Development of a convolutional neural network to accurately detect land use and land cover

The detection and modeling of Land Use and Land Cover (LULC) play pivotal roles in natural resource management, environmental modeling and assessment, and ecological connectivity management. However, addressing LULCC detection and modeling constitutes a complex data-driven process. In the present study, a Convolutional Neural Network (CNN) is employed due to its great potential in image classification. The development of these tools applies the deep learning method. A methodology has been developed that classifies the set of land uses in a natural area of special protection. This study area covers the Sierra del Cando (Galicia, northwest Spain), considered by the European Union as a Site of Community Interest and integrated in the Natura 2000 Network. The results of the CNN model developed show an accuracy of 91 % on training dataset and 88 % on test dataset. In addition, the model was tested on images of the study area, both from Sentinel-2 and PNOA. Despite some confusion especially in the residential class due to the characteristics in this area, CNNs prove to be a powerful classification tool.•Classifications based on a CNN model•LULC are classified into 10 different classes•Training and test accuracy are 91 % and 88 %, respectively.

Macrophyte coverage drives microbial community structure and interactions in a shallow sub-tropical lake

Shallow lakes are typically dominated by macrophytes, which have important functional roles regulating trophic conditions and creating biological habitat. Macrophytes have been shown to strongly influence water chemistry and shape microbial communities in shallow lakes. In Florida, many large, shallow lakes are dominated by alien invasive, submersed macrophytes, such as hydrilla (Hydrilla verticillata [L.F.] Royle) and are intensively managed to reduce infestations and contain the spread of these alien invasive macrophytes. In this study, we investigated the effects of large (40 ha) herbicidal and mechanical control treatments on a large lake located in Central Florida that resulted in the reduction of Hydrilla and concomitant changes in water chemistry and microbial communities (both bacteria and protists [microbial eukaryotes]). We observed a considerable decrease in macrophyte coverage associated with plant control treatments as well as a temporal change in macrophyte coverage in Lake Tohopekaliga. We found that changes in macrophyte coverage, regardless of treatment type, significantly affected the water chemistry of the lake, resulting in a sharp increase of chlorophyll a concentration as well as an increase in turbidity with the decrease of macrophyte coverage. Moreover, the decline in macrophytes led to decreases in microbial community diversity with over-representation of phototrophic functional groups. Specifically, we observed an increase in cyanobacteria with the decrease in macrophyte coverage. Our study highlights the advantages and disadvantages of macrophyte control. Although there was an initial decrease in macrophyte coverage associated with the chemical and mechanical control of aquatic plants, after a few months, we found a considerable increase in coverage. In addition, the increase of cyanobacterial relative abundance demonstrates the possible consequences of aquatic plant control such as cyanobacterial blooms if there is a continued decline of macrophytes.

Characterising the distribution of mangroves along the southern coast of Vietnam using multi-spectral indices and a deep learning model

Mangroves are an ecologically and economically valuable ecosystem that provides a range of ecological services, including habitat for a diverse range of plant and animal species, protection of coastlines from erosion and storms, carbon sequestration, and improvement of water quality. Despite their significant ecological role, in many areas, including in Vietnam, large scale losses have occurred, although restoration efforts have been underway. Understanding the scale of the loss and the efficacy of restoration requires high resolution temporal monitoring of mangrove cover on large scales. We have produced a time series of 10-m-resolution mangrove cover maps using the Multispectral Instrument on the Sentinel 2 satellites and with this tool measured the changes in mangrove distribution on the Vietnamese Southern Coast (VSC). We extracted the annual mangrove cover ranging from 2016 to 2023 using a deep learning model with a U-Net architecture based on 17 spectral indices. Additionally, a comparison of misclassification by the model with global products was conducted, indicating that the U-Net architecture demonstrated superior performance when compared to experiments including multispectral bands of Sentinel-2 and time-series of Sentinel-1 data, as shown by the highest performing spectral indices. The generated performance metrics, including overall accuracy, precision, recall, and F1-score were above 90 % for entire years. Water indices were investigated as the most important variables for mangrove extraction. Our study revealed some misclassifications by global products such as World Cover and Global Mangrove Watch and highlighted the significance of our study for local analysis. While we did observe a loss of 34,778 ha (42.2 %) of mangrove area in the region, 47,688 ha (57.8 %) of new mangrove area appeared, resulting in a net gain of 12,910 ha (15.65 %) over the eight-year period of the study. The majority of new mangrove areas were concentrated in Ca Mau peninsulas and within estuaries undergoing recovery programs and natural recovery processes. Mangrove loss occurred in regions where industrial development, wind farm projects, reclaimed land, and shrimp pond expansion is occurring. Our study provides a theoretical framework as well as up-to-date data for mapping and monitoring mangrove cover change that can be readily applied at other sites.

Coarse-resolution burned area datasets severely underestimate fire-related forest loss

Global coarse-resolution (≥250 m) burned area (BA) products have been used to estimate fire related forest loss, but we hypothesised that a significant part of fire impacts might be undetected because of the underestimation of small fires (<100 ha), especially in the tropics. In this paper, we analysed fire-related forest cover loss in sub-Saharan Africa (SSA) for 2016 and 2019 based on a BA product generated from Sentinel-2 data (20 m), which was observed to have significantly lower omission errors than the coarse-resolution BA products. Using these higher resolution BA datasets, we found that fires contribute to >46 % of total forest losses over SSA, more than twice the estimates from coarse-resolution BA products. In addition, burned forest areas showed more than twofold likelihood of subsequent loss compared to unburned ones. In moist tropical forests, the most fire-vulnerable biome, burning had even six times more chance to precede forest loss than unburned areas. We also found that fire-related characteristics, such as fire size and season, and forest fragmentation play a major role in the determination of tree cover fate. Our results reveal that medium-resolution BA detects more fires in late fire season, which tend to have higher impact on forests than early season ones. On the other hand, small fires represented the major driver of forest loss after fires and the vast majority of these losses occur in fragmented landscapes near forest edge (<260 m). Therefore medium-resolution BA products are required to obtain a more accurate evaluation of fire impacts in tropical ecosystems.

A comprehensive research on open surface drinking water resources in Istanbul using remote sensing technologies

Istanbul is a megacity with a population of 15.5 million and is one of the fastest-growing cities in Europe. Due to the rapidly increasing population and urbanization, Istanbul's daily water needs are constantly increasing. In this study, eight drinking water basins that supply water to Istanbul were comprehensively examined using remote sensing observations and techniques. Water surface area changes were determined monthly, and their relationships with meteorological parameters and climate change were investigated. Monthly water surface areas of natural lakes and dams were determined with the Normalized Difference Water Index (NDWI) applied to Sentinel-2 satellite images. Sentinel-1 Synthetic Aperture Radar (SAR) images were used in months when optical images were unavailable. The study was carried out using 3705 optical and 1167 SAR images on the Google Earth Engine (GEE) platform. Additionally, to determine which areas of water resources are shrinking, water frequency maps of the major drinking water resources were produced. Land use/land cover (LULC) changes that occurred over time were determined, and the effects of the increase in urbanization, especially on drinking water surface areas, were investigated. ESRI LULC data was used to determine LULC changes in watersheds, and the increase in urbanization areas from 2017 to 2022 ranged from 1 to 91.43%. While the basin with the least change was in Istranca, the highest increase in the artificial surface was determined to be in the Büyükçekmece basin with 1833.03 ha (2.89%). While there was a 1-12.35% decrease in the surface areas of seven water resources from 2016 to 2022, an increase of 2.65-93% was observed in three water resources (Büyükçekmece, Sazlıdere, and Elmalı), each in different categories depending on their size. In the overall analysis, total WSA decreased by 62.33 ha from 2016 to 2022, a percentage change of 0.70%. Besides the areal change analysis, the algae contents of the drinking water resources over the years were examined for the major water basins using the Normalized Difference Chlorophyll Index (NDCI) and revealed their relationship with meteorological factors and urbanization.

Satellite imagery as a management tool for monitoring water clarity across freshwater ponds on Cape Cod, Massachusetts

Water clarity serves as both an indicator and a regulator of biological function in aquatic systems. Large-scale, consistent water clarity monitoring is needed for informed decision-making. Inland freshwater ponds and lakes across Cape Cod, a 100-km peninsula in Massachusetts, are of particular interest for water clarity monitoring. Secchi disk depth (SDD), a common measure of water clarity, has been measured intermittently for over 200 Cape Cod ponds since 2001. Field-measured SDD data were used to estimate SDD from satellite data, leveraging the NASA/USGS Landsat Program and Copernicus Sentinel-2 mission, spanning 1984 to 2022. Random forest machine learning models were generated to estimate SDD from satellite reflectance data and maximum pond depth. Spearman rank correlations (rs) were "strong" for Landsat 5 and 7 (rs = 0.78 and 0.79), and "very strong" for Landsat 8, 9, and Sentinel-2 (rs = 0.83, 0.86, and 0.80). Mean absolute error also indicated strong predictive capacity, ranging from 0.65 to 1.05 m, while average bias ranged from -0.20 to 0.06 m. Long- and recent short-term changes in satellite-estimated SDD were assessed for 193 ponds, selected based on surface area and the availability of maximum pond depth data. Long-term changes between 1984 and 2022 established a retrospective baseline using the Mann-Kendall test for trend and Theil-Sen slope. Generally, long-term water clarity improved across the Cape; 149 ponds indicated increasing water clarity, and 8 indicated deteriorating water clarity. Recent short-term changes between 2021 and 2022 identified ponds that may benefit from targeted management efforts using the Mann-Whitney U test. Between 2021 and 2022, 96 ponds indicated deteriorations in water clarity, and no ponds improved in water clarity. While the 193 ponds analyzed here constitute only one quarter of Cape Cod ponds, they represent 85% of its freshwater surface area, providing the most spatially and temporally comprehensive assessment of Cape Cod ponds to date. Efforts are focused on Cape Cod, but can be applied to other areas given the availability of local field data. This study defines a framework for monitoring and assessing change in satellite-estimated SDD, which is important for both local and regional management and resource prioritization.

