Individual canopy tree species maps for the National Ecological Observatory Network.

The ecology of forest ecosystems depends on the composition of trees. Capturing fine-grained information on individual trees at broad scales provides a unique perspective on forest ecosystems, forest restoration, and responses to disturbance. Individual tree data at wide extents promises to increase the scale of forest analysis, biogeographic research, and ecosystem monitoring without losing details on individual species composition and abundance. Computer vision using deep neural networks can convert raw sensor data into predictions of individual canopy tree species through labeled data collected by field researchers. Using over 40,000 individual tree stems as training data, we create landscape-level species predictions for over 100 million individual trees across 24 sites in the National Ecological Observatory Network (NEON). Using hierarchical multi-temporal models fine-tuned for each geographic area, we produce open-source data available as 1 km2 shapefiles with individual tree species prediction, as well as crown location, crown area, and height of 81 canopy tree species. Site-specific models had an average performance of 79% accuracy covering an average of 6 species per site, ranging from 3 to 15 species per site. All predictions are openly archived and have been uploaded to Google Earth Engine to benefit the ecology community and overlay with other remote sensing assets. We outline the potential utility and limitations of these data in ecology and computer vision research, as well as strategies for improving predictions using targeted data sampling.

A benchmark dataset for canopy crown detection and delineation in co-registered airborne RGB, LiDAR and hyperspectral imagery from the National Ecological Observation Network.

Broad scale remote sensing promises to build forest inventories at unprecedented scales. A crucial step in this process is to associate sensor data into individual crowns. While dozens of crown detection algorithms have been proposed, their performance is typically not compared based on standard data or evaluation metrics. There is a need for a benchmark dataset to minimize differences in reported results as well as support evaluation of algorithms across a broad range of forest types. Combining RGB, LiDAR and hyperspectral sensor data from the USA National Ecological Observatory Network's Airborne Observation Platform with multiple types of evaluation data, we created a benchmark dataset to assess crown detection and delineation methods for canopy trees covering dominant forest types in the United States. This benchmark dataset includes an R package to standardize evaluation metrics and simplify comparisons between methods. The benchmark dataset contains over 6,000 image-annotated crowns, 400 field-annotated crowns, and 3,000 canopy stem points from a wide range of forest types. In addition, we include over 10,000 training crowns for optional use. We discuss the different evaluation data sources and assess the accuracy of the image-annotated crowns by comparing annotations among multiple annotators as well as overlapping field-annotated crowns. We provide an example submission and score for an open-source algorithm that can serve as a baseline for future methods.

Species-level tree crown maps improve predictions of tree recruit abundance in a tropical landscape.

Predicting forest recovery at landscape scales will aid forest restoration efforts. The first step in successful forest recovery is tree recruitment. Forecasts of tree recruit abundance, derived from the landscape-scale distribution of seed sources (i.e., adult trees), could assist efforts to identify sites with high potential for natural regeneration. However, previous work revealed wide variation in the effect of seed sources on seedling abundance, from positive to no effect. We quantified the relationship between adult tree seed sources and tree recruits and predicted where natural recruitment would occur in a fragmented, tropical, agricultural landscape. We integrated species-specific tree crown maps generated from hyperspectral imagery and property ownership data with field data on the spatial distribution of tree recruits from five species. We then developed hierarchical Bayesian models to predict landscape-scale recruit abundance. Our models revealed that species-specific maps of tree crowns improved recruit abundance predictions. Conspecific crown area had a much stronger impact on recruitment abundance (8.00% increase in recruit abundance when conspecific tree density increases from zero to one tree; 95% credible interval (CI): 0.80% to 11.57%) than heterospecific crown area (0.03% increase with the addition of a single heterospecific tree, 95% CI: -0.60% to 0.68%). Individual property ownership was also an important predictor of recruit abundance: The best performing model had varying effects of conspecific and heterospecific crown area on recruit abundance, depending on individual property ownership. We demonstrate how novel remote sensing approaches and cadastral data can be used to generate high-resolution and landscape-level maps of tree recruit abundance. Spatial models parameterized with field, cadastral, and remote sensing data are poised to assist decision support for forest landscape restoration.

Monitoring tropical forest succession at landscape scales despite uncertainty in Landsat time series.

