Repeatable methods capable of quantifying Arctic surface water extent at high resolutions are important, but still require development. Here, we present a study using very-high resolution (VHR) X-band Synthetic Aperture Radar (SAR) imagery from Capella Space for fine-scale semantic segmentation of Arctic surface water features. Our study proposes a modified U-Net encoder-decoder model for this task, optimized using the Nadam algorithm. Otsu thresholding was leveraged to rapidly generate 512 x 512-pixel patches for the U-Net, resulting in an efficient and automated training pipeline. Within this study, we also quantitatively compared the deep learning (DL) U-Net to a shallow machine learning (ML) algorithm, XGBoost (XGB), and evaluated the Capella Space imagery against spatially and temporally coincident Sentinel-1 C-band. Performance evaluations showed the U-Net outperforms XGB measured under several statistical metrics, reaching an Intersection over Union (IoU) of 0.955. An explainability analysis was conducted to complement this finding, using Gradient-weighted Class Activation Mapping (Grad-Cam). Visual analysis also underscored the extreme detail of small water features captured by Capella Space imagery, which are at times omitted or lack clarity in conventional Sentinel-1. This research makes several contributions to Arctic surface water mapping, demonstrating the effectiveness of combining VHR SAR imagery with DL. Le d & eacute;veloppement des m & eacute;thodes reproductibles permettant de quantifier l'& eacute;tendue des eaux de surface de l'Arctique & agrave; fine r & eacute;solution est important. Nous pr & eacute;sentons ici une & eacute;tude utilisant des images radar & agrave; synth & egrave;se d'ouverture (SAR) en bande X & agrave; tr & egrave;s haute r & eacute;solution spatiale (THR) de Capella Space pour la segmentation s & eacute;mantique & agrave; & eacute;chelle fine des caract & eacute;ristiques des eaux de surface de l'Arctique. Notre & eacute;tude propose un mod & egrave;le codeur-d & eacute;codeur U-Net modifi & eacute; pour cette t & acirc;che, optimis & eacute; par l'algorithme Nadam. Le seuillage Otsu a & eacute;t & eacute; utilis & eacute; pour g & eacute;n & eacute;rer des imagettes de 512 x 512 pixels pour le mod & egrave;le U-Net, ce qui a permis de cr & eacute;er un pipeline d'apprentissage efficace et automatis & eacute;. Dans le cadre de cette & eacute;tude, nous avons & eacute;galement compar & eacute; quantitativement l'algorithme d'apprentissage profond U-Net (DL: Deep Learning) & agrave; un algorithme d'apprentissage automatique (ML: Machine Learning), XGBoost (XGB). De plus, nous avons compar & eacute; les r & eacute;sultats obtenus & agrave; partir de l'imagerie Capella Space avec ceux obtenus & agrave; partir des images Sentinel-1 en bande C, pour des r & eacute;solutions spatiales et temporelles similaires. Les & eacute;valuations ont montr & eacute; que les r & eacute;sultats issus du model U-Net surpassent ceux du mod & egrave;le XGB pour plusieurs m & eacute;triques statistiques, atteignant une intersection sur l'union (IoU: Intersection over Union) de 0,955. Une analyse plus approfondie a & eacute;t & eacute; r & eacute;alis & eacute;e pour mieux interpr & eacute;ter ces r & eacute;sultats, & agrave; l'aide de la cartographie d'activation de classe pond & eacute;r & eacute;e par gradient (Grad-Cam: Gradient-weighted Class Activation Mapping). L'analyse visuelle a & eacute;galement mis en & eacute;vidence comment l'imagerie Capella Space peut faire ressortir des petites caract & eacute;ristiques aquatiques, qui sont parfois omises ou manquent de clart & eacute; avec l'usage des images de Sentinel-1. Cette recherche apporte plusieurs contributions & agrave; la cartographie des eaux de surface de l'Arctique, d & eacute;montrant l'efficacit & eacute; de la combinaison de l'imagerie SAR THR avec un mod & egrave;le DL.