Incorporating high-resolution climate, remote sensing and topographic data to map annual forest growth in central and eastern Europe

To enhance our understanding of forest carbon sequestration, climate change mitigation and drought impact on forest ecosystems, the availability of high-resolution annual forest growth maps based on tree-ring width (TRW) would provide a significant advancement to the field. Site-specific characteristics, which can be approximated by high-resolution Earth observation by satellites (EOS), emerge as crucial drivers of forest growth, influencing how climate translates into tree growth. EOS provides information on surface reflectance related to forest characteristics and thus can potentially improve the accuracy of forest growth models based on TRW. Through the modelling of TRW using EOS, climate and topography data, we showed that species-specific models can explain up to 52 % of model variance (Quercus petraea), while combining different species results in relatively poor model performance (R2 = 13 %). The integration of EOS into models based solely on climate and elevation data improved the explained variance by 6 % on average. Leveraging these insights, we successfully generated a map of annual TRW for the year 2021. We employed the area of applicability (AOA) approach to delineate the range in which our models are deemed valid. The calculated AOA for the established forest-type models was 73 % of the study region, indicating robust spatial applicability. Notably, unreliable predictions predominantly occurred in the climate margins of our dataset. In conclusion, our large-scale assessment underscores the efficacy of combining climate, EOS and topographic data to develop robust models for mapping annual TRW. This research not only fills a critical void in the current understanding of forest growth dynamics but also highlights the potential of integrated data sources for comprehensive ecosystem assessments.

Comparison between U-shaped structural deep learning models to detect landslide traces

Landslides endanger lives and public infrastructure in mountainous areas. Monitoring landslide traces in real-time is difficult for scientists, sometimes costly and risky because of the harsh terrain and instability. Nowadays, modern technology may be able to identify landslide-prone locations and inform locals for hours or days when the weather worsens. This study aims to propose indicators to detect landslide traces on the fields and remote sensing images; build deep learning (DL) models to identify landslides from Sentinel-2 images automatically; and apply DL-trained models to detect this natural hazard in some particular areas of Vietnam. Nine DL models were trained based on three U-shaped architectures, including U-Net, U2-Net, and U-Net3+, and three options of input sizes. The multi-temporal Sentinel-2 images were chosen as input data for training all models. As a result, the U-Net, using an input image size of 32 × 32 and a performance of 97 % with a loss function of 0.01, can detect typical landslide traces in Vietnam. Meanwhile, the U-Net (64 × 64) can detect more considerable landslide traces. Based on multi-temporal remote sensing data, a different case study in Vietnam was chosen to see landslide traces over time based on the trained U-Net (32 × 32) model. The trained model allows mountain managers to track landslide occurrences during wet seasons. Thus, landslide incidents distant from residential areas may be discovered early to warn of flash floods.

Fine-scale monitoring of lake ice phenology by synthesizing remote sensed and climatologic features based on high-resolution satellite constellation and modeling

Lake ice, as a crucial component of the cryosphere, serves as a sensitive indicator of climate change. Fine-scale monitoring of spatiotemporal patterns in lake ice phenology holds significant importance in scientific research and environmental management. However, the rapid and dynamic nature of the freeze-thaw process of lake ice poses challenges to existing methods, resulting in their limited application in small lakes. In this study, we propose a novel approach of investigating ice phenology of lakes in various sizes. We conducted a case study in Hoh Xil, known for its vulnerability to climate change and a wide distribution of small lakes, to analyze the ice phenology of 372 lakes (>1 km2) during 2017-2021. Firstly, ensemble machine-learning model was developed for lake ice identification from Landsat-8/9 and Sentinel-2 A/B imagery. The accuracy evaluation reveals the overall good performance for ice extraction results based on Landsat-8/9 (97.03 %) and Sentinel-2 A/B (96.89 %). Next, the XGBoost models were employed to reconstruct ice coverages on unobserved dates for the freezeup and breakup periods, respectively. Totally, 744 XGBoost models were constructed for the study lakes, and the majority of them perform well. Based on the reconstructed daily ice coverage, phenology parameters could be extracted for examining the spatiotemporal characteristics of ice cover and possible relationships with lake sizes and terrains. From early-October to early-November, the Hoh Xil lakes freeze from the northwest to the southeast, while the breakup period starts in late-March and lasts until late-June. Moreover, the results indicate relatively small variability in freezeup-end dates among lakes, but significant differences in breakup dates, showing a greater sensitivity to temperature variations. Furthermore, ice phenology in small lakes exhibit stronger consistency with subtle climatic fluctuations. The results highlight the significant role of ice phenology in small lakes, as they dominate the overall tendency of ice phenology in Hoh Xil.

Dynamic monitoring and analysis of chlorophyll-a concentrations in global lakes using Sentinel-2 images in Google Earth Engine

Remote estimation of Chlorophyll-a (Chl-a) has long been used to investigate the responses of aquatic ecosystems to global climate change. High-spatiotemporal-resolution Sentinel-2 satellite images make it possible to routinely monitor and trace the spatial distributions of lake Chl-a if reliable retrieval algorithms are available. In this study, Sentinel-2 images and in-situ measured data were used to develop a Chl-a retrieval algorithm based on 13 optical water types (OWTs) with a satisfying performance (R2 = 0.74, RMSE = 0.42 mg/m3, MAE = 0.33 mg/m3, and MAPE = 55.56 %). After removing the disturbance of algal blooms and other factors, the distribution of Chl-a in 3067 of the largest global lakes (≥50 km2) was mapped using the Google Earth Engine (GEE). From 2019 to 2021, the average Chl-a concentration was 16.95 ± 5.95 mg/m3 for the largest global lakes. During the COVID-19 pandemic, global lake-averaged Chl-a concentration reached its lowest value in 2020. From the perspective of spatial distribution, lakes with low Chl-a concentrations were mainly distributed in high-latitude, high-elevation, or economically underdeveloped areas. Among all the potential influencing factors, lake surface temperature had the largest contribution to Chl-a and showed a positive correlation with Chl-a in approximately 92.39 % of the lakes. Conversely, factors such as precipitation and tree cover area around the lake were negatively correlated with Chl-a concentration in nearly 61.44 % of the lakes.

Spatial and remote sensing monitoring shows the end of the bark beetle outbreak on Belgian and north-eastern France Norway spruce (Picea abies) stands

In 2022, Europe emerged from eight of the hottest years on record, leading to significant spruce mortality across Europe. The particularly dry weather conditions of 2018 triggered an outbreak of bark beetles (Ips typographus), causing the loss of thousands of hectares of Norway spruce stands, including in Wallonia and North-eastern France. A methodology for detecting the health status of spruce was developed based on a dense time series of satellite imagery (Sentinel-2). The time series of satellite images allowed the modelling of the spectral response of healthy spruce forests over the seasons: a decrease in photosynthetic activity of the forest canopy causes deviations from this normal seasonal vegetation index trajectory. These anomalies are caused by a bark beetle attack and are detected automatically. The method leads in the production of an annual spruce health map of Wallonia and Grand-Est. The goal of this paper is to assess the damage caused by bark beetle using the resulting spruce health maps. A second objective was to compare the influence of basic variables on the mortality of spruce trees in these two regions. Lasted 6 years (2017-2022), bark beetle has destroyed 12.2% (23,674 ha) of the spruce area in Wallonia and Grand-Est of France. This study area is composed of three bioclimatic areas: Plains, Ardennes and Vosges, which have not been equally affected by bark beetle attacks. The plains were the most affected, with 50% of spruce forests destroyed, followed by the Ardennes, which lost 11.3% of its spruce stands. The Vosges was the least affected bioclimatic area, with 5.6% of spruce stands lost. For the most problematic sites, Norway spruce forestry should no longer be considered.

Leveraging Google Earth Engine for a More Effective Grassland Management: A Decision Support Application Perspective

Grasslands cover a substantial portion of the earth's surface and agricultural land and is crucial for human well-being and livestock farming. Ranchers and grassland management authorities face challenges in effectively controlling herders' grazing behavior and grassland utilization due to underdeveloped infrastructure and poor communication in pastoral areas. Cloud-based grazing management and decision support systems (DSS) are needed to address this issue, promote sustainable grassland use, and preserve their ecosystem services. These systems should enable rapid and large-scale grassland growth and utilization monitoring, providing a basis for decision-making in managing grazing and grassland areas. In this context, this study contributes to the objectives of the EU LIFE IMAGINE project, aiming to develop a Web-GIS app for conserving and monitoring Umbria's grasslands and promoting more informed decisions for more sustainable livestock management. The app, called "Praterie" and developed in Google Earth Engine, utilizes historical Sentinel-2 satellite data and harmonic modeling of the EVI (Enhanced Vegetation Index) to estimate vegetation growth curves and maturity periods for the forthcoming vegetation cycle. The app is updated in quasi-real time and enables users to visualize estimates for the upcoming vegetation cycle, including the maximum greenness, the days remaining to the subsequent maturity period, the accuracy of the harmonic models, and the grassland greenness status in the previous 10 days. Even though future additional developments can improve the informative value of the Praterie app, this platform can contribute to optimizing livestock management and biodiversity conservation by providing timely and accurate data about grassland status and growth curves.

