Propagating variational model uncertainty for bioacoustic call label smoothing.

Along with propagating the input toward making a prediction, Bayesian neural networks also propagate uncertainty. This has the potential to guide the training process by rejecting predictions of low confidence, and recent variational Bayesian methods can do so without Monte Carlo sampling of weights. Here, we apply sample-free methods for wildlife call detection on recordings made via passive acoustic monitoring equipment in the animals' natural habitats. We further propose uncertainty-aware label smoothing, where the smoothing probability is dependent on sample-free predictive uncertainty, in order to downweigh data samples that should contribute less to the loss value. We introduce a bioacoustic dataset recorded in Malaysian Borneo, containing overlapping calls from 30 species. On that dataset, our proposed method achieves an absolute percentage improvement of around 1.5 points on area under the receiver operating characteristic (AU-ROC), 13 points in F1, and 19.5 points in expected calibration error (ECE) compared to the point-estimate network baseline averaged across all target classes.

Optimizing the automated recognition of individual animals to support population monitoring.

Reliable estimates of population size and demographic rates are central to assessing the status of threatened species. However, obtaining individual-based demographic rates requires long-term data, which is often costly and difficult to collect. Photographic data offer an inexpensive, noninvasive method for individual-based monitoring of species with unique markings, and could therefore increase available demographic data for many species. However, selecting suitable images and identifying individuals from photographic catalogs is prohibitively time-consuming. Automated identification software can significantly speed up this process. Nevertheless, automated methods for selecting suitable images are lacking, as are studies comparing the performance of the most prominent identification software packages. In this study, we develop a framework that automatically selects images suitable for individual identification, and compare the performance of three commonly used identification software packages; Hotspotter, I3S-Pattern, and WildID. As a case study, we consider the African wild dog, Lycaon pictus, a species whose conservation is limited by a lack of cost-effective large-scale monitoring. To evaluate intraspecific variation in the performance of software packages, we compare identification accuracy between two populations (in Kenya and Zimbabwe) that have markedly different coat coloration patterns. The process of selecting suitable images was automated using convolutional neural networks that crop individuals from images, filter out unsuitable images, separate left and right flanks, and remove image backgrounds. Hotspotter had the highest image-matching accuracy for both populations. However, the accuracy was significantly lower for the Kenyan population (62%), compared to the Zimbabwean population (88%). Our automated image preprocessing has immediate application for expanding monitoring based on image matching. However, the difference in accuracy between populations highlights that population-specific detection rates are likely and may influence certainty in derived statistics. For species such as the African wild dog, where monitoring is both challenging and expensive, automated individual recognition could greatly expand and expedite conservation efforts.

Limits to the accurate and generalizable use of soundscapes to monitor biodiversity.

Although eco-acoustic monitoring has the potential to deliver biodiversity insight on vast scales, existing analytical approaches behave unpredictably across studies. We collated 8,023 audio recordings with paired manual avifaunal point counts to investigate whether soundscapes could be used to monitor biodiversity across diverse ecosystems. We found that neither univariate indices nor machine learning models were predictive of species richness across datasets but soundscape change was consistently indicative of community change. Our findings indicate that there are no common features of biodiverse soundscapes and that soundscape monitoring should be used cautiously and in conjunction with more reliable in-person ecological surveys.

Tropical forests post-logging are a persistent net carbon source to the atmosphere.

Logged and structurally degraded tropical forests are fast becoming one of the most prevalent land-use types throughout the tropics and are routinely assumed to be a net carbon sink because they experience rapid rates of tree regrowth. Yet this assumption is based on forest biomass inventories that record carbon stock recovery but fail to account for the simultaneous losses of carbon from soil and necromass. Here, we used forest plots and an eddy covariance tower to quantify and partition net ecosystem CO2 exchange in Malaysian Borneo, a region that is a hot spot for deforestation and forest degradation. Our data represent the complete carbon budget for tropical forests measured throughout a logging event and subsequent recovery and found that they constitute a substantial and persistent net carbon source. Consistent with existing literature, our study showed a significantly greater woody biomass gain across moderately and heavily logged forests compared with unlogged forests, but this was counteracted by much larger carbon losses from soil organic matter and deadwood in logged forests. We estimate an average carbon source of 1.75 ± 0.94 Mg C ha-1 yr-1 within moderately logged plots and 5.23 ± 1.23 Mg C ha-1 yr-1 in unsustainably logged and severely degraded plots, with emissions continuing at these rates for at least one-decade post-logging. Our data directly contradict the default assumption that recovering logged and degraded tropical forests are net carbon sinks, implying the amount of carbon being sequestered across the world's tropical forests may be considerably lower than currently estimated.