In Canada's Arctic tundra region, permafrost is continuous, and the landscape is rich in patterned features. Polygonal terrain, which includes both high- and low-centered features and their wet trenches below, is considered to be high-latitude wetlands in the continuous permafrost region. These prominent hydrological features retain and transport water within widespread ice-wedge networks and govern many ecosystem dynamics. Due to the meter-scale spatial gradients of these processes, mapping of polygonal wetland networks necessitates high-resolution imagery. To date, most studies have used optical imagery for this task; however, these sensors are affected by cloud cover and polar darkness, limiting image availability and repeatability. Thus, our overall objective was to evaluate high-resolution hybrid compact polarimetric (HCP) imagery from the recently launched Radarsat Constellation Mission (RCM), in fusion with ArcticDEM topographic data, for Arctic landscape mapping with a focus on polygonal wetlands. RCM's 5 m Stripmap beam mode, which has yet to be studied for such a task, represents an innovative HCP synthetic aperture radar (SAR) data source since it allows for polarimetric decomposition methods, despite being a dual-pol system. Within this study, we present a seven-input channel Convolutional Neural Network (CNN) model for the classification of ice-wedge dominated landscapes. A range of model hyperparameters as well as the effect of SAR speckle filtering on classification accuracy, have been examined. The optimized CNN achieved a high classification accuracy (0.931 mean Intersection Over Union; mIOU) for three semantic classes representative of the study area, namely polygonal wetlands, open water, and uplands. These results were superior in comparison to a benchmark machine learning (ML) Random Forest (RF) algorithm, thus demonstrating the proposed CNN's potential for regional-scale permafrost feature mapping. Notably, the optimal CNN architecture used unfiltered SAR data as input, underscoring the importance of spatial resolution when classifying polygonal terrain with deep learning (DL). These findings have important implications regarding the design, tuning, and sensitivity of CNN algorithms, and on the efficacy of HCP SAR, for mapping Arctic regions.

Earth observation (EO) plays a pivotal role in understanding our planet’s rapidly changing environment. Recently, geospatial technologies used to analyse EO data have made remarkable progress, in particular from innovations in Artificial Intelligence (AI) and scalable cloud-computing resources. This chapter presents a brief overview of these developments, with a focus on geospatial “big data.” A case study is presented where Google Earth Engine (GEE) was used to upscale airborne active layer thickness (ALT) measurements over an extensive permafrost region. GEE’s machine learning (ML) capabilities were leveraged for upscaling measurements to several multi-source satellite EO datasets. Novel Explainable Artificial Intelligence (XAI) techniques were also used for model feature selection and interpretation. The optimized ML model achieved an R2 of 0.476, although performance varied by ecosystem. This chapter highlights the capabilities of new RS sensors and geospatial technologies for better understanding permafrost environments, which is important in the face of climate change.

The Arctic-Boreal zone (ABZ) covers over 26 million km2 and is home to numerous duck species; however, understanding the spatiotemporal distribution of their populations across this vast landscape is challenging, in part due to extent and data scarcity. Species abundance models for ducks in the ABZ commonly use static (time invariant) habitat covariates to inform predictions, such as wetland type and extent maps. For the first time in this region, we developed species abundance models using high-resolution, time-varying wetland inundation data produced using satellite remote sensing methods. This data captured metrics of surface water extent and inundated vegetation in the Peace Athabasca Delta, Canada, which is within the NASA Arctic Boreal Vulnerability Experiment core domain. We used generalized additive mixed models to demonstrate the improved predictive value of this novel data set over time-invariant data. Our findings highlight both the potential complementarity and efficacy of dynamic wetland inundation information for improving estimation of duck abundance and distribution at high latitudes. Further, these data can be an asset to spatial targeting of biodiversity conservation efforts and developing model-based metrics of their success under rapidly changing climatic conditions.

The Arctic-Boreal zone (ABZ) is a vast landscape, supporting many waterfowl species. However, because of spatial extent and data scarcity, modelling waterfowl populations across the ABZ is challenging. Species abundance models (SAMs) for waterfowl typically use time-static habitat covariates to make predictions, such as wetland type maps. Here, for the first time, we used remote sensing methods to create time-varying ABZ wetland inundation covariates, which helped improve our waterfowl SAMs. SAMs were tested over the Peace Athabasca Delta (PAD), within the NASA Arctic Boreal Vulnerability Experiment (ABoVE) domain. Generalized Additive Mixed Models (GAMM) were used to demonstrate the efficacy and additive value of this novel inundation data, which was derived using Google Earth Engine (GEE) and Sentinel-1 satellite imagery. Timely inundation data like these provide an opportunity to better understand the spatio-temporal variation in ABZ waterfowl under changing climatic conditions.

