Remote Sensing in Ecology and Conservation

Wildlife re-identification (re-ID) is a widely used and powerful tool with diverse applications in animal ecology and conservation. Current automated methods typically operate on single images of a single body part of the animal. However, a single encounter may contain multiple images capturing different body regions, each providing complementary individual-specific information. In contrast to automated approaches, researchers often manually select the most suitable images and regions for identification based on factors like visibility, occlusion and image quality. This creates a mismatch between automated methods and field practice, limiting the practical adoption of current automated re-ID pipelines. Here, we address this by introducing an encounter-based, multi-body-part re-ID framework, using sea turtles as a model taxon. Our framework combines three elements: (1) An orientation-aware deep learning model, TurtleDetector, that in addition to the full bodies, it also automatically segments key body regions, i.e. heads, front and hind flippers, from images within an encounter; (2) a hybrid body-part-specific retrieval method, that sequentially combines a fast global-feature model (MiewID or DINOv3) with a more accurate but costlier local-feature model (ALIKED with LightGlue); and (3) a merged identity-prediction strategy that selects the highest calibrated similarity score across all available body parts and images of an encounter. We evaluate the framework on three long-term re-ID datasets spanning three species, loggerheads, greens, and hawksbill turtles, under an evaluation protocol that mirrors real-world, time-aware re-ID workflows. Across datasets, combining multiple body regions consistently improved identification performance over the best-performing single body region, resulting to an increase of 4-6% in top-1 accuracy. Interestingly, body regions traditionally underused in sea turtle re-ID, such as the hind flippers and carapaces, provided complementary identifying information that improved encounter-level re-ID when integrated through the hybrid retrieval method. Our findings demonstrate that automated wildlife re-ID can benefit from moving beyond single-image, single-body-part identification towards encounter-level integration of all available visual evidence. Our work further suggests that, where feasible, field photo-acquisition protocols should aim to capture multiple informative views of an individual during each encounter. Importantly, many species and taxa, including elephants, primates, cetaceans, and other large vertebrates, possess such individual-specific features across multiple body regions, highlighting the broad potential applicability of our framework.

High-resolution monitoring of forest structure and productivity is essential for effective natural resource management. However, monitoring approaches such as field-based forest inventories or extensive lidar campaigns are costly, time-intensive, and spatially limited. Therefore, inexpensive and accessible methods are needed. SatCHM (Satellite Canopy Height Model) was developed to be an accessible and open-source tool for researchers, allowing for site-specific and temporally flexible predictions of canopy height with limited computational resources. SatCHM requires four inputs: panchromatic satellite imagery, solar and sensor angle metadata of satellite imagery, digital elevation models (DEMs), and lidar-produced CHMs for an area of interest. After SatCHM pre-processes inputs, data is loaded into a collection of convolutional neural networks (CNNs) for image-to-image regression. This ensemble cooperates to yield high-resolution predictions (up to 0.5-meter) of three-dimensional tree structure with discernible tree crowns across a broader defined area of interest. After calculating the mean absolute error for each prediction output, the median of these mean absolute errors was 6.06 meters.

Coral reef fish assemblages are widely used as indicators of ecosystem condition, yet manual annotation of underwater video remains a major bottleneck for scalable biodiversity monitoring. Despite rapid progress in automated detection, ecologically realistic and publicly available datasets remain scarce, particularly for the Western Indian Ocean. Here, we present WIO-ReefFish, a reef fish detection dataset derived from diver-operated line-intercept transects and designed for ecological monitoring under natural survey conditions. WIO-ReefFish comprises 1,000 ultra-high-definition images (3840 $\times$ 2160 pixels) and 6,768 exhaustive bounding-box annotations spanning 24 taxonomic categories, thereby preserving full-frame assemblage structure in complex reef scenes. We also establish a standardized benchmark across nine object detection models under two complementary protocols: class-aware detection and class-agnostic fish localization. Detection performance was consistently higher under the class-agnostic protocol. The best-performing model (RT-DETR) improved from 0.48 mAP50 in the class-aware setting to 0.70 mAP50 when taxonomic constraints were removed, indicating that taxonomic discrimination remains substantially more challenging than fish localisation in reef imagery. Spatially independent evaluation revealed a pronounced generalisation gap, particularly for taxonomic detection, whereas class-agnostic fish localisation remained substantially more robust across transects and countries. Together, these results establish WIO-ReefFish as a realistic benchmark for automated reef fish detection and provide a foundation for more robust computer-vision tools in coral reef biodiversity monitoring. The WIO-ReefFish dataset and associated benchmarking resources are publicly available.

