Convolutional neural network for high-resolution wetland mapping with open data: Variable selection and the challenges of a generalizable model

Landscape scale wetland conservation requires accurate, up-to-date wetland maps. The most useful approaches to creating such maps are automated, spatially generalizable, temporally repeatable, and can be applied at large spatial scales. However, mapping wetlands with predictive models is challenging due to the highly variable characteristics of wetlands in both space and time. Currently, most approaches are limited by coarse resolution, commercial data, and geographic specificity. Here, we trained a deep learning model and evaluated its ability to automatically map wetlands at landscape scale in a variety of geographies. We trained a U-Net architecture to map wetlands at 1-meter spatial resolution with the following remotely sensed covariates: multispectral data from the National Agriculture Imagery Program and the Sentinel-2 satellite system, and two LiDAR-derived datasets, intensity and geomorphons. The full model mapped wetlands accurately (94 % accuracy, 96.5 % precision, 95.2 % AUC) at 1-meter resolution. Post hoc model evaluation showed that the model correctly predicted wetlands even in areas that had incorrect label/training data, which penalized the recall rate (90.2 %). Applying the model in a new geography resulted in poor performance (precision = ~80 %, recall = 48 %). However, limited retraining in this geography improved model performance substantially, demonstrating an effective means to create a spatially generalizable model. We demonstrate wetlands can be mapped at high-resolution (1 m) using free data and efficient deep-learning models that do not require manual feature engineering. Including LiDAR and geomorphons as input data improved model accuracy by 2 %, and where these data are unavailable a simpler model can efficiently map wetlands. Given the dynamic nature of wetlands and the important ecosystem services they provide, high-resolution mapping can be a game changer in terms of informing restoration and development decisions.

Matching expert range maps with species distribution model predictions

Species' range maps based on expert opinion are a critical resource for conservation planning. Expert maps are usually accompanied by species descriptions that specify sources of internal range heterogeneity, such as habitat associations, but these are rarely considered when using expert maps for analyses. We developed a quantitative metric (expert score) to evaluate the agreement between an expert map and a habitat probability surface obtained from a species distribution model. This method rewards both the avoidance of unsuitable sites and the inclusion of suitable sites in the expert map. We obtained expert maps of 330 butterfly species from each of 2 widely used North American sources (Glassberg [1999, 2001] and Scott [1986]) and computed species-wise expert scores for each. Overall, the Glassberg maps secured higher expert scores than Scott (0.61 and 0.41, respectively) due to the specific rules (e.g., Glassberg only included regions where the species was known to reproduce whereas Scott included all areas a species expanded to each year) they used to include or exclude areas from ranges. The predictive performance of expert maps was almost always hampered by the inclusion of unsuitable sites, rather than by exclusion of suitable sites (deviance outside of expert maps was extremely low). Map topology was the primary predictor of expert performance rather than any factor related to species characteristics such as mobility. Given the heterogeneity and discontinuity of suitable landscapes, expert maps drawn with more detail are more likely to agree with species distribution models and thus minimize both commission and omission errors.

Contrasting responses to climate change at Himalayan treelines revealed by population demographics of two dominant species

Alpine treelines are expected to shift upward due to recent climate change. However, interpretation of changes in montane systems has been problematic because effects of climate change are frequently confounded with those of land use changes. The eastern Himalaya, particularly Langtang National Park, Central Nepal, has been relatively undisturbed for centuries and thus presents an opportunity for studying climate change impacts on alpine treeline uncontaminated by potential confounding factors.We studied two dominant species, Abies spectabilis (AS) and Rhododendron campanulatum (RC), above and below the treeline on two mountains. We constructed 13 transects, each spanning up to 400 m in elevation, in which we recorded height and state (dead or alive) of all trees, as well as slope, aspect, canopy density, and measures of anthropogenic and animal disturbance.All size classes of RC plants had lower mortality above treeline than below it, and young RC plants (<2 m tall) were at higher density above treeline than below. AS shows little evidence of a position change from the historic treeline, with a sudden extreme drop in density above treeline compared to below. Recruitment, as measured by size-class distribution, was greater above treeline than below for both species but AS is confined to ~25 m above treeline whereas RC is luxuriantly growing up to 200 m above treeline. Synthesis. Evidence suggests that the elevational limits of RC have shifted upward both because (a) young plants above treeline benefited from facilitation of recruitment by surrounding vegetation, allowing upward expansion of recruitment, and (b) temperature amelioration to mature plants increased adult survival. We predict that the current pure stand of RC growing above treeline will be colonized by AS that will, in turn, outshade and eventually relegate RC to be a minor component of the community, as is the current situation below the treeline.