Lakes and ponds are extensive features throughout the circumpolar region, spanning a broad range of environmental conditions which controls their hydro-ecological processes and spatiotemporal distribution. The physical characteristics of these freshwater ecosystems, including their extent and depth, are particularly responsive to climatic conditions. Thus, having the ability to efficiently map and monitor these elements is crucial as the climate continues to warm, especially over large spatial extents. In this study, satellite derived bathymetry (SDB) methods were implemented to model regional inland Arctic water depths within the open-source and cloud-based Google Earth Engine (GEE) platform. High-resolution (10 m) spectral reflectance data from Sentinel-2 was used as covariates in non-parametric and non-linear machine learning (ML) models, namely random forest (RF), support vector regression (SVR), and classification and regression trees (CART). These three ML models were also compared to a more classical and parametric multiple linear regression (MLR) model. All models were calibrated using in situ bathymetric data collected near the Toolik Field Station, in the Alaskan Arctic tundra. With such a large sample repository available, the effects of training set size, noise, and multicollinearity on model performance was comprehensively investigated. Results clearly demonstrated the superior efficacy and robustness of the RF algorithm for predicting water depths, achieving a best coefficient of determination (R-2) of 0.74, mean absolute error (MAE) of 2.12 m, and root mean square error (RMSE) of 3.04 m. Overall, this study highlights the potential of advanced and flexible ML algorithms within the GEE, and demonstrates the capabilities of the polar-orbiting Copernicus Sentinel-2 satellite for this Arctic application. In the future, this ML-driven methodology can be applied using GEE's cloud infrastructure to produce updated, cost-effective, and accurate bathymetry maps over large expanses of the Arctic tundra. (c) 2023 COSPAR. Published by Elsevier B.V. All rights reserved.

Climate-driven permafrost degradation and an intensification of the hydrological cycle are rapidly altering the intricate ecohydrological processes of Arctic wetlands, threatening their long-term carbon sequestration capabilities. Addressing this concern through effective management holds immense potential for climate regulation, mitigation, and adaptation efforts. As such, there is growing need for timely spatial inventory data identifying Arctic wetlands with sufficient accuracy, resolution, and detail. Wetland mapping at large scales necessitates the processing of large volumes of Earth observation (EO) data, a challenge known as “Big Data”. Consequently, in this study, we present a cloud-based methodology exploiting the remarkable collection of EO data and computational power of Google Earth Engine (GEE) to map Arctic wetlands at 10 m spatial resolution. Our workflow evaluated temporally aggregated optical and radar satellite imagery and novel hydro-physiographic layers as inputs into a robust Random Forest (RF) machine learning (ML) algorithm. Both pixel and object-based classification approaches were assessed, whereby ML models were calibrated with a training dataset of sufficient and comprehensive samples. The study was conducted over Canada’s Southern Arctic ecozone (830,000 km2). GEE enabled the efficient preprocessing and classification of large volumes of EO data and resulted in excellent yet similar statistical performance for both pixel and object-based approaches, achieving overall accuracies of > 89 % and mean F1-scores of > 0.79. Moreover, McNemar tests indicated that these classifications were not statistically different, which has significant implications regarding computing time and processing efficiencies. These results demonstrate the efficacy and scalability of our cloud-based GEE methodology, and as such can support future endeavors around Pan-Arctic wetland mapping and monitoring.

