Sentinel-2 Multispectral Images Research Articles

Graph Neural Networks Extract High-Resolution Cultivated Land Maps From Sentinel-2 Image Series

Maintaining farm sustainability through optimizing the agricultural management practices helps build more planet-friendly environment. The emerging satellite missions can acquire multi- and hyperspectral imagery which captures more detailed spectral information concerning the scanned area, hence allows us to benefit from subtle spectral features during the analysis process in agricultural applications. We introduce an approach for extracting 2.5 m cultivated land maps from 10 m Sentinel-2 multispectral image series which benefits from a compact graph convolutional neural network. The experiments indicate that our models not only outperform classical and deep machine learning techniques through delivering higher-quality segmentation maps, but also dramatically reduce the memory footprint when compared to U-Nets (almost 8k trainable parameters of our models, with up to 31M parameters of U-Nets). Such memory frugality is pivotal in the missions which allow us to uplink a model to the AI-powered satellite once it is in orbit, as sending large nets is impossible due to the time constraints.

Comparison of Proximal and Remote Sensing for the Diagnosis of Crop Status in Site-Specific Crop Management.

The presented paper deals with the issue of selecting a suitable system for monitoring the winter wheat crop in order to determine its condition as a basis for variable applications of nitrogen fertilizers. In a four-year (2017–2020) field experiment, 1400 ha of winter wheat crop were monitored using the ISARIA on-the-go system and remote sensing using Sentinel-2 multispectral satellite images. The results of spectral measurements of ISARIA vegetation indices (IRMI, IBI) were statistically compared with the values of selected vegetation indices obtained from Sentinel-2 (EVI, GNDVI, NDMI, NDRE, NDVI and NRERI) in order to determine potential hips. Positive correlations were found between the vegetation indices determined by the ISARIA system and indices obtained by multispectral images from Sentinel-2 satellites. The correlations were medium to strong (r = 0.51–0.89). Therefore, it can be stated that both technologies were able to capture a similar trend in the development of vegetation. Furthermore, the influence of climatic conditions on the vegetation indices was analyzed in individual years of the experiment. The values of vegetation indices show significant differences between the individual years. The results of vegetation indices obtained by the analysis of spectral images from Sentinel-2 satellites varied the most. The values of winter wheat yield varied between the individual years. Yield was the highest in 2017 (7.83 t/ha), while the lowest was recorded in 2020 (6.96 t/ha). There was no statistically significant difference between 2018 (7.27 t/ha) and 2019 (7.44 t/ha).

Optical Remote Sensing Indexes of Soil Moisture: Evaluation and Improvement Based on Aircraft Experiment Observations

Optical remote sensing (about 0.4~2.0 μm) indexes of soil moisture (SM) are valuable for some specific applications such as monitoring agricultural drought and downscaling microwave SM, due to their abundant data sources, higher spatial resolution, and easy-to-use features, etc. In this study, we evaluated thirteen typical optical SM indexes with aircraft and in situ observed SM from two field campaigns, the Soil Moisture Active Passive Validation Experiment 2012 (SMAPVEX12) and 2016 (SMAPVEX16) conducted in Manitoba, Canada. MODIS surface reflectance products (MOD09A1) and Sentinel-2 multispectral imager Level-1C data were utilized to calculate the optical SM indexes. The evaluation results demonstrated that (1) the Visible and Shortwave Infrared Drought Index (VSDI) and Optical TRApezoid Model (OPTRAM) outperform the other eleven optical SM indexes as compared with aircraft and in situ observed SM. They also presented well consistence in temporal variation with the in situ observed SM. (2) The VSDI achieved comparable performance with the OPTRAM while the former has very simple calculation expression and the latter requires complex process to determine the dry and wet boundaries. (3) Both the VSDI and OPTRAM utilize two sensitive bands of soil and vegetation moisture, i.e., Red and SWIR bands, whereas the other eleven SM indexes only employ one sensitive band. This may be the main reason of the evaluation results. (4) Based on this recognition, improvements of the VSDI and OPTRAM were created and validated in this study through adding more sensitive band to VSDI and combining NDVI and modified VSDI into a new feature space for calculating the optical SM index as with OPTRAM. The results are conducive to selecting and utilizing the current numerous optical SM indexes for SM and drought monitoring.

