Sentinel-1 Datasets Research Articles

Transfer learning approach based on satellite image time series for the crop classification problem

This paper presents a transfer learning approach to the crop classification problem based on time series of images from the Sentinel-2 dataset labeled for two regions: Brittany (France) and Vojvodina (Serbia). During preprocessing, cloudy images are removed from the input data, the time series are interpolated over the time dimension, and additional remote sensing indices are calculated. We chose TransformerEncoder as the base model for knowledge transfer from source to target domain with French and Serbian data, respectively. Even more, the accuracy of the base model with the preprocessing step is improved by 2% when trained and evaluated on the French dataset. The transfer learning approach with fine-tuning of the pre-trained weights on the French dataset outperformed all other methods in terms of overall accuracy 0.94 and mean class recall 0.907 on the Serbian dataset. Our partially fine-tuned model improved recall of crop types that were poorly classified by the base model. In the case of sugar beet, class recall is improved by 85.71%.

Identification of cotton and corn plant areas by employing deep transformer encoder approach and different time series satellite images: A case study in Diyarbakir, Turkey

It is very important to determine the crops in the agricultural field in a short time and accurately. Thanks to the satellite images obtained from remote sensing sensors, information can be obtained on many subjects such as the detection and development of agricultural products and annual product forecasting. In this study, it is aimed to automatically detect agricultural crops (corn and cotton) by using Sentinel-1 and Landsat-8 satellite image indexes via a new deep learning approach (Deep Transformer Encoder).This work was carried out in several stages, respectively. In the first stage, a pilot area was determined to obtain Sentinel-1 and Landsat-8 satellite images of agricultural crops used in this study. In the second stage, the coordinates of 100 sample points from this pilot area were taken with the help of GPS and these coordinates were then transferred to Sentinel-1 and Landsat-8 satellite images. In the next step, reflection and backscattering values were obtained from the pixels of the satellite images corresponding to the sample points of these agricultural crops. While creating the data sets of satellite images, the months of June, July, August and September for the years 2016–2021, when the development and harvesting times of agricultural products are close to each other, were preferred. The image data set used in the study consists of a total of 434 images for Sentinel-1 satellite and a total of 693 images for Landsat-8.At the last stage, the datasets obtained from different satellite images were evaluated in three different categories for crop identification with the aid of Deep Transformer Encoder approach. These are: (1-) Crop identification with only Sentinel-1 dataset, (2-) Crop identification only with Landsat-8 dataset, (3-) Crop identification with both Sentinel-1 and Landsat-8 datasets. The results showed that 85%, 95% and 87.5% accuracy values were obtained from the band parameters of Sentinel-1 dataset, Landsat-8 dataset and Sentinel-1&Landsat-8 datasets, respectively.

Satellite-derived bathymetry combined with Sentinel-2 and ICESat-2 datasets using machine learning

Most satellite-derived bathymetry (SDB) methods developed thus far from passive remote sensing data have required in situ water depth, thus limiting their utility in areas with no in situ data. Recently, new Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) observations have shown great potential in providing high-precision a priori water depth benefits from range-resolved lidar. In this study, we propose a combined active and passive remote sensing SDB method using only satellite data. An adaptive ellipse DBSCAN (AE-DBSCAN) algorithm is introduced to derive a priori bathymetric data from ICESat-2 data to automatically adapt to the terrain change complexity, and then we use these a priori bathymetric data in Sentinel-2 images to help build a model between remote sensing reflectance (Rrs) and water depth. Three machine learning (ML) methods are then employed, and the performances compared with conventional empirical SDB models are presented. After that, the results using different Sentinel-2 Rrs band combinations and the effects with and without atmospheric correction on ML-based SDB are discussed. The results showed that our AE-DBSCAN method performs better than the standard DBSCAN method, and the ML-based SDB can achieve an overall RMSE of less than 1.5 m in St. Thomas better than the traditional SDB method. They also indicate that ML-based SDB can obtain a relatively high-precision water depth without atmospheric correction, which helps to improve processing efficiency by avoiding a complex atmospheric correction process.

