Lichen Cladonia is the primary source of forage for caribou (Rangifer tarandus) during winter. Accurate, intelligent mapping of Cladonia lichens is essential for effective habitat management across northern Canada’s diverse ecozones. This study develops a robust deep learning framework for the semantic segmentation of lichen Cladonia from ultra-high-resolution UAV imagery (1 cm), addressing key challenges related to label scarcity and domain adaptation. A comprehensive annotated dataset was constructed from UAV orthomosaics in the Boreal Shield ecozone using Gaussian Mixture Model clustering refined by expert editing. Three U-Net-based architectures, Attention U-Net, Recurrent Residual U-Net (R2U-Net), and Recurrent Residual Attention U-Net (R2AU-Net), were trained and evaluated. The R2AU-Net model achieved the highest performance on independent Boreal Shield test sites (F1-score: 83.28%, IoU: 71.34%). To enhance model transferability to ecologically distinct regions, an active domain adaptation framework was implemented, combining entropy-based uncertainty sampling with fine-tuning on selected Taiga Shield imagery. Fine-tuning increased the model’s F1-score from 59.70% to 81.61% and reduced misclassification of grey lichen and spectrally similar substrates such as bare ground and rock, while saving over 90% of annotation costs. The results demonstrate the feasibility of scalable, adaptable lichen mapping in complex northern landscapes, providing a transferable methodology for supporting large-scale ecological monitoring and habitat management.
The application of image processing techniques for plant variety identification has grown substantially over the past two decades, driven by the increasing availability of digital plant imagery and global focus on biodiversity conservation. The present work aims to review recent studies and research in agriculture that utilize various machine/deep learning algorithms, image acquisition techniques, image databases, and accuracy assessment methods for plant variety classification. The PRISMA process was followed to systematically review 83 highly-relevant studies on image processing techniques for plant variety identification, published in the last decade (2015–2025). Deep learning algorithms, especially convolutional neural network variants, are predominantly employed and have demonstrated superior accuracy in plant variety identification. Integrating traditional machine learning techniques with optimization algorithms and handcrafted features further enhances model robustness, paving the way for scalable, real-world agricultural applications. Image data acquisition in most studies was conducted either through publicly available image databases or via smartphone-based imaging. Over the past decade, cereal crop varieties have been the most frequently studied in classification research. Overall, this global review offers valuable insights for researchers, practitioners, and policymakers to enhance the efficiency and accuracy of data-driven plant variety classification.
Satellite data from Landsat and Sentinel-2 are widely used to model forest structural attributes at the pixel level, supporting applications in forest management, habitat assessment, carbon cycle analysis, and climate impact mitigation. However, many management and modeling frameworks rely on stand-level forest inventories derived from ocular interpretation of aerial imagery. To address this information need, a Satellite-Based Forest Inventory (SBFI) was developed for Canada’s forested ecosystems (∼650 Mha), providing spatially exhaustive and nationally consistent stand-level attributes derived from a multi-decadal series of satellite data. A key SBFI attribute is total treed aboveground biomass (AGB). We present a new dataset that further partitions this total SBFI AGB into species-specific biomass components for stemwood, bark, branches, and foliage, and also includes estimates of AGB for non-treed vegetation pools. This level of detail supports applications requiring biomass accounting beyond total AGB, including carbon reporting, productivity assessment, wildfire fuel characterization, and modeling approaches that depend on biomass components. We outline the development, modeling framework, and structure of this expanded biomass component database. This remote-sensing-informed, stand-level product delivers national coverage of species-specific component AGB across more than 25 million forest stands and the dataset is openly available to support analyses from local to national scales.
This study compares terrain-corrected polarimetric variables derived from L-band ALOS-2 PALSAR-2 acquired over a forested and mountainous region of southeast British Columbia (BC) during a drier and a wetter date. Backscatter, polarimetric decomposition parameters, and polarimetric discriminators were compared across 5 broad fuel types (forest, sparse, unforested, incomplete burn, complete burn) to examine if values differ between fuel moisture conditions as characterized by the value of the Drought Code (DC). Polarimetric variables suitable for use in fuel moisture modeling were identified using a one-way analysis of variance (ANOVA) test of significance applied to the normalized differences of these variables between 2 images. Values of linearly polarized backscatter, circularly polarized backscatter, and the parameters of the Van Zyl decomposition had relatively large normalized differences between 2 distinct DC values. The values of parameters of the Cloude-Pottier and Neumann decompositions did not differ between distinct DC values. Relatively large normalized differences occurred with polarimetric discriminators including the total power, maximum and minimum of scattering intensity, the maximum of power received, the maximum and minimum of the completely polarized component, and the maximum of the completely unpolarized component. Results differed by fuel type. Cette & eacute;tude compare les variables polarim & eacute;triques corrig & eacute;es du relief, d & eacute;riv & eacute;es des donn & eacute;es en bande L des capteurs ALOS-2 PALSAR-2 acquises au-dessus d'une r & eacute;gion foresti & egrave;re et montagneuse du sud-est de la Colombie-Britannique au Canada lors d'une p & eacute;riode s & egrave;che et d'une p & eacute;riode humide. La r & eacute;trodiffusion, les param & egrave;tres de d & eacute;composition polarim & eacute;trique et les discriminateurs polarim & eacute;triques ont & eacute;t & eacute; compar & eacute;s pour cinq grands types de combustibles (for & ecirc;t, clairsem & eacute;, non bois & eacute;, br & ucirc;lis incomplet, br & ucirc;lis complet) afin d'examiner si les valeurs diff & egrave;rent selon l'humidit & eacute; du combustible, caract & eacute;ris & eacute;e par la valeur de l'indice de s & eacute;cheresse (IS). Les variables polarim & eacute;triques pertinentes pour la mod & eacute;lisation de l'humidit & eacute; du combustible ont & eacute;t & eacute; identifi & eacute;es par une analyse de variance (ANOVA) & agrave; un facteur appliqu & eacute; aux diff & eacute;rences normalis & eacute;es de ces variables entre les deux images. Les valeurs de la r & eacute;trodiffusion polaris & eacute;e lin & eacute;airement, de la r & eacute;trodiffusion polaris & eacute;e circulairement et les param & egrave;tres de la d & eacute;composition de Van Zyl pr & eacute;sentaient des diff & eacute;rences normalis & eacute;es relativement importantes entre les deux niveaux distincts d'IS. Les valeurs des param & egrave;tres des d & eacute;compositions de Cloude-Pottier et de Neumann ne diff & eacute;raient pas entre les valeurs d'IS. Des diff & eacute;rences normalis & eacute;es relativement importantes ont & eacute;t & eacute; observ & eacute;es avec les discriminateurs polarim & eacute;triques, notamment la puissance totale, les valeurs maximales et minimales de l'intensit & eacute; de diffusion, la puissance maximale re & ccedil;ue, le maximum et minimum de la composante totalement polaris & eacute;e et le maximum de la composante totalement non polaris & eacute;e. Les r & eacute;sultats variaient selon le type de combustible.