Remote Sensing Evaluation Drone Herbicide Application Effectiveness for Controlling Echinochloa spp. in Rice Crop in Valencia (Spain)

Rice (Oryza sativa L.) is a staple cereal in the diet of more than half of the world's population. Within the European Union, Spain is a leader in rice production due to its climate and tradition, accounting for 26% of total EU production in 2020. The Valencian rice area covers around 15,000 hectares and is strongly influenced by biotic and abiotic factors. An important biotic factor affecting rice production is weeds, which compete with rice for sunlight, water and nutrients. The dominant weed in Spain is Echinochloa spp., although wild rice is becoming increasingly important. Rice cultivation in Valencia takes place in the area of L'Albufera de Valencia, which is a natural park, i.e., a special protection area. In this natural area, the use of phytosanitary products is limited, so it is necessary to use the minimum amount possible. Therefore, the objective of this work is to evaluate the possibility of using remote sensing effectively to determine the effectiveness of the application of the herbicide cyhalofop-butyl by drone for the control of Echinochloa spp. in rice crops in Valencia. The results will be compared with those obtained by using sterilisation machines (electric backpack sprayers) to apply the herbicide. To evaluate the effectiveness of the application, the reflectance obtained by the satellite sensors in the red and near infrared (NIR) wavelengths, as well as the normalised difference vegetation index (NDVI), were used. The remote sensing results were analysed and complemented by the number of rice plants and weeds per area, plant dry weight, leaf area, BBCH phenological state, SPAD index values, chlorophyll content and relative growth rate. Remote sensing is validated as an effective tool for determining the efficacy of an herbicide in controlling weeds applied by both the drone and the electric backpack sprayer. The weeds slowed down their development after the treatment. Depending on the phenological state of the crop and the active ingredient of the herbicide, these results are applicable to other areas with different climatic and environmental conditions.

Deep learning-based burned forest areas mapping via Sentinel-2 imagery: a comparative study

In order to evaluate the effects of forest fires on the dynamics of the function and structure of ecosystems, it is necessary to determine burned forest areas with high accuracy, effectively, economically, and practically using satellite images. Extraction of burned forest areas utilizing high-resolution satellite images and image classification algorithms and assessing the successfulness of varied classification algorithms has become a prominent research field. This study aims to indicate on the capability of the deep learning-based Stacked Autoencoders method for the burned forest areas mapping from Sentinel-2 satellite images. The Stacked Autoencoders, used in this study as an unsupervised learning method, were compared qualitatively and quantitatively with frequently used supervised learning algorithms (k-Nearest Neighbors (k-NN), Subspaced k-NN, Support Vector Machines, Random Forest, Bagged Decision Tree, Naive Bayes, Linear Discriminant Analysis) on two distinct burnt forest zones. By selecting burned forest zones with contrasting structural characteristics from one another, an objective assessment was achieved. Manually digitized burned areas from Sentinel-2 satellite images were utilized for accuracy assessment. For comparison, different classification performance and quality metrics (Overall Accuracy, Mean Squared Error, Correlation Coefficient, Structural Similarity Index Measure, Peak Signal-to-Noise Ratio, Universal Image Quality Index, and KAPPA metrics) were used. In addition, whether the Stacked Autoencoders method produces consistent results was examined through boxplots. In terms of both quantitative and qualitative analysis, the Stacked Autoencoders method showed the highest accuracy values.

Scaling up benthic primary productivity estimates in a large intertidal estuary using remote sensing

As two main primary producers in temperate intertidal regions, seagrass and microphytobenthos (MPB) support estuarine ecosystem functions in multiple ways including stabilizing food webs and regulating sediment resuspension among others. Monitoring estuary productivity at large scales can inform ecosystem scale responses to environmental stressors (climate change, pollution and habitat degradation). Here we use a case study to show how Sentinel-2 data can be used to estimate estuary-wide emerged and submerged gross primary productivity (GPP) on intertidal flats by coupling a new machine learning model to map seagrass and unvegetated habitats with literature-derived photosynthesis-irradiance (P - I) relationships. The model consisted of (1) supervised classification with random forest to delineate seagrass and unvegetated areas and (2) artificial neural network (ANN) regression to predict % seagrass coverage. Our seagrass delineation by supervised classification had an overall accuracy of 0.96, while the ANN regression on seagrass coverage provided high predictive accuracy (R2 = 0.71 and RMSE = 0.11). The estimated GPP showed seagrass contributed slightly more to intertidal benthic productivity than MPB in the case-study estuary over the 3-year study period. This model can be used to predict the response of seagrass and MPB GPP to sea level rise, which shows that the future state may be very sensitive to increased turbidity. For example, by the year 2100, the model shows a sharp decline in productivity with sea level rise, assuming current turbidity trends, (loss of up to 52-53 % for seagrass and 23-45 % for MPB, a function of whether shoreward migration of seagrass is incorporated). However, GPP under conditions of unchanging turbidity (and no seagrass migration), exhibits minimal negative impact of sea level rise (loss of 3 % for seagrass and increase of 29 % for MPB). Therefore, controlling water turbidity might be an efficient solution to maintaining the current GPP as sea level rises.

Retrieval of suspended sediment concentration (SSC) in the Arabian Gulf water of arid region by Sentinel-2 data

Suspended sediment concentration (SSC) in water increases temperature and turbidity, limits the photosynthesis of aquatic plants, and reduces biologically available oxygen. It is important to study SSC in the coastal waters of the Arabian Gulf. Thus, this study mapped the SSC of coastal water between Al Arish and Al Ghariyah in northern Qatar using the spectral bands of the MultiSpectral Imager (MSI) of Sentinel-2 by calculating the Normalized Difference Suspended Sediment Index and Normalized Suspended Material Index. The results are studied using the Normalized Difference Turbidity Index and Modified Normalized Difference Water Index. The mapping of SSC in the water using NDSSI showed the presence of a high concentration of suspended sediments between Al Arish and Al Mafjar and a low concentration between Al Mafjar and Al Ghariyah. The mapping of NSMI showed values between 0.012 (clear water) and 0.430 (more suspended material) for the occurrence of suspended materials and supported the results of NDSSI. The study of turbidity using an NDTI image showed turbidity index values ranging from -0.44 (clear water) to 0.12 (high turbidity) and confirmed the occurrence and distribution of suspended sediments and materials in the water. The MNDWI image was able to discriminate clear water with bright pixels from silty sand and mud flats. The relationships between NDSSI, NSMI, and NDTI were correlated with in-situ measurements and studied to find suitable indices to map SSC. Regression analyses showed the strongest relationship between NSMI and NDTI (R2 = 0.95) next to NDSSI and NDTI, where NDTI had the strongest effect on NDSSI (R2 = 0.86). The satellite data results were evaluated by studying the physical parameters and spatial distribution of suspended sediments in the surface and bottom waters. In addition, the grain size distributions, mineral identification, and chemical element concentrations in the bottom sediment samples were studied.

Sen-2 LULC: Land use land cover dataset for deep learning approaches

Land Use Land Cover (LULC) classification is pivotal to sustainable environment and natural resource management. It is critical in planning, monitoring, and management programs at various local and national levels. Monitoring changes in LULC patterns over time is crucial for understanding evolving landscapes. Traditionally, LULC classification has been achieved through satellite data by remote sensing, geographic information system (GIS) techniques, machine learning classifiers, and deep learning models. Semantic segmentation, a technique for assigning land cover classes to individual pixels in an image, is commonly employed for LULC mapping. In recent years, the deep learning revolution, particularly Convolutional Neural Networks (CNNs), has reshaped the field of computer vision and LULC classification. Deep architectures have consistently outperformed traditional methods, offering greater accuracy and efficiency. However, the availability of high-quality datasets has been a limiting factor. Bridging the gap between modern computer vision and remote sensing data analysis can revolutionize our understanding of the environment and drive breakthroughs in urban planning and ecosystem change research. The "Sen-2 LULC Dataset" has been created to facilitate this convergence. This dataset comprises of 213,761 pre-processed 10 m resolution images representing seven LULC classes. These classes encompass water bodies, dense forests, sparse forests, barren land, built-up areas, agricultural land, and fallow land. Importantly, each image may contain multiple coexisting land use and land cover classes, mirroring the real-world complexity of landscapes. The dataset is derived from Sentinel-2 satellite imagery sourced from the Copernicus Open Access Hub (https://scihub.copernicus.eu/) platform. It includes spectral bands B4, B3, and B2, corresponding to red, green, and blue (RGB) channels, and offers a spectral resolution of 10 m. The dataset also provides an equal number of mask images. Structured into six folders, the dataset offers training, testing, and validation sets for images and masks. Researchers across various domains can leverage this resource to advance LULC classification in the context of the Indian region. Additionally, it catalyzes fostering collaboration between remote sensing and computer vision communities, enabling novel insights into environmental dynamics and urban planning challenges.

Recognizing protected and anthropogenic patterns in landscapes using interpretable machine learning and satellite imagery

The accurate and comprehensive mapping of land cover has become a central task in modern environmental research, with increasing emphasis on machine learning approaches. However, a clear technical definition of the land cover class is a prerequisite for learning and applying a machine learning model. One of the challenging classes is naturalness and human influence, yet mapping it is important due to its critical role in biodiversity conservation, habitat assessment, and climate change monitoring. We present an interpretable machine learning approach to map patterns related to territorial protected and anthropogenic areas as proxies of naturalness and human influence using satellite imagery. To achieve this, we train a weakly-supervised convolutional neural network and subsequently apply attribution methods such as Grad-CAM and occlusion sensitivity mapping. We propose a novel network architecture that consists of an image-to-image network and a shallow, task-specific head. Both sub-networks are connected by an intermediate layer that captures high-level features in full resolution, allowing for detailed analysis with a wide range of attribution methods. We further analyze how intermediate layer activations relate to their attributions across the training dataset to establish a consistent relationship. This makes attributions consistent across different scenes and allows for a large-scale analysis of remote sensing data. The results highlight that our approach is a promising way to observe and assess naturalness and territorial protection.