Coastal forests are increasingly threatened by saturated soil and elevated salinity levels resulting from sea level rise, saltwater intrusion, and storm surges. In response to rising salinization and flooding, healthy coastal forests that rely on freshwater (both wetland forests and low-elevation upland forests) are transitioning into landscapes dominated by dead or dying trees, known as ghost forests. Situated among salt-tolerant shrubs and grasses, ghost forests eventually become marshes or open water. Here, our main objective was to quantify the dynamics and pathways of these forest landscape conversions, as well as the factors contributing to the changes, which is vital for understanding the progression of coastal ecosystem degradation and forecasting future changes. We focused first on identifying the best method to track forest landscape change by exploring the role of multiple remote sensing indices (i.e., multispectral, bi-seasonal, topographical, and phenological metrics) in enhancing the performance of deep learning models (convolutional neural networks, CNNs) for land cover classification in the coastal plain of North Carolina using surface reflectance of Landsat 8 and Sentinel-2 images. Then, we used the best available data (Landsat 8) to understand long-term change and identify patterns of land cover change from 1985 to 2021. Our study reveals that incorporating phenology and topographical indices enhances the separability of the ghost forests class from all other vegetation classes. In our assessment, the higher-resolution Sentinel-2 data (F1 Score = 96.3) outperformed Landsat images (F1 score = 93.4) for the 2021 co-available year. However, Landsat remains an important tool used due to its long-term data record. Therefore, we used Landsat to determine that 21% of forests were lost between 1985 and 2021, and that the rate of loss is increasing. Between 2010 and 2021, 23,876 ha of forest were converted to marsh, ghost forest, and shrub, which is 1.5 times higher than the 16,968 ha lost between 1985 and 2010. These conversions from forest to ghost forest and marshes were driven primarily by proximity to the channel, salinity, and the increasing rate of relative sea level rise (RSLR), which are the key environmental drivers of observed changes. By quantifying these changes, we highlight regions most vulnerable to environmental stressors, providing a basis for targeted conservation strategies.

The tidal wetlands of Louisiana comprise about 25% of those found throughout the conterminous United States yet estimates of wetland loss rates in the region between 1932 and 2016 have exceeded 60 km2 yr-1. To mitigate further degradation and wetland loss in the region, a globally unprecedented $50B, 50-year plan for coastal Louisiana is driving restoration efforts, and demand exists from multiple stakeholders for regularly updated, regional-scale, accurate land cover information. We used machine learning (random forests; RF) and cloud computing to develop a new Landsat-based, marsh vegetation community geospatial dataset. The dataset depicts wetland vegetation community types defined in a previous study at annual (1985-2025) time steps at 30-m resolution. An RF algorithm was used to integrate training samples with feature variables derived from Landsat imagery, and the resulting geospatial data product achieved an overall correct classification rate of 78%. The approach for development of the land cover dataset presented here has potential for application in other coastal wetland habitats throughout the world.