Smartphone-based photogrammetry offers a low-cost approach to indoor reconstruction but typically requires extensive ground control point (GCP) networks. This study provides a controlled proof-of-concept that commercial ultrasonic real-time location system (RTLS) constraints can reduce ground control requirements while meeting the U.S. Institute of Building Documentation Level of Accuracy 20 (LOA20) specification of 2-5 cm for architectural documentation. A ZeroKey Quantum RTLS supplied camera poses for two experiments evaluated against terrestrial laser scanner references. In the sparse experiment, the recommended configuration achieved 2.6 mm inter-target distance root mean square error (RMSE) under position-only constraints-comparable to the 2.9 mm GCP baseline-and 11.5 mm under combined position and orientation constraints, both within LOA20. Other position-only configurations produced a linear scale error; combined constraints additionally introduced higher-order error components consistent with an uncalibrated boresight matrix. Where scale error was present, it was correctable through minimal ground control: a scale bar for reconstructions in the RTLS local frame, or a minimum of three GCPs for external alignment. The dense reconstruction illustrated this: GCP-based scale correction during registration reduced the check point RMSE from 32.1 mm to 20.5 mm, within the LOA20 specification, against a 10.1 mm GCP baseline. La photogramm & eacute;trie & agrave; l'aide de smartphone offre une solution & eacute;conomique pour la reconstruction de surfaces int & eacute;rieures, mais n & eacute;cessite g & eacute;n & eacute;ralement un vaste r & eacute;seau de points de contr & ocirc;le au sol (GCP: Ground Control Point). Cette & eacute;tude d & eacute;montre que l'application d'une m & eacute;thode contr & ocirc;l & eacute;e peut r & eacute;duire les besoins en GCP au sol associ & eacute; aux contraintes d'un syst & egrave;me de localization en temps r & eacute;el (RTLS) ultrasonique commercial, tout en respectant le niveau de pr & eacute;cision 20 (LOA20) de 2 & agrave; 5 cm sp & eacute;cifi & eacute; par l'Institut am & eacute;ricain de documentation du b & acirc;timent pour les documents en architecture. Deux exp & eacute;riences ont & eacute;t & eacute; r & eacute;alis & eacute;es avec la cam & eacute;ra RTLS ZeroKey Quantum pour comparer les r & eacute;sultats & agrave; des r & eacute;f & eacute;rences issues de scans de laser terrestre. Dans l'exp & eacute;rience avec un r & eacute;seau clairsem & eacute;, la configuration recommand & eacute;e a permis d'atteindre une erreur quadratique moyenne (RMSE) de 2,6 mm entre les cibles, sous contraintes de position uniquement (comparable & agrave; la valeur de r & eacute;f & eacute;rence de 2,9 mm obtenue avec les GCP), et de 11,5 mm sous contraintes combin & eacute;es de position et d'orientation, les deux valeurs restant dans les limites du LOA20. D'autres configurations, bas & eacute;es uniquement sur la position, ont produit une erreur d'ordre lin & eacute;aire; les contraintes combin & eacute;es ont en outre introduit des composantes d'erreur d'ordre sup & eacute;rieur, compatibles avec une matrice de vis & eacute;e non calibr & eacute;e. Les erreurs d'ordre lin & eacute;aire sont corrigibles par un minimum de GCP au sol comme une & eacute;chelle de r & eacute;f & eacute;rence pour les reconstructions dans le rep & egrave;re local du syst & egrave;me de localization temps r & eacute;el (RTLS), ou au moins trois GCP pour l'alignement externe. La reconstruction dense l'a illustr & eacute;: la correction d'& eacute;chelle bas & eacute;e sur les GCP lors de l'enregistrement a r & eacute;duit le RMSE des points de contr & ocirc;le de 32,1 mm & agrave; 20,5 mm, conform & eacute;ment & agrave; la sp & eacute;cification LOA20, par rapport & agrave; une valeur de r & eacute;f & eacute;rence de 10,1 mm pour les GCP.