Forecasting rates of forest succession at landscape scales will aid global efforts to restore tree cover to millions of hectares of degraded land. While optical satellite remote sensing can detect regional land cover change, quantifying forest structural change is challenging. We developed a state-space modeling framework that applies Landsat satellite data to estimate variability in rates of natural regeneration between sites in a tropical landscape. Our models work by disentangling measurement error in Landsat-derived spectral reflectance from process error related to successional variability. We applied our modeling framework to rank rates of forest succession between 10 naturally regenerating sites in Southwestern Panama from about 2001 to 2015 and tested how different models for measurement error impacted forecast accuracy, ecological inference, and rankings of successional rates between sites. We achieved the greatest increase in forecasting accuracy by adding intra-annual phenological variation to a model based on Landsat-derived normalized difference vegetation index (NDVI). The best-performing model accounted for inter- and intra-annual noise in spectral reflectance and translated NDVI to canopy height via Landsat-lidar fusion. Modeling forest succession as a function of canopy height rather than NDVI also resulted in more realistic estimates of forest state during early succession, including greater confidence in rank order of successional rates between sites. These results establish the viability of state-space models to quantify ecological dynamics from time series of space-borne imagery. State-space models also provide a statistical approach well-suited to fusing high-resolution data, such as airborne lidar, with lower-resolution data that provides better temporal and spatial coverage, such as the Landsat satellite record. Monitoring forest succession using satellite imagery could play a key role in achieving global restoration targets, including identifying sites that will regain tree cover with minimal intervention.

Dispersal limitation drives successional pathways in Central Siberian forests under current and intensified fire regimes.

Fire is a primary driver of boreal forest dynamics. Intensifying fire regimes due to climate change may cause a shift in boreal forest composition toward reduced dominance of conifers and greater abundance of deciduous hardwoods, with potential biogeochemical and biophysical feedbacks to regional and global climate. This shift has already been observed in some North American boreal forests and has been attributed to changes in site conditions. However, it is unknown if the mechanisms controlling fire-induced changes in deciduous hardwood cover are similar among different boreal forests, which differ in the ecological traits of the dominant tree species. To better understand the consequences of intensifying fire regimes in boreal forests, we studied postfire regeneration in five burns in the Central Siberian dark taiga, a vast but poorly studied boreal region. We combined field measurements, dendrochronological analysis, and seed-source maps derived from high-resolution satellite images to quantify the importance of site conditions (e.g., organic layer depth) vs. seed availability in shaping postfire regeneration. We show that dispersal limitation of evergreen conifers was the main factor determining postfire regeneration composition and density. Site conditions had significant but weaker effects. We used information on postfire regeneration to develop a classification scheme for successional pathways, representing the dominance of deciduous hardwoods vs. evergreen conifers at different successional stages. We estimated the spatial distribution of different successional pathways under alternative fire regime scenarios. Under intensified fire regimes, dispersal limitation of evergreen conifers is predicted to become more severe, primarily due to reduced abundance of surviving seed sources within burned areas. Increased dispersal limitation of evergreen conifers, in turn, is predicted to increase the prevalence of successional pathways dominated by deciduous hardwoods. The likely fire-induced shift toward greater deciduous hardwood cover may affect climate-vegetation feedbacks via surface albedo, Bowen ratio, and carbon cycling.

A hyperspectral image can predict tropical tree growth rates in single-species stands.

Remote sensing is increasingly needed to meet the critical demand for estimates of forest structure and composition at landscape to continental scales. Hyperspectral images can detect tree canopy properties, including species identity, leaf chemistry and disease. Tree growth rates are related to these measurable canopy properties but whether growth can be directly predicted from hyperspectral data remains unknown. We used a single hyperspectral image and light detection and ranging-derived elevation to predict growth rates for 20 tropical tree species planted in experimental plots. We asked whether a consistent relationship between spectral data and growth rates exists across all species and which spectral regions, associated with different canopy chemical and structural properties, are important for predicting growth rates. We found that a linear combination of narrowband indices and elevation is correlated with standardized growth rates across all 20 tree species (R2  = 53.70%). Although wavelengths from the entire visible-to-shortwave infrared spectrum were involved in our analysis, results point to relatively greater importance of visible and near-infrared regions for relating canopy reflectance to tree growth data. Overall, we demonstrate the potential for hyperspectral data to quantify tree demography over a much larger area than possible with field-based methods in forest inventory plots.