Detecting and mapping the spatial distribution of Chromoleana odorata invasions in communal areas of South Africa using Sentinel-2 multispectral remotely sensed data

Alien Invasive Plants (AIPs) have been proclaimed to be a significant non-climatic driver of global change. The Siam weed (Chromoleana odorata) is highly invasive in South African rangelands, with serious environmental, agricultural and socio-economic consequences. Therefore, a comprehensive analysis of C.odorata's spatial distribution is necessary for implementing relevant mitigation and management approaches to vulnerable and invaded landscapes. Remotely sensed data offer a viable opportunity for detecting and mapping AIPs spatial extents. Hence, this study sought to compare the value of algorithms (i.e. maximum likelihood and random forest) in detecting and mapping C.odorata's spatial distribution in relation to other land use land cover classes (LULC) using the freely available Sentinel-2 multispectral image data. The findings of the study revealed that the vegetation red edge and near-infrared spectral bands of Sentinel-2 multispectral imagery (MSI) sensor were the most important spectral variables for discriminating C.odorata from other land cover classes. The random forest algorithm yielded the highest overall classification accuracy of 83%, outperforming the traditional maximum likelihood classifier (75%) in classifying C.odorata's spatial distribution. Overall, results demonstrate that the combination of freely available Sentinel-2 MSI satellite data and machine learning random forest algorithm produces the highest accuracies for detecting the spatial distribution of C.odorata, particularly in heterogeneous environments. These results could be beneficial to sustainable rangeland management and the adoption of site-specific mitigation approaches in areas invaded by C.odorata.

The Spatial Distribution and Prediction of Soil Heavy Metals Based on Measured Samples and Multi-Spectral Images in Tai Lake of China

Soil is an important natural resource. The excessive amount of heavy metals in soil can harm and threaten human health. Therefore, monitoring of soil heavy metal content is urgent. Monitoring soil heavy metals by traditional methods requires many human and material resources. Remote sensing has shown advantages in the field of monitoring heavy metals. Based on 971 heavy metal samples and Sentinel-2 multi-spectral images in Tai Lake, China, we analyzed the correlation between six heavy metals (Cd, Hg, As, Pb, Cu, Zn) and spectral factors, and selected As and Hg as the input factors of inversion model. The correlation coefficient of the best model of As was 0.53 (p < 0.01), and of Hg was 0.318 (p < 0.01). We used the methods of partial least squares regression (PLSR) and back propagation neural network (BPNN) to establish inversion models with different combinations of spectral factors by using 649 measured samples. In addition, 322 measured samples were used for accuracy evaluation. Compared with the PLSR model, the BP neural network builds the model with higher accuracy, and B1-B4 combined with LnB1-LnB4 builds the model with the highest accuracy. The accuracy of the best model was verified, with an average error of 19% for As and 45% for Hg. Analyzing the spatial distribution of heavy metals by using the interpolation method of Kriging and IDW. The overall distribution trend of the two interpolations is similar. The concentration of As elements tends to increase from north to south, and the relatively high value of Hg elements is distributed in the east and west of the study area. The factories in the study area are distributed along rivers and lakes, which is consistent with the spatial distribution of heavy metal enrichment areas. The relatively high-value areas of heavy metal elements are related to the distribution of metal products factories, refractory porcelain factories, tile factories, factories and mining enterprises, etc., indicating that factory pollution is the main reason for the enrichment of heavy metals.

Improved Filtering of ICESat-2 Lidar Data for Nearshore Bathymetry Estimation Using Sentinel-2 Imagery

