Data science competition for cross-site individual tree species identification from airborne remote sensing data.

Data on individual tree crowns from remote sensing have the potential to advance forest ecology by providing information about forest composition and structure with a continuous spatial coverage over large spatial extents. Classifying individual trees to their taxonomic species over large regions from remote sensing data is challenging. Methods to classify individual species are often accurate for common species, but perform poorly for less common species and when applied to new sites. We ran a data science competition to help identify effective methods for the task of classification of individual crowns to species identity. The competition included data from three sites to assess each methods' ability to generalize patterns across two sites simultaneously and apply methods to an untrained site. Three different metrics were used to assess and compare model performance. Six teams participated, representing four countries and nine individuals. The highest performing method from a previous competition in 2017 was applied and used as a baseline to understand advancements and changes in successful methods. The best species classification method was based on a two-stage fully connected neural network that significantly outperformed the baseline random forest and gradient boosting ensemble methods. All methods generalized well by showing relatively strong performance on the trained sites (accuracy = 0.46-0.55, macro F1 = 0.09-0.32, cross entropy loss = 2.4-9.2), but generally failed to transfer effectively to the untrained site (accuracy = 0.07-0.32, macro F1 = 0.02-0.18, cross entropy loss = 2.8-16.3). Classification performance was influenced by the number of samples with species labels available for training, with most methods predicting common species at the training sites well (maximum F1 score of 0.86) relative to the uncommon species where none were predicted. Classification errors were most common between species in the same genus and different species that occur in the same habitat. Most methods performed better than the baseline in detecting if a species was not in the training data by predicting an untrained mixed-species class, especially in the untrained site. This work has highlighted that data science competitions can encourage advancement of methods, particularly by bringing in new people from outside the focal discipline, and by providing an open dataset and evaluation criteria from which participants can learn.

Animal seed dispersal recovery during passive restoration in a forested landscape.

Seed dispersal by animals is key for restoration of tropical forests because it maintains plant diversity and accelerates community turnover. Therefore, changes in seed dispersal during forest restoration can indicate the recovery of species interactions, and yet these changes are rarely considered in forest restoration planning. In this study, we examined shifts in the importance of different seed dispersal modes during passive restoration in a tropical chronosequence spanning more than 100 years, by modelling the proportion of trees dispersed by bats, small birds, large birds, flightless mammals and abiotic means as a function of forest age. Contrary to expectations, tree species dispersed by flightless mammals dominated after 20 years of regeneration, and tree richness and abundance dispersed by each mode mostly recovered to old growth levels between 40 and 70 years post-abandonment. Seed dispersal by small birds declined over time during regeneration, while bat dispersal played a minor role throughout all stages of succession. Results suggest that proximity to old growth forests, coupled with low hunting, explained the prevalence of seed dispersal by animals, especially by flightless mammals at this site. We suggest that aspects of seed dispersal should be monitored when restoring forest ecosystems to evaluate the reestablishment of species interactions. This article is part of the theme issue 'Understanding forest landscape restoration: reinforcing scientific foundations for the UN Decade on Ecosystem Restoration'.

Bats and their vital ecosystem services: a global review.

Bats play crucial ecosystem services as seed dispersers, pollinators, controllers of insects, and nutrient recyclers. However, there has not been a thorough global review evaluating these roles in bats across all biogeographical regions of the world. We reviewed the literature published during the last two decades and identified 283 relevant studies: 78 dealt with the control of potential insect pests by bats, 80 related to the suppression of other arthropods, 60 on the dispersal of native or endemic seeds, 11 dealt with the dispersal of seeds of introduced plants, 29 on the pollination of native or endemic plants, 1 study on pollination of introduced plants, and 24 on the use of guano as fertilizer. Our literature search showed that queries combining the terms "seed dispersal," "insectivorous bats," "nectarivorous bats," "use of guano," and "ecosystem services" returned 577 studies, but half were experimental in nature. We found that the evaluation of ecosystem services by bats has been mostly conducted in the Neotropical and Palearctic regions. To detect differences across relevant studies, and to explain trends in the study of ecosystem services provided by bats, we performed generalized linear mixed models (GLMM) fitted with a Poisson distribution to analyze potential differences among sampling methods. We identified 409 bat species that provide ecosystem services, 752 insect species consumed by bats and 549 plant species either dispersed or pollinated by bats. Our review summarizes the importance of conserving bat populations and the ecological services they provide, which is especially important during the current pandemic.

Local canopy disturbance as an explanation for long-term increases in liana abundance.

Canopy disturbance explains liana abundance and distribution within tropical forests and thus may also explain the widespread pattern of increasing liana abundance; however, this hypothesis remains untested. We used a 10-year study (2007-2017) of 117,100 rooted lianas in an old-growth Panamanian forest to test whether local canopy disturbance explains increasing liana abundance. We found that liana density increased 29.2% and basal area 12.5%. The vast majority of these increases were associated with clonal stem proliferation following canopy disturbance, particularly in liana-dense, low-canopy gaps, which had far greater liana increases than did undisturbed forest. Lianas may be ecological niche constructors, arresting tree regeneration in gaps and thus creating a high-light environment that favours sustained liana proliferation. Our findings demonstrate that liana abundance is increasing rapidly and their ability to proliferate via copious clonal stem production in canopy gaps explains much of their increase in this and possibly other tropical forests.