Seasonal evaluation and mapping of aboveground biomass in natural rangelands using Sentinel-1 and Sentinel-2 data

Rangelands play a vital role in developing countries' biodiversity conservation and economic growth, since most people depend on rangelands for their livelihood. Aboveground-biomass (AGB) is an ecological indicator of the health and productivity of rangeland and provides an estimate of the amount of carbon stored in the vegetation. Thus, monitoring seasonal AGB is important for understanding and managing rangelands' status and resilience. This study assesses the impact of seasonal dynamics and fire on biophysical parameters using Sentinel-1 (S1) and Sentinel-2 (S2) image data in the mesic rangeland of Limpopo, South Africa. Six sites were selected (3/area), with homogenous vegetation (10 plots/site of 30m2). The seasonal measurements of LAI and biomass were undertaken in the early summer (December 2020), winter (July-August 2021), and late summer (March 2022). Two regression approaches, random forest (RF) and stepwise multiple linear regression (SMLR), were used to estimate seasonal AGB. The results show a significant difference (p < 0.05) in AGB seasonal distribution and occurrence between the fire (ranging from 0.26 to 0.39 kg/m2) and non-fire areas (0.24-0.35 kg/m2). In addition, the seasonal predictive models derived from random forest regression (RF) are fit to predict disturbance and seasonal variations in mesic tropical rangelands. The S1 variables were excluded from all models due to high moisture content. Hence, this study analyzed the time series to evaluate the correlation between seasonal estimated and field AGB in mesic tropical rangelands. A significant correlation between backscattering, AGB and ecological parameters was observed. Therefore, using S1 and S2 data provides sufficient data to obtain the seasonal changes of biophysical parameters in mesic tropical rangelands after disturbance (fire) and enhanced assessments of critical phenology stages.

Crop productivity estimation by integrating multisensor satellite, in situ, and eddy covariance data into efficiency-based model

Accurate and quantitative regional estimates of the carbon budget require an integration of eddy covariance (EC) flux-tower observations and remote sensing in ecosystem models. In this study, a simple remote sensing driven light use efficiency (LUE) model was used to estimate the primary productivity for major cropping systems using multi-temporal satellite data over the Saharanpur district in India.The model is based on radiation absorption and its conversion into biomass. The LUE model was implemented for major crop rotations derived from the time-series of Sentinel-2 and Landsat 8 with monthly satellite-based spatially explicit fields of photosynthetically active radiation (PAR), fraction of absorbed PAR (fAPAR) and down-regulated light use efficiency. Incident PAR and fAPAR were estimated on monthly basis from the ground-calibrated empirical equation using INSAT-3D insolation product and remote sensing-based vegetation indices, respectively. Spatial LUE maps created by down-regulating maximum LUE (EC tower-based) with water and temperature stressors derived from land surface water index (LSWI) and EC-based cardinal temperature, respectively. LUE-based modeled GPP over the sugarcane-wheat system was found higher than the rice-wheat system in Saharanpur district. This is because C4 crop (sugarcane) has very high photosynthetic efficiency compared to C3 crops (rice and wheat). Modeled GPP over the sugarcane-wheat system was found in good agreement with observed EC tower-based GPP (Index of Agreement = 0.93). Further regionally calibrated remote sensing-based LUE model well captures gross photosynthesis rates (GPP) over cropland ecosystem compared to globally modeled MODIS GPP product.

Machine learning for modeling forest canopy height and cover from multi-sensor data in Northwestern Ethiopia

Continuous mapping of the height and canopy cover of forests is vital for measuring forest biomass, monitoring forest degradation and restoration. In this regard, the contribution of Light Detection and Ranging (LiDAR) sensors, which were developed to obtain detailed data on forest composition across large geographical areas, is immense. Accordingly, this study aims to predict forest canopy cover and height in tropical forest areas utilizing Global Ecosystem Dynamics Investigation (GEDI) LIDAR, multisensor images, and random forest regression. To achieve this, we gathered predictor variables from the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), Sentinel-2 multispectral datasets, and Sentinel-1 synthetic aperture radar (SAR) backscatters. The model's accuracy was evaluated based on a validation dataset of GEDI Level 2A and Level 2B. The random forest method was used the combination of data layers from Sentinel-1, Sentinel-2, and topographic measurements to model forest canopy cover and height. The produced canopy height and cover maps had a resolution of 30 m with R2 = 0.86 and an RMSE of 3.65 m for forest canopy height and R2 = 0.87 and an RMSE of 0.15 for canopy cover for the year 2022. These results suggest that combining multiple variables and data sources improves canopy cover and height prediction accuracy compared to relying on a single data source. The output of this study could be helpful in creating forest management plans that support sustainable utilization of the forest resources.

Where does floating Sargassum in the East China Sea come from?

Floating macroalgae of Sargassum horneri (S. horneri) in the East China Sea (ECS) has increased in recent years, with ocean warming being one of the driving factors. Yet their possible origins, based on a literature review, are unclear. Here, using multi-sensor high-resolution remote sensing data and numerical experiments for the period of 2015-2023, we show two possible origins of the ECS floating S. horneri, one being local near the Zhejiang coast with initiation in January-February and the other being remote (> 800 km from the first) in the Bohai Sea with initiation in June-November. While their drifting pathways are revealed in the sequential remote sensing imagery, numerical experiments suggest that S. horneri from the remote origin (Bohai Sea) can hardly meander through the strong Yangtze River frontal zone, which may serve as a "wall" to prevent trespassing of surface floating seaweed to the south of the frontal zone, where S. horneri has a local origin. PLAIN LANGUAGE SUMMARY: Sargassum horneri (S. horneri) is a brown macroalgae (seaweed) abundant in surface waters of the East China Sea (ECS), which can serve as a moving habitat, but can also cause major beaching events and environmental problems. Knowledge of its origins is important to help implement mitigation strategies and understand possible ecological impacts along its drifting pathways. Using high-resolution remote sensing images and numerical experiments, we track floating S. horneri in space and time between 2015 and 2023. Two possible origins are identified, one being far away from the ECS and the other being local, both of which are known to have benthic S. horneri. The study also reveals how S. horneri are transported from their source regions resulting in large-scale distributions previously observed in medium-resolution satellite imagery.

Spatial and Temporal Analysis of Water Quality in High Andean Lakes with Sentinel-2 Satellite Automatic Water Products

The water of high Andean lakes is strongly affected by anthropic activities. However, due to its complexity this ecosystem is poorly researched. This study analyzes water quality using Sentinel-2 (S2) images in high Andean lakes with apparent different eutrophication states. Spatial and temporal patterns are assessed for biophysical water variables from automatic products as obtained from versions of C2RCC (Case 2 Regional Coast Color) processor (i.e., C2RCC, C2X, and C2X-COMPLEX) to observe water characteristics and eutrophication states in detail. These results were validated using in situ water sampling. C2X-COMPLEX appeared to be an appropriate option to study bodies of water with a complex dynamic of water composition. C2RCC was adequate for lakes with high transparency, typical for lakes of highlands with excellent water quality. The Yambo lake, with chlorophyll-a concentration (CHL) values of 79.6 ± 5 mg/m3, was in the eutrophic to hyper-eutrophic state. The Colta lake, with variable values of CHL, was between the oligotrophic to mesotrophic state, and the Atillo lakes, with values of 0.16 ± 0.1 mg/m3, were oligotrophic and even ultra-oligotrophic, which remained stable in the last few years. Automatic S2 water products give information about water quality, which in turn makes it possible to analyze its causes.

Water distribution based on SAR and optical data to improve hazard mapping

Climate projections foresee intense precipitation and long-term drought events is increasing with consequent rapid changes in surface water bodies in a short period. In areas with drastic hydrological changes, achieving accurate and rapid mapping of these phenomena in combination with hydrologic variability characteristics is a key of effective emergency management and disaster risk reduction plans. This study presents an automatic method for mapping drought and flood hazards, particularly in regions with significant hydrological changes. We use Sentinel-1/2 and Landsat data to extract surface water and classify permanent and seasonal water bodies in historical periods, which serve as the basis for identifying flood or drought areas. The water extraction method combines index-based analysis for optical data and the region-Otsu method for radar data, ensuring accurate identification of water. The effectiveness of this approach is demonstrated through comparisons with existing products in Poyang Lake (China), the Po River Plain (Italy), and the Indus River Plain (Pakistan). Findings show a high similarity between the two, and our results can provide more specific details. Our method is particularly well-suited for areas with fluctuating hydrological conditions, can also map quickly without optical data. By effectively identifying areas affected by drought and flood hazards while mitigating errors from natural hydrological dynamics, this methodology contributes valuable insights to enhance emergency management and disaster risk reduction plans.

