Remote Sensing

Military aviation training noise remains understudied despite its widespread impacts across urban, rural, and wilderness areas. The predominance of low-frequency noise and repetitive training can create pervasive noise pollution, yet past research often fails to capture the full range of health and quality-of-life effects. This study analyzed two complaint datasets related to Whidbey Island Naval Air Station noise: U.S. Navy records (2017-2020) and Quiet Skies Over San Juan County data (2021-2023). We analyzed and mapped sentiment intensity from noise complaints relative to modeled annual noise exposure, developed a typology to classify impacts, and modeled the environmental and operational factors influencing complaints. Findings revealed widespread negative sentiment and anger, often beyond the bounds of estimated noise contours, suggesting that annual cumulative noise models inadequately estimate community impacts. Complaints consistently highlighted sleep disturbance, hearing and health concerns, and compromised home environments due to shaking, vibration, and disruption of daily life. Residents also reported significant social, recreational, and work disruptions, along with feelings of fear, helplessness, and concern for childrens well-being. The number of complaints were strongly associated with training schedules, with late-night sessions being the strongest predictor. A delayed response pattern suggests residents reach a frustration threshold before filing complaints. Overall, our findings demonstrate persistent negative sentiment and diverse impacts from military aviation noise. Results highlight the need for improved noise metrics, modeling and operational adjustments to mitigate the most disruptive effects.

Terrestrial Laser Scanning (TLS) captures fine-scaled three-dimensional measurements of ecosystem structure, supporting monitoring of the Essential Biodiversity Variables (EBVs). Yet employing TLS across landscapes remains challenging in remote and topographically complex areas. Remote sensing provides a potential pathway for upscaling TLS-derived structural metrics, but to what extent is unquantified particularly in heterogenous environments, like oceanic islands. Here, we investigated the ability of remote sensing to estimate TLS-derived habitat structure across three contrasting habitats (lowland rainforest, montane cloud forest, and subalpine summit scrub) on La Reunion island. Sentinel-1, Sentinel-2, and Aerial LiDAR (ALS) data were acquired over plots where TLS was completed. We derived defined indices of backscatter coefficients, vegetation indices, and LiDAR metrics and assessed their alignment with TLS measurements using a Procrustes analysis. Subsequently, we used General Additive Models to estimate TLS habitat structure from remote sensing variables. Sentinel-2 exhibited the highest multivariate alignment with TLS (r = 0.51). TLS measurements of horizontal and vertical structure were estimated with the highest cross-validated predictive accuracy (R2 0.39 - 0.73), whilst structural complexity metrics were estimated with greater difficulty (R2 0.02 - 0.20). Multi-sensor models outperformed all single-sensor models in prediction estimates. Model performance also varied across habitats, with the highest agreement between predicted and observed values in the lowland rainforest (r = 0.38), and the lowest agreement (r = 0.35) in the montane cloud forest. Yet the dominant structural feature of each habitat was most accurately captured with remote sensing. Our results demonstrate the potential of integrating multi-sensor remote sensing data to upscale key dimensions of TLS-derived ecosystem structure but remains challenging for fine-scale structural complexity. These findings highlight both the potential and constraints of remote sensing for developing scalable, long-term monitoring frameworks for EBVs, especially in structurally complex and underrepresented island ecosystems.