The interpretation of increasing vegetation, also called greening, using the normalized difference vegetation index (NDVI) remains challenging in sparsely vegetated regions. Vegetation changes in higher latitude regions have mostly been assessed through circumpolar studies, which place its magnitude in the context of more productive, lower-latitude ecosystems. However, absolute greening trends (Delta NDVImax) may underrepresent ecologically meaningful variation where plant cover is low. Here, we used four decades (1984-2023) of Landsat-based NDVImax time series to assess vegetation change across the Canadian Arctic Archipelago combining the commonly used Absolute Greening Index (AGI) and the Relative Greening Index (RGI; percent change relative to initial NDVImax based on 1984 data). Both metrics indicated widespread but low-magnitude greening, with stronger increases at higher latitudes. However, RGI revealed clearer differences among bioclimatic zones and emphasized vegetation change in sparsely vegetated northern regions, while AGI suggested more homogeneous trends. Specific examples showed that RGI highlighted subtle NDVI increases along coastal and patchy vegetation areas, whereas AGI provided a more conservative depiction of change. Our results suggest that using both metrics provide complementary insights into vegetation dynamics, improving interpretation of greening patterns across the Canadian Arctic Archipelago.
Understanding landscape change and its impact on ecosystem services such as biodiversity and carbon sequestration is essential for developing effective management strategies that balance conservation and economic objectives. Reliable information on landscape change is critical for building this understanding and assessing impacts at the landscape scale. To this end, we applied a modified version of a previously developed change detection method to generate annual land cover change data across the Canadian Prairies from 1984 to 2022. The method captures from-to transitions among ten land cover classes, encompassing 90 possible change classes (excluding no-change). We assessed the accuracy of these change classes and produced summaries to identify various spatial and temporal patterns of change. Many of the change classes achieved F1 scores above 70%, indicating good model performance. Sources of error included mixed pixels, small landscape features, gradual transitions, and confusion between similar classes. The timing of detected changes was generally within +/- 1 year; however, precise reference timing is often uncertain due to gradual or complex transitions and interpretation challenges with Landsat. Change summaries reveal key trends, examples included net forest change, grassland loss, expansion of built-up areas, and conversion of wetlands and water bodies to cropland.
Carbon dioxide (CO2), the major anthropogenic greenhouse gas, plays a critical role in climate change. Using Atmospheric Infrared Sounder (AIRS) CO2 retrievals and auxiliary datasets, this study investigated the spatiotemporal characteristics of mid-tropospheric CO2 and its driving factors over Central Asia during 2003-2011. Pixel-based linear trend analysis, empirical orthogonal function (EOF), singular value decomposition (SVD), and correlation analysis were applied. The results show that mid-tropospheric CO2 exhibited a distinct low-high-low spatial pattern from north to south across Central Asia. Low-CO2 centers were mainly associated with vegetation and terrain, whereas high-CO2 centers were related to land surface type, atmospheric circulation, and topography. Mid-tropospheric CO2 increased at an average rate of 2.1 ppm yr(-1) during the study period. Pronounced seasonal fluctuations were also observed, primarily driven by seasonal variations in carbon exchange between terrestrial ecosystems and the atmosphere. In addition, the response of mid-tropospheric CO2 to seasonal carbon exchange showed an approximately four-month lag. These findings provide a regional perspective on the distribution, trend, and seasonal dynamics of mid-tropospheric CO2 over Central Asia and improve understanding of CO2 variability in arid inland regions influenced by long-range atmospheric transport.
Peatland degradation in tropical regions threatens carbon storage and land stability. This study monitored peatland subsidence using the Small Baseline Subset (SBAS) InSAR technique with Sentinel-1 C-band SAR data (2017-2022) in Sub-Peat Hydrological Zone 1. Time-lagged correlation analysis evaluated relationships between vertical displacement and environmental variables (groundwater level, rainfall, and Enhanced Vegetation Index). Vertical displacement rates ranged from -12.6 to 3.6 cm/year, where negative values indicate subsidence while positive values indicate relative surface uplift. The most significant subsidence was concentrated around drainage canals and infrastructure. A Pearson correlation revealed a moderate negative relationship between groundwater levels and subsidence (r = -0.44, p < 0.001), while rainfall and vegetation showed negligible correlations. Cross-correlation analysis revealed that groundwater significantly influences subsidence dynamics, with a persistent negative correlation peaking at r = -0.6 with a 10-month lag. In contrast, rainfall showed delayed indirect effects (r = -0.33 at lag 4), mediated by groundwater recharge, while vegetation showed consistently negligible correlation. These findings confirm groundwater level as the primary hydrological control on subsidence dynamics. Nevertheless, subsidence dynamics were not fully explained by the examined variables, suggesting that additional factors such as peat thickness heterogeneity, fire history, and land-use intensity warrant investigation in future studies.