The accurate estimation of nearshore bathymetry is necessary for multiple aspects of coastal research and practices. The traditional shipborne single-beam/multi-beam echo sounders and Airborne Lidar bathymetry (ALB) have a high cost, are inefficient, and have sparse coverage. The Satellite-derived bathymetry (SDB) method has been proven to be a promising tool in obtaining bathymetric data in shallow water. However, current empirical SDB methods for multispectral imagery data usually rely on in situ depths as control points, severely limiting their spatial application. This study proposed a satellite-derived bathymetry method without requiring a priori in situ data by merging active and passive remote sensing (SDB-AP). It realizes rapid bathymetric mapping with only satellite remotely sensed data, which greatly extends the spatial coverage and temporal scale. First, seafloor photons were detected from the ICESat-2 raw photons based on an improved adaptive Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, which could calculate the optimal detection parameters for seafloor photons by adaptive iteration. Then, the bathymetry of the detected seafloor photons was corrected because of the refraction that occurs at the air–water interface. Afterward, the outlier photons were removed by an outlier-removal algorithm to improve the retrieval accuracy. Subsequently, the high spatial resolution (0.7 m) ICESat-2 derived bathymetry data were gridded to match the Sentinel-2 data with a lower spatial resolution (10 m). All of the ICESate-2 gridded data were randomly separated into two parts: 80% were employed to train the empirical bathymetric model, and the remaining 20% were used to quantify the inversion accuracy. Finally, after merging the ICESat-2 data and Sentinel-2 multispectral images, the bathymetric maps over St. Thomas of the United States Virgin Islands, Acklins Island in the Bahamas, and Huaguang Reef in the South China Sea were produced. The ICESat-2-derived results were compared against in situ data over the St. Thomas area. The results showed that the estimated bathymetry reached excellent inversion accuracy and the corresponding RMSE was 0.68 m. In addition, the RMSEs between the SDB-AP estimated depths and the ICESat-2 bathymetry results of St. Thomas, Acklins Island, and Huaguang Reef were 0.96 m, 0.91 m, and 0.94 m, respectively. Overall, the above results indicate that the SDB-AP method is effective and feasible for different shallow water regions. It has great potential for large-scale and long-term nearshore bathymetry in the future.

Ship Detection in Sentinel 2 Multi-Spectral Images with Self-Supervised Learning

Automatic ship detection provides an essential function towards maritime domain awareness for security or economic monitoring purposes. This work presents an approach for training a deep learning ship detector in Sentinel-2 multi-spectral images with few labeled examples. We design a network architecture for detecting ships with a backbone that can be pre-trained separately. By using self supervised learning, an emerging unsupervised training procedure, we learn good features on Sentinel-2 images, without requiring labeling, to initialize our network’s backbone. The full network is then fine-tuned to learn to detect ships in challenging settings. We evaluated this approach versus pre-training on ImageNet and versus a classical image processing pipeline. We examined the impact of variations in the self-supervised learning step and we show that in the few-shot learning setting self-supervised pre-training achieves better results than ImageNet pre-training. When enough training data are available, our self-supervised approach is as good as ImageNet pre-training. We conclude that a better design of the self-supervised task and bigger non-annotated dataset sizes can lead to surpassing ImageNet pre-training performance without any annotation costs.

Improving Estimates of Soil Salt Content by Using Two-Date Image Spectral Changes in Yinbei, China

Soil salt content (SSC) is normally featured with obvious spatiotemporal variations in arid and semi-arid regions. Space factors such as elevation, temperature, and spatial locations are usually used as input variables for a model to estimate the SSC. However, whether temporal patterns of salt-affected soils (identified as temporal spectral patterns) can indicate the SSC level and be applied as a covariate in a model to estimate the SSC remains unclear. Hence, temporal changes in soil spectral patterns need to be characterized and explored as to their use as an input variable to improve SSC estimates. In this study, a total of 54 field samples and a time-series of Sentinel-2 multispectral images taken at monthly intervals (from October 2017 to April 2018) were collected in the Yinbei area of western China. Then, two-date satellite images were used to quantify significant spectral changes over time using spectral change vector analysis, and four two-date-based index methods were used to characterize soil spectral changes. Lastly, the optimal two-date-based spectral indices and multispectral bands were used as input variables to build the estimation models using a random forest algorithm. Results showed that the two-date-based spectral index could be applied as an input variable to improve the accuracy of SSC estimation at a regional scale. Temporal changes in salt-induced spectral patterns can be indicated by the band difference in the wavelength range from 400 nm to 900 nm. Three two-date-based indices designated as D28a (i.e., the band difference between band 2 from an image acquired in April 2018 and band 8a from an image acquired in December 2017), D22, and D28 were the optimal parameters for characterizing salt-induced spectral changes, which were dominated by the total brightness, chloride, and sulfate accumulation of the soils. The model did not yield satisfactory estimation results (RPD = 1.49) when multispectral bands were used as the input variables. Multispectral bands coupled with two two-date-based indices (D22 and D28a) used as the input variables produced the best estimation result (R2 = 0.92, RPD = 3.27). Incorporating multispectral bands and two-date-based indices into the random forest model provides a remotely-sensed strategy that effectively supports the monitoring of soil salt content.