Logged tropical forests have amplified and diverse ecosystem energetics.

Old-growth tropical forests are widely recognized as being immensely important for their biodiversity and high biomass1. Conversely, logged tropical forests are usually characterized as degraded ecosystems2. However, whether logging results in a degradation in ecosystem functions is less clear: shifts in the strength and resilience of key ecosystem processes in large suites of species have rarely been assessed in an ecologically integrated and quantitative framework. Here we adopt an ecosystem energetics lens to gain new insight into the impacts of tropical forest disturbance on a key integrative aspect of ecological function: food pathways and community structure of birds and mammals. We focus on a gradient spanning old-growth and logged forests and oil palm plantations in Borneo. In logged forest there is a 2.5-fold increase in total resource consumption by both birds and mammals compared to that in old-growth forests, probably driven by greater resource accessibility and vegetation palatability. Most principal energetic pathways maintain high species diversity and redundancy, implying maintained resilience. Conversion of logged forest into oil palm plantation results in the collapse of most energetic pathways. Far from being degraded ecosystems, even heavily logged forests can be vibrant and diverse ecosystems with enhanced levels of ecological function.

Functional susceptibility of tropical forests to climate change.

Tropical forests are some of the most biodiverse ecosystems in the world, yet their functioning is threatened by anthropogenic disturbances and climate change. Global actions to conserve tropical forests could be enhanced by having local knowledge on the forests' functional diversity and functional redundancy as proxies for their capacity to respond to global environmental change. Here we create estimates of plant functional diversity and redundancy across the tropics by combining a dataset of 16 morphological, chemical and photosynthetic plant traits sampled from 2,461 individual trees from 74 sites distributed across four continents together with local climate data for the past half century. Our findings suggest a strong link between climate and functional diversity and redundancy with the three trait groups responding similarly across the tropics and climate gradient. We show that drier tropical forests are overall less functionally diverse than wetter forests and that functional redundancy declines with increasing soil water and vapour pressure deficits. Areas with high functional diversity and high functional redundancy tend to better maintain ecosystem functioning, such as aboveground biomass, after extreme weather events. Our predictions suggest that the lower functional diversity and lower functional redundancy of drier tropical forests, in comparison with wetter forests, may leave them more at risk of shifting towards alternative states in face of further declines in water availability across tropical regions.

A protocol for a longitudinal, observational cohort study of infection and exposure to zoonotic and vector-borne diseases across a land-use gradient in Sabah, Malaysian Borneo: a socio-ecological systems approach.

Introduction. Landscape changes disrupt environmental, social and biological systems, altering pathogen spillover and transmission risks. This study aims to quantify the impact of specific land management practices on spillover and transmission rates of zoonotic and vector-borne diseases within Malaysian Borneo. This protocol describes a cohort study with integrated ecological sampling to assess how deforestation and agricultural practices impact pathogen flow from wildlife and vector populations to human infection and detection by health facilities. This will focus on malaria, dengue and emerging arboviruses (Chikungunya and Zika), vector-borne diseases with varying contributions of simian reservoirs within this setting. Methods. A prospective longitudinal observational cohort study will be established in communities residing or working within the vicinity of the Stability of Altered Forest Ecosystems (SAFE) Project, a landscape gradient within Malaysian Borneo encompassing different plantation and forest types. The primary outcome of this study will be transmission intensity of selected zoonotic and vector-borne diseases, as quantified by changes in pathogen-specific antibody levels. Exposure will be measured using paired population-based serological surveys conducted at the beginning and end of the two-year cohort study. Secondary outcomes will include the distribution and infection rates of Aedes and Anopheles mosquito vectors, human risk behaviours and clinical cases reported to health facilities. Longitudinal data on human behaviour, contact with wildlife and GPS tracking of mobility patterns will be collected throughout the study period. This will be integrated with entomological surveillance to monitor densities and pathogen infection rates of Aedes and Anopheles mosquitoes relative to land cover. Within surrounding health clinics, continuous health facility surveillance will be used to monitor reported infections and febrile illnesses. Models will be developed to assess spillover and transmission rates relative to specific land management practices and evaluate abilities of surveillance systems to capture these risks.