Using Sentinel images for analyzing water and land separability in an agricultural river basin

The presence or absence of water can result in floods or droughts, potentially impacting agricultural productivity to a great extent. With advancements in remote sensing technology, the reliability of identifying water bodies has significantly improved, particularly in terms of distinguishing between water and land. This study introduced remote sensing methods to improve the accuracy of differentiating water within the Dawenhe River basin. Various water body scenarios were examined, and the performance of these methods was evaluated to determine the proper approach for water-land separation. In applying water body indices to Sentinel-2 images, it was found that the normalized difference water index (NDWI) outperformed the modified normalized difference water index (MNDWI) in identifying water bodies. Consequently, histograms of frequency distribution for Sentinel-1 were generated, revealing that water and land were more distinguishable in VV polarization than in VH polarization. Using histogram thresholding on VV polarized images in Dongping Lake resulted in an overall classification accuracy of 97.58%, surpassing that of Otsu's method at 97.36%. To address the persisting misclassifications, this study identified three leading causes and proposed corresponding solutions. These solutions included (1) employing the morphological dilation algorithm to expand the water area, mitigating pixel mixing issues at the water-land boundary that caused the water bodies to appear smaller; (2) utilizing incidence angles and digital elevation model (DEM) to locate and remove shadows; and (3) slightly lowering the thresholds and manually correcting misclassifications. As a result, the average accuracy of the four areas increased from 95.56 to 96.94%.

Wildland-urban interface typologies prone to high severity fires in Spain

Due to complex interactions between climate and land use changes, large forest fires have increased in frequency and severity over the last decades, impacting dramatically on biodiversity and society. In southern European countries affected by demographic challenges, fire risk and danger play special relevance at the wildland-urban interfaces (WUIs), where decision-making and land management have strong socio-ecological implications. WUIs have been historically typified according to both fire occurrence probability and settlement vulnerability, but those classifications lack generality regarding fire regime components. We aim to develop an integrated and comprehensive scheme for identifying the WUI typologies most at risk to fire severity across large territories. We selected fourteen large wildfires (over than 500 ha) occurred in Spain (2016-2021) containing different WUI scenarios. First, based on a building cartography and a multi-temporal series of Sentinel-2 imagery, each WUI was delimited and spatially characterized according to building density and pre-fire fuel characteristics (type, amount, and structure). Afterwards, a decision tree regression model was applied to identify the most relevant pre-fire vegetation parameters driving burn severity. The combined effect of the selected pre-fire vegetation drivers and the building density patterns on fire severity was evaluated using linear mixed models. Finally, the WUI typologies most prone to high burn severity were recognized using Tukey post-hoc tests. Results indicated that building density, land cover class and vegetation cover fraction determined fire severity in areas close to human settlements. Specifically, isolated, scattered and sparsely clustered buildings enclosed in a high-cover shrub matrix were the WUI typologies most susceptible to high-severity fires. These findings contribute to the development of appropriate strategies to minimize the risk of severe fires in WUIs and avoid potential losses of multiple ecosystem services valuable for society.

Agronomic performance and remote sensing assessment of organic and mineral fertilization in rice fields

Introduction

Rice heavily relies on nitrogen fertilizers, posing environmental, resource, and geopolitical challenges. This study explores sustainable alternatives like animal manure and remote sensing for resource-efficient rice cultivation. It aims to assess the long-term impact of organic fertilization and remote sensing monitoring on agronomic traits, yield, and nutrition.

Methods

A six-year experiment in rice fields evaluated fertilization strategies, including pig slurry (PS) and chicken manure (CM) with mineral fertilizers (MIN), MIN-only, and zero-fertilization. Traits, yield, spectral responses, and nutrient content were measured. Sentinel-2 remote sensing tracked crop development.

Results

Cost-effective organic fertilizers (PS and CM) caused a 13% and 15% yield reduction but still doubled zero-fertilization yield. PS reduced nitrogen leaching. Heavy metals in rice grains were present at safe amounts. Organic-fertilized crops showed nitrogen deficiency at the late vegetative stages, affecting yield. Sentinel-2 detected nutrient deficiencies through NDVI.

Discussion

Organic fertilizers, especially PS, reduce nitrogen loss, benefiting the environment. However, they come with yield trade-offs and nutrient management challenges that can be managed and balanced with reduced additional mineral applications. Sentinel-2 remote sensing helps manage nutrient deficiencies. In summary, this research favors cost-effective organic fertilizers with improved nutrient management for sustainable rice production.

A fast Fourier convolutional deep neural network for accurate and explainable discrimination of wheat yellow rust and nitrogen deficiency from Sentinel-2 time series data

Introduction

Accurate and timely detection of plant stress is essential for yield protection, allowing better-targeted intervention strategies. Recent advances in remote sensing and deep learning have shown great potential for rapid non-invasive detection of plant stress in a fully automated and reproducible manner. However, the existing models always face several challenges: 1) computational inefficiency and the misclassifications between the different stresses with similar symptoms; and 2) the poor interpretability of the host-stress interaction.

Methods

In this work, we propose a novel fast Fourier Convolutional Neural Network (FFDNN) for accurate and explainable detection of two plant stresses with similar symptoms (i.e. Wheat Yellow Rust And Nitrogen Deficiency). Specifically, unlike the existing CNN models, the main components of the proposed model include: 1) a fast Fourier convolutional block, a newly fast Fourier transformation kernel as the basic perception unit, to substitute the traditional convolutional kernel to capture both local and global responses to plant stress in various time-scale and improve computing efficiency with reduced learning parameters in Fourier domain; 2) Capsule Feature Encoder to encapsulate the extracted features into a series of vector features to represent part-to-whole relationship with the hierarchical structure of the host-stress interactions of the specific stress. In addition, in order to alleviate over-fitting, a photochemical vegetation indices-based filter is placed as pre-processing operator to remove the non-photochemical noises from the input Sentinel-2 time series.

Results and discussion

The proposed model has been evaluated with ground truth data under both controlled and natural conditions. The results demonstrate that the high-level vector features interpret the influence of the host-stress interaction/response and the proposed model achieves competitive advantages in the detection and discrimination of yellow rust and nitrogen deficiency on Sentinel-2 time series in terms of classification accuracy, robustness, and generalization.

Radioactive remote signatures derived from Sentinel-2 images and field verification in West Sulawesi, Indonesia

Mamuju, West Sulawesi, is an area in Indonesia with high radiation levels. A high radiation dose rate was detected in Adang volcanic rock. Activities related to radioactive minerals contained in rocks and soil may release hazardous radiation into the environment. The study area generally exhibits a highly irregular morphology that provides limited access because of the high slope gradient of the volcanic terrain. Therefore, it is challenging to identify and map minerals carrying radioactive elements via Sentinel-2 remote sensing. Since optical satellite images are superior in land cover detection, we proposed a mapping technique for radioactive carrier minerals based on vegetation indices verified by chlorophyll field measurements. We applied band rationing to identify the distribution of hydrothermal alteration minerals and vegetation stress, while field chlorophyll measurements of Dryopteris marginalis and Nephrolepis exaltata were conducted. The Sentinel-2 color composite of images with 4/2, 8A/11, and 11/12 RGB band ratios revealed the distribution of iron oxide, ferromagnesian silicates, and clay minerals. High levels of uranium (U) were scattered in leucite basalt rocks, with a broad distribution of iron oxide minerals and small amounts of ferromagnesian minerals. In contrast, the presence of thorium was not affected by the presence of these minerals. In addition, band rationing of chlorophyll spectra captured by the red edge vegetation index (REVI) was used as the basis for vegetation stress mapping related to radiation exposure based on the chlorophyll content in ferns in the study area. The REVI image showed an anomalous vegetation stress concordant with the high radioactivity. To obtain more accurate results, ground measurements were also performed to identify the vegetation stress due to the presence of minerals carrying radioactive elements. The areas with radioactive mineralization and vegetation stress were located upstream of the Mamuju River and the Botteng and Ahu areas in Tapalang.

A method for estimating yield of maize inbred lines by assimilating WOFOST model with Sentinel-2 satellite data

Maize is the most widely planted food crop in China, and maize inbred lines, as the basis of maize genetic breeding and seed breeding, have a significant impact on China's seed security and food safety. Satellite remote sensing technology has been widely used for growth monitoring and yield estimation of various crops, but it is still doubtful whether the existing remote sensing monitoring means can distinguish the growth difference between maize inbred lines and hybrids and accurately estimate the yield of maize inbred lines. This paper explores a method for estimating the yield of maize inbred lines based on the assimilation of crop models and remote sensing data, initially solves the problem. At first, this paper analyzed the WOFOST(World Food Studies)model parameter sensitivity and used the MCMC(Markov Chain Monte Carlo) method to calibrate the sensitive parameters to obtain the parameter set of maize inbred lines differing from common hybrid maize; then the vegetation indices were selected to establish an empirical model with the measured LAI(Leaf Area Index) at three key development stages to obtain the remotely sensed estimated LAI; finally, the yield of maize inbred lines in the study area was estimated and mapped pixel by pixel using the EnKF(Ensemble Kalman Filter) data assimilation algorithm. Also, this paper compares a method of assimilation by setting a single parameter. Instead of the WOFOST parameter optimization process, a parameter representing the growth weakness of the inbred lines was set in WOFOST to distinguish the inbred lines from the hybrids. The results showed that the yield estimated by the two methods compared with the field measured yield data had R2: 0.56 and 0.18, and RMSE: 684.90 Kg/Ha and 949.95 Kg/Ha, respectively, which proved that the crop growth model of maize inbred lines established in this study combined with the data assimilation method could initially achieve the growth monitoring and yield estimation of maize inbred lines.