Statewide tracking of urban tree canopy change is essential for evaluating progress toward policy targets, but detecting real change requires both high-resolution mapping and rigorous uncertainty estimation. We produced a four-year canopy cover time series for all California census-designated places using 60-cm NAIP aerial imagery and a U-Net deep learning model trained with semi-automated LiDAR-derived labels and manually annotated tiles. Canopy cover and change were estimated using stratified, error-adjusted area estimation, enabling comparisons across years. Statewide canopy cover showed a modest negative trend from 2016 to 2022 (Sens slope: -0.60% per year), but confidence intervals included zero across all groups and climate zones, indicating that trends were not statistically distinguishable from no change. Urban canopy cover was consistently lower than non-urban canopy by approximately six percentage points, and canopy cover was highest in the Northern California Coast and lowest in the Southwest Desert. Residential parcels accounted for 55-56% of canopy within incorporated urban areas across all years, indicating that statewide canopy increase goals will require engagement with private landowners. Error adjustment substantially altered canopy estimates relative to raw pixel-count totals, with direct implications for AB 2251 canopy tracking where baselines and targets drawn from unadjusted maps may not reflect true canopy extent. This open-source workflow is transferable to future NAIP acquisition years and other U.S. states, providing a scalable framework for long-term urban forest monitoring.

Fire has acted as a major eco-evolutionary force since the evolutionary appearance of plants, shaping plant-traits, diversity, vegetation assembly, and ecosystem functioning. Its ecological role depends on long-term fire regimes. Anthropogenic land-use change and climate warming are disrupting these regimes, particularly in densely populated regions such as the Mediterranean Basin. In the Italian peninsula (Mediterranean region) fire activity peaks during the dry summer months and is projected to intensify under climate change scenarios. Recent methodological developments - based on emerging satellite data, ground-based observations combined with Random Forest (RF) habitat classification, and spectral indices such as the NDVI provide a robust framework for monitoring post-fire land-cover dynamics over time. In this study, we applied RF modelling to classify vegetation cover using a 2017-2024 satellite imagery time series of the Monte Pisano area (central Italy) to assess pre- and post-fire vegetation trajectories. Evergreen shrubs and trees exhibited rapid post-fire regrowth, whereas coniferous stands showed slower recovery rates. NDVI trends revealed an expected sharp decline immediately after the fire, followed by gradual recovery of broadleaf forests and shrubland communities. Moreover, our results indicated a progressive increase in the cover of native deciduous and evergreen species of high conservation value (listed under the Habitat Directive). The framework delivers spatially explicit insights into post-fire recovery, supporting targeted management, restoration under European Nature Restoration Regulation, and long-term monitoring in Mediterranean ecosystems. Incorporating fine-scale environmental variables may further improve classification accuracy and enhance assessments of vegetation resilience and ecosystem recovery following fire events. HighlightsO_LIRecurring fires strongly affect ecosystem structure and function in Mediterranean landscapes. C_LIO_LIIntegrating remote sensing with Random Forest models enables effective monitoring of post-fire vegetation recovery over time. C_LIO_LINDVI time series provide reliable proxies for tracking vegetation vigor and land-cover change. C_LIO_LIPost-fire recovery trajectories are shaped by fire severity, vegetation physiognomy, plant functional types, and soil conditions. C_LIO_LITargeted restoration and management interventions informed by spatial-temporal vegetation patterns are urgently needed. C_LIO_LIThe proposed framework aligns with objectives of the EU Nature Restoration Regulation for ecosystem and habitat recovery. C_LI

Road traffic accidents represent a global public safety crisis, necessitating advanced computational tools for accurate injury severity prediction and effective decision support. This study evaluates high-performing ensemble machine learning models, including AdaBoost, XGBoost, LightGBM, HistGBRT, CatBoost, Gradient Boosting, NGBoost, and Random Forest, using a comprehensive National Highway Traffic Safety Administration (NHTSA) dataset from 2018 to 2022. While all models demonstrated exceptional predictive accuracy, with HistGBRT achieving the highest overall accuracy of 92.26%, a defining achievement of this work is the perfect classification (100% precision and recall) of fatal injuries across all ensemble architectures. To bridge the gap between predictive performance and actionable intelligence, this research integrates SHapley Additive exPlanations (SHAP) to provide both global insights into dataset-wide risk factors and local, instance-specific rationales for individual crash events. The global analysis identified ethnicity, airbag deployment, and harmful event type as primary drivers of injury severity, while local force and waterfall plots revealed the precise "push and pull" of variables for specific incidents. The results offer a robust, interpretable framework for stakeholders tasked with improving traffic safety and mitigating crash-related harm.