Oil palm expansion increases the vectorial capacity of dengue vectors in Malaysian Borneo.

Changes in land-use and the associated shifts in environmental conditions can have large effects on the transmission and emergence of mosquito-borne disease. Mosquito-borne disease are particularly sensitive to these changes because mosquito growth, reproduction, survival and susceptibility to infection are all thermally sensitive traits, and land use change dramatically alters local microclimate. Predicting disease transmission under environmental change is increasingly critical for targeting mosquito-borne disease control and for identifying hotspots of disease emergence. Mechanistic models offer a powerful tool for improving these predictions. However, these approaches are limited by the quality and scale of temperature data and the thermal response curves that underlie predictions. Here, we used fine-scale temperature monitoring and a combination of empirical, laboratory and temperature-dependent estimates to estimate the vectorial capacity of Aedes albopictus mosquitoes across a tropical forest-oil palm plantation conversion gradient in Malaysian Borneo. We found that fine-scale differences in temperature between logged forest and oil palm plantation sites were not sufficient to produce differences in temperature-dependent demographic trait estimates using published thermal performance curves. However, when measured under field conditions a key parameter, adult abundance, differed significantly between land-use types, resulting in estimates of vectorial capacity that were 1.5 times higher in plantations than in forests. The prediction that oil palm plantations would support mosquito populations with higher vectorial capacity was robust to uncertainties in our adult survival estimates. These results provide a mechanistic basis for understanding the effects of forest conversion to agriculture on mosquito-borne disease risk, and a framework for interpreting emergent relationships between land-use and disease transmission. As the burden of Ae. albopictus-vectored diseases, such as dengue virus, increases globally and rising demand for palm oil products drives continued expansion of plantations, these findings have important implications for conservation, land management and public health policy at the global scale.

The macroecology of landscape ecology.

One of landscape ecology's main goals is to unveil how biodiversity is impacted by habitat transformation. However, the discipline suffers from significant context dependency in observed spatial and temporal trends, hindering progress towards understanding the mechanisms driving species declines and preventing the development of accurate estimates of future biodiversity change. Here, we discuss recent evidence that populations' and species' responses to habitat change at the landscape scale are modulated by factors and processes occurring at macroecological scales, such as historical disturbance rates, distance to geographic range edges, and climatic suitability. We suggest that placing landscape ecology studies in a macroecological lens will help to explain seemingly inconsistent results and will ultimately create better predictive models to help mitigate the biodiversity crisis.

How index selection, compression, and recording schedule impact the description of ecological soundscapes.

Acoustic indices derived from environmental soundscape recordings are being used to monitor ecosystem health and vocal animal biodiversity. Soundscape data can quickly become very expensive and difficult to manage, so data compression or temporal down-sampling are sometimes employed to reduce data storage and transmission costs. These parameters vary widely between experiments, with the consequences of this variation remaining mostly unknown.We analyse field recordings from North-Eastern Borneo across a gradient of historical land use. We quantify the impact of experimental parameters (MP3 compression, recording length and temporal subsetting) on soundscape descriptors (Analytical Indices and a convolutional neural net derived AudioSet Fingerprint). Both descriptor types were tested for their robustness to parameter alteration and their usability in a soundscape classification task.We find that compression and recording length both drive considerable variation in calculated index values. However, we find that the effects of this variation and temporal subsetting on the performance of classification models is minor: performance is much more strongly determined by acoustic index choice, with Audioset fingerprinting offering substantially greater (12%-16%) levels of classifier accuracy, precision and recall.We advise using the AudioSet Fingerprint in soundscape analysis, finding superior and consistent performance even on small pools of data. If data storage is a bottleneck to a study, we recommend Variable Bit Rate encoded compression (quality = 0) to reduce file size to 23% file size without affecting most Analytical Index values. The AudioSet Fingerprint can be compressed further to a Constant Bit Rate encoding of 64 kb/s (8% file size) without any detectable effect. These recommendations allow the efficient use of restricted data storage whilst permitting comparability of results between different studies.

Fine root dynamics across pantropical rainforest ecosystems.