Synergy of Sentinel-1 and Sentinel-2 Time Series for Cloud-Free Vegetation Water Content Mapping with Multi-Output Gaussian Processes

Optical Earth Observation is often limited by weather conditions such as cloudiness. Radar sensors have the potential to overcome these limitations, however, due to the complex radar-surface interaction, the retrieving of crop biophysical variables using this technology remains an open challenge. Aiming to simultaneously benefit from the optical domain background and the all-weather imagery provided by radar systems, we propose a data fusion approach focused on the cross-correlation between radar and optical data streams. To do so, we analyzed several multiple-output Gaussian processes (MOGP) models and their ability to fuse efficiently Sentinel-1 (S1) Radar Vegetation Index (RVI) and Sentinel-2 (S2) vegetation water content (VWC) time series over a dry agri-environment in southern Argentina. MOGP models not only exploit the auto-correlations of S1 and S2 data streams independently but also the inter-channel cross-correlations. The S1 RVI and S2 VWC time series at the selected study sites being the inputs of the MOGP models proved to be closely correlated. Regarding the set of assessed models, the Convolutional Gaussian model (CONV) delivered noteworthy accurate data fusion results over winter wheat croplands belonging to the 2020 and 2021 campaigns (NRMSEwheat2020 = 16.1%; NRMSEwheat2021 = 10.1%). Posteriorly, we removed S2 observations from the S1 & S2 dataset corresponding to the complete phenological cycles of winter wheat from September to the end of December to simulate the presence of clouds in the scenes and applied the CONV model at the pixel level to reconstruct spatiotemporally-latent VWC maps. After applying the fusion strategy, the phenology of winter wheat was successfully recovered in the absence of optical data. Strong correlations were obtained between S2 VWC and S1 & S2 MOGP VWC reconstructed maps for the assessment dates (R2¯wheat−2020 = 0.95, R2¯wheat−2021 = 0.96). Altogether, the fusion of S1 SAR and S2 optical EO data streams with MOGP offers a powerful innovative approach for cropland trait monitoring over cloudy high-latitude regions.

Interferometric Synthetic Aperture Radar versus Inclinometer Ground Surface Deformations at Active Landslide

Rockfalls and landslides commonly occur in the Hawaiian islands. When these events occur along a thoroughfare, they can result in the loss of lives, cause significant damage to infrastructure, and force full or partial road closures, resulting in traffic congestion and commuting delays, sometimes for months. The Hawaii Department of Transportation (HDOT) is interested in historic slope movements along its roads and highways, as this history can guide in prioritizing spending on slope remediation. This has prompted HDOT to be interested in the use of interferometric synthetic aperture radar (InSAR) imagery for detecting historical slope movements and identifying areas that might pose concern for landslide or rockfall activity. To provide a proof of concept for the implementation of this technology in Hawaii, an InSAR case study of a landslide on the island of O’ahu was conducted to compare the results with available inclinometer data. Two-dimensional decomposition was implemented for opposing orbits of descending and ascending Sentinel-1 datasets. Overall, the results show that when both ascending and descending datasets are used to derive line-of-sight displacements that are resolved in a direction along the direction of movement (typically perpendicular to the slope contours), InSAR analysis can effectively capture inclinometer trends in areas experiencing relatively no or little displacement over time (<30 mm/year) but the accuracy diminishes in fast-moving slides (≥270 mm/year).

Performance of Multiple Models for Estimating Rodent Activity Intensity in Alpine Grassland Using Remote Sensing