Rice production in the Vietnamese Mekong Delta (VMD) faces rising climate and ecological challenges, prompting national policies to reduce cropping intensity and restore floodplain functions. Synthetic Aperture Radar (SAR) is an effective tool for continuous rice production monitoring, as it is less affected by cloud cover. As such, it can capture complete phenological cycles of rice growth and enable differentiation of rice typologies. This research developed an object-based classification framework using SAR backscatter time series data and Dynamic Time Warping (DTW) to map the Triple Rice and Double Rice production typologies. Landscape segmentation was implemented to delineate homogeneous units for object-based analysis. Sentinel-1 backscatter time series were generated for each unit to capture annual temporal dynamics. Using t-SNE and HDBSCAN clustering, the predominant temporal patterns within the VMD were identified. By associating these with field survey labels, reference temporal profiles for Double and Triple Rice were defined. Finally, landscape units were assigned to rice typologies with the most similar temporal profiles, and the similarity was measured using DTW. This framework was implemented for two years (2019 and 2022). Its accuracy exceeded 83% for 2022. A comparison between 2019 and 2022 indicates a clear transition toward lower cropping intensity. La production rizicole dans le delta du M & eacute;kong vietnamien (DMV) est confront & eacute;e & agrave; des d & eacute;fis climatiques et & eacute;cologiques croissants, ce qui a incit & eacute; les autorit & eacute;s nationales & agrave; mettre en oe uvre des politiques visant & agrave; r & eacute;duire l'intensit & eacute; des cultures et & agrave; restaurer les fonctions des plaines inondables. Le radar & agrave; synth & egrave;se d'ouverture (RSO) est un outil efficace pour le suivi continu de la production rizicole, car il est moins sensible & agrave; la couverture nuageuse. Il permet ainsi de saisir l'int & eacute;gralit & eacute; des cycles ph & eacute;nologiques de la croissance du riz et de diff & eacute;rencier les typologies de riz. Notre recherche a permis de d & eacute;velopper un cadre de classification orient & eacute;-objet utilisant des donn & eacute;es de s & eacute;ries temporelles de r & eacute;trodiffusion RSO du capteur Sentinel-1 et la technique de d & eacute;formation temporelle dynamique (DTD) afin de cartographier les typologies de production de riz triple et de riz double.Une segmentation du paysage a & eacute;t & eacute; appliqu & eacute;e afin de d & eacute;limiter des unit & eacute;s homog & egrave;nes pour une analyse orient & eacute;e-objet. Des s & eacute;ries temporelles de donn & eacute;es de r & eacute;trodiffusion Sentinel-1 ont & eacute;t & eacute; g & eacute;n & eacute;r & eacute;es pour chaque unit & eacute; afin d'analyser les dynamiques temporelles annuelles. Gr & acirc;ce aux algorithmes de clustering t-SNE et HDBSCAN, les principaux sch & eacute;mas temporels au sein du DMV ont & eacute;t & eacute; identifi & eacute;s. En les associant aux & eacute;tiquettes des relev & eacute;s de terrain, des profils temporels de r & eacute;f & eacute;rence pour le riz double et le riz triple ont & eacute;t & eacute; d & eacute;finis. Enfin, les unit & eacute;s paysag & egrave;res ont & eacute;t & eacute; rattach & eacute;es aux typologies de riz pr & eacute;sentant les profils temporels les plus similaires, et la similarit & eacute; a & eacute;t & eacute; mesur & eacute;e & agrave; l'aide de la DTD. Cette analyse a & eacute;t & eacute; appliqu & eacute;e sur deux ann & eacute;es (2019 et 2022). Sa pr & eacute;cision a d & eacute;pass & eacute; 83 % en 2022. Une comparaison entre 2019 et 2022 indique une nette transition vers une intensit & eacute; de culture plus faible.
Rising Arctic temperatures are making polygonal tundra increasingly vulnerable, primarily due to high ground ice contents. These landscapes form through soil-hydrology interactions, leading to ice wedge formation and degradation. Understanding the future of ice wedge polygon (IWP) landscapes requires detailed land cover classifications, as soil properties vary significantly across IWP sub-features like rims and centers. Existing classifications often distinguish between high-centered (HCP) and low-centered (LCP) polygons but fail to capture finer sub-feature distributions. This study provides high-resolution land cover datasets for two IWP sites on the Canadian Beaufort coast using WorldView-3 imagery. Ptarmigan Bay features well-defined landforms, while Komakuk Beach exhibits greater permafrost degradation. We compare two land cover-mapping approaches: object-based image analysis (OBIA) with segmentation and random forest classification, and a deep learning U-net model. Results show that the OBIA-random forest method performed better, and substantial differences in the landform type distribution between study areas and methods exist. Both methods identify 60% of HCP centers at Ptarmigan Bay and 50% at Komakuk Beach, but mapped IWP sub-feature proportions (HCP troughs, LCP centers, LCP rims) vary across areas and methods, reflecting classification uncertainties. Furthermore, the transferability of models between regions is constrained when there are pronounced differences in degradation of landforms. La hausse des temp & eacute;ratures arctiques rend la toundra polygonale graduellement plus vuln & eacute;rable, principalement en raison de la forte teneur en glace & agrave; la surface du sol. Ces paysages se forment par le biais d'interactions entre le sol et l'hydrologie, entra & icirc;nant la formation et la d & eacute;gradation de coins de glace. Comprendre l'& eacute;volution des paysages de polygones & agrave; coins de glace (PCG) n & eacute;cessite des classifications d & eacute;taill & eacute;es de l'occupation du sol, car les propri & eacute;t & eacute;s du sol varient consid & eacute;rablement selon les sousstructures des PCG, comme les bordures et les centres. Les classifications existantes font souvent la distinction entre les polygones & agrave; centre & eacute;lev & eacute; (PCE) et les polygones & agrave; centre bas (PCB), mais ne parviennent pas & agrave; saisir les distributions plus fines des sous-structures. Cette & eacute;tude utilise des ensembles de donn & eacute;es d'occupation du sol fine r & eacute;solution spatiale pour deux sites de PCG sur la c & ocirc;te canadienne de la mer de Beaufort & agrave; l'aide d'images WorldView-3. La baie de Ptarmigan pr & eacute;sente des formes de relief bien d & eacute;finies, tandis que la plage de Komakuk pr & eacute;sente une plus grande d & eacute;gradation du perg & eacute;lisol. Nous comparons deux approches d'analyse d'images pour cartographier l'occupation du sol : (1) l'approche orient & eacute;e objet (OBIA : Object-based Image Analysis) avec segmentation et classification par la technique statistique random forest, et (2) l'approche par un mod & egrave;le U-Net d'apprentissage profond. Les r & eacute;sultats montrent que la m & eacute;thode OBIA-random forest est plus performante, alors que des diff & eacute;rences significatives existent dans la distribution des types de relief entre les zones d'& eacute;tude et les m & eacute;thodes. Les deux m & eacute;thodes identifient 60 % des centres PCE & agrave; Ptarmigan Bay et 50 % & agrave; Komakuk Beach, mais les proportions des sous-& eacute;l & eacute;ments PCG cartographi & eacute;s (creux PCE, centres PCB, bordures PCB) varient selon les zones et les m & eacute;thodes, refl & eacute;tant les incertitudes de classification. De plus, la transf & eacute;rabilit & eacute; des mod & egrave;les entre les r & eacute;gions est limit & eacute;e en cas de diff & eacute;rences marqu & eacute;es dans la d & eacute;gradation des reliefs.