Landscape-scale consequences of differential tree mortality from catastrophic wind disturbance in the Amazon.

Wind disturbance can create large forest blowdowns, which greatly reduces live biomass and adds uncertainty to the strength of the Amazon carbon sink. Observational studies from within the central Amazon have quantified blowdown size and estimated total mortality but have not determined which trees are most likely to die from a catastrophic wind disturbance. Also, the impact of spatial dependence upon tree mortality from wind disturbance has seldom been quantified, which is important because wind disturbance often kills clusters of trees due to large treefalls killing surrounding neighbors. We examine (1) the causes of differential mortality between adult trees from a 300-ha blowdown event in the Peruvian region of the northwestern Amazon, (2) how accounting for spatial dependence affects mortality predictions, and (3) how incorporating both differential mortality and spatial dependence affect the landscape level estimation of necromass produced from the blowdown. Standard regression and spatial regression models were used to estimate how stem diameter, wood density, elevation, and a satellite-derived disturbance metric influenced the probability of tree death from the blowdown event. The model parameters regarding tree characteristics, topography, and spatial autocorrelation of the field data were then used to determine the consequences of non-random mortality for landscape production of necromass through a simulation model. Tree mortality was highly non-random within the blowdown, where tree mortality rates were highest for trees that were large, had low wood density, and were located at high elevation. Of the differential mortality models, the non-spatial models overpredicted necromass, whereas the spatial model slightly underpredicted necromass. When parameterized from the same field data, the spatial regression model with differential mortality estimated only 7.5% more dead trees across the entire blowdown than the random mortality model, yet it estimated 51% greater necromass. We suggest that predictions of forest carbon loss from wind disturbance are sensitive to not only the underlying spatial dependence of observations, but also the biological differences between individuals that promote differential levels of mortality.

Indirect interactions among tropical tree species through shared rodent seed predators: a novel mechanism of tree species coexistence.

The coexistence of numerous tree species in tropical forests is commonly explained by negative dependence of recruitment on the conspecific seed and tree density due to specialist natural enemies that attack seeds and seedlings ('Janzen-Connell' effects). Less known is whether guilds of shared seed predators can induce a negative dependence of recruitment on the density of different species of the same plant functional group. We studied 54 plots in tropical forest on Barro Colorado Island, Panama, with contrasting mature tree densities of three coexisting large seeded tree species with shared seed predators. Levels of seed predation were far better explained by incorporating seed densities of all three focal species than by conspecific seed density alone. Both positive and negative density dependencies were observed for different species combinations. Thus, indirect interactions via shared seed predators can either promote or reduce the coexistence of different plant functional groups in tropical forest.

Estimation of the distribution of Tabebuia guayacan (Bignoniaceae) using high-resolution remote sensing imagery.

Species identification and characterization in tropical environments is an emerging field in tropical remote sensing. Significant efforts are currently aimed at the detection of tree species, of levels of forest successional stages, and the extent of liana occurrence at the top of canopies. In this paper we describe our use of high resolution imagery from the Quickbird Satellite to estimate the flowering population of Tabebuia guayacan trees at Barro Colorado Island (BCI), in Panama. The imagery was acquired on 29 April 2002 and 21 March 2004. Spectral Angle Mapping via a One-Class Support Vector machine was used to detect the presence of 422 and 557 flowering tress in the April 2002 and March 2004 imagery. Of these, 273 flowering trees are common to both dates. This study presents a new perspective on the effectiveness of high resolution remote sensing for monitoring a phenological response and its use as a tool for potential conservation and management of natural resources in tropical environments.

Changes in floodplain hydrology following serial damming of the Tocantins River in the eastern Amazon.