On improved nearshore bathymetry estimates from satellites using ensemble and machine learning approaches

In this article, two popular linear empirical methods viz., log-ratio model (LRM) and log-linear model (LLM) are used to derive water depths in shallow nearshore waters from Sentinel-2 multi-spectral images. Based on these empirical models, a multi-scene ensemble and a non-linear Support Vector Regression (SVR) machine learning approaches are applied to improve the accuracies from these traditional methods. In this analysis, firstly the best scene is selected from a set of six Sentinel-2 A&B images using noise-equivalent reflectance NEΔRrs (sr−1), optimal bands for LRM and LLM are selected using Optimal band ratio analysis (OBRA) and by computing Pearson Correlation Coefficient (R) between each band respectively. A total of 80% depth data points obtained from JetSki based echo-sounding measurements are used for training and the remaining 20% are used for testing each approach. The overall errors during the test phase for the range of depth 0–12 m are 0.79 m (0.67 m), 0.94 m (0.66 m) and 0.57 m (0.39 m) using traditional empirical LRM (LLM) methods from the best image, empirical-based ensemble and SVR approaches respectively. Irrespective of the approach, the LLM produced smoother and relatively accurate bathymetry as compared to the LRM. The LLM based SVR ML approach provides the best performance over the entire depth range as compared to all the methods considered in this study. Therefore, this method can be used for efficiently estimating nearshore depths and produce updated high-resolution bathymetry maps for many coastal applications.

Sentinel-2 görüntüleri ve ICESat-2 ATL03 foton yükseklik verilerinin kombinasyonu ile batimetri haritası üretilebilirliğinin araştırılması

Bathymetry maps can be produced with high accuracy by using active acoustic and laser sensing systems. Various acoustic systems such as side-scan sonar, single and multi-beam echosounder are difficult to operate in shallow waters, and measurement with laser systems depends on the presence of appropriate optical transparency. Disadvantages such as workload, time, and economic costs should also be mentioned in the implementation of the two systems. On the other hand, bathymetry maps can be obtained in shallow waters up to a depth of 15 m at an affordable cost with multispectral images covering larger areas and remote/inaccessible locations. This method, known as the experimental approach, requires however sufficient field depth data. In this study, bathymetric map of a study area was generated by combining external depth measurements obtained from ICESat-2 ATL03 photon heights and Sentinel-2 multispectral images. This study aimed to test the accuracy of this bathymetric map against an existing map. Reflectance values measured in Sentinel-2’s coastal aerosol, blue and green bands were modelled under multiple linear regression by using both log-linear and log-rational approaches, and as a result a relative depth map was obtained. The resultant non-metric map is scaled by refraction corrected ICESat-2 depth values building the model’s training set. The bathymetry map was calculated with the multiple linear regression model obtained from this training set and compared with the existing side-scan sonar bathymetric map of the test area. The comparison tests revealed that the atmospheric correction did not improve the results, and better accuracy values were achieved with the log-linear approach compared to the log-rational approach. A root mean square error of ±1.6 m was calculated from the comparisons to the acoustic bathymetry map. The number of differences having absolute values of 5 m and below constituted 94% of the total comparison points. Although the expected accuracy for navigation is not met, bathymetric maps of shallow areas with high optical transparency can still be obtained at a low cost by combining two independent satellite-based datasets for research and other applications.

Coniferous Plantations Growing Stock Volume Estimation Using Advanced Remote Sensing Algorithms and Various Fused Data