Projected distribution and climate refugia of endangered Kashmir musk deer Moschus cupreus in greater Himalaya, South Asia

Kashmir musk deer Moschus cupreus (KMD) are the least studied species of musk deer. We compiled genetically validated occurrence records of KMD to construct species distribution models using Maximum Entropy. We show that the distribution of KMD is limited between central Nepal on the east and north-east Afghanistan on the west and is primarily determined by precipitation of driest quarter, annual mean temperature, water vapor, and precipitation during the coldest quarter. Precipitation being the most influential determinant of distribution suggests the importance of pre-monsoon moisture for growth of the dominant vegetation, Himalayan birch Betula utilis and Himalayan fir Abies spectabilis, in KMD's preferred forests. All four Representative Concentration Pathway Scenarios result an expansion of suitable habitat in Uttarakhand, India, west Nepal and their associated areas in China in 2050s and 2070s but a dramatic loss of suitable habitat elsewhere (Kashmir region and Pakistan-Afghanistan border). About 1/4th of the current habitat will remain as climate refugia in future. Since the existing network of protected areas will only include a tiny fraction (4%) of the climatic refugia of KMD, the fate of the species will be determined by the interplay of more urgent short-term forces of poaching and habitat degradation and long-term forces of climate change.

Detecting interaction networks in the human microbiome with conditional Granger causality

Human microbiome research is rife with studies attempting to deduce microbial correlation networks from sequencing data. Standard correlation and/or network analyses may be misleading when taken as an indication of taxon interactions because "correlation is neither necessary nor sufficient to establish causation"; environmental filtering can lead to correlation between non-interacting taxa. Unfortunately, microbial ecologists have generally used correlation as a proxy for causality although there is a general consensus about what constitutes a causal relationship: causes both precede and predict effects. We apply one of the first causal models for detecting interactions in human microbiome samples. Specifically, we analyze a long duration, high resolution time series of the human microbiome to decipher the networks of correlation and causation of human-associated microbial genera. We show that correlation is not a good proxy for biological interaction; we observed a weak negative relationship between correlation and causality. Strong interspecific interactions are disproportionately positive, whereas almost all strong intraspecific interactions are negative. Interestingly, intraspecific interactions also appear to act at a short timescale causing vast majority of the effects within 1-3 days. We report how different taxa are involved in causal relationships with others, and show that strong interspecific interactions are rarely conserved across two body sites whereas strong intraspecific interactions are much more conserved, ranging from 33% between the gut and right-hand to 70% between the two hands. Therefore, in the absence of guiding assumptions about ecological interactions, Granger causality and related techniques may be particularly helpful for understanding the driving factors governing microbiome composition and structure.

Ecological correlates of Himalayan musk deer Moschus leucogaster

Himalayan musk deer (Moschus leucogaster; hereafter musk deer) are endangered as a result of poaching and habitat loss. The species is nocturnal, crepuscular, and elusive, making direct observation of habitat use and behavior difficult. However, musk deer establish and repeatedly use the same latrines for defecation. To quantify musk deer habitat correlates, we used observational spatial data based on presence-absence of musk deer latrines, as well as a range of fine spatial-scale ecological covariates. To determine presence-absence of musk deer, we exhaustively searched randomly selected forest trails using a 20-m belt transect in different study sites within the Neshyang Valley in the Annapurna Conservation Area. In a subsequent way, study sites were classified as habitat or nonhabitat for musk deer. A total of 252 plots, 20 × 20 m, were systematically established every 100 m along 51 transects (each ~0.5 km long) laid out at different elevations to record a range of ecological habitat variables. We used mixed-effect models and principal component analysis to characterize relationships between deer presence-absence data and habitat variables. We confirmed musk deer use latrines in forests located at higher elevations (3,200-4,200 m) throughout multiple seasons and years. Himalayan birch (Betula utilis) dominated forest, mixed Himalayan fir (Abies spectabilis), and birch forest were preferred over pure Himalayan fir and blue pine (Pinus wallichiana) forest. Greater crown cover and shrub diversity were associated with the presence of musk deer whereas tree height, diameter, and diversity were weakly correlated. Topographical attributes including aspect, elevation, distance to water source, and slope were also discriminated by musk deer. Over- and understory forest management can be used to protect forests likely to have musk deer as predicted by the models to ensure long-term conservation of this rare deer.