1. Passive acoustic monitoring (PAM) is increasingly used for ecological research, biodiversity monitoring, assessment, and reporting. Automated species classifiers make it feasible to process large audio datasets but generate numerous detections that often need validation before use in downstream analyses or formal outputs. 2. Method development in PAM has focused on classifier building and downstream models that account for imperfect detection, yet the practical step between these - post-classification validation - remains weakly supported and is often implemented through ad hoc workflows. This increases manual handling, creates scope for transcription or consolidation errors, limits transparency and makes it difficult to document what was reviewed. 3. We introduce PAMalytics, an open-source, no-code, local browser-based application to support post-classification validation as a standardised workflow stage. PAMalytics ingests detections from any classifier, allows users to define how detections are sampled for review, and presents selected detections alongside their spectrograms with audio playback in one unified interface. Sampling strategy and review decisions are tracked alongside reviewer identity improving traceability and reproducibility across the validation workflow. 4. Case studies with Conservation International Cambodia and Imperial College London demonstrate PAMalytics in two validation settings. In Cambodia, gibbon predictions from a large, uneven dataset were sampled within sites, with likely classifier errors prioritised for validation. At Imperial, Amazon bird detections were sampled across each species classifier-confidence range before biodiversity metrics were derived. In both cases, PAMalytics reduced manual handling and validation time. By turning an ad hoc step into an accessible, structured workflow for conservation practitioners, PAMalytics fills a practical gap in the PAM bioacoustics pipeline and strengthens the link between automated detections and evidence used in biodiversity monitoring and reporting.

Understanding how animals respond to their physical environment requires environmental observations at the scale at which behavioural decisions are made. For soaring birds, the coarse resolution of weather products has long hindered the analysis of their behavioural response to fine-scale atmospheric dynamics, forcing uplift sources to be inferred largely from behaviour itself. Here, we combined high-resolution movement data from 24 golden eagles with the kilometre-scale COSMO weather model. We first classified thermal, orographic, and gravity-wave uplifts using independent atmospheric predictors and then quantified the birds' use of each uplift type and their fine-scale behavioural responses. Eagles relied predominantly on thermals, but opportunistically adjusted their use of uplift sources seasonally. The birds' flight behaviour could not reliably indicate which uplift type was primarily used, and thus suggests that both atmospheric processes and behavioural responses are better described as continua than discrete categories. Finally, we compared vertical wind velocities derived from eagles soaring behaviour with those modelled by the COSMO weather model, showing that most of the thermals exploited by eagles remain unresolved at kilometre-scale model resolution. Our results demonstrate how high-resolution weather models provide new insights into bird movement decisions, while also highlighting the potential of soaring birds as biologically embedded atmospheric sensors that could help closing the resolution gap in atmospheric models.

Non-invasive, high-throughput phenotyping tools are needed that can identify environmental effects on plant structure and function to diagnose factors responsible for reduced growth in commercial and non-commercial settings. In this study, we explored whether the integration of 3D-multispectral (3D) and 2D-hyperspectral imaging (HSI), aided by machine learning (ML), could be used to identify environmental stress treatments imposed during plant growth. Controlled environment-grown Nicotiana Benthamiana plants were subjected to a range of abiotic treatments - including different growth irradiances, heat treatment and drought stress - with the treatments resulting in differences in shoot height, biomass, leaf area and spectral reflectance. ML models were trained to identify these treatments using morphological and spectral traits measured at 27, 29, 31, and 34 days after sowing (DAS). A 3D-multispectral scanner was used to obtain information on plant height, biomass, and leaf area. A visible and near-infrared (VNIR) HSI camera provided detailed spectral information for deriving spectral indices including the Normalised Difference Vegetation Index (NDVI), Photochemical Reflectance Index (PRI) and Normalized Difference Red Edge (NDRE). Manual measurements provided baseline comparative data. The 3D-multispectral scanner reliably estimated above-ground traits, with high correlations between manual and scanner-derived measurements. The ML models accurately differentiated among environmental stress treatments, with the fused 3D+HSI model achieving the best overall predictive performance across all evaluated metrics compared with models based on either imaging modality alone. Results demonstrated the effectiveness of combining 3D-multispectral and 2D-HSI data with ML analyses for non-destructive, high-throughput phenotyping. The integration of these techniques enabled non-destructive, high-throughput identification of environmental stress treatments imposed during plant growth.