Rodents are a vital part of the natural succession chain of the alpine grassland ecosystem, and rodent activities have an important impact on alpine grassland ecology. Moderate rodent population activities positively improve soil permeability, promote nutrient cycling, and promote biodiversity. However, too much rodent population or excessive activity intensity will bring negative effects on the ecological environment. Therefore, it is of great significance to accurately grasp the rodent activity intensity (RAI) in alpine grassland to cope with the changes in rodent populations and maintain the stability of the alpine grassland ecosystem. The Zoige alpine grassland was used as the study area in this study. In addition, UAV was sent to sample the rodent activity area in the alpine grassland. With the aid of field survey data, the surface information of rodent activity in the experimental area was identified, and the RAI index in the sample plot was calculated. Then, based on Sentinel-2A satellite remote sensing multi-spectral data and spectral index, multiple linear regression (MLR), multi-layer perceptron neural networks (MPL neural nets), random forest (RF), and support vector regression (SVR) were used to construct four models for RAI and Sentinel-2 datasets. The accuracy of the four models was compared and analyzed. The results showed that the RF model had the highest prediction accuracy (R2 = 0.8263, RWI = 0.8210, LCCC = 0.8916, RMSE = 0.0840, MAE = 0.0549), followed by the SVR model, the MLP neural nets model, and the MLR model. Overall, the nonlinear relationship between rodent activity intensity and satellite remote sensing images is obvious. Machine learning with strong nonlinear fitting ability can better characterize the RAI in alpine grassland. The RF model, with the best accuracy, can quantitatively estimate RAI in the alpine grassland, providing theoretical and technical support for monitoring RAI and rodent control in the alpine grassland.

Assessing Glacier Boundaries in the Ala-Archa Valley of Kyrgyzstan by Using Sentinel-1 SAR Dataset and High-Resolution UAV Imagery

The significant retreat of glaciers in terms of climate change compels researchers to increase the frequency of studies regarding the transformations occurring in glacier boundaries. In this study, we provided glacier area delineation of Ala-Archa valley glaciers by using a Sentinel-1 SAR dataset and the InSAR Coherence technique. Since glaciers have specific patterns of movement, the low coherence method signals the presence of ice. The analysis used the pair of Sentinel-1 datasets for the summer, to ensure the lowest coherence and provide an areal estimation during the peak of ablation. The independence of the SAR images from cloud and light conditions permits us to acquire the images in a timely manner, which highly affects the results of glacier monitoring. This method has shown high potential in the mapping of debris-covered ice and the indication of dead ice. To identify and separate areas of low coherence, such as glacier lakes and unstable slopes, we used object-based mapping by using the geomorphological features of the ice. In this study, we defined a coherence value of less than 0.3 in the glacier area. Our research identified a number of 56 glaciers within the study area of 31.45 km2 and obtained highly accurate glacier maps for the glaciers with a smooth terminus. The analysis shows that automatic and manual delineation of the glaciers’ boundaries have certain limitations, but using the advantages of both scientific approaches, further studies will generate more accurate results.

High Spatiotemporal Flood Monitoring Associated with Rapid Lake Shrinkage Using Planet Smallsat and Sentinel-1 Data

High spatiotemporal flood monitoring is critical for flood control, mitigation, and management purposes in areas where tectonic and geological events significantly exacerbate flood disasters. For example, the rapid lake shrinkage resulting from the transformations of enclosed seas into lakes by the rapid land movement in the collision zone dramatically increases the flood risks in Indonesia, which requires frequent and detailed monitoring and assessment. This study primarily quantified the detailed flood disasters associated with the rapid lake shrinkage in Gorontalo Regency in Gorontalo Province, Indonesia using high spatiotemporal monitoring with a combination of PlanetScope smallsat constellations, Sentinel-1, and surface water datasets. Based on the findings that indicated its volume, distribution, pace, and pattern, the flood event that occurred in Gorontalo in November 2022 was demonstrated within a short interval of 2–12 days. The results also indicate both direct and indirect floodwater overflow from different water resources. Combining these results with the surface water occurrences from 1984 to 2021, our findings reveal the historical major flood-prone areas associated with the rapid lake shrinkage. These findings are expected to aid in the timely high spatiotemporal monitoring of rapid environmental change-induced flood disasters, even in tropical regions with high cloud coverage. Furthermore, these are also expected to be integrated into the flood hazard mitigation and management strategies associated with local-specific tectonic and geological systems.