Spatial distribution prediction of growing stock volume (GSV) for supporting the sustainable management of forest ecosystems, is one of the most widespread applications of remote sensing. For this purpose, remote sensing data were used as predictor variables in combination with ground data obtained from field sample plots. However, with the increase in forest GSV values, the spectral reflectance of remote sensing imagery is often saturated or less sensitive to the GSV changes, making accurate estimation difficult. To improve this, we examined the GSV estimation performance and data saturation of four optical remote sensing image datasets (Landsat 8, Sentinel-2, ZiYuan-3, and GaoFen-2) in the subtropical region of Central South China. First, various feature variables were extracted and three optimization methods were used to select optimal feature variable combinations. Subsequently, k-nearest-neighbor (kNN), random forest regression, and categorical boosting algorithms were employed to build the GSV estimation models, and evaluate the GSV estimation accuracy and saturation. Second, Gram Schmidt (GS) and NNDiffuse pan sharpening (NND) methods were employed to fuse the optimal multispectral images and explore various image fusion schemes suitable for GSV estimation. We proposed an adaptive stacking (AdaStacking) model ensemble algorithm to further improve GSV estimation performance. The results indicated that Sentinel-2 had the highest GSV estimation accuracy exhibiting a minimum relative root mean square error of 20.06% and saturation of 434 m3/ha, followed by GaoFen-2 with a minimum relative root mean square error of 22.16% and a saturation of 409 m3/ha. Among the four fusion images, the NND-B2 image—obtained by fusing the GaoFen-2 green band and Sentinel-2 multispectral image with the NND method—had the best estimation accuracy. The estimated optimal RMSEs of NND-B2 were 24.4% and 16.5% lower than those of GaoFen-2 and Sentinel-2, respectively. Therefore, the fused image data based on GF-2 and Sentinel-2 can effectively couple the advantages of the two images and significantly improve the GSV estimation performance. Moreover, the proposed adaptive stacking model is more effective in GSV estimation than a single model. The GSV estimation saturation value of the AdaStacking model based on NND-B2 was 5.4% higher than that of the KNN-Maha model. The GSV distribution map estimated by AdaStacking model used the NND-B2 dataset corresponded accurately with the field observations. This study provides some insights into the optical image fusion scheme, feature selection, and adaptive modeling algorithm in GSV estimation for coniferous forest.

Derivation of spatially detailed lentic habitat map and inventory at a basin scale by integrating multispectral Sentinel-2 satellite imagery and USGS Digital Elevation Models

Aquatic habitats are essential for aquaculture, agriculture, and municipal water supply. The geographical location and quantitative information about the aquatic habitats are critical for inventorying, monitoring, and managing these valuable resources. This paper presents a novel method that integrates the object-oriented image analysis and DEM-based stream network analysis to map and quantify lentic aquatic habitats, using freely available ESA Sentinel-2 multispectral imagery and USGS DEMs. This integrated method has been applied to the entire Mobile River Basin, USA, and lentic habitats at the basin scale have been delineated and inventoried at the 10-m spatial resolution. The surface water pixels of aquatic habitats are extracted through the Iterative Self-Organizing Data Analysis Technique (ISODATA) unsupervised classification of the Sentinel-2 multispectral images. An object-oriented image analysis approach is then used to group contiguous water pixels to form and map discrete aquatic habitat objects. The stream network in the basin has been hydrologically derived from the USGS DEMs at the 10-m spatial resolution. The combination of the geometric and shape information from the object-oriented image analysis and hydrological information from the DEM-based stream network analysis enables the separation of lotic and lentic habitats. A set of planimetric, geometric, and shape attributes has been numerically derived for each lentic habitat object. The abundance pattern, shape pattern, and spatial distribution pattern of the lentic habitats in the basin have been analyzed. The vast majority of the aquatic habitat objects identified in the Mobile River Basin are lentic habitats. The most striking spatial pattern of lentic habitats is that small, oblong lakes and fish ponds are densely distributed along the Black Prairie Belt in the basin, laying the foundation for the catfish aquacultural industry in the regions of mid-west Alabama and east Mississippi. This research demonstrates that the integration of the recently emerged Sentinel-2A/B satellite multispectral imagery with USGS DEMs represents an effective approach to the inventory and environmental investigation of lentic habitats at a basin scale.

Estimating aboveground net primary productivity of reforested trees in an urban landscape using biophysical variables and remotely sensed data

Recently, urban reforestation programs have emerged as potential carbon sinks and climate mitigates in urban landscapes. Thus, spatially explicit information on net primary productivity (NPP) of reforested trees in urban environments is central to understanding the value of reforestation initiatives in the global carbon budget and climate regulation potential. To date, numerous studies have mainly focused on natural and commercial forests NPP at a regional scale based on coarse spatial resolution remotely sensed data. Generally, local scale NPP studies based on fine spatial resolution data are limited. Therefore, this study sought to estimate aboveground NPP of an urban reforested landscape using biophysical and Sentinel-2 Multispectral Imager data derived variables. Using the MOD17 model, results showed that mean NPP ranged between 6.24 Mg C ha−1 with high coefficient of determination (R2: 0.92) and low RMSE (0.82 Mg ha−1) across all reforested trees within the study area. Results also showed a considerable variation in NPP among the reforested trees, with deciduous Acacia and Dalbergia obovate species showing the highest NPP (7.62 Mg C ha−1 and 7.58 Mg C ha-1, respectively), while the evergreen Syzygium cordatum and shrub Artemisia afra had the lowest NPP (4.54 Mg C ha−1 and 5.26 Mg C ha−1). Furthermore, the multiple linear regression analysis showed that vegetation specific biophysical variables (i.e. leaf area index, Normalized Difference Vegetation Index and Fraction of Photosynthetically Active Radiation) significantly improved the estimation of reforested aboveground NPP at a fine-scale resolution. These findings demonstrate the effectiveness of biophysical and remotely sensed variables in determining NPP (as carbon sequestration surrogate) at fine-scaled reforested urban landscape. Furthermore, the utility of species biometric measurements and MOD17 model offers unprecedented opportunity for improved local scale reforestation assessment and monitoring schedules.