Spatially explicit accuracy assessment of deep learning-based, fine-resolution built-up land data in the United States

Geospatial datasets derived from remote sensing data by means of machine learning methods are often based on probabilistic outputs of abstract nature, which are difficult to translate into interpretable measures. For example, the Global Human Settlement Layer GHS-BUILT-S2 product reports the probability of the presence of built-up areas in 2018 in a global 10 m × 10 m grid. However, practitioners typically require interpretable measures such as binary surfaces indicating the presence or absence of built-up areas or estimates of sub-pixel built-up surface fractions. Herein, we assess the relationship between the built-up probability in GHS-BUILT-S2 and reference built-up surface fractions derived from a highly reliable reference database for several regions in the United States. Furthermore, we identify a binarization threshold using an agreement maximization method that creates binary built-up land data from these built-up probabilities. These binary surfaces are input to a spatially explicit, scale-sensitive accuracy assessment which includes the use of a novel, visual-analytical tool which we call focal precision-recall signature plots. Our analysis reveals that a threshold of 0.5 applied to GHS-BUILT-S2 maximizes the agreement with binarized built-up land data derived from the reference built-up area fraction. We find high levels of accuracy (i.e., county-level F-1 scores of almost 0.8 on average) in the derived built-up areas, and consistently high accuracy along the rural-urban gradient in our study area. These results reveal considerable accuracy improvements in human settlement models based on Sentinel-2 data and deep learning, as compared to earlier, Landsat-based versions of the Global Human Settlement Layer.

Mapping cover crop species in southeastern Michigan using Sentinel-2 satellite data and Google Earth Engine

Cover crops are a critical agricultural practice that can improve soil quality, enhance crop yields, and reduce nitrogen and phosphorus losses from farms. Yet there is limited understanding of the extent to which cover crops have been adopted across large spatial and temporal scales. Remote sensing offers a low-cost way to monitor cover crop adoption at the field scale and at large spatio-temporal scales. To date, most studies using satellite data have mapped the presence of cover crops, but have not identified specific cover crop species, which is important because cover crops of different plant functional types (e.g., legumes, grasses) perform different ecosystem functions. Here we use Sentinel-2 satellite data and a random forest classifier to map the cover crop species cereal rye and red clover, which represent grass and legume functional types, in the River Raisin watershed in southeastern Michigan. Our maps of agricultural landcover across this region, including the two cover crop species, had moderate to high accuracies, with an overall accuracy of 83%. Red clover and cereal rye achieved F1 scores that ranged from 0.7 to 0.77, and user's and producer's accuracies that ranged from 63.3% to 86.2%. The most common misclassification of cover crops was fallow fields with remaining crop stubble, which often looked similar because these cover crop species are typically planted within existing crop stubble, or interseeded into a grain crop. We found that red-edge bands and images from the end of April and early July were the most important for classification accuracy. Our results demonstrate the potential to map individual cover crop species using Sentinel-2 imagery, which is critical for understanding the environmental outcomes of increasing crop diversity on farms.

NDVI from Sentinel-2 Images for Crop Identification

Today, machine learning applied to remote sensing data is used for crop detection. This makes it possible to not only monitor crops but also to detect pests, a lack of irrigation, or other problems. For systems that require high accuracy in crop identification, a large amount of data is required to generate reliable models. The more plots of and data on crop evolution used over time, the more reliable the models. Here, a study has been carried out to analyse neural network models trained with the Sentinel satellite's 12 bands, compared to models that only use the NDVI, in order to choose the most suitable model in terms of the amount of storage, calculation time, accuracy, and precision. This study achieved a training time gain of 59.35% for NDVI models compared with 12-band models; however, models based on 12-band values are 1.96% more accurate than those trained with the NDVI alone when it comes to making predictions. The findings of this study could be of great interest to administrations, businesses, land managers, and researchers who use satellite image data mining techniques and wish to design an efficient system, particularly one with limited storage capacity and response times.

Natural grasslands across mainland France: A dataset including a 10 m raster and ground reference points

The data provided here include the first 10 m raster of natural grasslands across mainland France and related ground reference points. The latter consist of 1770 field observations that describe natural and artificial grasslands from respectively a compilation of hundreds of field-based vegetation maps and the European Union Land Parcel Identification System (LPIS). Based on analysis of aerial images, ground reference points were manually extracted from grassland polygons of the field-based vegetation maps and the LPIS within herbaceous areas larger than 30 × 30 m. The raster data of natural grasslands were derived from five annual 10 m land cover maps of France from 2016-2020. Pixels classified as ``grassland'' every year from 2016-2020 were considered natural grasslands, while those classified as ``crop'' at least once were considered artificial grasslands. Validation using the ground reference points revealed that natural and artificial grasslands were accurately mapped (overall accuracy = 86%). The ground reference points, publicly available in GeoJSON vector format, can be used as training or test samples for spatial modeling. The natural grassland map, publicly available in GeoTIFF raster format, can be used as a predictor variable for spatial modeling or as a base map for landscape ecology analyses.

Early-season and refined mapping of winter wheat based on phenology algorithms - a case of Shandong, China

Winter wheat is one of the major food crops in China, and timely and effective early-season identification of winter wheat is crucial for crop yield estimation and food security. However, traditional winter wheat mapping is based on post-season identification, which has a lag and relies heavily on sample data. Early-season identification of winter wheat faces the main difficulties of weak remote sensing response of the vegetation signal at the early growth stage, difficulty of acquiring sample data on winter wheat in the current season in real time, interference of crops in the same period, and limited image resolution. In this study, an early-season refined mapping method with winter wheat phenology information as priori knowledge is developed based on the Google Earth Engine cloud platform by using Sentinel-2 time series data as the main data source; these data are automated and highly interpretable. The normalized differential phenology index (NDPI) is adopted to enhance the weak vegetation signal at the early growth stage of winter wheat, and two winter wheat phenology feature enhancement indices based on NDPI, namely, wheat phenology differential index (WPDI) and normalized differential wheat phenology index (NDWPI) are developed. To address the issue of " different objects with the same spectra characteristics" between winter wheat and garlic, a plastic mulched index (PMI) is established through quantitative spectral analysis based on the differences in early planting patterns between winter wheat and garlic. The identification accuracy of the method is 82.64% and 88.76% in the early overwintering and regreening periods, respectively, These results were consistent with official statistics (R2 = 0.96 and 0.98, respectively). Generalization analysis demonstrated the spatiotemporal transferability of the method across different years and regions. In conclusion, the proposed methodology can obtain highly precise spatial distribution and planting area information of winter wheat 4_6 months before harvest. It provides theoretical and methodological guidance for early crop identification and has good scientific research and application value.

Graph Neural Network-Based Method of Spatiotemporal Land Cover Mapping Using Satellite Imagery

Multispectral satellite imagery offers a new perspective for spatial modelling, change detection and land cover classification. The increased demand for accurate classification of geographically diverse regions led to advances in object-based methods. A novel spatiotemporal method is presented for object-based land cover classification of satellite imagery using a Graph Neural Network. This paper introduces innovative representation of sequential satellite images as a directed graph by connecting segmented land region through time. The method's novel modular node classification pipeline utilises the Convolutional Neural Network as a multispectral image feature extraction network, and the Graph Neural Network as a node classification model. To evaluate the performance of the proposed method, we utilised EfficientNetV2-S for feature extraction and the GraphSAGE algorithm with Long Short-Term Memory aggregation for node classification. This innovative application on Sentinel-2 L2A imagery produced complete 4-year intermonthly land cover classification maps for two regions: Graz in Austria, and the region of Portorož, Izola and Koper in Slovenia. The regions were classified with Corine Land Cover classes. In the level 2 classification of the Graz region, the method outperformed the state-of-the-art UNet model, achieving an average F1-score of 0.841 and an accuracy of 0.831, as opposed to UNet's 0.824 and 0.818, respectively. Similarly, the method demonstrated superior performance over UNet in both regions under the level 1 classification, which contains fewer classes. Individual classes have been classified with accuracies up to 99.17%.

Towards precision forestry: Drought response from remote sensing-based disturbance monitoring and fine-scale soil information in Central Europe

Prolonged drought and susceptibility to biotic stressors induced an extensive calamity in Norway spruce (Picea abies (L.) Karst.) and widespread crown defoliation in European beech (Fagus sylvatica L.) in Central Europe. For future management decisions, it is crucial to link changes in canopy cover to site conditions. However, current knowledge on the role of soil properties for drought-induced forest disturbance is limited due to the scarcity and low spatial resolution of soil information. We present a fine-scale assessment on the role of soil properties for forest disturbance in Norway spruce and European beech derived from optical remote sensing. A forest disturbance modeling framework based on Sentinel-2 time series was applied on 340 km² in low mountain ranges of Central Germany. Spatio-temporal information on forest disturbance was calculated at 10 m spatial resolution in the period 2019-2021 and intersected with high-resolution soil information (1:10,000) based on roughly 2850 soil profiles. We found distinct differences in disturbed area, depending on soil type, texture, stoniness, effective rooting depth and available water capacity (AWC). For spruce, we found a polynomial relationship between AWC (R² = 0.7) and disturbance, with highest disturbed area (65 %) for AWC between 90 and 160 mm. Interestingly, we found no evidence for generally higher disturbance on shallow soils, although stands on the deepest soils were significantly less affected. Noteworthy, sites affected first did not necessarily exhibit highest proportions of disturbed area post-drought, indicating recovery or adaptation. We conclude that site- and species-specific understanding of drought impacts benefits from a combination of remote sensing and fine-scale soil information. Since our approach revealed which sites were affected first and most, it qualifies for prioritizing in situ monitoring activities to most vulnerable stands in acute drought conditions as well as for developing long-term strategies for reforestation and site-specific risk assessment for precision forestry.

