No place to hide: Rare plant detection through remote sensing

Foliar spectra accurately distinguish most temperate tree species and show strong phylogenetic signal

PREMISE

Spectroscopy is a powerful remote sensing tool for monitoring plant biodiversity over broad geographic areas. Increasing evidence suggests that foliar spectral reflectance can be used to identify trees at the species level. However, most studies have focused on only a limited number of species at a time, and few studies have explored the underlying phylogenetic structure of leaf spectra. Accurate species identifications are important for reliable estimations of biodiversity from spectral data.

METHODS

Using over 3500 leaf-level spectral measurements, we evaluated whether foliar reflectance spectra (400-2400 nm) can accurately differentiate most tree species from a regional species pool in eastern North America. We explored relationships between spectral, phylogenetic, and leaf functional trait variation as well as their influence on species classification using a hurdle regression model.

RESULTS

Spectral reflectance accurately differentiated tree species (κ = 0.736, ±0.005). Foliar spectra showed strong phylogenetic signal, and classification errors from foliar spectra, although present at higher taxonomic levels, were found predominantly between closely related species, often of the same genus. In addition, we find functional and phylogenetic distance broadly control the occurrence and frequency of spectral classification mistakes among species.

CONCLUSIONS

Our results further support the link between leaf spectral diversity, taxonomic hierarchy, and phylogenetic and functional diversity, and highlight the potential of spectroscopy to remotely sense plant biodiversity and vegetation response to global change.

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.

Seeing from space makes sense: Novel earth observation variables accurately map species distributions over Himalaya

Topical advances in earth observation have enabled spatially explicit mapping of species' fundamental niche limits that can be used for nature conservation and management applications. This study investigates the possibility of applying functional variables of ecosystem retrieved from Moderate Resolution Imaging Spectroradiometer (MODIS) onboard sensor data to map the species distribution of two alpine treeline species, namely Betula utilis D.Don and Rhododendron campanulatum D.Don over the Himalayan biodiversity hotspot. In this study, we have developed forty-nine Novel Earth Observation Variables (NEOVs) from MODIS products, an asset to the present investigation. To determine the effectiveness and ecological significance of NEOVs combinations, we built and compared four different models, namely, a bioclimatic model (BCM) with bioclimatic predictor variables, a phenology model (PhenoM) with earth observation derived phenological predictor variables, a biophysical model (BiophyM) with earth observation derived biophysical predictor variables, and a hybrid model (HM) with a combination of selected predictor variables from BCM, PhenoM, and BiophyM. All models utilized topographical variables by default. Models that include NEOVs were competitive for focal species, and models without NEOVs had considerably poor model performance and explanatory strength. To ascertain the accurate predictions, we assessed the congruence of predictions by pairwise comparisons of their performance. Among the three machine learning algorithms tested (artificial neural networks, generalised boosting model, and maximum entropy), maximum entropy produced the most promising predictions for BCM, PhenoM, BiophyM, and HM. Area under curve (AUC) and true skill statistic (TSS) scores for the BCM, PhenoM, BiophyM, and HM models derived from maximum entropy were AUC ≥0.9 and TSS ≥0.6 for the focal species. The overall investigation revealed the competency of NEOVs in the accurate prediction of species' fundamental niches, but conventional bioclimatic variables were unable to achieve such a level of precision. A principal component analysis of environmental spaces disclosed that niches of focal species substantially overlapped each other. We demonstrate that the use of satellite onboard sensors' biotic and abiotic variables with species occurrence data can provide precision and resolution for species distribution mapping at a scale that is relevant ecologically and at the operational scale of most conservation and management actions.

Field observations and remote assessment identify climate change, recreation, invasive species, and livestock as top threats to critically imperiled rare plants in Nevada

Introduction

Rare plant species comprise >36.5% of the world’s flora and disproportionately support ecosystem function and resilience. However, rare species also lead global plant extinctions, and unique ecological characteristics can make them vulnerable to anthropogenic pressure. Despite their vulnerability, many rare plants receive less monitoring than is needed to inform conservation efforts due to limited capacity for field surveys.