Integrating spatio-temporal-spectral information for downscaling Sentinel-3 OLCI images

Sentinel-3 is a newly launched satellite implemented by the European Space Agency (ESA) for global observation. The Ocean and Land Colour Imager (OLCI) sensor onboard Sentinel-3 provides 21 band images with a fine spectral resolution and is of great value for ocean, land and atmospheric monitoring. The two platforms (Sentinel-3A and -3B) can provide OLCI images at an almost daily temporal resolution. The coarse spatial resolution of the 21 band OLCI images (i.e., 300 m), however, limits greatly their utility for local, precise monitoring. Sentinel-2, another satellite provided by ESA, carries the Multispectral Imager (MSI) sensor which can supply much finer spatial resolution (e.g., 10 m and 20 m) images. This paper introduces a new fusion framework integrating spatio-temporal-spectral information for downscaling Sentinel-3 OLCI images, which has two parts. Based on bands with similar wavelengths (i.e., bands 2, 3, 4 and 8a for Sentinel-2 and bands Oa4, Oa6, Oa8 and Oa17 for Sentinel-3), the four Sentinel-3 bands are first downscaled to the spatial resolution of Sentinel-2 images by applying spatio-temporal fusion to Sentinel-2 MSI and Sentinel-3 OLCI images. Then, to take full advantage of all 21 available OLCI bands of the Sentinel-3 images, the extended image pair-based spatio-spectral fusion (EIPSSF) method is proposed in this paper to downscale the other 17 bands. EIPSSF is performed based on the new concept of the extended image pair (EIP) and by exploiting existing spatio-temporal fusion approaches. The framework consisting of spatio-temporal and spatio-spectral fusion is entirely general, which provides a practical solution for comprehensive downscaling of Sentinel-3 OLCI images for fine spatial, temporal and spectral resolution monitoring.

Complementary water quality observations from high and medium resolution Sentinel sensors by aligning chlorophyll-a and turbidity algorithms

High resolution imaging spectrometers are prerequisite to address significant data gaps in inland optical water quality monitoring. In this work, we provide a data-driven alignment of chlorophyll-a and turbidity derived from the Sentinel-2 MultiSpectral Imager (MSI) with corresponding Sentinel-3 Ocean and Land Colour Instrument (OLCI) products. For chlorophyll-a retrieval, empirical ‘ocean colour’ blue-green band ratios and a near infra-red (NIR) band ratio algorithm, as well as a semi-analytical three-band NIR-red ratio algorithm, were included in the analysis. Six million co-registrations with MSI and OLCI spanning 24 lakes across five continents were analysed. Following atmospheric correction with POLYMER, the reflectance distributions of the red and NIR bands showed close similarity between the two sensors, whereas the distribution for blue and green bands was positively skewed in the MSI results compared to OLCI. Whilst it is not possible from this analysis to determine the accuracy of reflectance retrieved with either MSI or OLCI results, optimizing water quality algorithms for MSI against those previously derived for the Envisat Medium Resolution Imaging Spectrometer (MERIS) and its follow-on OLCI, supports the wider use of MSI for aquatic applications. Chlorophyll-a algorithms were thus tuned for MSI against concurrent OLCI observations, resulting in significant improvements against the original algorithm coefficients. The mean absolute difference (MAD) for the blue-green band ratio algorithm decreased from 1.95 mg m−3 to 1.11 mg m−3, whilst the correlation coefficient increased from 0.61 to 0.80. For the NIR-red band ratio algorithms improvements were modest, with the MAD decreasing from 4.68 to 4.64 mg m−3 for the empirical red band ratio algorithm, and 3.73 to 3.67 for the semi-analytical 3-band algorithm. Three implementations of the turbidity algorithm showed improvement after tuning with the resulting distributions having reduced bias. The MAD reduced from 0.85 to 0.72, 1.22 to 1.10 and 1.93 to 1.55 FNU for the 665, 708 and 778 nm implementations respectively. However, several sources of uncertainty remain: adjacent land showed high divergence between the sensors, suggesting that high product uncertainty near land continues to be an issue for small water bodies, while it cannot be stated at this point whether MSI or OLCI results are differentially affected. The effect of spectrally wider bands of the MSI on algorithm sensitivity to chlorophyll-a and turbidity cannot be fully established without further availability of in situ optical measurements.