The morphology of cyanobacterial colonies plays a key role in harmful cyanobacterial blooms, with implications for their vertical migration, resistance against grazing, and light availability. In this study, we introduce the use of Optical Coherence Tomography (OCT) to investigate the three-dimensional morphology of cyanobacterial colonies. The technique enables non-invasive 3D imaging of colonies up to several millimeters in size, providing access to detailed mesoscale morphological features. Gas vesicles inside cells were shown to strongly improve image quality. We describe the sample preparation and image acquisition protocol, as well as an image processing pipeline that extracts mesoscale morphological features and provides a volumetric visualization of colonies. The method was tested for representative colonies of different cyanobacterial species while a dataset of volumetric images and measured mesoscale features was acquired for natural colonies of Microcystis. We demonstrate the utility of 3D imaging by quantifying the effects of irregular colony morphologies on their flotation velocity and the light availability within colonies. We anticipate OCT to become a key imaging technique to monitor populations of cyanobacterial colonies and investigate colony formation, with potential extensions to other colonial and aggregated organisms in freshwater and marine environments.

Coral reef monitoring needs scalable, non-invasive tools to complement resource-intensive traditional survey methods. Passive Acoustic Monitoring (PAM) offers a promising supplement, but its effectiveness is limited by the difficulty of attributing recorded sounds to species outside of previously well-characterized taxa. Using Omnidirectional Underwater Passive Acoustic Cameras (UPAC-360), we identified sounds from 31 reef fish species across 14 families on the Kona coast of Hawaii Island, including 13 not previously documented as soniferous. By releasing video and audio specimens, we have created the largest open-access collection of in-situ reef fish sounds to date for the Pacific. A subset of acoustically distinctive taxa--such as Hawaiian Dascyllus (Dascyllus albisella), Lei Triggerfish (Sufflamen bursa), soldierfishes (Myripristis spp.), wrasses, and herbivorous grazers--were identifiable in PAM recordings through manual acoustic and spectrogram review. Through identifying particular sounds linked to species with different ecological roles, these sounds have the potential to serve as indicators of reef function to increase the information and value coming from PAM surveys of Hawaiian and Pacific coral reefs.

1. Forest canopies harbour strong microclimatic gradients that shape plant performance, species interactions and ecosystem processes. Yet, despite renewed interest sparked by global change, forest canopies remain difficult-to-access experimental spaces. 2. With the goal to expand access to tree canopies as experimental arenas, we designed, built, and tested TreeTOP, a standardized experimental platform that opens canopy space for manipulative ecological experiments, specifically with potted plants. TreeTOP features lightweight aluminum frames placed in mature tree canopies non-invasively, allowing potted plants to be placed in three different heights, ground level, shade canopy, and sun canopy. 3. We implemented TreeTOP using two contrasting infrastructure concepts to demonstrate its applicability in both highly equipped canopy research facilities and forests without permanent canopy infrastructure. One installation relied on a canopy crane, grid power and fully automated irrigation, whereas the second was built by certified tree climbers and was equipped with an autonomous solar-powered, battery-operated irrigation system. At both sites, environmental sensor networks monitor the experiment. 4. TreeTOP successfully reproduced characteristic canopy microclimatic gradients, including increasing light availability, daytime air temperatures and thermal extremes with canopy height. Despite differing infrastructures, both implementations generated comparable microclimatic patterns, demonstrating that standardized canopy experiments are feasible in forests with or without permanent canopy access. By opening canopy space for manipulative experiments, TreeTOP provides a transferable framework for investigating plant performance, phenology, species interactions and microbiome assembly under realistic forest conditions.