Convolutional Neural Networks (CNNs) have become increasingly prominent in biodiversity monitoring due to their strong performance in accurately detecting species from sound recordings, overcoming some limitations of traditional methods such as point-counts. Yet, their use in urban ecosystems remains limited, highlighting the need for frameworks that identify modelling strategies to optimize their performance in these complex soundscapes. Here, we evaluated how preprocessing and labelling strategies, detection thresholds, sample size, and architecture affect the performance of CNNs for bird identification in urban tropical ecosystems. We also assessed its potential by comparing CNN-derived biodiversity estimates with those from point-counts and acoustic indices. For this, we used one week of recordings collected along urbanization gradients in five Colombian Andes cities to developed 11 multiclass CNN models varying in spectral representation, labelling strategies, training data source and backbone architecture. The best-performing model, evaluated with F1-scores, combined Log-Mel spectrograms, multispecies labels, ecosystem-specific recordings, a probability threshold of 0.3 and a ConvNeXt backbone with its performance generally improving with sample size. Although CNNs and point counts detected partially distinct assemblages, CNN-derived species richness was comparable to that estimated from point-counts. In addition, the Normalized Difference Soundscape Index (NDSI) was positively associated with richness, suggesting its potential as a biodiversity proxy in tropical urban soundscapes. Overall, by identifying effective modelling designs and monitoring strategies, our study advances the development of robust biodiversity assessment frameworks in urbanized ecosystems in the Neotropics whilst also providing methodological guidance for future research and practical insights for wildlife monitoring and conservation.

Image-based crop and pest recognition is considered useful for reducing the delay and cost of manual field scouting, therefore supporting timely intervention in precision-agriculture workflows. However, the real field imagery remains challenging due to the cluttered backgrounds, occlusions, illumination changes, and strong scale variation that are frequently observed across crops. The symptoms are often small or low-contrast, and pests may be partially hidden, which reduces the reliability when the setting is outside controlled environments. A unified multi-class crop-pest/condition recognition framework is presented, where a ResNet-50 backbone is utilized and enhanced with a Multi-Scale Contextual Attention (MSCA) module. The novelty is mainly considered to be achieved through the integration of explicit multi-scale contextual aggregation with lightweight joint channel and spatial attention by means of residual fusion, while the empirical evaluation was kept controlled under a fixed and reproducible protocol. A curated dataset of 21,404 field-style images covering 15 crop and pest/condition classes was compiled, and a leakage-aware fixed split with a held-out test set was adopted to support reproducibility. Augmentation was applied only to the training subset to improve robustness, although the validation data was not augmented in the same manner. On the held-out test set, balanced performance was achieved by the proposed approach, with about 0.93 accuracy and a macro-F1 score close to 0.94 being obtained, while established baselines such as EfficientNet, Vision Transformer, and attention-based CNN models were outperformed under identical evaluation settings. Controlled ablations were used to isolate the contribution of MSCA and augmentation under the same training configuration. These results indicate that lightweight multi-scale contextual attention is effective for crop and pest recognition under realistic field conditions, although some visually similar classes remained difficult.

The US NEXRAD radar network has monitored the aerosphere over the US and its territories continuously since the 1990s and archived nearly 300 million radar volume scans. These data contain a wealth of information about the movements of birds, bats, and insects. Historically, this biological information was difficult to access due to the amount of data and challenges in analyzing it. In the last 15 years, fueled by computational and methodological advances, large-scale aeroecology research has blossomed. However, comprehensive analyses of the NEXRAD archive remain very costly. We collected measurements of biological activity from every volume scan in the NEXRAD archive--nearly 300 million data files total--to assemble a dataset of aerial biomass over the US from 1995 to 2025. The core data are vertical profiles, which summarize biological activity at different heights above the radar station for each volume scan. We also provide time series data products that aggregate vertical profiles to point measurements at radar stations across time. These data products can support a range of aeroecology analyses at significantly reduced effort.