Uniting niche differentiation and dispersal limitation predicts tropical forest succession.

Tropical secondary forests are increasingly important for carbon sequestration and biodiversity conservation worldwide; yet, we still cannot accurately predict community turnover during secondary succession. We propose that integrating niche differentiation and dispersal limitation will generate an improved theoretical explanation of tropical forest succession. The interaction between seed sources and dispersers regulates seed movement throughout succession, and recent technological advances in animal tracking and molecular analyses enable us to accurately monitor seed movement as never before. We propose a framework to bridge the gap between niche differentiation and dispersal limitation. The Source-Disperser Limitation Framework (SDLF) provides a way to better predict secondary tropical forest succession across gradients of landscape disturbance by integrating seed sources and frugivore behavior.

Lianas do not reduce tree biomass accumulation in young successional tropical dry forests.

Young successional tropical forests are crucial in the global carbon cycle because they can quickly sequester large quantities of atmospheric carbon. However, lianas (woody vines) can significantly decrease biomass accumulation in young regenerating forests. Lianas are abundant in tropical dry forests, and thus we hypothesized that lianas reduce biomass accretion in dry forests. Lianas may be particularly detrimental to the growth of young trees, which are vulnerable to competition from lianas. Alternatively, lianas may have a stronger negative effect on the largest trees because lianas seek the high-light environment at the top of the forest canopy. We tested these hypotheses using a liana-removal experiment in 13 dry forest stands that ranged from 1 to 70 years in southwestern Panama. We measured biomass accumulation annually for more than 10,000 stems from 2013 to 2017. Contrary to our expectations, liana removal had no effect on tree biomass accumulation across our successional forests and throughout our study period. Liana removal did not benefit smaller trees or larger trees. Lianas did not increase biomass accumulation on recruits, and did not increase biomass loss due to mortality. Surprisingly, removing lianas had a negative effect on three out of 41 tree species. Lianas had no effect on biomass accumulation and loss, possibly because: (1) trees allocated resources to roots instead of stems, (2) trees and lianas partitioned water, (3) higher irradiance after liana removal reduced soil moisture, or (4) low water availability might have been such a strong stressor that it reduced plant-plant competition.

The response of lianas to 20 yr of nutrient addition in a Panamanian forest.

Over the past two decades, liana density and basal area have been increasing in many tropical forests, which has profound consequences for forest diversity and functioning. One hypothesis to explain increasing lianas is elevated nutrient deposition in tropical forests resulting from fossil fuels, agricultural fertilizer, and biomass burning. We tested this hypothesis by surveying all lianas ≥1 cm in diameter (n = 3,967) in 32 plots in a fully factorial nitrogen (N), phosphorus (P), and potassium (K) addition experiment in a mature tropical forest in central Panama. We conducted the nutrient-addition experiment from 1998 until present and we first censused lianas in 2013 and then again in 2018. After 20 yr of nutrient addition (1998-2018), liana density, basal area, and rarefied species richness did not differ significantly among any of the nutrient-addition and control treatments. Moreover, nutrient addition in the most recent 5 yr of the experiment did not affect liana relative growth, recruitment, or mortality rates. From 2013 until 2018, liana density, basal area, and species richness increased annually by 1.6%, 1.4%, and 2.4%, respectively. Nutrient addition did not influence these increases. Our findings indicate that nutrient deposition does not explain increasing lianas in this tropical forest. Instead, increases in tree mortality and disturbance, atmospheric carbon dioxide, drought frequency and severity, and hunting pressure may be more likely explanations for the increase in lianas in tropical forests.

Lianas reduce biomass accumulation in early successional tropical forests.

Early successional tropical forests could mitigate climate change via rapid accumulation of atmospheric carbon. However, liana (woody vine) abundance and biomass has been increasing in many tropical forests over the past decades, which may slow the speed at which secondary forests accumulate biomass. Lianas decrease biomass accumulation in tropical forests, and may have a particularly strong effect on young forests by stalling tree growth. As forests mature, trees may outgrow or shed lianas, thus escaping some of the negative effects of lianas. Alternatively, lianas may have the strongest effect in older successional forests if the effect of lianas is commensurate with their density, which increases dramatically in the first decades of forest succession. We tested these two hypotheses using a landscape liana-removal experiment in 30 forest stands that ranged from 10 to 35 yr old in Central Panama. We measured tree growth and biomass accumulation in the stands every year from 2014 to 2017. We found that the effect of liana removal on large trees (≥20-cm diameter) decreased with forest age, supporting the hypothesis that lianas have the strongest negative effects on trees, and thus biomass uptake and carbon storage, in very young successional forests. Large trees accumulated more biomass in the absence of lianas in younger forests than in older forests (compared to controls) even after accounting for the effect of canopy completeness and crown illumination, implying that the detrimental effects of lianas go well beyond resource availability and crown health. There was no significant effect of lianas on small trees (1-20-cm diameter), likely because lianas seek light and thus do not deploy their leaves on small trees that are trapped in the forest understory. Our results show that high liana density early in forest succession reduces forest biomass accumulation by negatively impacting large trees, thus decreasing the capacity of young secondary forests to mitigate climate change. Although the negative effects of lianas on forest biomass diminish as forests age, they do not disappear, and thus lianas are an important component of tropical forest carbon budgets throughout succession.