Semi-empirical models for remote estimating colored dissolved organic matter (CDOM) in a productive tropical estuary

Inland waters are important components of the global carbon cycle as they regulate the flow of terrestrial carbon to the oceans. In this context, remote monitoring of Colored Dissolved Organic Matter (CDOM) allows for analyzing the carbon content in aquatic systems. In this study, we develop semi-empirical models for remote estimation of the CDOM absorption coefficient at 400 nm (a_CDOM) in a tropical estuarine-lagunar productive system using spectral reflectance data. Two-band ratio models usually work well for this task, but studies have added more bands to the models to reduce interfering signals, so in addition to the two-band ratio models, we tested three- and four-band ratios. We used a genetic algorithm (GA) to search for the best combination of bands, and found that adding more bands did not provide performance gains, showing that the proper choice of bands is more important. NIR-Green models outperformed Red-Blue models. A two-band NIR-Green model showed the best results (R² = 0.82, RMSE = 0.22 m^-1, and MAPE = 5.85%) using field hyperspectral data. Furthermore, we evaluated the potential application for Sentinel-2 bands, especially using the B5/B3, Log(B5/B3) and Log(B6/B2) band ratios. However, it is still necessary to further explore the influence of atmospheric correction (AC) to estimate the a_CDOM using satellite data.

Ecodatacube.eu: analysis-ready open environmental data cube for Europe

The article describes the production steps and accuracy assessment of an analysis-ready, open-access European data cube consisting of 2000-2020+ Landsat data, 2017-2021+ Sentinel-2 data and a 30 m resolution digital terrain model (DTM). The main purpose of the data cube is to make annual continental-scale spatiotemporal machine learning tasks accessible to a wider user base by providing a spatially and temporally consistent multidimensional feature space. This has required systematic spatiotemporal harmonization, efficient compression, and imputation of missing values. Sentinel-2 and Landsat reflectance values were aggregated into four quarterly averages approximating the four seasons common in Europe (winter, spring, summer and autumn), as well as the 25th and 75th percentile, in order to retain intra-seasonal variance. Remaining missing data in the Landsat time-series was imputed with a temporal moving window median (TMWM) approach. An accuracy assessment shows TMWM performs relatively better in Southern Europe and lower in mountainous regions such as the Scandinavian Mountains, the Alps, and the Pyrenees. We quantify the usability of the different component data sets for spatiotemporal machine learning tasks with a series of land cover classification experiments, which show that models utilizing the full feature space (30 m DTM, 30 m Landsat, 30 m and 10 m Sentinel-2) yield the highest land cover classification accuracy, with different data sets improving the results for different land cover classes. The data sets presented in the article are part of the EcoDataCube platform, which also hosts open vegetation, soil, and land use/land cover (LULC) maps created. All data sets are available under CC-BY license as Cloud-Optimized GeoTIFFs (ca. 12 TB in size) through SpatioTemporal Asset Catalog (STAC) and the EcoDataCube data portal.

Comparison of PlanetScope, Sentinel-2, and landsat 8 data in soybean yield estimation within-field variability with random forest regression

Accurate timely and early-season crop yield estimation within the field variability is important for precision farming and sustainable management applications. Therefore, the ability to estimate the within-field variability of grain yield is crucial for ensuring food security worldwide, especially under climate change. Several Earth observation systems have thus been developed to monitor crops and predict yields. Despite this, new research is required to combine multiplatform data integration, advancements in satellite technologies, data processing, and the application of this discipline to agricultural practices. This study provides further developments in soybean yield estimation by comparing multisource satellite data from PlanetScope (PS), Sentinel-2 (S2), and Landsat 8 (L8) and introducing topographic and meteorological variables. Herein, a new method of combining soybean yield, global positioning systems, harvester data, climate, topographic variables, and remote sensing images has been demonstrated. Soybean yield shape points were obtained from a combine-harvester-installed GPS and yield monitoring system from seven fields over the 2021 season. The yield estimation models were trained and validated using random forest, and four vegetation indices were tested. The result showed that soybean yield can be accurately predicted at 3-, 10-, and 30-m resolutions with mean absolute error (MAE) value of 0.091 t/ha for PS, 0.118 t/ha for S2, and 0.120 t/ha for L8 data (root mean square error (RMSE) of 0.111, 0.076). The combination of the environmental data with the original bands provided further improvements and an accurate yield estimation model within the soybean yield variability with MAE of 0.082 t/ha for PS, 0.097 t/ha for S2, and 0.109 t/ha for L8 (RMSE of 0.094, 0.069, and 0.108 t/ha). The results showed that the optimal date to predict the soybean yield within the field scale was approximately 60 or 70 days before harvesting periods during the beginning bloom stage. The developed model can be applied for other crops and locations when suitable training yield data, which are critical for precision farming, are available.

Evaluating the relationship between groundwater quality and land use in an urbanized watershed

Understanding the impact of urbanization on groundwater quality is critical. Effective water management requires understanding the relationship between land use and water quality. The study's goals were to compare the effects of land use, identify the types of land that impact hydrochemistry, and define how different land use affects water quality. For this purpose, the comparative relationship between groundwater quality, land use classes and landscape metrics were established for the years 2016 and 2021. Water samples were collected from 42 wells, and different hydro-chemical variables were considered to calculate the water quality index (WQI). The WQI value in 2016 ranged from 26.49 to 151.03 and 29.65 to 155.62 in 2021. The results indicate that the water quality in most parts of the study area is moderate for drinking and domestic purpose use. The google earth engine platform was used and radiometrically corrected and orthorectified Sentinel-2 satellite images were processed to classify land use classes for selected years. Five buffer zones were established within a 2-km watershed along each well site, and the effects of land use types and landscape metrics on water quality in the buffer zones were analyzed. Results revealed that the effects of land use types on water quality were mainly reflected in buffer 1 (B1), buffer 4 (B4), buffer 5 (B5) in 2016 and B1, buffer 3 (B3), and B5 in 2021. The impacts of landscape-level metrics on water quality are mainly reflected in buffer 2 (B2) and B3 in 2021, while at the class-level, they are mainly reflected in B1 and B4 in 2021. The redundancy analysis revealed that different hydro-chemical variables behaved differently with the land use classes and landscape metrics in the various buffer zones.

A New and Automated Method for Improving Georeferencing in Nighttime Thermal ECOSTRESS Imagery

Georeferencing accuracy plays a crucial role in providing high-quality ready-to-use remote sensing data. The georeferencing of nighttime thermal satellite imagery conducted by matching to a basemap is challenging due to the complexity of thermal radiation patterns in the diurnal cycle and the coarse resolution of thermal sensors in comparison to sensors used for imaging in the visual spectral range (which is typically used for creating basemaps). The presented paper introduces a novel approach for the improvement of the georeferencing of nighttime thermal ECOSTRESS imagery: an up-to-date reference is created for each to-be-georeferenced image, derived from land cover classification products. In the proposed method, edges of water bodies are used as matching objects, since water bodies exhibit a relatively high contrast with adjacent areas in nighttime thermal infrared imagery. The method was tested on imagery of the East African Rift and validated using manually set ground control check points. The results show that the proposed method improves the existing georeferencing of the tested ECOSTRESS images by 12.0 pixels on average. The strongest source of uncertainty for the proposed method is the accuracy of cloud masks because cloud edges can be mistaken for water body edges and included in fitting transformation parameters. The georeferencing improvement method is based on the physical properties of radiation for land masses and water bodies, which makes it potentially globally applicable, and is feasible to use with nighttime thermal infrared data from different sensors.

Identification of time-varying wetlands neglected in Pakistan through remote sensing techniques

Aside from Ramsar Convention awareness programs, the concept of wetlands is mostly ignored in developing countries. Wetland ecosystems are essential to hydrological cycles, ecosystem diversity, climatic change, and economic activity. Under the Ramsar Convention, there are 2414 wetlands that are internationally recognized, and Pakistan is home to 19 of them. The major goal of this study is to use the satellite image technology to locate Pakistan's underutilized wetlands (Borith, Phander, Upper Kachura, Satpara, and Rama Lakes). The other goals are to understand how these wetlands are affected by climate change, ecosystem change, and water quality. We used analytical techniques including supervised classification and Tasseled Cap Wetness to identify the wetlands. To find changes caused by climate change, Quick Bird high-resolution images was used to create the change detection index. Tasseled Cap Greenness and the Normalized Difference Turbidity Index were also used to assess the water quality and changes in the ecology in these wetlands. Sentinel-2 was used to analyze data from 2010 and 2020. ASTER DEM was also used to do a watershed analysis. The land surface temperature (°C) of a few selected wetlands was calculated using Modis data. Rainfall (mm) data was taken from PERSIANN (precipitation estimation from remotely sensed information using artificial neural networks) databases. Results indicated that in 2010, the water content of Borith, Phander, Upper Kachura, Satpara, and Rama Lakes was 22.83%, 20.82%, 22.26%, 24.40%, and 22.91%. While in 2020, these lakes' water ratios are 21.33%, 20.65%, 21.76%, 23.85%, and 22.59%, respectively. Therefore, the competent authorities must take precautions to ensure that these wetlands are preserved in the future in order to improve the dynamics of the ecosystem.

Mapping global distribution of mangrove forests at 10-m resolution

Mangrove forests deliver incredible ecosystem goods and services and are enormously relevant to sustainable living. An accurate assessment of the global status of mangrove forests warrants the necessity of datasets with sufficient information on spatial distributions and patch patterns. However, existing datasets were mostly derived from ∼30 m resolution satellite imagery and used pixel-based image classification methods, which lacked spatial details and reasonable geo-information. Here, based on Sentinel-2 imagery, we created a global mangrove forest dataset at 10-m resolution, namely, High-resolution Global Mangrove Forests (HGMF_2020), using object-based image analysis and random forest classification. We then analyzed the status of global mangrove forests from the perspectives of conservation, threats, and resistance to ocean disasters. We concluded the following: (1) globally, there were 145,068 km² mangrove forests in 2020, among which Asia contained the largest coverage (39.2%); at the country level, Indonesia had the largest amount of mangrove forests, followed by Brazil and Australia. (2) Mangrove forests in South Asia were estimated to be in the better status due to the higher proportion of conservation and larger individual patch size; in contrast, mangrove forests in East and Southeast Asia were facing intensive threats. (3) Nearly, 99% of mangrove forest areas had a patch width greater than 100 m, suggesting that nearly all mangrove forests were efficient in reducing coastal wave energy and impacts. This study reports an innovative and up-to-date dataset and comprehensive information on mangrove forests status to contribute to related research and policy implementation, especially for supporting sustainable development.