While tree pollen is a major trigger of allergic respiratory conditions and different taxa exhibit varying allergenic potentials, the lack of high-resolution, taxon-specific exposure metrics have limited our ability to identify which local pollen taxa are primarily responsible for respiratory illness. Traditional pollen monitoring networks, which have an intermittent sampling schedule, are not ideal for examining the delayed effects of pollen exposure due to the missing days. In this study, we developed a modeling framework integrating atmospheric dispersion effects, taxa-specific phenology, and machine learning to predict daily counts of 13 tree taxa in the five-county Metro Atlanta area, Georgia at a 1-km resolution from 2020 to 2024. Machine learning model performance was validated with daily pollen counts collected by a multi-site monitoring network equipped with automated pollen sensors. Findings showed that Betula and Quercus pollens exhibited higher predictive performance than other taxa, with R2 values ranging from 0.69 to 0.92 and from 0.71 to 0.89, respectively. Our 1-kilometer prediction data provides gapless exposure metrics to understand the spatial and temporal variability in pollen exposure, can facilitate investigation of urban pollen hotspots and support epidemiologic studies of pollen-related respiratory outcomes.

Hyperspectral imaging is an imaging technique that allows for acquisition of high-resolution spectral information beyond that of the visible spectrum. When applied to plants, it effectively enables non-invasive characterization of physiological status and has been widely used in agricultural settings. Marchantia is a model bryophyte species whose flat morphology and visually distinct stress-response phenotypes makes it an ideal candidate for imaging studies. Here, we provide a comprehensive protocol for hyperspectral imaging for Marchantia plants, which encompasses hardware configuration, data acquisition, and computations processing. This protocol features a streamlined data processing pipeline hosted on a web-based development platform that automates 1) the segmentation of plant area into spatially distinct regions for localized analysis of intra-specimen physiological gradients, and 2) classification of plant pixels based on their spectral signatures. All results are exported as structured CSV files for ease of further analysis as desired by the user.

Premise: Rapid and accurate plant species identification is a critical challenge exacerbated by the taxonomic impediment. Although portable near-infrared (Micro NIR) spectroscopy represents a promising solution, the current absence of standardized protocols and a fundamental understanding of how critical acquisition and analysis parameters influence accuracy remain significant barriers. This study focused on the systematic optimization and validation of a comprehensive workflow designed to maximize the reliability of plant identification using this technology. To ensure methodological robustness across diverse foliar matrices, four vascular plant species were strategically selected as a representative test set to encompass morphological extremes, including significant variations in leaf thickness, pubescence, and surface texture. Methods: Using a portable spectrometer on herbarium specimens (exsiccate) of four vascular plant species, we systematically tested five spectral backgrounds, seven pre-processing methods, and four classification models. Subsequently, we optimized the number of spectral readings and evaluated the influence of the leaf scanning surface (adaxial vs. abaxial) on model accuracy. Results: The highest-performing combination was a Shiny Aluminum background, Second Derivative pre-processing, and a Random Forest model, which achieved a mean cross-validated accuracy of 99%. An average of just three spectral readings from the adaxial (upper) leaf face was sufficient to saturate model performance, proving statistically superior to other approaches (p < 0.001). Discussion: This study establishes a validated, high-accuracy protocol for plant species identification from herbarium specimens using portable NIR, offering a powerful tool for biodiversity studies. Direct applicability to fresh plants in the field requires future validation to account for the spectral influence of moisture variability.

