle and low elevations. In these three elevation categories, pattern in composition of species was examined for endemism in Western Ghats–Sri Lanka biodiversity hotspot, uniqueness to an elevation zone, adaptations of adults and modes of breeding.Results Species richness at the alpha level increased linearly with elevation, while at the gamma level there were three peaks. Maximum species richness was observed at the highest elevation (1200 m) at both the alpha and the gamma levels. The observed patterns differed significantly from mid‐domain null predictions. The multi‐modal pattern in species richness was a consequence of overlapping species range boundaries. Soil temperature was the best single measure in explaining the majority of variation in species richness at the alpha level (r2 = 0.846, P < 0.01). However, soil moisture was the best predictor when both the alpha and the gamma sites were pooled (r2 = 0.774, P < 0.01). Anuran body size decreased with an increase in elevation. The highest proportions of endemic and unique species were found at high elevations (> 700 m). The proportion of arboreal anurans increased from low to high elevation. Anurans exhibiting direct development were predominantly found at high elevations.Main conclusions Geometric constraints did not influence anuran species richness along the elevational gradient. Overlapping range boundaries influenced species richness at the gamma level. Abiotic factors such as soil temperature and moisture influenced anuran species richness in the mountain range. The ‘Massenerhebung effect’ could be responsible for range restriction and endemism of anurans, differences in guilds and mode of reproduction. These findings highlight the importance of cloud forests for endemic anurans.

Multiple determinants of Australian tropical frog biodiversity

Most studies examining continental-to-global patterns of species richness rely on the overlaying of extent-of-occurrence range maps. Because a species does not occur at all locations within its geographic range, range-map-derived data represent actual distributional patterns only at some relatively coarse and undefined resolution. With the increasing availability of high-resolution climate and land-cover data, broad-scale studies are increasingly likely to estimate richness at high resolutions. Because of the scale dependence of most ecological phenomena, a significant mismatch between the presumed and actual scale of ecological data may arise. This may affect conclusions regarding basic drivers of diversity and may lead to errors in the identification of diversity hotspots. Here, we examine avian range maps of 834 bird species in conjunction with geographically extensive survey data sets on two continents to determine the spatial resolutions at which range-map data actually characterize species occurrences and patterns of species richness. At resolutions less than 2 degrees ( approximately 200 km), range maps overestimate the area of occupancy of individual species and mis-characterize spatial patterns of species richness, resulting in up to two-thirds of biodiversity hotspots being misidentified. The scale dependence of range-map accuracy poses clear limitations on broad-scale ecological analyses and conservation assessments. We suggest that range-map data contain less information than is generally assumed and provide guidance about the appropriate scale of their use.

Question:What is the relationship between species richness of vascular plants, bryophytes and macrolichens, and two important gradients in the alpine environment, altitude and local topography?Location:Northernmost Fennoscandia, 250–1525 m a.s.l. corresponding to the range between timberline and mountain top.Methods:The vegetation was sampled in six mountain areas. For each 25 vertical metres, the local topographic gradient from wind‐blown ridge to snowbed was sampled in quadrats of 0.8 m × 0.8 m. Patterns in species richness were explored using Poisson regression (Generalized Linear Models). Functional groups of species, i.e. evergreen and deciduous dwarf‐shrubs, forbs, graminoids, mosses, hepatics and lichens were investigated separately.Results:Functional groups showed markedly different patterns with respect to both altitude and topography. Species richness of all vascular plants showed a unimodal relationship with altitude. The same was true for graminoids, forbs and lichens analysed separately, but forb richness peaked at much higher altitudes than total richness. The richness of dwarf‐shrubs decreased monotonically with altitude, whereas richness of mosses and liverworts showed an increasing trend. Significant interactions between altitude and local topography were present for several groups. The unimodal pattern for total plant species richness was interpreted in terms of local productivity, physical disturbance, trophic interactions, and in terms of species pool effects.Conclusions:Patterns in local species richness result from the action of two opposing forces: declining species pool and decreasing intensity of competition with altitude.

We introduce a novel framework for conceptualising, quantifying and unifying discordant patterns of species richness along geographical gradients. While not itself explicitly mechanistic, this approach offers a path towards understanding mechanisms. In this study, we focused on the diverse patterns of species richness on mountainsides. We conjectured that elevational range midpoints of species may be drawn towards a single midpoint attractor - a unimodal gradient of environmental favourability. The midpoint attractor interacts with geometric constraints imposed by sea level and the mountaintop to produce taxon-specific patterns of species richness. We developed a Bayesian simulation model to estimate the location and strength of the midpoint attractor from species occurrence data sampled along mountainsides. We also constructed midpoint predictor models to test whether environmental variables could directly account for the observed patterns of species range midpoints. We challenged these models with 16 elevational data sets, comprising 4500 species of insects, vertebrates and plants. The midpoint predictor models generally failed to predict the pattern of species midpoints. In contrast, the midpoint attractor model closely reproduced empirical spatial patterns of species richness and range midpoints. Gradients of environmental favourability, subject to geometric constraints, may parsimoniously account for elevational and other patterns of species richness.

Patterns and drivers of plant biodiversity in Chinese university campuses

Aim To analyse the patterns in species richness and endemism of the native European riverine fish fauna, in the light of the Messinian salinity crisis and the Last Glacial Maximum (LGM). Location European continent. Methods After gathering native fish faunistic lists of 406 hydrographical networks, we defined large biogeographical regions with homogenous fish fauna, based on a hierarchical cluster analysis. Then we analysed and compared the patterns in species richness and endemism among these regions, as well as species‐area relationships. Results Among the 233 native species present in the data set, the Cyprinidae family was strongly dominant (> 50% of the total number of species). Seven biogeographical regions were defined: Western Peri-Mediterranea, Central Peri-Mediterranea, Eastern Peri-Mediterranea, Ponto-Caspian Europe, Northern Europe, Central Europe and Western Europe. The highest regional species richness was observed for Central Peri-Mediterranea and Ponto-Caspian Europe. The highest endemic richness was found in Central Peri-Mediterranea. Species‐area relationships were characterized by high slope values for Peri-Mediterranean Europe and low values for Central and Western Europe. Main conclusions The results were in agreement with the ‘Lago Mare’ hypothesis explaining the specificity of Peri-Mediterranean fish fauna, as well as with the history of recolonization of Central and Western Europe from Ponto-Caspian Europe following the LGM. The results also agreed with the mechanisms of speciation and extinction influencing fish diversity in hydrographical networks. We advise the use of the seven biogeographical regions for further studies, and suggest considering