Geology Reports⌕ Search
Home/Archive/USGS

USGS · 70259638

Propagating observation errors to enable scalable and rigorous enumeration of plant population abundance with aerial imagery

Andrii Zaiats·Trevor Caughlin·Jennyffer Cruz·David S. Pilliod·Megan E Cattau·Rongsong Liu·Richard Rachman·Maisha Maliha·Donna M. Delparte·John DF Clare
View original source ↗Export metadata ↓

Abstract

Estimating and monitoring plant population size is fundamental for ecological research, as well as conservation and restoration programs. High-resolution imagery has potential to facilitate such estimation and monitoring. However, remotely sensed estimates typically have higher uncertainty than field measurements, risking biased inference on population status. We present a model that accounts for false negative (missed plants) and false positive (misclassified or double-counted plants) error in counts from high-resolution imagery via integration with ground data. We apply it to estimate the abundance of a foundational shrub species in post-wildfire landscapes in the western United States. In these landscapes, plant recruitment is crucial for ecological recovery but locally patchy, motivating the use of spatially extensive measurements from unoccupied aerial systems (UAS). Integrating >16 ha of UAS imagery with >700 georeferenced field plots, we fit our model to generate insights into the prevalence and drivers of observation errors associated with classification algorithms used to distinguish individual plants, relationships between abundance and landscape context, and to generate spatially explicit maps of shrub abundance. Raw counts of plant abundance in high-resolution imagery resulted in substantial false negative and false positive observation errors. The probability of detecting ( p ) adult plants (≥0.25 m tall) varied between sites within 0.52 < p̂adult < 0.82, whereas the detection of smaller plants (<0.25 m) was lower, 0.03 < p̂small < 0.3. On average, we estimate that 19% of all detected plants were false positive errors, which varied spatially in relation to topographic predictors. Abundance declined toward the interior of previous wildfires and was positively associated with terrain roughness. Our study demonstrates that integrated models accounting for imperfect detection improve estimates of plant population abundance derived from inherently imperfect UAS imagery. We believe such models will further improve inference on plant population dynamics—relevant to restoration, wildlife habitat and related objectives—and echo previous calls for remote sensing applications to better differentiate between ecological and observational processes.

Explore related subjects

Methods in Ecology and EvolutionArticle

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Andrii Zaiats, Trevor Caughlin, Jennyffer Cruz, David S. Pilliod, Megan E Cattau, Rongsong Liu, Richard Rachman, Maisha Maliha, Donna M. Delparte, John DF Clare. 2024-10-13. Propagating observation errors to enable scalable and rigorous enumeration of plant population abundance with aerial imagery. https://doi.org/10.1111/2041-210x.14421

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2025

Simulated soundscapes and transfer learning boost the performance of acoustic classifiers under data scarcity

1. The biodiversity crisis necessitates spatially extensive methods to monitor multiple taxonomic groups for evidence of change in response to evolving environmental conditions. Programs that combine passive acoustic monitoring and machine learning are increasingly used to meet this need. These methods require large, annotated datasets, which are time-consuming and expensive to produce, creating potential barriers to adoption in data- and funding-poor regions. Recently released pre-trained avian acoustic classification models provide opportunities to reduce the need for manual labelling and accelerate the development of new acoustic classification algorithms through transfer learning. Transfer learning is a strategy for developing algorithms under data scarcity that uses pre-trained models from related tasks to adapt to new tasks. 2. Our primary objective was to develop a transfer learning strategy using the feature embeddings of a pre-trained avian classification model to train custom acoustic classification models in data-scarce contexts. We used three annotated avian acoustic datasets to test whether transfer learning and soundscape simulation-based data augmentation could substantially reduce the annotated training data necessary to develop performant custom acoustic classifiers. We also conducted a sensitivity analysis for hyperparameter choice and model architecture. We then assessed the generalizability of our strategy to increasingly novel non-avian classification tasks. 3. With as few as two training examples per class, our soundscape simulation data augmentation approach consistently yielded new classifiers with improved performance relative to the pre-trained classification model and transfer learning classifiers trained with other augmentation approaches. Performance increases were evident for three avian test datasets, including single-class and multi-label contexts. We observed that the relative performance among our data augmentation approaches varied for the avian datasets and nearly converged for one dataset when we included more training examples. 4. We demonstrate an efficient approach to developing new acoustic classifiers leveraging open-source sound repositories and pre-trained networks to reduce manual labelling. With very few examples, our soundscape simulation approach to data augmentation yielded classifiers with performance equivalent to those trained with many more examples, showing it is possible to reduce manual label-ling while still achieving high-performance classifiers and, in turn, expanding the potential for passive acoustic monitoring to address rising biodiversity monitoring needs.

Methods in Ecology and Evolution↗
USGS2025

Spatial close-kin mark-recapture models applied to terrestrial species with continuous natal dispersal

Close-kin mark–recapture (CKMR) methods use information on genetic relatedness among individuals to estimate demographic parameters. An individual's genotype can be considered a ‘recapture’ of each of its parent's genotype, and the frequency of kin-pair matches detected in a population sample can directly inform estimates of abundance. CKMR inference procedures require analysts to define kinship probabilities in functional forms, which inevitably involve simplifying assumptions. Among others, population structure can have a strong influence on how kinship probabilities are formulated. Many terrestrial species are philopatric or face barriers to dispersal, and not accounting for dispersal limitation in kinship probabilities, can create substantial bias if sampling is also spatially structured (e.g. via harvest). We present a spatially explicit formulation of CKMR that corrects for incomplete mixing by incorporating natal dispersal distances and spatial distribution of individuals into the kinship probabilities. We used individual-based simulations to evaluate the accuracy of abundance estimates obtained with one spatially naïve and two spatially explicit CKMR models across six scenarios with distinct spatial patterns of relative abundance and sampling probability. Estimates of abundance obtained with a CKMR model naïve to spatial structure were negatively biased when sampling was spatially biased. Incorporating patterns of natal dispersal in the kinship probabilities helped address this bias, but estimates were not always accurate depending on the model used and the scenario considered. Incorporating natal dispersal into spatially structured CKMR models can address the bias created by population structure and heterogeneous sampling but will often require additional assumptions and auxiliary data (e.g. relative abundance indices). The models shown here were designed for terrestrial species with continuous patterns of natal dispersal and high year-to-year site fidelity but could be extended to other species.

Methods in Ecology and Evolution↗
USGS2025

Sampling mass mortality events to enable diagnoses: A protocol using freshwater mussels

Many taxa around the globe are threatened by often unexplained mass mortality events (MMEs), which can decimate populations and compromise key ecosystem functions. One example of a highly threatened taxon facing frequent MMEs is freshwater mussels (Unionida). There has been a recent increase in interest in understanding the causes of freshwater mussel MMEs, but standardised methodologies for how best to respond to them to facilitate diagnoses are unavailable. When an MME is observed, swift and appropriate sample collection is imperative owing to the transient nature of these phenomena. Here we provide structured guidance that will facilitate rapid and appropriate sampling of MMEs, using freshwater mussels as an example. We set out standardised procedures for sample collection, preparation and preservation. The procedures we outline will improve our capacity for diagnostic investigations of MMEs and other mortality events, not only in freshwater mussels but also across many other taxa. This, in turn, can inform appropriate management responses.

Methods in Ecology and Evolution↗