OverviewThis dataset originates from a preliminary respirometry study on carabid beetles from the Elbe Estuary (Northern Germany), encompassing species from freshwater and saltmarsh habitats along a salinity gradient. The study was designed to establish and validate a workflow for measuring oxygen consumption, including chamber setup, sensor recording, drift correction, and calculation of absolute and mass-specific metabolic rates. Oxygen consumption was measured for five species (Carabus auratus, Carabus granulatus, Limodromus assimilis, Poecilus versicolor and Pterostichus niger) using sealed glass vials connected to an optical oxygen system. The dataset provides individual-level measurements and serves primarily as a methodological reference for future respirometry studies on ground-dwelling arthropods. The O2 consumption rates of carabid beetles showed interspecific differences and followed metabolic scaling theory, revealing an inverse relationship between body mass and mass-specific metabolic rates across species (Figure 3). O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=127 SRC="FIGDIR/small/720111v1_fig3.gif" ALT="Figure 3"> View larger version (17K): org.highwire.dtl.DTLVardef@f41f27org.highwire.dtl.DTLVardef@12939eeorg.highwire.dtl.DTLVardef@19a4630org.highwire.dtl.DTLVardef@17611ba_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOFigure 3:C_FLOATNO Oxygen consumption rates of Carabid species per (a) animal in [ml O2 h-1] and as (b) mass-specific consumption rate [ml O2 h-1 g-1]. Points represent mean oxygen consumption per individual (C. auratus: n = 2; L. assimilis: n = 6; P. versicolor: n = 7; P. niger: n = 6). C_FIG

Early, precise, and non-destructive stress detection is essential for maintaining crop productivity, particularly in high-density plant growth systems like controlled environment agriculture (CEA), where manual monitoring is often impractical. Using plant motion as a proxy for growth and plant health, we demonstrate a method for early, non-invasive stress detection through quantitative leaf-movement analysis in lettuce and five other CEA relevant crops. Leaf-movement dynamics under stress were imaged with a low-cost, scalable Raspberry Pi imaging setup and quantified using a repurposed open-source motion estimation algorithm; Tracking Rhythms in Plants (TRiP). Our system detected stress-induced changes in leaf-movement within 1 hour of stress, with the timing dependent on the nature of the stress. Sustained reductions in leaf-movement coincide with decreased biomass accumulation. This approach offers a non-invasive, rapid, scalable, and cost-effective solution for continuous crop monitoring, with potential for application in both terrestrial and space farming CEA systems. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=138 SRC="FIGDIR/small/732190v1_ufig1.gif" ALT="Figure 1"> View larger version (54K): org.highwire.dtl.DTLVardef@19ee20eorg.highwire.dtl.DTLVardef@b0804org.highwire.dtl.DTLVardef@3b3fa8org.highwire.dtl.DTLVardef@1d04026_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical abstract:C_FLOATNO Quantification of leaf-movement dynamics as a high-throughput proxy for plant physiological status, enabling early stress detection and timely intervention to mitigate yield penalties in CEA settings (image made with biorender.org). C_FIG