Crested wheatgrass (Agropyron cristatum) is an introduced invasive grass species in North American grasslands. It was initially seeded to increase the grazing duration but is now threatening native grassland biodiversity. In this study we investigated which biophysical and spectral properties (hyperspectral and simulated multispectral) distinguish crested wheatgrass from native grasses. We further assessed whether spectral indices, linked to biophysical traits, could distinguish crested wheatgrass from native grasses. We collected field data of hyperspectral reflectance, leaf area index (LAI), biomass, and vegetation cover. Crested wheatgrass sites have higher grass cover, height, LAI, and biomass (total, live, and dead), but lower bare ground than native grasses. Hyperspectral data showed lower visible and SWIR reflectance for crested wheatgrass than for native grasses. SWIR spectral indices for hyperspectral data, SWIR and visible spectral indices for simulated Sentinel-2A data and visible spectral indices for PlanetScope SuperDove data discriminated crested wheatgrass from native grasses. These findings will advance invasive grass monitoring by shifting the focus from phenology to biophysical properties-based detection, supporting management of crested wheatgrass and restoration of native grassland ecosystems. L'agropyre & agrave; cr & ecirc;te (Agropyron cristatum) est une esp & egrave;ce de gramin & eacute;e envahissante introduite dans les prairies nord-am & eacute;ricaines. Initialement sem & eacute;e pour prolonger la dur & eacute;e du p & acirc;turage, elle menace d & eacute;sormais la biodiversit & eacute; des prairies indig & egrave;nes. Cette & eacute;tude examine les propri & eacute;t & eacute;s biophysiques et spectrales (hyperspectrales et multispectrales simul & eacute;es) qui distinguent l'agropyre & agrave; cr & ecirc;te des gramin & eacute;es indig & egrave;nes. Elle & eacute;value & eacute;galement si des indices spectraux, li & eacute;s & agrave; des caract & eacute;ristiques biophysiques, permettent de diff & eacute;rencier l'agropyre & agrave; cr & ecirc;te des gramin & eacute;es indig & egrave;nes. Des donn & eacute;es de terrain portant sur la r & eacute;flectance hyperspectrale, l'indice de surface foliaire (ISF), la biomasse et le couvert v & eacute;g & eacute;tal ont & eacute;t & eacute; collect & eacute;es. Les sites infest & eacute;s d'agropyre & agrave; cr & ecirc;te pr & eacute;sentent un couvert herbac & eacute;, une hauteur, un ISF et une biomasse (totale, vivante et morte) plus & eacute;lev & eacute;s, mais une surface de sol nu plus faible que les sites domin & eacute;s par les gramin & eacute;es indig & egrave;nes. Les donn & eacute;es hyperspectrales r & eacute;v & egrave;lent une r & eacute;flectance visible et infrarouge & agrave; ondes courtes (SWIR : ShortWave InfraRed) plus faible pour l'agropyre & agrave; cr & ecirc;te que pour les gramin & eacute;es indig & egrave;nes. Les indices spectraux SWIR des donn & eacute;es hyperspectrales, les indices spectraux SWIR et visibles des donn & eacute;es simul & eacute;es Sentinel-2A et les indices spectraux visibles des donn & eacute;es PlanetScope SuperDove ont permis de distinguer l'agropyre & agrave; cr & ecirc;te des gramin & eacute;es indig & egrave;nes. Ces r & eacute;sultats contribueront & agrave; am & eacute;liorer la surveillance des gramin & eacute;es envahissantes en d & eacute;pla & ccedil;ant l'attention de la ph & eacute;nologie vers une d & eacute;tection bas & eacute;e sur les propri & eacute;t & eacute;s biophysiques, ce qui facilitera la gestion de l'agropyre & agrave; cr & ecirc;te et la restauration des & eacute;cosyst & egrave;mes de prairies indig & egrave;nes.