SNOWED: Automatically Constructed Dataset of Satellite Imagery for Water Edge Measurements

Monitoring the shoreline over time is essential to quickly identify and mitigate environmental issues such as coastal erosion. Monitoring using satellite images has two great advantages, i.e., global coverage and frequent measurement updates; but adequate methods are needed to extract shoreline information from such images. To this purpose, there are valuable non-supervised methods, but more recent research has concentrated on deep learning because of its greater potential in terms of generality, flexibility, and measurement accuracy, which, in contrast, derive from the information contained in large datasets of labeled samples. The first problem to solve, therefore, lies in obtaining large datasets suitable for this specific measurement problem, and this is a difficult task, typically requiring human analysis of a large number of images. In this article, we propose a technique to automatically create a dataset of labeled satellite images suitable for training machine learning models for shoreline detection. The method is based on the integration of data from satellite photos and data from certified, publicly accessible shoreline data. It involves several automatic processing steps, aimed at building the best possible dataset, with images including both sea and land regions, and correct labeling also in the presence of complicated water edges (which can be open or closed curves). The use of independently certified measurements for labeling the satellite images avoids the great work required to manually annotate them by visual inspection, as is done in other works in the literature. This is especially true when convoluted shorelines are considered. In addition, possible errors due to the subjective interpretation of satellite images are also eliminated. The method is developed and used specifically to build a new dataset of Sentinel-2 images, denoted SNOWED; but is applicable to different satellite images with trivial modifications. The accuracy of labels in SNOWED is directly determined by the uncertainty of the shoreline data used, which leads to sub-pixel errors in most cases. Furthermore, the quality of the SNOWED dataset is assessed through the visual comparison of a random sample of images and their corresponding labels, and its functionality is shown by training a neural model for sea-land segmentation.

Learning-Based Near-Infrared Band Simulation with Applications on Large-Scale Landcover Classification

Multispectral sensors are important instruments for Earth observation. In remote sensing applications, the near-infrared (NIR) band, together with the visible spectrum (RGB), provide abundant information about ground objects. However, the NIR band is typically not available on low-cost camera systems, which presents challenges for the vegetation extraction. To this end, this paper presents a conditional generative adversarial network (cGAN) method to simulate the NIR band from RGB bands of Sentinel-2 multispectral data. We adapt a robust loss function and a structural similarity index loss (SSIM) in addition to the GAN loss to improve the model performance. With 45,529 multi-seasonal test images across the globe, the simulated NIR band had a mean absolute error of 0.02378 and an SSIM of 89.98%. A rule-based landcover classification using the simulated normalized difference vegetation index (NDVI) achieved a Jaccard score of 89.50%. The evaluation metrics demonstrated the versatility of the learning-based paradigm in remote sensing applications. Our simulation approach is flexible and can be easily adapted to other spectral bands.

Remote Sensing Temporal Reconstruction of the Flooded Area in "Tablas de Daimiel" Inland Wetland 2000-2021

Tablas de Daimiel National Park (TDNP) is a unique inland wetland located in the Mancha plain (Spain). It is recognized at the international level, and it is protected by different figures, such as Biosphere Reserve. However, this ecosystem is endangered due to aquifer overexploitation, and it is at risk of losing its protection figures. The objective of our study is to analyze the evolution of the flooded area between the year 2000 and 2021 by Landsat (5, 7 and 8) and Sentinel-2 images, and to assess the TDNP state through an anomaly analysis of the total water body surface. Several water indices were tested, but the NDWI index for Sentinel-2 (threshold -0.20), the MNDWI for Landsat-5 (threshold -0.15), and the MNDWI for Landsat-8 (threshold -0.25) showed the highest accuracy to calculate the flooded surface inside the protected area's limits. During the period 2015-2021, we compared the performance of Landsat-8 and Sentinel-2 and an R² value of 0.87 was obtained for this analysis, indicating a high correspondence between both sensors. Our results indicate a high variability of the flooded areas during the analyzed period with significant peaks, the most notorious in the second quarter of 2010. Minimum flooded areas were observed with negative precipitation index anomalies since fourth quarter of 2004 to fourth quarter of 2009. This period corresponds to a severe drought that affected this region and caused important deterioration. No significant correlation was observed between water surface anomalies and precipitation anomalies, and the significant correlation with flow and piezometric anomalies was moderate. This can be explained because of the complexity of water uses in this wetland, which includes illegal wells and the geological heterogeneity.

Iranian wetland inventory map at a spatial resolution of 10 m using Sentinel-1 and Sentinel-2 data on the Google Earth Engine cloud computing platform

Detailed wetland inventories and information about the spatial arrangement and the extent of wetland types across the Earth's surface are crucially important for resource assessment and sustainable management. In addition, it is crucial to update these inventories due to the highly dynamic characteristics of the wetlands. Remote sensing technologies capturing high-resolution and multi-temporal views of landscapes are incredibly beneficial in wetland mapping compared to traditional methods. Taking advantage of the Google Earth Engine's computational power and multi-source earth observation data from Sentinel-1 multi-spectral sensor and Sentinel-2 radar, we generated a 10 m nationwide wetlands inventory map for Iran. The whole country is mapped using an object-based image processing framework, containing SNIC superpixel segmentation and a Random Forest classifier that was performed for four different ecological zones of Iran separately. Reference data was provided by different sources and through both field and office-based methods. Almost 70% of this data was used for the training stage and the other 30% for evaluation. The whole map overall accuracy was 96.39% and the producer's accuracy for wetland classes ranged from nearly 65 to 99%. It is estimated that 22,384 km² of Iran are covered with water bodies and wetland classes, and emergent and shrub-dominated are the most common wetland classes in Iran. Considering the water crisis that has been started in Iran, the resulting ever-demanding map of Iranian wetland sites offers remarkable information about wetland boundaries and spatial distribution of wetland species, and therefore it is helpful for both governmental and commercial sectors.

Mapping dinoflagellate blooms (Noctiluca and Alexandrium) in aquaculture production areas in the NW Iberian Peninsula with the Sentinel-2/3 satellites

The Galician Rías (northwestern Spain) are periodically affected by harmful algal blooms (HABs), mostly dinoflagellates, which pose a challenge to aquaculture activities due to the accumulation of biotoxins in shellfish. Typically, reddish blooms in the Rías are associated with non-toxic species like Noctiluca scintillans, with a few exceptions such as Alexandrium minutum, a producer of paralytic shellfish toxins (PST). Here, a useful approach is presented for monitoring reddish blooms through satellite imagery based on three case studies, two of them belonged to monospecific blooms of red Noctiluca scintillans, and the third to a bloom of Alexandrium spp. dominated by A. tamarense. In every case, a propulsive index was evaluated using Sentinel-2A/B satellites, which provide high spatial and spectral resolutions, combined with adequate atmospheric and sunglint correction by using the ACOLITE and C2RCC processors. This approach offers a simple and feasible method to accurately and timely map blooms of red N. scintillans and Alexandrium spp. in the study area, useful to detect the distribution of reddish blooms with synoptic observations for monitoring and aquaculture management purposes. Conversely, Sentinel-3A/B satellites with a relatively coarser spatial resolution, lacking adequate visualization and mapping of the extent of small blooms, did not accurately detect bloom footprints in the coastal bay region, although this sensor displays a set of suitable multispectral bands.

Mapping intertidal macrophytes in fjords in Southwest Greenland using Sentinel-2 imagery

Changes in the distribution of coastal macrophytes in Greenland, and elsewhere in the Arctic are difficult to quantify as the region remains challenging to access and monitor. Satellite imagery, in particular Sentinel-2 (S2), may enable large-scale monitoring of coastal areas in Greenland but its use is impacted by the optically complex environments and the scarcity of supporting data in the region. Additionally, the canopies of the dominant macrophyte species in Greenland do not extend to the sea surface, limiting the use of indices that exploit the reflection of near-infrared radiation by vegetation due to its absorption by seawater. Three hypotheses are tested: I) 10-m S2 imagery and commonly used detection methods can identify intertidal macrophytes that are exposed at low tide in an optically complex fjord system in Greenland impacted by marine and land terminating glaciers; II) detached and floating macrophytes accumulate in patches that are sufficiently large to be detected by 10-m S2 images; III) iceberg scour and/or turbid meltwater runoff shape the spatial distribution of intertidal macroalgae in fjord systems with marine-terminating glaciers. The NDVI produced the best results in optically complex fjord systems in Greenland. 12 km² of exposed intertidal macrophytes were identified in the study area at low tide. Floating mats of macrophytes ranged in area from 400 m² to 326,800 m² and were most common at the mouth of the fjord. Icebergs and turbidity appear to play a role in structuring the distribution of intertidal macrophytes and the retreat of marine terminating glaciers could allow macrophytes cover to expand. The challenges and solutions presented here apply to most fjords in Greenland and, therefore, the methodology may be extended to produce a Greenland-wide estimate of intertidal macrophytes.