High-throughput 3D multispectral plant phenotyping platforms generate large volumes of point cloud files, but trait extraction is typically performed by sensor-bundled software whose internal algorithms are not publicly documented, which limits reproducibility and integration into custom research pipelines. Here we present PhytoScan3D, an open-source Python pipeline that extracts morphological and spectral phenotypic traits, spanning plant height, 3D leaf area, digital biomass, convex hull volume, leaf inclination, canopy geometry, NDVI, hue, and vegetation indices, from both PLY and PCD point cloud files generated by Phenospex PlantEye F500 and F600 sensors, and is portable to point clouds from any acquisition platform. PhytoScan3D was validated against HortControl (PhenoSpex) ground-truth measurements on 936 barley (Hordeum vulgare) pot-date observations from the growth chamber trial (20 Norwegian cultivars, 12 scan dates, Septemenr 2025 to January 2026), achieving Pearson r = 0.913 to 0.999 and ratio approximately 1.000 for Plant Height Max, 3D Leaf Area, and NDVI Average. A vectorised mesh face filtering implementation achieved a 120x speed improvement, increasing valid 3D Leaf Area coverage from 0.6% to 100% of files. Cross-format validation on 223 PlantEye F600 PCD files from the ICRISAT LeasyScan platform (four legume species: mungbean, cowpea, lima bean, and common bean; 1,523 plant observations) yielded r = 0.884 against independent cuboid annotation heights. The systematic positive bias (mean +27.2 mm, ratio = 1.44) is attributable to PhytoScan3D computing height from raw point cloud Z-range while cuboid annotations are fitted to segmented plant points only, with the offset consistent across all four species (per-species r = 0.880 to 0.888). Cross-dataset processing of 1,180 PLY files from the Crops3D benchmark (8 species, 3 acquisition methods) confirmed zero extraction errors. PhytoScan3D is available at "github.com/kovimallik/phytoscan3d" under the MIT licence and processes 1,651 files across three independent datasets in under 12 minutes on GPU hardware. HighlightsO_LIPhytoScan3D is the first open-source Python pipeline for batch extraction of phenotypic traits, including plant height, 3D leaf area, digital biomass, convex hull volume, leaf inclination, NDVI, and excess green index, from both PLY and PCD point cloud files generated by Phenospex PlantEye sensors. C_LIO_LIPrimary validation against HortControl ground-truth measurements on 936 barley pot-date observations achieved Pearson r = 0.913-0.999 for Plant Height Max, 3D Leaf Area, and NDVI Average. C_LIO_LIA 120x computational speedup in mesh face filtering (vectorised NumPy vs. set-based loop) increased the coverage of valid 3D Leaf Area extraction from 0.6% to 100% of files. C_LIO_LICross-format validation on 223 PlantEye F600 PCD files from ICRISAT LeasyScan (four legume species, 1,523 plants) achieved r = 0.884 against independent cuboid annotation heights. The systematic +27.2 mm bias reflects a methodological difference (raw Z-range vs. soil-segmented annotations), is consistent and predictable across all four species (per-species r = 0.880-0.888), and is correctable by a single linear factor. C_LIO_LICross-dataset processing of 1,180 PLY files from the Crops3D benchmark (8 species, 3 acquisition methods) confirmed zero extraction errors. C_LIO_LISignificant scan-unit variation was detected for Plant Height Max (F = 5.71, p < 0.001, 2 = 0.138) and Canopy Width X (F = 6.32, p < 0.001, 2 = 0.150), demonstrating the biological utility of extracted traits. C_LI

1Non-structural carbohydrates (NSCs) are central to plant carbon allocation and physiological regulation, yet their quantification typically relies on destructive biochemical assays that lack spatial resolution. Here, we developed a shortwave infrared (SWIR) hyperspectral imaging workflow for non-destructive estimation and spatial reconstruction of starch-associated variation in strawberry leaves. The workflow combined automated hyperspectral segmentation, spectral preprocessing, Partial Least Squares Regression (PLSR), and constrained wavelength selection. Sample-level spectra extracted from 114 strawberry leaf samples grown across three different metabolic conditions were paired with destructive starch measurements and used to train models across the 900-1750 nm spectral range. A constrained greedy band-selection strategy revealed that predictive performance approached a plateau at approximately 12 wavelengths, indicating substantial spectral redundancy within the full hyperspectral dataset. The final reduced-band model achieved a cross-validated coefficient of determination (R2) of 0.771 {+/-} 0.066 and a root mean squared error (RMSE) of 0.743 {+/-} 0.098 mg g-1 fresh weight using repeated stratified 5-fold cross-validation. Pixel-wise application of the final model generated spatial starch-associated maps that preserved pronounced intra-leaf heterogeneity, including vein-associated spatial structure. These results demonstrate that starch-associated spectral information can be reconstructed from a constrained reduced-band SWIR framework while retaining sufficient predictive performance for spatial mapping. The identified wavelength reduction supports the feasibility of deployable multispectral systems for non-destructive carbohydrate sensing in plant phenotyping applications.