The combination of unoccupied aerial vehicles and deep learning offers a promising approach for mapping invasive alien plant species (IAPS), though its effectiveness in early detection case remains uncertain. In this study, we evaluated the suitability of this approach based on a convolutional neural network for mapping the location of common reed (Phragmites australis subsp. australis) within Parc national des & Icirc;les-de-Boucherville located in southern Qu & eacute;bec, Canada. We collected data on six distinct dates (July-October 2022) during the growing season, covering environments with different levels of reed invasion (Dense, Establishing and Post-treatment). Overall, model performance was high for the different dates and zones, especially for recall (mean of 0.89). The results showed an increase in performance, reaching a peak following the appearance of the inflorescence in September (highest F1-score at 0.98). Despite challenges associated with common reed mapping in a post-treatment monitoring context with poorer detection, this approach has the potential to serve as an effective tool for speeding up the work of biologists in the field and ensuring better management of IAPS. To this end, we provide a comprehensive dataset of high-resolution imagery and deep learning models enabling the detection of common reed across its phenological cycle. Combiner les drones et l'apprentissage profond offre une approche prometteuse pour cartographier les esp & egrave;ces v & eacute;g & eacute;tales exotiques envahissantes (EVEE), bien que leur efficacit & eacute; pour d & eacute;tecter les repousses & agrave; un stade pr & eacute;coce reste & agrave; d & eacute;montrer. Dans cette & eacute;tude, nous avons & eacute;valu & eacute; la pertinence de cette approche bas & eacute;e sur l'utilisation d'un r & eacute;seau neuronal convolutif pour cartographier l'emplacement du roseau commun (Phragmites australis subsp. australis) & agrave; l'int & eacute;rieur du Parc national des & Icirc;les-de-Boucherville situ & eacute; dans le sud du Qu & eacute;bec, Canada. Nous avons collect & eacute; des donn & eacute;es & agrave; six dates distinctes (juillet-octobre 2022) durant la saison de croissance, couvrant des environnements pr & eacute;sentant diff & eacute;rents niveaux d'envahissement par le roseau (dense, en voie d'& eacute;tablissement et post-traitement). De fa & ccedil;on g & eacute;n & eacute;rale, la performance du mod & egrave;le & eacute;tait & eacute;lev & eacute;e pour les diff & eacute;rentes dates et zones, surtout au niveau du rappel (moyenne globale de 0.89). Les r & eacute;sultats ont montr & eacute; une augmentation de la performance pour atteindre un sommet & agrave; la suite de l'apparition de l'inflorescence en septembre (F1-score le plus haut & agrave; 0.98). Malgr & eacute; des d & eacute;fis associ & eacute;s & agrave; la cartographie du roseau commun dans un contexte de gestion post-traitement avec un taux de d & eacute;tection plus faible, cette approche pourrait & ecirc;tre un outil efficace pour acc & eacute;l & eacute;rer le travail des biologistes sur le terrain et assurer une meilleure gestion des EVEE. & Agrave; cette fin, nous fournissons un jeu de donn & eacute;es complet comprenant des images haute r & eacute;solution et des mod & egrave;les d'apprentissage profond permettant la d & eacute;tection du roseau commun tout au long de son cycle ph & eacute;nologique.
Characterizing the hydrological effects of forest disturbance and recovery is complex. Process-based hydrological models can address this challenge, but rely on accurate parameterization, which is often limited by data availability. Leaf area index (LAI) is a key parameter in many process-based models however, few datasets provide sufficient spatiotemporal resolution to represent forest recovery. This study addresses this limitation using two remote sensing datasets: a monthly 30 m satellite-derived LAI time series and an airborne laser scanning (ALS)-derived LAI dataset. These data were combined to characterize forest recovery across the catchment and within ecozones, enabling the development of new LAI model parameterizations. Daily streamflow simulations using these parameterizations were compared with a baseline. ALS-derived LAI differed by up to 40% from the satellite product, which underestimated LAI in dense forests. In five years of post-disturbance, the LAI reached one third of pre-disturbance values. After 10 years, LAI recovery ranged from 45% to 73% across ecozones. The data-driven LAI parameterizations resulted in notable differences in daily and mean annual flow, as well as 2-year and 20-year peak flows. This work shows that remotely sensed datasets can constrain LAI parameters for improved hydrological modeling of forest recovery. La caract & eacute;risation de l'hydrologie & agrave; partir des perturbations et de la r & eacute;g & eacute;n & eacute;ration foresti & egrave;res est complexe. Les mod & egrave;les hydrologiques bas & eacute;s sur les processus permettent de relever ce d & eacute;fi, mais ils reposent sur une param & eacute;trisation pr & eacute;cise, souvent limit & eacute;e par la disponibilit & eacute; des donn & eacute;es. L'indice de surface foliaire (ISF) est un param & egrave;tre cl & eacute; dans de nombreux mod & egrave;les bas & eacute;s sur les processus, ce qui peut compenser pour le peu de jeux de donn & eacute;es qui offrent une r & eacute;solution spatio-temporelle suffisante pour repr & eacute;senter la r & eacute;g & eacute;n & eacute;ration foresti & egrave;re. Notre & eacute;tude s'est pench & eacute;e sur cette limitation en utilisant deux jeux de donn & eacute;es de t & eacute;l & eacute;d & eacute;tection : une s & eacute;rie temporelle mensuelle d'ISF d & eacute;riv & eacute;e de donn & eacute;es satellitaires & agrave; 30 m et un jeu de donn & eacute;es d'ISF d & eacute;riv & eacute; de donn & eacute;es lidar & agrave; partir d'un capteur laser a & eacute;roport & eacute; (ALS). Ces donn & eacute;es ont & eacute;t & eacute; combin & eacute;es pour caract & eacute;riser la r & eacute;g & eacute;n & eacute;ration foresti & egrave;re & agrave; l'& eacute;chelle du bassin versant et au sein des & eacute;cozones, permettant ainsi d'introduire des valeurs d'ISF comme param & egrave;tre au mod & egrave;le hydrologique. Les simulations de d & eacute;bit journalier utilisant cette nouvelle param & eacute;trisation ont & eacute;t & eacute; compar & eacute;es & agrave; une valeur de r & eacute;f & eacute;rence. L'ISF d & eacute;riv & eacute; de l'ALS diff & eacute;rait jusqu'& agrave; 40% du produit satellitaire, qui sous-estimait l'ISF des for & ecirc;ts denses. Cinq ans apr & egrave;s des perturbations, l'ISF atteignait un tiers de ses valeurs initiales. Apr & egrave;s 10 ans, la r & eacute;cup & eacute;ration de l'ISF variait de 45% & agrave; 73% selon les & eacute;cozones. Les param & eacute;trisations bas & eacute;es sur ces donn & eacute;es d'IFS ont r & eacute;v & eacute;l & eacute; des diff & eacute;rences notables dans les d & eacute;bits journaliers et annuels moyens, ainsi que dans les d & eacute;bits de pointe aux 2 ans et aux 20 ans. Notre & eacute;tude d & eacute;montre que les donn & eacute;es de t & eacute;l & eacute;d & eacute;tection permettent de contraindre les param & egrave;tres de l'ISF pour une mod & eacute;lisation hydrologique am & eacute;lior & eacute;e de la r & eacute;g & eacute;n & eacute;ration foresti & egrave;re.