Synergism of Multi-Modal Data for Mapping Tree Species Distribution—A Case Study from a Mountainous Forest in Southwest China

Accurately mapping tree species is crucial for forest management and conservation. Most previous studies relied on features derived from optical imagery, and digital elevation data and the potential of synthetic aperture radar (SAR) imagery and other environmental factors have, generally, been underexplored. Therefore, the aim of this study is to evaluate the potential of fusing freely available multi-modal data for accurately mapping tree species. Sentinel-2, Sentinel-1, and various environmental datasets over a large mountainous forest in Southwest China were obtained and analyzed using Google Earth Engine (GEE). Seven data cases considering the individual or joint performance of different features, and four additional cases considering a novel clustering-based feature selection method, were analyzed. All 11 cases were assessed using three machine learning algorithms, including random forest (RF), support vector machine (SVM), and extreme gradient boosting tree (XGBoost). The best performance, with an overall accuracy of 77.98%, was attained from the case with all features and the random forest classifier. Sentinel-2 data alone exhibited similar performance as environmental data in terms of overall accuracy. Similar species, such as oak and birch, cannot be spectrally discriminated based on Sentinel-2-based features alone. The addition of SAR features improved discrimination, especially when distinguishing between some coniferous and deciduous species, but also decreased accuracy for oak. The analysis based on different data cases and feature importance rankings indicated that environmental features are important. The random forest outperformed other models, and a better prediction was achieved for planted tree species compared to that for the natural forest. These results suggest that accurately mapping tree species over large mountainous areas is feasible with freely accessible multi-modal data, especially when considering environmental factors.

Soil Erosion Modelling and Accumulation Using RUSLE and Remote Sensing Techniques: Case Study Wadi Baysh, Kingdom of Saudi Arabia

This study examines the sediment retention in Wadi Baysh using the Revised Universal Soil Loss Equation (RUSLE) and TerrSet models, accompanied by integrated remote sensing and Geographic Information System (GIS) techniques. The contribution of this study is mainly associated with the application of TerrSet integrated with high resolution datasets to precisely estimate sediments load, which provide useful information to operate dams and improve the operational efficiency of dams. The Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR) data are utilized to delineate the basin and have been used as an input to the TerrSet model. The rainfall erosivity (R factor) was calculated using the Climate Hazards Center Infrared Precipitation with Stations (CHIRPS) in the research area during 2015–2020. The soil erodibility (K factor) and LULC categorization are calculated using the digital soil map of the world (DSMW) and Sentinel-2 datasets, respectively. The R factor calculated for Wadi Baysh ranges between 91.35 and 115.95 MJ mm/ha/h/year, while the estimated K factor ranges from 0.139 to 0.151 t ha h/ha M. The Support Vector Machine (SVM) method categorized LULC of the study area into four major classes including barren land (81% of the total area), built-up area (11%), vegetation (8%), and water bodies (1%). Results from the sediment retention module (TerrSet) indicated that each year, 57.91 million tons of soil loss occurred in the basin. The data show that soil loss is greater in the northeast and south, whereas it is typical in the middle of Wadi Baysh. It is concluded from the current analyses that the dam lake of Wadi Baysh, located downstream, will be filled soon in the coming few years if sediment loads are carried to the lake at the same rate. Surface dam operators can obtain a full understanding of sedimentation and take proactive measures to reduce its influence on dam operations by leveraging TerrSet’s sophisticated capabilities.

HyFormer: Hybrid Transformer and CNN for Pixel-Level Multispectral Image Land Cover Classification.

To effectively solve the problems that most convolutional neural networks cannot be applied to the pixelwise input in remote sensing (RS) classification and cannot adequately represent the spectral sequence information, we propose a new multispectral RS image classification framework called HyFormer based on Transformer. First, a network framework combining a fully connected layer (FC) and convolutional neural network (CNN) is designed, and the 1D pixelwise spectral sequences obtained from the fully connected layers are reshaped into a 3D spectral feature matrix for the input of CNN, which enhances the dimensionality of the features through FC as well as increasing the feature expressiveness, and can solve the problem that 2D CNN cannot achieve pixel-level classification. Secondly, the features of the three levels of CNN are extracted and combined with the linearly transformed spectral information to enhance the information expression capability, and also used as the input of the transformer encoder to improve the features of CNN using the powerful global modelling capability of the Transformer, and finally the skip connection of the adjacent encoders to enhance the fusion between different levels of information. The pixel classification results are obtained by MLP Head. In this paper, we mainly focus on the feature distribution in the eastern part of Changxing County and the central part of Nanxun District, Zhejiang Province, and conduct experiments based on Sentinel-2 multispectral RS images. The experimental results show that the overall accuracy of HyFormer for the study area classification in Changxing County is 95.37% and that of Transformer (ViT) is 94.15%. The experimental results show that the overall accuracy of HyFormer for the study area classification in Nanxun District is 95.4% and that of Transformer (ViT) is 94.69%, and the performance of HyFormer on the Sentinel-2 dataset is better than that of the Transformer.