The automated classification of animals from photos is important in ecology and conservation biology for organizing and understanding the immense diversity of species, as well as facilitating effective conservation and management practices. It is equally important for disease surveillance systems, allowing prompt detection of anomalies in species distributions and boosting citizen-scientist platforms by making user-reported data more accurate and convenient. Image classification uses photos and can also rely on the geographical locations of animals to improve performance. While image classification models have difficulties in classifying low-quality images, unbalanced datasets, and with a small number of images, species distribution models have difficulty in classifying species that coexist in a given region. We propose here strategies for combining image classification models based on deep neural networks with species distribution models using genetic algorithms. The proposal is applied to a real-world dataset comprising fifteen classes of animals from the Brazilian fauna obtained from Fiocruzs citizen-scientist Wildlife Health Information System (SISS-Geo). The SISS-Geo photos portray the reality of animals in their environments, with varying quality, and pose numerous difficulties for classification. Experimental results demonstrate that the proposed integration consistently outperforms standalone models. While individual SDMs achieve Top-1 accuracies of 27.79% (MaxEnt) and 31.76% (Bioclim), and CNN-based classifiers reach 58.17% with ResNet50 and 64.13% with ResNet-152, the hybrid strategies yield substantial improvements. The genetic algorithm-based integration with a single global weight achieves up to 67.96% Top-1 accuracy, whereas the class-specific integration using fifteen parameters attains the best overall performance, reaching 69.03%.

1 - Automated in situ sensors - e.g., buried scanners - are transforming biodiversity monitoring by generating data at spatio-temporal resolutions unattainable through traditional sampling, including in cryptic environments such as soil that have remained largely inaccessible to existing methods. However, extracting ecologically meaningful information from these data streams requires substantial image processing effort that currently constitutes a critical bottleneck, particularly when the signal-to-noise ratio is low and annotated training data are scarce. 2 - Standard end-to-end deep learning detection pipelines offer unsatisfactory results due to the lack of training data and heterogeneity of the taxa of interest. We explore the potential of combining traditional computer vision algorithms with state-of-the-art deep learning models to build an efficient raw data processing pipelines from limited annotation effort. Specifically, based on the observation that the background barely changes, we focus on the differences between two consecutive images to turn the initial detection problem (with very low signal) into a simpler classification problem, which we solve by fine-tuning foundation models on limited annotated data. 3 - Our approach significantly reduces the annotation effort, allowing us to release an open dataset with about 600 soil scans and more than 8 000 labeled invertebrate occurrences across nine taxa. Using this dataset to train our models, we obtained population count estimates with relative errors ranging from 10% to 61% across taxa over a three-month period. Ecological validation through a land-use stability analysis showed full directional congruence between automated and expert-annotated classifications across all nine taxa examined, with effect-size discrepancies proportional to per-taxon classification accuracy. 4 - These results demonstrate that combining domain-specific heuristics with fine-tuned foundation models provides an effective and data-efficient strategy for automating ecological image processing workflows in low-signal, data-scarce contexts. The validated pipeline removes the manual annotation bottleneck that has historically limited scanner-based soil monitoring to short observational windows and restricted taxonomic scope, opening the way for continuous, large-scale tracking of soil invertebrate community dynamics at resolutions previously unachievable.