Methods

We used field observations and geospatial data to summarize how 128 imperiled, rare vascular plant species in Nevada are affected by various threats. We assessed correlations between threats predicted by geospatial data and threats observed on the ground and asked how historic and current threats compare.

Results

The most commonly observed threats were from recreation, invasive and non-native/alien species, and livestock farming and ranching. Threat prevalence varied by elevation (e.g., a greater variety of threats at lower elevations, greater threat from climate change observed at higher elevations) and land management. There was a 28.1% overall correlation between predicted and observed threats, which was stronger for some threats (e.g., development of housing and urban areas, livestock farming and ranching) than others. All species experienced extreme climatic differences during 1990-2020 compared to baseline conditions, with the most extreme change in southern Nevada. The average number of threats observed per occurrence increased by 0.024 each decade.

Discussion

While geospatial data did not perfectly predict observed threats, many of these occurrences have not been visited in over 30 years, and correlations may be stronger than we were able to detect here. Our approach can be used to help guide proactive monitoring, conservation, and research efforts for vulnerable species.

A Practical Assessment of Using sUASs (Drones) to Detect and Quantify Wright Fishhook Cactus (Sclerocactus wrightiae L.D. Benson) Populations in Desert Grazinglands

Obtaining accurate plant population estimates has been integral in listing, recovery, and delisting species under the U.S. Endangered Species Act of 1973 and for monitoring vegetation in response to livestock grazing. Obtaining accurate population estimates remains a daunting and labor-intensive task. Small unmanned aircraft systems (sUASs or drones) may provide an effective alternative to ground surveys for rare and endangered plants. The objective of our study was to evaluate the efficacy of sUASs (DJI Phantom 4 Pro) for surveying the Wright fishhook cactus (Sclerocactus wrightiae), a small (1–8 cm diameter) endangered species endemic to grazinglands in the southwest desert of Utah, USA. We assessed sUAS-based remotely sensed imagery to detect and count individual cacti compared to ground surveys and estimated optimal altitudes (10 m, 15 m, or 20 m) for collecting imagery. Our results demonstrated that low altitude flights provided the best detection rates (p < 0.001) and counts (p < 0.001) compared to 15 m and 20 m. We suggest that sUASs can effectively locate cactus within grazingland areas, but should be coupled with ground surveys for higher accuracy and reliability. We also acknowledge that these technologies may have limitations in effectively detecting small, low-growing individual plants such as the small and obscure fishhook cactus species.

Deep Species Distribution Modeling From Sentinel-2 Image Time-Series: A Global Scale Analysis on the Orchid Family

Species distribution models (SDMs) are widely used numerical tools that rely on correlations between geolocated presences (and possibly absences) and environmental predictors to model the ecological preferences of species. Recently, SDMs exploiting deep learning and remote sensing images have emerged and have demonstrated high predictive performance. In particular, it has been shown that one of the key advantages of these models (called deep-SDMs) is their ability to capture the spatial structure of the landscape, unlike prior models. In this paper, we examine whether the temporal dimension of remote sensing images can also be exploited by deep-SDMs. Indeed, satellites such as Sentinel-2 are now providing data with a high temporal revisit, and it is likely that the resulting time-series of images contain relevant information about the seasonal variations of the environment and vegetation. To confirm this hypothesis, we built a substantial and original dataset (called DeepOrchidSeries) aimed at modeling the distribution of orchids on a global scale based on Sentinel-2 image time series. It includes around 1 million occurrences of orchids worldwide, each being paired with a 12-month-long time series of high-resolution images (640 x 640 m RGB+IR patches centered on the geolocated observations). This ambitious dataset enabled us to train several deep-SDMs based on convolutional neural networks (CNNs) whose input was extended to include the temporal dimension. To quantify the contribution of the temporal dimension, we designed a novel interpretability methodology based on temporal permutation tests, temporal sampling, and temporal averaging. We show that the predictive performance of the model is greatly increased by the seasonality information contained in the temporal series. In particular, occurrence-poor species and diversity-rich regions are the ones that benefit the most from this improvement, revealing the importance of habitat's temporal dynamics to characterize species distribution.