Fine roots constitute a significant component of the net primary productivity (NPP) of forest ecosystems but are much less studied than aboveground NPP. Comparisons across sites and regions are also hampered by inconsistent methodologies, especially in tropical areas. Here, we present a novel dataset of fine root biomass, productivity, residence time, and allocation in tropical old-growth rainforest sites worldwide, measured using consistent methods, and examine how these variables are related to consistently determined soil and climatic characteristics. Our pantropical dataset spans intensive monitoring plots in lowland (wet, semi-deciduous, and deciduous) and montane tropical forests in South America, Africa, and Southeast Asia (n = 47). Large spatial variation in fine root dynamics was observed across montane and lowland forest types. In lowland forests, we found a strong positive linear relationship between fine root productivity and sand content, this relationship was even stronger when we considered the fractional allocation of total NPP to fine roots, demonstrating that understanding allocation adds explanatory power to understanding fine root productivity and total NPP. Fine root residence time was a function of multiple factors: soil sand content, soil pH, and maximum water deficit, with longest residence times in acidic, sandy, and water-stressed soils. In tropical montane forests, on the other hand, a different set of relationships prevailed, highlighting the very different nature of montane and lowland forest biomes. Root productivity was a strong positive linear function of mean annual temperature, root residence time was a strong positive function of soil nitrogen content in montane forests, and lastly decreasing soil P content increased allocation of productivity to fine roots. In contrast to the lowlands, environmental conditions were a better predictor for fine root productivity than for fractional allocation of total NPP to fine roots, suggesting that root productivity is a particularly strong driver of NPP allocation in tropical mountain regions.

Recovery of logged forest fragments in a human-modified tropical landscape during the 2015-16 El Niño.

The past 40 years in Southeast Asia have seen about 50% of lowland rainforests converted to oil palm and other plantations, and much of the remaining forest heavily logged. Little is known about how fragmentation influences recovery and whether climate change will hamper restoration. Here, we use repeat airborne LiDAR surveys spanning the hot and dry 2015-16 El Niño Southern Oscillation event to measure canopy height growth across 3,300 ha of regenerating tropical forests spanning a logging intensity gradient in Malaysian Borneo. We show that the drought led to increased leaf shedding and branch fall. Short forest, regenerating after heavy logging, continued to grow despite higher evaporative demand, except when it was located close to oil palm plantations. Edge effects from the plantations extended over 300 metres into the forests. Forest growth on hilltops and slopes was particularly impacted by the combination of fragmentation and drought, but even riparian forests located within 40 m of oil palm plantations lost canopy height during the drought. Our results suggest that small patches of logged forest within plantation landscapes will be slow to recover, particularly as ENSO events are becoming more frequent.

Major and persistent shifts in below-ground carbon dynamics and soil respiration following logging in tropical forests.

Soil respiration is the largest carbon efflux from the terrestrial ecosystem to the atmosphere, and selective logging influences soil respiration via changes in abiotic (temperature, moisture) and biotic (biomass, productivity, quantity and quality of necromass inputs) drivers. Logged forests are a predominant feature of the tropical forest landscape, their area exceeding that of intact forest. We quantified both total and component (root, mycorrhiza, litter, and soil organic matter, SOM) soil respiration in logged (n = 5) and old-growth (n = 6) forest plots in Malaysian Borneo, a region which is a global hotspot for emission from forest degradation. We constructed a detailed below-ground carbon budget including organic carbon inputs into the system via litterfall and root turnover. Total soil respiration was significantly higher in logged forests than in old-growth forests (14.3 ± 0.23 and 12.7 ± 0.60 Mg C ha-1  year-1 , respectively, p = 0.037). This was mainly due to the higher SOM respiration in logged forests (55 ± 3.1% of the total respiration in logged forests vs. 50 ± 3.0% in old-growth forests). In old-growth forests, annual SOM respiration was equal to the organic carbon inputs into the soil (difference between SOM respiration and inputs 0.18 Mg C ha-1  year-1 , with 90% confidence intervals of -0.41 and 0.74 Mg C ha-1  year-1 ), indicating that the system is in equilibrium, while in logged forests SOM respiration exceeded the inputs by 4.2 Mg C ha-1  year-1 (90% CI of 3.6 and 4.9 Mg C ha-1  year-1 ), indicating that the soil is losing carbon. These results contribute towards understanding the impact of logging on below-ground carbon dynamics, which is one of the key uncertainties in estimating emissions from forest degradation. This study demonstrates how significant perturbation of the below-ground carbon balance, and consequent net soil carbon emissions, can persist for decades after a logging event in tropical forests.

Forest conversion to oil palm compresses food chain length in tropical streams.