Assessing breeding phenology using singing intensity can potentially provide insights into a species' responses to climate or habitat change. Although male passerines are known to sing elaborate songs during the breeding season, females of several tropical passerines also sing. The Nilgiri Flycatcher (Eumyias albicaudatus) is an endemic dimorphic songbird found in the southern Western Ghats of India. Despite its limited distribution and its ecological significance as an indicator species of cloud forests, very little is known about its vocal behaviour and breeding ecology. To examine sex-specific differences in vocalisation patterns, we analysed 294 male and 102 female focal songs from two pairs of birds. We then used data from a year-long automated recorder placed near a breeding pair to study the species' singing phenology using the BirdNET Analyser. The two sexes broadly produced similar songs, although males' songs included more notes at a faster pace. The sex-specific detectors failed due to this high similarity among songs of the two sexes, but we were able to automate species-level detection (F1-score = 0.795 at a confidence threshold of 0.1). We found strong seasonality in vocal activity, with peaks in singing during the breeding season, and identified a diurnal peak at dawn. More data are required to parameterise differences in annual vocal activity between the sexes. These detection patterns provide a framework for a landscape-wide examination of Nilgiri Flycatcher phenology and occurrence. This study suggests that automatic detectors can be a reliable tool for assessing singing phenology in birds and provide insights into breeding periods.

Citizen science observations are abundant, but conservation requires turning uneven records into reliable predictions and directing new surveys to where information is missing. We developed a biodiversity platform for Japan that is updated monthly and integrates 2.32 million records to predict 8,297 species across seven taxonomic groups. Shared representation models outperformed species-specific models in four groups and extended predictions to species with few records. Five independent datasets, including structured monitoring, environmental DNA and complete forest inventories, confirmed that the models ranked observed species and occupied sites above alternatives, with median AUCs of 0.724 to 0.894 across sites and 0.650 to 0.841 across species. For any user-selected area, the platform returns candidate species, distribution predictions, a biodiversity map corrected for uneven observation effort, a conservation priority map for native species and a map recommending where to survey next. This map highlights places where species with few records are predicted to occur despite limited sampling. Independent observations showed that areas ranked highly by this predicted potential contained many such species, indicating that model predictions can help direct surveys toward knowledge gaps. New observations are incorporated into monthly updates, creating a national feedback system connecting citizen science, local conservation decisions and future surveys.

The ocean tides influence key Earth system processes at a range of spatial and temporal scales. It is known that the geometry of ocean basins is the leading controller of tidal energetics, so well-constrained palaeogeographic reconstructions and tidal properties for Earths past are imperative when investigating other Earth system processes. Here, we present a novel way to constrain both deep-time tidal model results and reconstructions, by combining palaeoecology with sedimentology. We compare new palaeo-tidal model simulations for the Cambrian period, significant for the early origin and radiation of major animal fauna, to tidal proxies. One of the most abundant soft-bodied organisms preserved during this time are cnidarian medusae (jellyfish). A total of 17 cnidarian medusae localities were obtained through the literature, which had an adequate global distribution and occurred at regular intervals throughout the period of study. In some locations there were also estimates of palaeo-tidal range. Our results show a good agreement between the simulations and proxy data. In the few locations where there is disagreement, it is proposed that the palaeogeographic reconstructions are missing details, e.g., island chains, and our results allow for the palaeogeographic reconstructions to be improved. The proxy method presented is promising and can be applied to other time-periods with different marine fossils, particularly at evolutionary and extinction periods where the marginal marine environment is of importance.

The integration of digital technologies for high-throughput field phenotyping is critical for accelerating crop improvement in agriculture. However, extracting traits from remote sensing data remains constrained by fragmented workflows, manual intervention, and limited interoperability among existing tools, resulting in delays that hinder timely biological insight and decision-making. To address these challenges, we present PhenoStream (Phenotyping Streaming), a scalable, end-to-end cyberinfrastructure designed to automate the full lifecycle of aerial imagery-based phenotyping, from data acquisition to plot- and genotype-level inference. The framework integrates automated data ingestion from distributed field sites, geospatial processing, and AI-enabled trait extraction within a unified, user-accessible graphical interface. Its modular and extensible architecture supports adaptable trait modeling and seamless integration of new data sources, enabling deployment across diverse crops, environments, and experimental designs. We demonstrate the system across a large multi-location field trial network of bioenergy crops, where it enables high-throughput characterization of spatiotemporal growth dynamics, genotype-by-environment (GxE) interactions, and predictive modeling of key agronomic traits. By significantly reducing processing latency and manual effort, the platform facilitates near-real-time analysis and reproducible workflows. This work establishes a generalizable and scalable pathway for operationalizing very-high-spatial resolution aerial phenotyping in agricultural research. By bridging data acquisition and analytics, the end-to-end cyberinfrastructure provides a foundation for integrating heterogeneous and unstructured data streams--including remote sensing, environmental, and management data--toward data-driven decision making in agriculture.