W-NetPan: Double-U network for inter-sensor self-supervised pan-sharpening

The increasing availability of remote sensing data allows dealing with spatial-spectral limitations by means of pan-sharpening methods. However, fusing inter-sensor data poses important challenges, in terms of resolution differences, sensor-dependent deformations and ground-truth data availability, that demand more accurate pan-sharpening solutions. In response, this paper proposes a novel deep learning-based pan-sharpening model which is termed as the double-U network for self-supervised pan-sharpening (W-NetPan). In more details, the proposed architecture adopts an innovative W-shape that integrates two U-Net segments which sequentially work for spatially matching and fusing inter-sensor multi-modal data. In this way, a synergic effect is produced where the first segment resolves inter-sensor deviations while stimulating the second one to achieve a more accurate data fusion. Additionally, a joint loss formulation is proposed for effectively training the proposed model without external data supervision. The experimental comparison, conducted over four coupled Sentinel-2 and Sentinel-3 datasets, reveals the advantages of W-NetPan with respect to several of the most important state-of-the-art pan-sharpening methods available in the literature. The codes related to this paper will be available at https://github.com/rufernan/WNetPan.

Monitoring nonlinear and fast deformation caused by underground mining exploitation using multi-temporal Sentinel-1 radar interferometry and corner reflectors: application, validation and processing obstacles

ABSTRACT In this study, Differential Interferometric Synthetic Aperture Radar Interferometry (DInSAR) of artificial Corner Reflectors (CRs) were validated in the area of fast and nonlinear deformation gradient caused by active coal longwall exploitation. Three Sentinel-1 datasets were processed using conventional DInSAR, Persistent Scatterer Interferometry (PSI), and Small BAseline Subset methods implemented in ENVI SARscape™. For evaluation, leveling and Global Navigation Satellite System (GNSS) measurements were used. Considering the challenge of snow cover, the removal of all winter images was not a successful strategy due to the long temporal baseline and strong movement, which cause phase unwrapping problems and underestimate the real deformation. The results indicate that only conventional DInSAR and SBAS with low network redundancy allow us to capture maximal deformation gradient and the root mean square error calculated between the CRs and the ground truth is on the level of 2–3 cm for the vertical and easting deformation component, respectively. For the small deformation gradient represented by the permanent GNSS station (4 cm/year), all SBAS techniques appeared to be more accurate than DInSAR, which corresponds to higher redundancy and better removal of the atmospheric signal. In contrast, DInSAR results allowed to capture information about two subsidence basins, which was not possible with SBAS and PSI approaches.

Mapping cropland extent using sentinel-2 datasets and machine learning algorithms for an agriculture watershed