In Southeast Asia, biodiversity-rich forests are being extensively logged and converted to oil palm monocultures. Although the impacts of these changes on biodiversity are largely well documented, we know addition to samples we collected in 201 little about how these large-scale impacts affect freshwater trophic ecology. We used stable isotope analyses (SIA) to determine the impacts of land-use changes on the relative contribution of allochthonous and autochthonous basal resources in 19 stream food webs. We also applied compound-specific SIA and bulk-SIA to determine the trophic position of fish apex predators and meso-predators (invertivores and omnivores). There was no difference in the contribution of autochthonous resources in either consumer group (70-82%) among streams with different land-use type. There was no change in trophic position for meso-predators, but trophic position decreased significantly for apex predators in oil palm plantation streams compared to forest streams. This change in maximum food chain length was due to turnover in identity of the apex predator among land-use types. Disruption of aquatic trophic ecology, through reduction in food chain length and shift in basal resources, may cause significant changes in biodiversity as well as ecosystem functions and services. Understanding this change can help develop more focused priorities for mediating the negative impacts of human activities on freshwater ecosystems.

Characterizing soundscapes across diverse ecosystems using a universal acoustic feature set.

Natural habitats are being impacted by human pressures at an alarming rate. Monitoring these ecosystem-level changes often requires labor-intensive surveys that are unable to detect rapid or unanticipated environmental changes. Here we have developed a generalizable, data-driven solution to this challenge using eco-acoustic data. We exploited a convolutional neural network to embed soundscapes from a variety of ecosystems into a common acoustic space. In both supervised and unsupervised modes, this allowed us to accurately quantify variation in habitat quality across space and in biodiversity through time. On the scale of seconds, we learned a typical soundscape model that allowed automatic identification of anomalous sounds in playback experiments, providing a potential route for real-time automated detection of irregular environmental behavior including illegal logging and hunting. Our highly generalizable approach, and the common set of features, will enable scientists to unlock previously hidden insights from acoustic data and offers promise as a backbone technology for global collaborative autonomous ecosystem monitoring efforts.

Imaging spectroscopy reveals the effects of topography and logging on the leaf chemistry of tropical forest canopy trees.

Logging, pervasive across the lowland tropics, affects millions of hectares of forest, yet its influence on nutrient cycling remains poorly understood. One hypothesis is that logging influences phosphorus (P) cycling, because this scarce nutrient is removed in extracted timber and eroded soil, leading to shifts in ecosystem functioning and community composition. However, testing this is challenging because P varies within landscapes as a function of geology, topography and climate. Superimposed upon these trends are compositional changes in logged forests, with species with more acquisitive traits, characterized by higher foliar P concentrations, more dominant. It is difficult to resolve these patterns using traditional field approaches alone. Here, we use airborne light detection and ranging-guided hyperspectral imagery to map foliar nutrient (i.e. P, nitrogen [N]) concentrations, calibrated using field measured traits, over 400 km2 of northeastern Borneo, including a landscape-level disturbance gradient spanning old-growth to repeatedly logged forests. The maps reveal that canopy foliar P and N concentrations decrease with elevation. These relationships were not identified using traditional field measurements of leaf and soil nutrients. After controlling for topography, canopy foliar nutrient concentrations were lower in logged forest than in old-growth areas, reflecting decreased nutrient availability. However, foliar nutrient concentrations and specific leaf area were greatest in relatively short patches in logged areas, reflecting a shift in composition to pioneer species with acquisitive traits. N:P ratio increased in logged forest, suggesting reduced soil P availability through disturbance. Through the first landscape scale assessment of how functional leaf traits change in response to logging, we find that differences from old-growth forest become more pronounced as logged forests increase in stature over time, suggesting exacerbated phosphorus limitation as forests recover.

Extinction filters mediate the global effects of habitat fragmentation on animals.

Habitat loss is the primary driver of biodiversity decline worldwide, but the effects of fragmentation (the spatial arrangement of remaining habitat) are debated. We tested the hypothesis that forest fragmentation sensitivity-affected by avoidance of habitat edges-should be driven by historical exposure to, and therefore species' evolutionary responses to disturbance. Using a database containing 73 datasets collected worldwide (encompassing 4489 animal species), we found that the proportion of fragmentation-sensitive species was nearly three times as high in regions with low rates of historical disturbance compared with regions with high rates of disturbance (i.e., fires, glaciation, hurricanes, and deforestation). These disturbances coincide with a latitudinal gradient in which sensitivity increases sixfold at low versus high latitudes. We conclude that conservation efforts to limit edges created by fragmentation will be most important in the world's tropical forests.