Regional-Scale Systematic Mapping of Archaeological Mounds and Detection of Looting Using COSMO-SkyMed High Resolution DEM and Satellite Imagery

“Tells” are archaeological mounds formed by deposition of large amounts of anthropogenic material and sediments over thousands of years and are the most important and prominent features in Near and Middle Eastern archaeological landscapes. In the last decade, archaeologists have exploited free-access global digital elevation model (DEM) datasets at medium resolution (i.e., up to 30 m) to map tells on a supra-regional scale and pinpoint tentative tell sites. Instead, the potential of satellite DEMs at higher resolution for this task was yet to be demonstrated. To this purpose, the 3 m resolution imaging capability allowed by the Italian Space Agency’s COSMO-SkyMed Synthetic Aperture Radar (SAR) constellation in StripMap HIMAGE mode was used in this study to generate DEM products of enhanced resolution to undertake, for the first time, a systematic mapping of tells and archaeological deposits. The demonstration is run at regional scale in the Governorate of Wasit in central Iraq, where the literature suggested a high density of sites, despite knowledge gaps about their location and spatial distribution. Accuracy assessment of the COSMO-SkyMed DEM is provided with respect to the most commonly used SRTM and ALOS World 3D DEMs. Owing to the 10 m posting and the consequent enhanced observation capability, the COSMO-SkyMed DEM proves capable to detect both well preserved and levelled or disturbed tells, standing out for more than 4 m from the surrounding landscape. Through the integration with CORONA KH-4B tiles, 1950s Soviet maps and recent Sentinel-2 multispectral images, the expert-led visual identification and manual mapping in the GIS environment led to localization of tens of sites that were not previously mapped, alongside the computation of a figure as up-to-date as February 2019 of the survived tells, with those affected by looting. Finally, this evidence is used to recognize hot-spot areas of potential concern for the conservation of tells. To this purpose, we upgraded the spatial resolution of the observations up to 1 m by using the Enhanced Spotlight mode to collect a bespoke time series. The change detection tests undertaken on selected clusters of disturbed tells prove how a dedicated monitoring activity may allow a regular observation of the impacts due to anthropogenic disturbance (e.g., road and canal constructions or ploughing).

Aircraft Detection above Clouds by Sentinel-2 MSI Parallax

Detection of aircrafts in satellite images is a challenging problem when the background is strongly reflective clouds with varying transparency. We develop a fast and effective detection algorithm that can find almost all aircrafts above and between clouds in Sentinel-2 multispectral images. It exploits the time delay of a few seconds between the recorded multispectral images such that a moving aircraft is observed at different positions due to parallax effects. The aircraft speed, heading and altitude are also calculated accurately. Analysing images over the English Channel during fall 2020, we obtain a detection accuracy of 80%, where the most of the remaining were covered by clouds. We also analyse images in the 1.38 μm water absorption band, where only 61% of the aircrafts are detected.

Response of supraglacial rivers and lakes to ice flow and surface melt on the northeast Greenland ice sheet during the 2017 melt season