Evaporative Stress Index (ESI) is a commonly used indicator for assessing ecosystem-level water stress, particularly for monitoring agricultural and forest droughts. This study assesses performance of two remote sensing ESI products: ECO4ESIPTJPL and ECO4ESIALEXI from the ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS), against in-situ data from 59 AmeriFlux sites across the continental United States. Results indicate that ECOSTRESS ESI, overall, underestimates site-based estimates, suggesting a tendency to indicate higher evaporative stress. When aggregated across all sites and time steps, ECO4ESIPTJPL underestimates ESI 55.1% of the time, while ECO4ESIALEXI underestimates 62.42%. However, under high evaporative stress conditions, both the products, particularly ECO4ESIPTJPL, overestimate ESI, meaning they indicate the stress is lower than it actually is. Further analysis reveals that ECOSTRESS-derived evapotranspiration (ET), rather than potential ET, is the primary source of bias in the ECOSTRESS ESI products. These findings emphasize the need for refining ECOSTRESS ESI products, particularly improving ECOSTRESS ET, to enhance accuracy of assessing evaporative stress across diverse landscapes and climatic conditions.
The aim of this paper is to employ satellite interferometry to analyze the spatial variability of mining-induced surface subsidence decay time after the end of underground mineral extraction. The tests were performed in the area of one of the Legnica-Glogow Copper Belt mines, in which copper ore is extracted at significant depths. The research was based on a modified methodology for determining the time coefficient c of the Knothe function based on SAR data, adapted to the Small Baseline Subset time series method. The results, validated against classic geodetic measurements, indicate that surface deformations can become visible as late as approximately eight years after the end of mining operations, with the surface deformation decay time being inversely proportional to the value of surface subsidence measured during the operations. Statistical tests showed that the distribution of the time coefficient c of the Knothe function corresponds to the Gaussian distribution, and that it can therefore be applied in predicting the decay time of mining deformations. The results may have significant implications for land-use planning in post-mining areas and may enable a precise determination of the deformation decay time across the entire subsidence trough. The research results indicate that surface change dynamics can be analyzed more precisely with the use of InSAR methods combined with the Knothe time function than with the use of traditional geodetic methods.
Landfast ice polynyas are important features near many northern coastal communities, and their automated detection from synthetic aperture radar (SAR) imagery is positioned to support on-ice travel safety under changing Arctic sea ice conditions. This research leveraged original datasets of over 5,000 Sentinel-1 SAR observations of wintertime polynyas mapped near the Canadian communities of Sanikiluaq and Nain to investigate deep learning-based landfast ice polynya detection. The Faster-RCNN object detection network was optimized for polynya detection through several modifications to network design elements and training strategies. Resulting detection models generalized well between regions and accurately detected polynyas with local backscatter contrasts above 5 dB, e.g. achieving 90% target recall at 24% precision. Polynyas smaller than 500 meters and with local backscatter contrasts less than 3 dB, constituting approximately half of all observations, were frequently missed. Precision scores below 30% were consistently incurred in attempts to optimize recall. Results highlight challenges to the consistent performance of single-image detectors due to variable and frequently weak polynya signatures in dual-polarized Sentinel-1 SAR backscatter. Future investigations into multi-temporal, multi-frequency, and/or higher-resolution SAR imagery could further support the delivery of robust hazard detection systems relevant to community sea ice safety and monitoring. Les polynies de glace c & ocirc;ti & egrave;re constituent un & eacute;l & eacute;ment important de la formation g & eacute;ographique & agrave; proximit & eacute; de nombreuses communaut & eacute;s c & ocirc;ti & egrave;res dans l'Arctique. Leur d & eacute;tection automatis & eacute;e & agrave; partir d'images radar & agrave; synth & egrave;se d'ouverture (SAR) soutiendrait la s & eacute;curit & eacute; des d & eacute;placements sur la glace dans un contexte temporellement variable des conditions de la banquise arctique. Cette recherche a exploit & eacute; des donn & eacute;es originales comprenant plus de 5 000 observations SAR Sentinel-1 de polynies hivernales cartographi & eacute;es pr & egrave;s des communaut & eacute;s canadiennes de Sanikiluaq et de Nain, afin d'& eacute;tudier la d & eacute;tection des polynies de glace c & ocirc;ti & egrave;re par apprentissage profond. Le r & eacute;seau de d & eacute;tection d'objets Faster-RCNN a & eacute;t & eacute; optimis & eacute; pour la d & eacute;tection des polynies gr & acirc;ce & agrave; plusieurs modifications apport & eacute;es & agrave; sa conception et & agrave; ses strat & eacute;gies d'entra & icirc;nement. Les mod & egrave;les de d & eacute;tection obtenus ont montr & eacute; une bonne g & eacute;n & eacute;ralisation