Conserving migratory birds effectively requires full annual cycle strategies that identify where on-the-ground efforts can have the greatest impact. Here, we present a hemispheric spatial framework to identify priority areas for 112 migratory bird species across the Americas. Building on full annual cycle prioritizations, we defined finer-scale spatial planning units that reflect differences in migratory and congregational behaviors between shorebirds and landbirds. We compiled population data for each planning unit and focal species and applied conservation planning tools to design area-efficient portfolios of sites and landscapes that secure 10% of each species population within the Americas flyways. The resulting minimum area portfolios include 175 shorebird sites and 80 landbird landscapes optimized to meet the species-specific 10% representation targets across breeding, non-breeding, and passage seasons. We also identified a broader set of complementary solutions, ranked by an importance score, to provide decision-makers with flexible options for strategic resource allocation. This framework provides the scientific foundation for the Americas Flyways Initiative (AFI), which aims to catalyze investment in nature-based solutions and bird-friendly infrastructure to enhance the conservation of migratory birds and strengthen the resilience of the Americas flyways by 2050.

Wildlife monitoring in arid environments is complicated by substantial temporal and spatial variation in animal activity driven by unpredictable environmental conditions. For highly mobile species, movements across the landscape may further influence activity recorded at fixed monitoring locations, raising questions about whether short-term surveys adequately represent patterns of site use. We examined temporal variation in acoustic activity of the threatened Pilbara Leaf-nosed Bat (Rhinonicteris aurantia Pilbara form), a highly mobile, cave-roosting insectivorous bat inhabiting the arid Pilbara region of north-western Australia. Acoustic activity was monitored continuously at two permanent diurnal roosts for up to two years, comprising 1601 detector-nights across three detector locations. We tested relationships between nightly activity and environmental conditions, detector location and anthropogenic disturbance. Activity varied substantially through time and among detector locations. Moon illumination and cumulative rainfall were strongly associated with activity, but their interaction differed between roosts: activity was highest under low moon illumination and low cumulative rainfall at Chateau Cave, whereas activity at Daltons was highest under low moon illumination and higher cumulative rainfall. Activity also differed markedly between detector positions within Chateau Cave, and humidity, temperature and artificial light at night had additional location-specific effects. Other measures of anthropogenic disturbance, including cave entry, blasting and mining activity, had comparatively weak support. The contrasting environmental relationships among locations demonstrate that acoustic activity at permanent roosts is strongly context dependent. For a species capable of extensive movements among roosts, temporal variation in activity may reflect behavioural responses and redistribution of individuals across the landscape rather than rapid demographic change. We suggest that facultative nomadism may provide a useful hypothesis for understanding temporal variation in roost use by Pilbara Leaf-nosed Bat. More broadly, our results demonstrate that short-duration acoustic surveys of highly mobile species in arid environments may provide an incomplete representation of longer-term site use and should incorporate temporal replication and environmental context when used for conservation and impact assessment.