Fast ice flow and substantial surface runoff form distinct and extensive supraglacial rivers and lakes on the northeast Greenland Ice Sheet (GrIS). These supraglacial rivers and lakes play an important role in controlling surface meltwater routing and storage. However, the hydromorphology of these complex supraglacial features remains poorly quantified. In this study, we mapped the multi-temporal supraglacial rivers and lakes on the northeast GrIS (area ~ 3 × 104 km2) using sixty-five 10 m Sentinel-2 multispectral satellite images acquired during the 2017 summer, and quantified their primary hydromorphology, including meltwater area fraction, meltwater volume, supraglacial river length, river sinuosity, river depth, supraglacial lake area, lake shape, and lake depth. The quantified hydromorphology was compared with variable ice flow regimes and regional climate model (RCM) runoff. The results indicate that widespread supraglacial rivers and lakes maximally cover 5.3% of the northeast GrIS, which is more than double the previous estimates and only slightly lower than that for the strongly melting southwest GrIS (~7.0%). As the surface runoff increases and the snowline retreats, supraglacial rivers and lakes migrate from the ice sheet margin (~400 m) to high elevations (~1400 m). The satellite-mapped surface meltwater area fraction for the entire study area is positively correlated with the RCM surface runoff before the timing of peak melt (August 1), but decreases relatively slowly (by 32–45%) compared to the rapidly decreasing runoff (by 84–96%) in August. Large elevational variations were found for the peak magnitude and timing of the meltwater extent and volume. High (800–1400 m) elevation bands achieve peak values over two weeks later than low (400–800 m) elevation bands, and the 800–1000 m elevation band yields maximum peak values due to the wide distribution of supraglacial rivers. Furthermore, the supraglacial drainage patterns vary with different ice flow regimes. There are long rivers and few lakes in areas where ice deformation controls the ice flow velocity while there are short rivers and large, narrow lakes in areas where basal sliding is a major component of the ice flow. Meanwhile, long, sinuous rivers and small lakes form on the floating ice. In general, this study conducts a preliminary investigation of supraglacial rivers and lakes on the poorly-studied northeast GrIS and reveals that surface melt and ice flow control the hydromorphology of supraglacial rivers and lakes.

Unsupervised GRNN flood mapping approach combined with uncertainty analysis using bi-temporal Sentinel-2 MSI imageries

ABSTRACT Floods occur frequently worldwide. The timely, accurate mapping of the flooded areas is an important task. Therefore, an unsupervised approach is proposed for automated flooded area mapping from bi-temporal Sentinel-2 multispectral images in this paper. First, spatial–spectral features of the images before and after the flood are extracted to construct the change magnitude image (CMI). Then, the certain flood pixels and non-flood pixels are obtained by performing uncertainty analysis on the CMI, which are considered reliable classification samples. Next, Generalized Regression Neural Network (GRNN) is used as the core classifier to generate the initial flood map. Finally, an easy-to-implement two-stage post-processing is proposed to reduce the mapping error of the initial flood map, and generate the final flood map. Different from other methods based on machine learning, GRNN is used as the classifier, but the proposed approach is automated and unsupervised because it uses samples automatically generated in uncertainty analysis for model training. Results of comparative experiments in the three sub-regions of the Poyang Lake Basin demonstrate the effectiveness and superiority of the proposed approach. Moreover, its superiority in dealing with uncertain pixels is further proven by comparing the classification accuracy of different methods on uncertain pixels.

Analysis of Salt Lake Volume Dynamics Using Sentinel-1 Based SBAS Measurements: A Case Study of Lake Tuz, Turkey

As one of the largest hypersaline lakes, Lake Tuz, located in the middle of Turkey, is a key waterbird habitat and is classified as a Special Environmental Protection Area in the country. It is a dynamic lake, highly affected by evaporation due to its wide expanse and shallowness (water depth <40 cm), in addition to being externally exploited by salt companies. Monitoring the dynamics of its changes in volume, which cause ecological problems, is required to protect its saline lake functions. In this context, a spatially homogeneous distributed gauge could be critical for monitoring and rapid response; however, the number of gauge stations and their vicinity is insufficient for the entire lake. The present study focuses on assessing the feasibility of a time-series interferometric technique, namely the small baseline subset (SBAS), for monitoring volume dynamics, based on freely available Sentinel-1 data. A levelling observation was also performed to quantify the accuracy of the SBAS results. Regression analysis between water levels, which is one of the most important volume dynamics, derived by SBAS and levelling in February, April, July and October was 67%, 80%, 84%, and 95% respectively, for correlation in the range of 10–40 cm in water level, and was in line with levelling. Salt lake components such as water, vegetation, moist soil, dry soil, and salt, were also classified with Sentinel-2 multispectral images over time to understand the reliability of the SBAS measurements based on interferometric coherence over different surface types. The findings indicate that the SBAS method with Sentinel-1 is a good alternative for measuring lake volume dynamics, including the monitoring of water level and salt movement, especially for the dry season. Even though the number of coherent, measurable, samples (excluding water) decrease during the wet season, there are always sufficient coherent samples (>0.45) over the lake.