Synthetic aperture radar (SAR) is a widely used tool for Earth observation activities. It is particularly effective during times of persistent cloud cover, low light conditions, or where in situ measurements are challenging. The intensity measured by a polarimetric SAR has proven effective for characterizing Arctic tundra landscapes due to the unique backscattering signatures associated with different cover types. However, recently, there has been increased interest in exploiting novel interferometric SAR (InSAR) techniques that rely on both the amplitude and absolute phase of a pair of acquisitions to produce coherence measurements, although the simultaneous use of both intensity and interferometric coherence in Arctic tundra image classification has not been widely tested. In this study, a time series of dual-polarimetric (VV, VH) Sentinel-1 SAR/InSAR data collected over one growing season, in addition to a digital elevation model (DEM), was used to characterize an Arctic tundra study site spanning a hydrologically dynamic coastal delta, open tundra, and high topographic relief from mountainous terrain. SAR intensity and coherence patterns based on repeat-pass interferometry were analyzed in terms of ecological structure (i.e., graminoid, or woody) and hydrology (i.e., wet, or dry) using machine learning methods. Six hydro-ecological cover types were delineated using time-series statistical descriptors (i.e., mean, standard deviation, etc.) as model inputs. Model evaluations indicated SAR intensity to have better predictive power than coherence, especially for wet landcover classes due to temporal decorrelation. However, accuracies improved when both intensity and coherence were used, highlighting the complementarity of these two measures. Combining time-series SAR/InSAR data with terrain derivatives resulted in the highest per-class F1 score values, ranging from 0.682 to 0.955. The developed methodology is independent of atmospheric conditions (i.e., cloud cover or sunlight) as it does not rely on optical information, and thus can be regularly updated over forthcoming seasons or annually to support ecosystem monitoring.

Arctic environments are remote and inaccessible, making conventional field-based data collection challenging. Thus, this study describes an efficient desktop-based methodology for deriving reference data to support large-scale remote sensing classification focusing on wetland ecosystems. Our study area was Canada's Southern Arctic Ecozone. Various Earth observation (EO) datasets, including optical, multi-spectral, and topographic, were used as a base to support a photointerpretation process for collecting reference data. Ten 30-by-30-kilometer sampling plots were established across the ecozone for this activity based on a suite of minimum criteria. Reference polygons were assigned to one of the five major wetland classes of the Canadian Wetland Classification System (CWCS), along with a detailed wetland type definition. It is anticipated this methodology will be applied later to other northern ecozones to support large-scale wetland classification updates and status and trends reporting.

The first Canadian wetland inventory (CWI) map, which was based on Landsat data, was produced in 2019 using the Google Earth Engine (GEE) big data processing platform. The proposed GEE-based method to create the preliminary CWI map proved to be a cost, time, and computationally efficient approach. Although the initial effort to produce the CWI map was valuable with a 71% overall accuracy (OA), there were several inevitable limitations (e.g., low-quality samples for the training and validation of the map). Therefore, it was important to comprehensively investigate those limitations and develop effective solutions to improve the accuracy of the Landsat-based CWI (L-CWI) map. Over the past year, the L-CWI map was shared with several governmental, academic, environmental nonprofit, and industrial organizations. Subsequently, valuable feedback was received on the accuracy of this product by comparing it with various in situ data, photo-interpreted reference samples, land cover/land use maps, and high-resolution aerial images. It was generally observed that the accuracy of the L-CWI map was lower relative to the other available products. For example, the average OA in four Canadian provinces using in situ data was 60%. Moreover, including reliable in situ data, using an object-based classification method, and adding more optical and synthetic aperture radar datasets were identified as the main practical solutions to improve the CWI map in the future. Finally, limitations and solutions discussed in this study are applicable to any large-scale wetland mapping using remote sensing methods, especially to CWI generation using optical satellite data in GEE.

This research presents the findings of a study where Sentinel-2 optical satellite imagery was assessed for its boreal open water mapping capabilities using a deep learning algorithm and an object-oriented image analysis approach [1]. Key components within the workflow of this study included the following: 1) segmentation of the optical imagery into meaningful and relatively homogenous image objects, 2) building a Convolutional Neural Network (CNN) deep learning model, 3) training the CNN, and 4) applying the CNN. This workflow produced probability heat maps of open water; an object-based approach was then used to iteratively assign final class labels (‘open water’, or ‘other’) based on various heatmap probabilities. Additionally, an important element of this research included the optimization of the CNN patch size and learning rate through a sensitivity assessment. The most optimal CNN model produced an overall accuracy of 96.2% (0.912 kappa coefficient), with an open water producer's accuracy of 98.1%.