Mapping and monitoring cropland extents are among the most appreciated element for resource management, monitoring, and assessment. India, being the most densely inhabited and agronomical country, mapping high-resolution cropland extent provides appreciated statistics and data for various resource management practices from an individual farm unit to larger administrative units. However, currently available cropland extent maps at global, continental, and regional extents are derived from coarser resolution satellite imageries. For smaller administrative unit with areas less than 10,000 Ha, these coarse resolution maps, often misclassify, resulting in uncertainties in accurate cropland accuracy. To overcome these limitations, in this study an effort is made to develop 10-m fine-resolution cropland extent maps using multi-temporal Sentinel-2 datasets for a watershed in Tadepalligudem, India. The selected study area is an agriculture dominant watershed covering an area of 5375 Ha. The satellite datasets used for the present study are selected with minimum cloud cover and are time-composited over three crop seasons (Kharif, Rabi, and Zaid). With the successful advent of machine learning methods in remote sensing applications, cropland extent maps are derived using Random Forest (RF) and Support Vector Machine (SVM) supervised machine learning algorithms in R-Studio, an R-programming interface. RF and SVM are CART-based machine learning algorithms, where the performance of RF depends on the number of trees (“ntrees”) and the number of sampled features (“mtry”), while for SVM it depends on Kernel functions used. To rejoice the need to develop a fine resolution, a comparative analysis is performed between 10-m resolution Sentinel-2 derived cropland extent maps and commonly used 30-m resolution Landsat-8 OLI-derived products. Training and testing samples collected from the field observations and high-resolution Google Earth Pro-is divided in a ratio of 75:25. In the present study, the accuracy of the classified cropland extent maps are validated with ground truth for both satellite sources and machine learning methods. The accuracy of the maps is assessed using Overall Accuracy (OA), and kappa coefficient derived from the confusion matrix. With an OA and kappa co-efficient of 92.8% and 0.86 (RF) respectively and 94.3% and 0.89 (SVM) respectively, classified cropland extent maps from Sentinel-2 datasets have been shown to outperform Landsat-8 derived maps. From the comparative analysis between the two machine learning methods, classified cropland extent maps obtained using the SVM algorithm, have been shown to outperform maps obtained using RF with greater OA and kappa coefficient values. The methodology adopted in this study finds its potential in delineating cropland from non-cropland regions predominantly in small agriculture watersheds with an accuracy of 0.01 ha per pixel.

Cross-modal change detection flood extraction based on convolutional neural network

Flood events are often accompanied by rainy weather, which limits the applicability of optical satellite images, whereas synthetic aperture radar (SAR) is less sensitive to weather and sunlight conditions. Although remarkable progress has been made in flood detection using heterogeneous multispectral and SAR images, there is a lack of publicly available large-scale datasets and more efforts are required for exploiting deep neural networks in heterogeneous flood detection. This study constructed a pre-disaster Sentinel-2 and post-disaster Sentinel-1 heterogeneous flood mapping dataset named CAU-Flood containing 18 study plots with careful image preprocessing and human annotation. A new deep convolutional neural network (CNN), named cross-modal change detection network (CMCDNet), was also proposed for flood detection using multispectral and SAR images. The proposed network employs a encoder-decoder structure and performs feature fusion at multiple stages using gating and self-attention modules. Furthermore, the network overcomes the feature misalignment issue during decoding by embedding a feature alignment module in the upsampling operation. The proposed CMCDNet outperformed SOTA methods in terms of flood detection accuracy and achieved an intersection over union (IoU) of 89.84%. The codes and datasets are available at: https://github.com/CAU-HE/CMCDNet.

Exploring the use of Sentinel-2 datasets and environmental variables to model wheat crop yield in smallholder arid and semi-arid farming systems

Low levels of agricultural productivity are associated with the persistence of food insecurity, poverty, and other socio-economic stresses. Mapping and monitoring agricultural dynamics and production in real-time at high spatial resolution are essential for ensuring food security and shaping policy interventions. However, an accurate yield estimation might be challenging in some arid and semi-arid regions since input datasets are generally scarce, and access is restricted due to security challenges. This work examines how well Sentinel-2 satellite sensor-derived data, topographic and climatic variables, can be used as covariates to accurately model and predict wheat crop yield at the farm level using statistical models in low data settings of arid and semi-arid regions, using Sulaimani governorate in Iraq as an example. We developed a covariate selection procedure that assessed the correlations between the covariates and their relationships with wheat crop yield. Potential non-linear relationships were investigated in the latter case using regression splines. In the absence of substantial non-linear relationships between the covariates and crop yield, and residual spatial autocorrelation, we fitted a Bayesian multiple linear regression model to model and predict crop yield at 10 m resolution. Out of the covariates tested, our results showed significant relationships between crop yield and mean cumulative NDVI during the growing season, mean elevation, mean end of the season, mean maximum temperature and mean the start of the season at the farm level. For in-sample prediction, we estimated an R2 value of 51 % for the model, whereas for out-of-sample prediction, this was 41 %, both of which indicate reasonable predictive performance. The calculated root-mean-square error for out-of-sample prediction was 69.80, which is less than the standard deviation of 89.23 for crop yield, further showing that the model performed well by reducing prediction variability. Besides crop yield estimates, the model produced uncertainty metrics at 10 m resolution. Overall, this study showed that Sentinel-2 data can be valuable for upscaling field measurement of crop yield in arid and semi-arid regions. In addition, the environmental covariates can strengthen the model predictive power. The method may be applicable in other areas with similar environments, particularly in conflict zones, to increase the availability of agricultural statistics.