Species undersampling in tropical bat surveys: effects on emerging biodiversity patterns.

Undersampling is commonplace in biodiversity surveys of species-rich tropical assemblages in which rare taxa abound, with possible repercussions for our ability to implement surveys and monitoring programmes in a cost-effective way. We investigated the consequences of information loss due to species undersampling (missing subsets of species from the full species pool) in tropical bat surveys for the emerging patterns of species richness (SR) and compositional variation across sites. For 27 bat assemblage data sets from across the tropics, we used correlations between original data sets and subsets with different numbers of species deleted either at random, or according to their rarity in the assemblage, to assess to what extent patterns in SR and composition in data subsets are congruent with those in the initial data set. We then examined to what degree high sample representativeness (r ≥ 0·8) was influenced by biogeographic region, sampling method, sampling effort or structural assemblage characteristics. For SR, correlations between random subsets and original data sets were strong (r ≥ 0·8) with moderate (ca. 20%) species loss. Bias associated with information loss was greater for species composition; on average ca. 90% of species in random subsets had to be retained to adequately capture among-site variation. For nonrandom subsets, removing only the rarest species (on average c. 10% of the full data set) yielded strong correlations (r > 0·95) for both SR and composition. Eliminating greater proportions of rare species resulted in weaker correlations and large variation in the magnitude of observed correlations among data sets. Species subsets that comprised ca. 85% of the original set can be considered reliable surrogates, capable of adequately revealing patterns of SR and temporal or spatial turnover in many tropical bat assemblages. Our analyses thus demonstrate the potential as well as limitations for reducing survey effort and streamlining sampling protocols, and consequently for increasing the cost-effectiveness in tropical bat surveys or monitoring programmes. The dependence of the performance of species subsets on structural assemblage characteristics (total assemblage abundance, proportion of rare species), however, underscores the importance of adaptive monitoring schemes and of establishing surrogate performance on a site by site basis based on pilot surveys.

Climate, habitat, and species interactions at different scales determine the structure of a Neotropical bat community.

Climate, habitat, and species interactions are factors that control community properties (e.g., species richness, abundance) across various spatial scales. Usually, researchers study how a few properties are affected by one factor in isolation and at one scale. Hence, there are few multi-scale studies testing how multiple controlling factors simultaneously affect community properties at different scales. We ask whether climate, habitat structure, or insect resources at each of three spatial scales explains most of the variation in six community properties and which theory best explains the distribution of selected community properties across a rainfall gradient. We studied a Neotropical insectivorous bat ensemble in the Isthmus of Panama with acoustic monitoring techniques. Using climatological data, habitat surveys, and insect captures in a hierarchical sampling design we determined how much variation of the community properties was explained by the three factors employing two approaches for variance partitioning. Our results revealed that most of the variation in species richness, total abundance, and feeding activity occurred at the smallest spatial scale and was explained by habitat structure. In contrast, climate at large scales explained most of the variation in individual species' abundances. Although each species had an idiosyncratic response to the gradient, species richness peaked at intermediate levels of precipitation, whereas total abundance was very similar across sites, suggesting density compensation. All community properties responded in a different manner to the factor and scale under consideration.

Flying high--assessing the use of the aerosphere by bats.

Bats feature prominently among organisms that occupy the aerosphere as they extensively use this environment for foraging, but also for dispersal, migration, and behavioral interactions. Differential use of the aerosphere is an important factor structuring bat assemblages, with species exhibiting distinct morphological, physiological, and sensory adaptations to different habitat types. This necessitates comprehensive sampling methodologies such as combined ground-level and canopy-level mist netting as well as acoustic monitoring to assess the presence, diversity, and activity of different functional groups of species adequately. Recent technological advances in acoustic detection and in methods of analysis, coupled with the expansion of libraries of echolocation calls for species identification, now allow for the reliable quantification of species numbers and activity of the scarcely known group of aerial insectivorous bats, particularly in species-rich tropical assemblages. We provide a brief, exemplary overview of recent studies on bats conducted in Panamá to demonstrate the necessity of comprehensive sampling methods and application of new technologies in order to adequately depict assemblage composition and responses of bats to structural changes in habitats induced by fragmentation. In addition to acoustic methods, miniaturization of radio transmitters has provided new insights into the patterns of spatial use of the aerosphere by bats and has identified species-specific differences in mobility as one of the important traits that determines bats' reactions to anthropogenic alterations of the landscape. Following the goals of the symposium on aeroecology, we propose new avenues of research for probing the aerosphere. We discuss how integration of a diverse array of remote sensing tools with data on species distribution and species traits, such as mobility and edge-sensitivity, might provide novel opportunities for the development, and application of conservation-oriented monitoring systems.