entre les r & eacute;gions et ont d & eacute;tect & eacute; avec pr & eacute;cision les polynies pr & eacute;sentant des contrastes de r & eacute;trodiffusion locaux sup & eacute;rieurs & agrave; 5 dB, atteignant par exemple un rappel de 90 % pour la classe cible, associ & eacute;e & agrave; une pr & eacute;cision de 24 %. Les polynies de moins de 500 m & egrave;tres et pr & eacute;sentant des contrastes de r & eacute;trodiffusion locaux inf & eacute;rieurs & agrave; 3 dB, qui repr & eacute;sentent environ la moiti & eacute; des observations, ont souvent & eacute;t & eacute; manqu & eacute;es. Des scores de pr & eacute;cision inf & eacute;rieurs & agrave; 30 % ont & eacute;t & eacute; fr & eacute;quemment obtenus lors des tentatives d'optimization du rappel. Les r & eacute;sultats mettent en & eacute;vidence les difficult & eacute;s rencontr & eacute;es pour assurer des r & eacute;sultats uniformes par des images uniques, en raison de la variabilit & eacute; et de la faible intensit & eacute; fr & eacute;quente des signatures des polynies par r & eacute;trodiffusion SAR & agrave; double polarization de Sentinel-1. De plus amples recherches sur l'imagerie SAR multi-temporelle, multi-fr & eacute;quence et/ou & agrave; plus haute r & eacute;solution contribueront & agrave; la mise en place de syst & egrave;mes de d & eacute;tection robustes des polynies et, par cons & eacute;quent, & agrave; la s & eacute;curit & eacute; et & agrave; la surveillance des glaces de mer.
Canada's remaining intact native Prairie grasslands are highly fragmented and difficult to distinguish from seeded forage using conventional satellite methods due to their phenological similarity. This study evaluated the potential of compact polarimetric (CP) C-band synthetic aperture radar (SAR) from the RADARSAT Constellation Mission (RCM) (3 m resolution) to discriminate between native grassland and seeded forage land cover types across five Southern Alberta sites representing diverse plant communities, soils and management regimes. Random forest classification with recursive feature elimination was applied to multitemporal CP-SAR features, including polarimetric decompositions and Stokes parameters. Results demonstrated that C-band CP-SAR effectively captures structural differences between native and seeded sites, with classification accuracies exceeding 92% in Dry and Moist Mixedgrass sites, with slightly lower accuracy (90-91.5%) in Foothills Fescue and Aspen Parkland sites. Early (May) and late-season (August-September) acquisitions, as well as imagery following large regional rainfall events, were particularly important for separability. Stokes parameters and degree of polarization consistently ranked highest among most informative features. The findings highlight the value of high-resolution CP-SAR for regional-scale mapping of rangelands, particularly where optical sensors are limited by cloud cover, providing new opportunities for operational monitoring of Canada's native grasslands. Les prairies indig & egrave;nes du Canada sont fragment & eacute;es et difficiles & agrave; distinguer des fourrages sem & eacute;s par les m & eacute;thodes satellitaires conventionnelles en raison de leur similarit & eacute; ph & eacute;nologique. Cette & eacute;tude a & eacute;valu & eacute; le potentiel de la polarim & eacute;trie compacte (PC) & agrave; partir du radar & agrave; synth & egrave;se d'ouverture (RSO) en bande C de la mission Constellation RADARSAT (MCR) (r & eacute;solution de 3 m) pour identifier les types de couverture v & eacute;g & eacute;tale (prairies indig & egrave;nes et fourrages sem & eacute;es) sur cinq sites du sud de l'Alberta, repr & eacute;sentatifs de diverses communaut & eacute;s v & eacute;g & eacute;tales, sols et r & eacute;gimes de gestion. Une classification par la technique random forest avec un & eacute;lagage r & eacute;cursif a & eacute;t & eacute; appliqu & eacute;e aux donn & eacute;es multitemporelles du PC-RSO, incluant les d & eacute;compositions polarim & eacute;triques et les param & egrave;tres de Stokes. Les r & eacute;sultats ont d & eacute;montr & eacute; que le PC-RSO en bande C identifie efficacement les diff & eacute;rences structurelles entre les prairies indig & egrave;nes et les fourrages sem & eacute;s, avec une pr & eacute;cision de classification sup & eacute;rieure & agrave; 92 % dans les sites de prairies mixtes s & egrave;ches et humides, et l & eacute;g & egrave;rement inf & eacute;rieure (90-91,5 %) dans les sites de f & eacute;tuque des piedmonts et les zones naturelles avec trembles. Les acquisitions r & eacute;alis & eacute;es en d & eacute;but (mai) et en fin de saison (ao & ucirc;t-septembre), ainsi que les images prises apr & egrave;s d'importants & eacute;pisodes de fortes pr & eacute;cipitations r & eacute;gionales, se sont av & eacute;r & eacute;es particuli & egrave;rement pertinentes pour la s & eacute;parabilit & eacute; des types de prairies. Les param & egrave;tres de Stokes et le degr & eacute; de polarisation les caract & eacute;ristiques les plus utiles. Ces r & eacute;sultats soulignent l'int & eacute;r & ecirc;t de l'imagerie PC-RSO haute r & eacute;solution pour la cartographie & agrave; l'& eacute;chelle r & eacute;gionale des prairies, notamment l & agrave; o & ugrave; la couverture nuageuse limite l'utilisation des capteurs optiques, offrant ainsi de nouvelles perspectives pour la surveillance op & eacute;rationnelle des prairies indig & egrave;nes du Canada.