The main objective in our study was to derive an accurate wetland inventory of the Dinaga Wek'ehodi region, Northwest Territories, while also enhancing our previously established wetland mapping workflow. Our methods used multidate optical and radar satellite imagery and fused these data with ArcticDEM topographic variables. Additionally, few studies to date have assessed the ArcticDEM for wetland mapping; our research helps fill this critical gap in the literature. A machine-learning, object-based approach was employed to classify the fused data stacks and included both mean and standard deviation image-object feature extractions. In this study, 18 random forest models were tested, each including various sensor inputs and feature extractions. The highest accuracy was achieved using a fusion of optical, radar, and ArcticDEM data and included both the mean and standard deviation of image objects (88.17% overall accuracy and kappa 0.858). Vegetated wetlands had producer accuracies ranging from 74% to 86%, whereas open water was 92%. Feature importance rankings indicated that 16 of the top 20 variables were derived from optical data, three from radar, and one from the ArcticDEM. The results of our study will be used to assist governments and other interested parties in advancing conservation initiatives for this significant high-latitude region. (C) 2020 Society of Photo-Optical Instrumentation Engineers (SPIE)

In this study, Sentinel-2 optical satellite imagery was acquired over the Peace Athabasca Delta and assessed for its open water classification capabilities using an object-oriented deep learning algorithm . The workflow involved segmenting the satellite data into meaningful image objects, building a Convolutional Neural Network (CNN), training the CNN, and lastly applying the CNN, resulting in probability heat maps of open water (with score values ranging from 0-1). Using the vector segmentation, heat maps were then iteratively assigned final class labels ('open water' or 'other') based on various probability thresholding. The ensuing open water classifications were assessed against a large validation dataset, and a highest overall accuracy of 96.2% (0.912 kappa coefficient) was achieved, with an open water producer's accuracy of 98.1%. These results were then compared against a Random Forest (RF) classification, and results indicated that the CNN algorithm outperforms RF in this study site. Additionally, an important component of this study was the optimization of several CNN configurations, including patch size and learning rate; the latter which plays a critical role in model adaptation. The optimized object-oriented CNN and associated results can be used to provide resource managers with accurate surface water extent maps at 10 m resolution.

Wetlands have and continue to undergo rapid environmental and anthropogenic modification and change to their extent, condition, and therefore, ecosystem services. In this first part of a two-part review, we provide decision-makers with an overview on the use of remote sensing technologies for the ‘wise use of wetlands’, following Ramsar Convention protocols. The objectives of this review are to provide: (1) a synthesis of the history of remote sensing of wetlands, (2) a feasibility study to quantify the accuracy of remotely sensed data products when compared with field data based on 286 comparisons found in the literature from 209 articles, (3) recommendations for best approaches based on case studies, and (4) a decision tree to assist users and policymakers at numerous governmental levels and industrial agencies to identify optimal remote sensing approaches based on needs, feasibility, and cost. We argue that in order for remote sensing approaches to be adopted by wetland scientists, land-use managers, and policymakers, there is a need for greater understanding of the use of remote sensing for wetland inventory, condition, and underlying processes at scales relevant for management and policy decisions. The literature review focuses on boreal wetlands primarily from a Canadian perspective, but the results are broadly applicable to policymakers and wetland scientists globally, providing knowledge on how to best incorporate remotely sensed data into their monitoring and measurement procedures. This is the first review quantifying the accuracy and feasibility of remotely sensed data and data combinations needed for monitoring and assessment. These include, baseline classification for wetland inventory, monitoring through time, and prediction of ecosystem processes from individual wetlands to a national scale.

Mapping and monitoring surface water features is important for sustainably managing this critical natural resource that is in decline due to numerous natural and anthropogenic pressures. Satellite Synthetic Aperture Radar is a popular and inexpensive solution for such exercises over large scales through the application of thresholds to distinguish water from non-water. Despite improvements to threshold methods, threshold selection is traditionally manual, which introduces subjectivity and inconsistency over large scales. This study presents a novel method for objectively determining and applying a threshold to determine water masks from Synthetic Aperture Radar (SAR) imagery on a scene-by-scene basis. The method was applied to Radarsat-2 and simulated Radarsat Constellation Mission scenes, and validated against two independent validation sources with high accuracy (Kappa ranging from 0.85 to 0.93). Expectedly, greatest misclassification occurs near shorelines, which are often ecologically important zones. Comparisons between Radarsat-2 and Radarsat Constellation Mission thresholds and outputs suggest that the latter is a capable successor for surface water applications. This work represents a foundational step toward objectivity and consistency in large-scale water mapping and monitoring.