AI4Boundaries: an open AI-ready dataset to map field boundaries with Sentinel-2 and aerial photography

Abstract. Field boundaries are at the core of many agricultural applications and are a key enabler for the operational monitoring of agricultural production to support food security. Recent scientific progress in deep learning methods has highlighted the capacity to extract field boundaries from satellite and aerial images with a clear improvement from object-based image analysis (e.g. multiresolution segmentation) or conventional filters (e.g. Sobel filters). However, these methods need labels to be trained on. So far, no standard data set exists to easily and robustly benchmark models and progress the state of the art. The absence of such benchmark data further impedes proper comparison against existing methods. Besides, there is no consensus on which evaluation metrics should be reported (both at the pixel and field levels). As a result, it is currently impossible to compare and benchmark new and existing methods. To fill these gaps, we introduce AI4Boundaries, a data set of images and labels readily usable to train and compare models on field boundary detection. AI4Boundaries includes two specific data sets: (i) a 10 m Sentinel-2 monthly composites for large-scale analyses in retrospect and (ii) a 1 m orthophoto data set for regional-scale analyses, such as the automatic extraction of Geospatial Aid Application (GSAA). All labels have been sourced from GSAA data that have been made openly available (Austria, Catalonia, France, Luxembourg, the Netherlands, Slovenia, and Sweden) for 2019, representing 14.8 M parcels covering 376 K km2. Data were selected following a stratified random sampling drawn based on two landscape fragmentation metrics, the perimeter/area ratio and the area covered by parcels, thus considering the diversity of the agricultural landscapes. The resulting “AI4Boundaries” dataset consists of 7831 samples of 256 by 256 pixels for the 10 m Sentinel-2 dataset and of 512 by 512 pixels for the 1 m aerial orthophoto. Both datasets are provided with the corresponding vector ground-truth parcel delineation (2.5 M parcels covering 47 105 km2), and with a raster version already pre-processed and ready to use. Besides providing this open dataset to foster computer vision developments of parcel delineation methods, we discuss the perspectives and limitations of the dataset for various types of applications in the agriculture domain and consider possible further improvements. The data are available on the JRC Open Data Catalogue: http://data.europa.eu/89h/0e79ce5d-e4c8-4721-8773-59a4acf2c9c9 (European Commission, Joint Research Centre, 2022).

A methodology to characterize 4D post-failure slope instability dynamics using remote sensing measurements: A case study of the Aniangzhai landslide in Sichuan, Southwest China

A massive landslide often causes long-lasting instability dynamics that need to be analyzed in detail for risk management and mitigation. Multiple satellite remote sensing observations, in-situ measurements, and geophysical approaches have been jointly implemented to monitor and interpret the life cycle of landslides and their failure mechanisms from various perspectives. In this work, we propose a framework where satellite optical and synthetic aperture radar (SAR) remote sensing techniques are combined with feature extractions using independent component analysis (ICA) and a mathematical relaxation model to assess the complete four-dimensional (4D) spatiotemporal patterns of post-failure slope evolution. The large, deep-seated Aniangzhai landslide in Southwest China that occurred on 17 June 2020 is comprehensively analyzed and characterized for its post-failure mechanism. Time series of Planet high-resolution optical images are first explored to derive the large horizontal motions for the first six months after the failure. Spatiotemporal dynamics of line-of-sight (LOS) displacement in the landslide body are then derived between November 2020 and February 2022 by combining 40 TerraSAR-X (TSX) High-resolution Spotlight (HS) images and 76 medium-resolution Sentinel-1 (S1) SAR datasets using Multi-temporal InSAR (MTI) method. The InSAR-derived results are subsequently analyzed with ICA to find common deformation components of points between optical and MTI results, indicating the same temporal evolution in the deformation pattern. Finally, the complete 4D deformation field for the whole post-failure period is modeled using a decaying exponential model representing stress relaxation after the failure by integrating multiple remote sensing datasets. Cross-correlation analysis of Planet imagery shows a decaying exponential pattern of post-failure displacements with an approximately 94% reduction in the deformation rate after six months with respect to the co-failure event. MTI analysis suggests a maximum LOS displacement rate of approximately 30 cm/year over the main failure body from November 2020 to February 2022; while the high-resolution TSX datasets show irreplaceable advantages in choosing the number of measurement points in MTI analysis with the number of measurement points being five times larger than those obtained by S1 datasets. The ICA analysis reveals three main types of kinematic patterns in the temporal evolution of post-failure deformation in MTI results, the dominant one being an exponential declining pattern similar to the results from Planet observations. Integrated 4D deformation modeling suggests that the most significant post-failure displacement mainly occurred toward the west, amounting to 28 m during the entire post-failure acquisitions from June 2020 until February 2022. Additionally, maximum displacements of 17 m and 19 m occurred in this period toward the north and downward, respectively.