Snow algae are major biological drivers of snow darkening in polar and high-alpine environments. However, the direct contribution of algal pigmentation to snow reflectance has remained difficult to quantify because field observations cannot disentangle the effects of pigmentation from variation in biomass, species composition, and snow physical properties. Here, we characterized the optical effects of pigmentation using hyperspectral spectroradiometry to compare green, orange, and red cyst-like cells of a snow-derived Haematococcus isolate while controlling for developmental stage and cell abundance. Cysts became more red with increasing astaxanthin concentrations while chlorophyll-a concentrations remained relatively constant. Relative to green cysts, mean reflectance decreased by approximately 30% in orange cysts and 40% in red cysts. Integrated reflectance across the visible spectrum (350-800 nm) was negatively correlated with astaxanthin concentration. These results provide direct experimental evidence that algal pigmentation alone substantially reduces reflectance after controlling for cell abundance and developmental stage, and indicate that differences in snow physical properties may partly obscure this effect under natural field conditions. Our findings identify astaxanthin accumulation as an intrinsic driver of biological snow darkening and suggest that algal pigmentation, which may vary with species identity and physiological state, should be considered alongside biomass when predicting the radiative effects of snow algal blooms.

The ability to model the complex and uncertain population dynamics of endangered species has improved dramatically in recent decades. However, approaches to identify optimal decisions often require a simplified representation of population dynamics. This leads to a conundrum where managers may be unsure about the output of dynamic decision models because they rely on simplified assumptions of the underlying population dynamics. Here, by pairing integrated population models (IPM) that synthesize diverse ecological data with deep reinforcement learning (DRL) capable of optimizing decisions with high-dimensional uncertainty, we introduce a framework that delivers data-driven and ecologically detailed adaptive management strategies. We demonstrate its utility through application to the supplementation program for the endangered Rio Grande silvery minnow. Using our IPM-DRL framework, we developed an adaptive decision model that selects production and distribution decisions of the supplementation program in response to the observed demographic, hydrological, and genetic environment. The decision model outperformed all heuristic approaches in the simulation across management objectives that weighed persistence and effective population size-related genetic impact differently. For example, the currently deployed supplementation strategy performed 5.3% worse than the decision model under the persistence-focused objective scoring and 185% worse under the genetics-focused one. Analysis of the models decisions in relation to demographic and environmental covariates revealed that minimum sub-population size and total population size were primary drivers of the models decisions. The results demonstrate that the IPM-DRL framework offers a high-performing and interpretable decision-support tool for managing endangered species. SignificanceConservation problems, like imperiled species management, are often challenging because the system dynamics are complex and uncertain. We demonstrate how combining an integrated population model that infers key demographic processes from noisy ecological data with a deep reinforcement learning framework that optimizes management actions addresses these challenges by generating high-performing supplementation strategies for a conservation-dependent species. Our approach embeds two decades of monitoring data within a multi-objective decision-making environment that accounts for ecological uncertainty. The result is a generalizable framework that links ecological inference directly to actionable policy outcomes, enabling scientists and managers to move beyond describing system states and processes toward identifying optimal management actions.

Madagascar hosts exceptional biodiversity and endemism, yet its native bee fauna remains poorly characterised. We assembled and cleaned the first comprehensive occurrence dataset for Malagasy bees, reducing 10,721 raw records to 4,071 validated occurrences covering 218 georeferenced species of the 224 checklist species across six families (89.3% endemic). Despite near-complete checklist coverage, the dataset remains critically sparse for Madagascar's size, with most species documented by only a few records. Sampling was highly uneven across taxa: most genera were underrepresented while a few were disproportionately sampled due to ecological prevalence, detectability, and collector specialisation. Spatial concentration within limited grid cells amplified these biases. Temporally, effort varied markedly, with historical peaks driven by individual collectors and a post-2010 shift toward Apidae-dominated records. Spatially, 79.9% of 25x25 km grid cells intersecting Madagascar held no bee records, and 77.1% of records fell within 2.5 km of roads, mirroring global sampling patterns. Sampling clustered near major cities, with common species consistently found near roads and rare species spread across wider distance ranges. Among 231 Key Biodiversity Areas (KBAs), 68.4% were entirely unsampled; sampled KBAs held only 26.7% of all records, and sampling remained uneven even there, leaving substantial undetected diversity across most sites. These results reveal pervasive temporal, spatial, taxonomic, and collector-driven biases, underscoring the need for targeted surveys within KBAs and beyond roadsides to improve coverage of data-deficient species and strengthen conservation assessments.