The following review is the second part of a two part series on the use of remotely sensed data for quantifying wetland extent and inferring or measuring condition for monitoring drivers of change on wetland environments. In the first part, we introduce policy makers and non-users of remotely sensed data with an effective feasibility guide on how data can be used. In the current review, we explore the more technical aspects of remotely sensed data processing and analysis using case studies within the literature. Here we describe: (a) current technologies used for wetland assessment and monitoring; (b) the latest algorithmic developments for wetland assessment; (c) new technologies; and (d) a framework for wetland sampling in support of remotely sensed data collection. Results illustrate that high or fine spatial resolution pixels (≤10 m) are critical for identifying wetland boundaries and extent, and wetland class, form and type, but are not required for all wetland sizes. Average accuracies can be up to 11% better (on average) than medium resolution (11–30 m) data pixels when compared with field validation. Wetland size is also a critical factor such that large wetlands may be almost as accurately classified using medium-resolution data (average = 76% accuracy, stdev = 21%). Decision-tree and machine learning algorithms provide the most accurate wetland classification methods currently available, however, these also require sampling of all permutations of variability. Hydroperiod accuracy, which is dependent on instantaneous water extent for single time period datasets does not vary greatly with pixel resolution when compared with field data (average = 87%, 86%) for high and medium resolution pixels, respectively. The results of this review provide users with a guideline for optimal use of remotely sensed data and suggested field methods for boreal and global wetland studies.

The authors evaluated multiple remotely sensed datasets for their contributions to operational wetland mapping in a subarctic, boreal cordillera study site in Yukon, Canada. They assessed Sentinel-2 optical imagery, Sentinel-1 C-band and ALOS PALSAR L-band synthetic aperture radar (SAR) imagery, and topographical data from the territorial digital elevation model (DEM) using an object-based image analysis (OBIA) approach. Three machine-learning algorithms were tested, namely random forest (RF), support vector machine (SVM) and k-nearest neighbor (KNN), using various data combinations (11 model scenarios). RF produced the most accurate results when incorporating all optical, SAR and DEM data (86.5%, kappa 0.84), with open water (100% producer accuracy, PA), marsh (75% PA) and swamps (85.7% PA) being detected most accurately. When assessed in isolation, Sentinel-2 optical data consistently generated more accurate classifications than either SAR platform or DEM data. RF variable importance metrics provided further explanation to these results, indicating the 8 most powerful variables to be optical. Variable reduction tests also produced comparable accuracies, indicating that an optimal RF model can be built based on predictive power rankings. The results can be used to inform resource managers on the efficacy of current datasets and their applications to wetland mapping in northern, subarctic environments.

The objective of this paper is to assess the accuracy of C-band synthetic aperture radar (SAR) datasets in mapping peatland types over a region of Canada's subarctic boreal zone. This paper assessed contributions of quad-polarization linear backscatter intensities (σ°HH, σ°HV, σ°VV), image textures, and two polarimetric scattering decompositions: 1) Cloude–Pottier, and 2) Freeman–Durden. Four quad-polarimetric RADADSAT-2 images were studied at incidence angles of 19.4°, 23.1°, 45.8°, and 48.1°. The influence of combining dual-angular information acquired within a short temporal span was also assessed. These C-band SAR data were used to classify peatlands according to isolated flat bogs (bogs), channel fens (fens), raised peat plateaus (plateaus), and forested uplands (uplands) using a supervised support vector machine (SVM) classifier. Numerous classifications were examined to compare the unique contributions of these variables to classification accuracy. Results suggest linear backscatter variables in isolation produce comparable classification results with those of the Freeman–Durden and Cloude–Pottier decompositions. Combining polarimetric decomposition and texture data into classifications with linear backscatter data resulted in only minor (∼1–3%) improvement. Combining classifications from small and large incidence angles (dual-angular) significantly improved classification results over those of a single image. Classification accuracy was the highest for isolated bogs and open water surfaces, whereas fens, uplands, and plateaus had lower accuracies. The highest accuracy classification (84% and kappa coefficient of 0.80) used a dual-angular approach, with additional decomposition and texture information. However, it is noted that texture information rarely improved classification results across all tests. This approach identified isolated flat bogs, channel fens, and raised peat plateaus with >76% producer's accuracies.