National high-resolution cropland classification of Japan with agricultural census information and multi-temporal multi-modality datasets

Multi-modality datasets offer advantages for processing frameworks with complementary information, particularly for large-scale cropland mapping. Extensive training datasets are required to train machine learning algorithms, which can be challenging to obtain. To alleviate the limitations, we extract the training samples from the agricultural census information. We focus on Japan and demonstrate how agricultural census data in 2015 can map different crop types for the entire country. Due to the lack of Sentinel-2 datasets in 2015, this study utilized Sentinel-1 and Landsat-8 collected across Japan and combined observations into composites for different prefecture periods (monthly, bimonthly, seasonal). Recent deep learning techniques have been investigated the performance of the samples from agricultural census information. Finally, we obtain nine crop types on a countrywide scale (around 31 million parcels) and compare our results to those obtained from agricultural census testing samples as well as those obtained from recent land cover products in Japan. The generated map accurately represents the distribution of crop types across Japan and achieves an overall accuracy of 87% for nine classes in 47 prefectures. Our findings highlight the importance of using multi-modality data with agricultural census information to evaluate agricultural productivity in Japan. The final products are available at https://doi.org/10.5281/zenodo.7519274.

Understanding the potential of Sentinel-2 for monitoring methane point emissions

Abstract. The use of satellite instruments to detect and quantify methane emissions from fossil fuel production activities is highly beneficial to support climate change mitigation. Different hyperspectral and multispectral satellite sensors have recently shown potential to detect and quantify point-source emissions from space. The Sentinel-2 (S2) mission, despite its limited spectral design, supports the detection of large emissions with global coverage and high revisit frequency thanks to coarse spectral coverage of methane absorption lines in the shortwave infrared. Validation of S2 methane retrieval algorithms is instrumental in accelerating the development of a systematic and global monitoring system for methane point sources. Here, we develop a benchmarking framework for such validation. We first develop a methodology to generate simulated S2 datasets including methane point-source plumes. These benchmark datasets have been created for scenes in three oil and gas basins (Hassi Messaoud, Algeria; Korpeje, Turkmenistan; Permian Basin, USA) under different scene heterogeneity conditions and for simulated methane plumes with different spatial distributions. We use the simulated methane plumes to validate the retrieval for different flux rate levels and define a minimum detection threshold for each case study. The results suggest that for homogeneous and temporally invariant surfaces, the detection limit of the proposed S2 methane retrieval ranges from 1000 to 2000 kg h−1, whereas for areas with large surface heterogeneity and temporal variations, the retrieval can only detect plumes in excess of 500 kg h−1. The different sources of uncertainty in the flux rate estimates have also been examined. Dominant quantification errors are either wind-related or plume mask-related, depending on the surface type. Uncertainty in wind speed, both in the 10 m wind (U10) and in mapping U10 to the effective wind (Ueff) driving plume transport, is the dominant source of error for quantifying individual plumes in homogeneous scenes. For heterogeneous and temporally variant scenes, the surface structure underlying the methane plume affects the plume masking and can become a dominant source of uncertainty.