Southern leaf blight (SLB) is a foliar disease of maize (Zea mays L.) caused by the necrotrophic fungal pathogen Cochliobolus heterostrophus. Genetic resistance is the most effective control method for SLB. Developing disease resistant maize lines requires field trials during which disease phenotypes must be visually assessed. Remote sensing using drones is an emerging technology that can be leveraged for high-throughput phenotyping of disease severity that is otherwise labor-intensive and subjective. This project used a deep learning approach to estimate SLB disease severity of single-row maize plots from drone imagery. Over 26,000 plot-level images produced from flights conducted across three growing seasons were labeled with in-field visual scores taken contemporaneously by expert raters. Variation in environmental conditions contributed to a labeled image dataset that reflects the complexity of agronomic field experiments. We assessed the ability of nine deep learning models from three architectural families to estimate disease severity. The best-performing model, EVA-02-B, achieved strong cross year generalization (R2 = 0.697). Error analysis found that performance was more strongly associated with seasonal disease progression and flight-score time offset than with image-level noise. UAV-based deep learning estimated SLB severity with comparable precision to expert raters. This study lays the groundwork for integrating automated phenotypes into genetic studies of disease resistance. PLAIN LANGUAGE SUMMARYSouthern leaf blight (SLB) of maize is a disease that causes yield loss worldwide and developing resistant varieties offers the best hope for controlling the disease. Studying SLB resistance requires plant pathologists to visually score severity in the field, a labor-intensive method that requires expertise. To address these challenges, we asked whether SLB severity scoring could be automated using drone images and artificial intelligence (AI). We trained AI models using three years of image and score data then compared the results to visual scores taken by five plant pathologists. The best performing AI model showed a similar level of consistency to the experts and proved capable of scoring severity despite unpredictable and uncontrollable conditions that affect field imaging experiments such as weeds or shadows. These findings provide a validated method that improves the efficiency of maize disease research, a critical area of study for agricultural sustainability and productivity.

Waterbirds serve as important indicators of both aquatic and terrestrial ecosystem health, making effective monitoring essential for tracking population health and identifying potential causes of decline. Drones have provided opportunities to overcome historic waterbird monitoring challenges, but the expertise and time required for manual image analysis creates a major bottleneck. Recent advances in deep learning-based object detection have enabled rapid, automatic detection of features in complex ecological imagery, though applications have largely been limited to single-species colonies, and practitioners lack quantitative comparisons of annotation time and accuracy across different levels of automation. We systematically compared four waterbird monitoring approaches using identical survey areas from Chester Island, a mixed-species colony in Matagorda Bay, Texas, in 2025: (1) traditional ground-based counts, (2) manual drone imagery-based counts, (3) computer-assisted counts using pre-annotations from an object detector with manual human verification (Human+ML), and (4) fully automated counts using object detector annotations (ML-only). We trained a YOLOv10 object detection model on manually annotated imagery of Chester Island in 2021 and applied it to the 2025 imagery. Manual drone annotation detected 6,530 birds in 40.5 hr and served as the primary reference standard. Human+ML detected 5,826 birds (89% of manual) in 7.7 hr, an 81% reduction in annotation time. ML-only detected 5,679 birds (87% of manual) in approximately 46 min, a 98% reduction. Ground counts recorded 5,868 birds (90% of manual). Detection generalized well across species while classification depended heavily on training data and morphological distinctiveness. The Human+ML workflow emerged as a practical middle ground, providing practitioners with empirical data to evaluate partial versus full automation strategies based on monitoring objectives.

Artificial intelligence (AI)-based image analysis can efficiently estimate fish length, but differences in devices, imaging conditions, operators, and AI models limit comparability among surveys. We propose a standardization framework that estimates a bin-specific error matrix from paired reference measurements and AI-derived lengths and applies it to standardize (correct) AI-derived length-frequency distributions. The Richardson-Lucy expectation-maximization algorithm was used, with the number of iterations selected via cross-validation. Simulations based on empirical length-frequency data from 110 species showed that standardization reduced relative bias and distributional discrepancy; median relative-bias and root mean square error ratios were below 1, and the performance was more affected by the amount of paired data than by the number of cross-validation folds. In real data from 957 Japanese jack mackerel, standardized AI-derived distributions approached human-observer histograms, although discrepancies remained in the range of 160-230 mm. The proposed framework provides a practical approach for improving the comparability of image-derived length-frequency data using paired calibration data, without retraining the underlying AI model.