A Stoichioproteomic Analysis of Samples from the Human Microbiome Project

Ecological stoichiometry (ES) uses organism-specific elemental content to explain differences in species life histories, species interactions, community organization, environmental constraints and even ecosystem function. Although ES has been successfully applied to a range of different organisms, most emphasis on microbial ecological stoichiometry focuses on lake, ocean, and soil communities. With the recent advances in human microbiome research, however, large amounts of data are being generated that describe differences in community composition across body sites and individuals. We suggest that ES may provide a framework for beginning to understand the structure, organization, and function of human microbial communities, including why certain organisms exist at certain locations, and how they interact with both the other microbes in their environment and their human host. As a first step, we undertake a stoichioproteomic analysis of microbial communities from different body sites. Specifically, we compare and contrast the elemental composition of microbial protein samples using annotated sequencing data from 690 gut, vaginal, oral, nares, and skin samples currently available through the Human Microbiome Project. Our results suggest significant differences in both the median and variance of the carbon, oxygen, nitrogen, and sulfur contents of microbial protein samples from different locations. For example, whereas proteins from vaginal sites are high in carbon, proteins from skin and nasal sites are high in nitrogen and oxygen. Meanwhile, proteins from stool (the gut) are particularly high in sulfur content. We interpret these differences in terms of the local environments at different human body sites, including atmospheric exposure and food intake rates.

Timing Effects of Heat-Stress on Plant Ecophysiological Characteristics and Growth

Heat-waves with higher intensity and frequency and longer durations are expected in the future due to global warming, which could have dramatic impacts in agriculture, economy and ecology. This field study examined how plant responded to heat-stress (HS) treatment at different timing in naturally occurring vegetation. HS treatment (5 days at 40.5°C) were applied to 12 1 m2 plots in restored prairie vegetation dominated by a warm-season C4 grass, Andropogon gerardii, and a warm-season C3 forb, Solidago canadensis, at different growing stages. During and after each heat stress (HS) treatment, temperature were monitored for air, canopy, and soil; net CO2 assimilation (Anet), quantum yield of photosystem II (ΦPSII), stomatal conductance (gs), and internal CO2 level (Ci), specific leaf area (SLA), and chlorophyll content of the dominant species were measured. One week after the last HS treatment, all plots were harvested and the biomass of above-ground tissue and flower weight of the two dominant species were determined. HS decreased physiological performance and growth for both species, with S. canadensis being affected more than A. gerardii, indicated by negative HS effect on both physiological and growth responses for S. canadensis. There were significant timing effect of HS on the two species, with greater reductions in the net photosynthetic rate and productivity occurred when HS was applied at later-growing season. The reduction in aboveground productivity in S. canadensis but not A. gerardii could have important implications for plant community structure by increasing the competitive advantage of A. gerardii in this grassland. The present experiment showed that HS, though ephemeral, may promote long-term effects on plant community structure, vegetation dynamics, biodiversity, and ecosystem functioning of terrestrial biomes when more frequent and severe HS occur in the future.

Effects of N on plant response to heat-wave: a field study with prairie vegetation

More intense, more frequent, and longer heat-waves are expected in the future due to global warming, which could have dramatic ecological impacts. Increasing nitrogen (N) availability and its dynamics will likely impact plant responses to heat stress and carbon (C) sequestration in terrestrial ecosystems. This field study examined the effects of N availability on plant response to heat-stress (HS) treatment in naturally-occurring vegetation. HS (5 d at ambient or 40.5 degrees C) and N treatments (+/-N) were applied to 16 1 m(2) plots in restored prairie vegetation dominated by Andropogon gerardii (warm-season C4 grass) and Solidago canadensis (warm-season C3 forb). Before, during, and after HS, air, canopy, and soil temperature were monitored; net CO2 assimilation (P(n)), quantum yield of photosystem II (Phi(PSII)), stomatal conductance (g(s)), and leaf water potential (Psi(w)) of the dominant species and soil respiration (R(soil)) of each plot were measured daily during HS. One week after HS, plots were harvested, and C% and N% were determined for rhizosphere and bulk soil, and above-ground tissue (green/senescent leaf, stem, and flower). Photosynthetic N-use efficiency (PNUE) and N resorption rate (NRR) were calculated. HS decreased P(n), g(s), Psi(w), and PNUE for both species, and +N treatment generally increased these variables (+/-HS), but often slowed their post-HS recovery. Aboveground biomass tended to decrease with HS in both species (and for green leaf mass in S. canadensis), but decrease with +N for A. gerardii and increase with +N for S. canadensis. For A. gerardii, HS tended to decrease N% in green tissues with +N, whereas in S. canadensis, HS increased N% in green leaves. Added N decreased NRR for A. gerardii and HS increased NRR for S. canadensis. These results suggest that heat waves, though transient, could have significant effects on plants, communities, and ecosystem N cycling, and N can influence the effect of heat waves.