Riparian forests are ecotones that link aquatic and terrestrial habitats, providing ecosystem services including sediment control and nutrient regulation. Riparian forest function is intimately linked to river hydrology and floodplain dynamics, which can be severely altered by dams. The Tocantins River in the eastern Amazon has six mega-dams along its course. To understand the large-scale and cumulative impacts of multiple dams on the Tocantins floodplain, we quantified landscape-scale changes in floodplain extent, hydroperiod, and flood timing on a 145-km stretch of the river downstream of five dams. We used water level data from 1985 to 2019 to compare daily floodplain inundation dynamics before and after damming. We also developed models to examine the impacts of climate and land use change on hydrology of the Tocantins River. After installation of the first dam in 1998, an average of 82.3 km2 (63%) of the floodplain no longer flooded, overall average hydroperiod decreased by 15 days (11%), and flooding started an average of five days earlier. After all five dams were installed, 72% of the average pre-dam flooded area no longer flooded, average hydroperiod had decreased by 35%, and average inundation onset occurred 12 days later. These changes in floodplain hydrology appeared to be driven primarily by dam operations as we found no significant changes in precipitation over the study period. Increasing loss of natural vegetation in the watershed may play a role in changed hydrology but cannot explain the abrupt loss of floodplain extent after the first dam was installed. This is one of few studies to quantify dam-induced floodplain alteration at a landscape scale and to investigate impacts of multiple dams on a landscape. Our results indicate that the Tocantins River floodplain is undergoing drastic hydrologic alteration. The impacts of multiple dams are cumulative and non-linear, especially for hydroperiod and flood timing.

A remote sensing derived data set of 100 million individual tree crowns for the National Ecological Observatory Network.

Forests provide biodiversity, ecosystem, and economic services. Information on individual trees is important for understanding forest ecosystems but obtaining individual-level data at broad scales is challenging due to the costs and logistics of data collection. While advances in remote sensing techniques allow surveys of individual trees at unprecedented extents, there remain technical challenges in turning sensor data into tangible information. Using deep learning methods, we produced an open-source data set of individual-level crown estimates for 100 million trees at 37 sites across the United States surveyed by the National Ecological Observatory Network's Airborne Observation Platform. Each canopy tree crown is represented by a rectangular bounding box and includes information on the height, crown area, and spatial location of the tree. These data have the potential to drive significant expansion of individual-level research on trees by facilitating both regional analyses and cross-region comparisons encompassing forest types from most of the United States.

Testing metabolic ecology theory for allometric scaling of tree size, growth and mortality in tropical forests.

The theory of metabolic ecology predicts specific relationships among tree stem diameter, biomass, height, growth and mortality. As demographic rates are important to estimates of carbon fluxes in forests, this theory might offer important insights into the global carbon budget, and deserves careful assessment. We assembled data from 10 old-growth tropical forests encompassing censuses of 367 ha and > 1.7 million trees to test the theory's predictions. We also developed a set of alternative predictions that retained some assumptions of metabolic ecology while also considering how availability of a key limiting resource, light, changes with tree size. Our results show that there are no universal scaling relationships of growth or mortality with size among trees in tropical forests. Observed patterns were consistent with our alternative model in the one site where we had the data necessary to evaluate it, and were inconsistent with the predictions of metabolic ecology in all forests.

Modeling the spatial distribution and fruiting pattern of a key tree species in a neotropical forest: methodology and potential applications.

BACKGROUND: The movement patterns of wild animals depend crucially on the spatial and temporal availability of resources in their habitat. To date, most attempts to model this relationship were forced to rely on simplified assumptions about the spatiotemporal distribution of food resources. Here we demonstrate how advances in statistics permit the combination of sparse ground sampling with remote sensing imagery to generate biological relevant, spatially and temporally explicit distributions of food resources. We illustrate our procedure by creating a detailed simulation model of fruit production patterns for Dipteryx oleifera, a keystone tree species, on Barro Colorado Island (BCI), Panama. METHODOLOGY AND PRINCIPAL FINDINGS: Aerial photographs providing GPS positions for large, canopy trees, the complete census of a 50-ha and 25-ha area, diameter at breast height data from haphazardly sampled trees and long-term phenology data from six trees were used to fit 1) a point process model of tree spatial distribution and 2) a generalized linear mixed-effect model of temporal variation of fruit production. The fitted parameters from these models are then used to create a stochastic simulation model which incorporates spatio-temporal variations of D. oleifera fruit availability on BCI. CONCLUSIONS AND SIGNIFICANCE: We present a framework that can provide a statistical characterization of the habitat that can be included in agent-based models of animal movements. When environmental heterogeneity cannot be exhaustively mapped, this approach can be a powerful alternative. The results of our model on the spatio-temporal variation in D. oleifera fruit availability will be used to understand behavioral and movement patterns of several species on BCI.