ExtremeEarth is a three-year H2020 ICT research and innovation project. Its main objective is to develop Artificial Intelligence and big data technologies that scale to the large volumes of big Copernicus data, information and knowledge, and apply these technologies in two of the European Space Agency (ESA) Thematic Exploitation Platforms (TEP): Food Security and Polar.

Bringing together a number of cutting-edge technologies that range from storing extremely large volumes of data all the way to developing scalable machine learning and deep learning algorithms in a distributed manner and having them operate over the same infrastructure poses unprecedented challenges. One of these challenges is the integration of European Space Agency (ESA)'s Thematic Exploitation Platforms (TEPs) and data information access service platforms with a data platform, namely Hopsworks, which enables scalable data processing, machine learning, and deep learning on Copernicus data, and development of very large training datasets for deep learning architectures targeting the classification of Sentinel images. In this article, we present the software architecture of ExtremeEarth that aims at the development of scalable deep learning and geospatial analytics techniques for processing and analyzing petabytes of Copernicus data. The ExtremeEarth software infrastructure seamlessly integrates existing and novel software platforms and tools for storing, accessing, processing, analyzing, and visualizing large amounts of Copernicus data. New techniques in the areas of remote sensing and artificial intelligence with an emphasis on deep learning are developed. These techniques and corresponding software presented in this article are to be integrated with and used in two ESA TEPs, namely Polar and Food Security TEPs. Furthermore, we present the integration of Hopsworks with the Polar and Food Security use cases and the flow of events for the products offered through the TEPs.

The potential of Sentinel-1 recordings, mainly the most extensive spatial coverage and the highest temporal resolution, is not fully exploited in polar areas because of thermal noise. The strong intensity of the noise in low backscattering areas makes it resemble ice, which deteriorates the sea-ice analysis. Moreover, in multi-looking modes, the thermal noise varies across the sub-swaths and introduces large stripes at their intersection. In this paper, we propose a noise removal approach based on wavelets. The experimental analysis conducted on several images shows the aptitude of the proposed method to effectively and efficiently eliminate the noise.

We explore new and existing convolutional neural network (CNN) architectures for sea ice classification using Sentinel-1 (S1) synthetic aperture radar (SAR) data by investigating two key challenges: binary sea ice versus open-water classification, and a multi-class sea ice type classification. The analysis of sea ice in SAR images is challenging because of the thermal noise effects and ambiguities in the radar backscatter for certain conditions that include the reflection of complex information from sea ice surfaces. We use manually annotated SAR images containing various sea ice types to construct a dataset for our Deep Learning (DL) analysis. To avoid contamination between classes we use a combination of near-simultaneous SAR images from S1 and fine resolution cloud-free optical data from Sentinel-2 (S2). For the classification, we use data augmentation to adjust for the imbalance of sea ice type classes in the training data. The SAR images are divided into small patches which are processed one at a time. We demonstrate that the combination of data augmentation and training of a proposed modified Visual Geometric Group 16-layer (VGG-16) network, trained from scratch, significantly improves the classification performance, compared to the original VGG-16 model and an ad hoc CNN model. The experimental results show both qualitatively and quantitatively that our models produce accurate classification results.

In this article, we propose a novelteacher–student-based label propagation deep semisupervised learning (TSLP-SSL) method for sea ice classification based on Sentinel-1 synthetic aperture radar data. For sea ice classification, labeling the data precisely is very time consuming and requires expert knowledge. Our method efficiently learns sea ice characteristics from a limited number of labeled samples and a relatively large number of unlabeled samples. Therefore, our method addresses the key challenge of using a limited number of precisely labeled samples to achieve generalization capability by discovering the underlying sea ice characteristics also from unlabeled data. We perform experimental analysis considering a standard dataset consisting of properly labeled sea ice data spanning over different time slots of the year. Both qualitative and quantitative results obtained on this dataset show that our proposed TSLP-SSL method outperforms deep supervised and semisupervised reference methods.

In this paper, we explore the potential of deep learning (DL) networks to produce reliable sea ice classification by analyzing data collected by synthetic aperture radar (SAR) sensors over polar regions. Taking advantage of their ability to extract features from complex datasets, DL schemes can be used to perform large scale data investigation, so to help manual interpretation conducted by experts in sea ice charting services. We highlighted the ability of different DL settings as well as their limits (mainly associated with scarce training data). Experimental results show the validation accuracy and the inference potential of three DL networks.

The additive system noise in synthetic aperture radar (SAR) imagery is a challenging problem for the operational use of SAR data for sea ice classification. This noise degrades the performance of the sea ice classification models. The most common way of dealing with this is to remove mean noise profiles from the backscatter intensities as a preprocessing step. In this study we investigate how including the nominal noise profiles as a feature directly into the model affects the classification. Our noise-aware approach can be used in conjunction with any other deep learning model for sea ice classification. Hence our findings pave the way for getting refined and smoother sea ice maps for ice charting. For experimentally evaluating our proposed approach, we train our noise-aware deep model using carefully labeled data consisting of both sea ice data and noise profile. We present validation results considering separate validation data. Our empirical study confirms the superior performance of the CNN model driven by noise-aware characteristics.

This dataset has been prepared for Ice types/Ice edge analysis based on deep neural networks. The dataset has been created based on 31 scenes in north of Svalbard based on labeled polygons. The dataset contains six classes including OpenWater, Leads with water, Brash/Pancake Ice, Thin Ice, Thick Ice-Flat and Thick Ice-Ridged. The data records, called patches, extracted all from inside of each polygon with stride 10 in different sizes, 10x10, 20x20, 32x32, 36x36, 46x46 pixels for each class

Copernicus is the European programme for monitoring the Earth. It consists of a set of systems that collect data from satellites and in-situ sensors, process this data and provide users with reliab ...

This paper shows initial results from estimating Doppler radial surface velocities (RVLs) over Arctic sea ice using the Sentinel-1A (S1A) satellite. Our study presents the first quantitative comparison between ice drift derived from the Doppler shifts and drift derived using time-series methods over comparable time scales. We compare the Doppler-derived ice velocities with global positioning system tracks from a drifting ice station as well as vector fields derived using traditional cross correlation between a pair of S1A and Radarsat-2 images with a time lag of only 25 min. A strategy is provided for precise calibration of the Doppler values in the context of the S1A level-2 ocean RVL product. When comparing the two methods, root-mean-squared errors (RMSEs) of 7 cm/s were found for the extra wide (EW4) and EW5 swaths, while the highest RMSE of 32 cm/s was obtained for the EW1 swath. Though the agreement is not perfect, our experiment demonstrates that the Doppler technique is capable of measuring a signal from the ice if the ice is fast moving. However, for typical ice speeds, the uncertainties quickly grow beyond the speeds we are trying to measure. Finally, we show how the application of an antenna pattern correction reduces a bias in the estimated Doppler offsets.

Using data from the Envisat Advanced Synthetic Aperture Radar (ASAR) instrument, this paper demonstrates how the high-precision radial surface velocity product, which will become available with the European Space Agency's Sentinel-1 satellite, can complement the analysis of sea ice motion. High-resolution Doppler frequency measurements are used to estimate the subsecond line-of-sight motion of drifting sea ice in Fram Strait. We compare the method with buoy measurements and a recent cross-correlation algorithm for tracking ice between pairs of images. Maximum speeds measured from the time series were on the order of 20 cm/s. Using our method, we measured instantaneous speeds reaching 40-60 cm/s.

This paper shows that Envisat ASAR data are degraded by a small periodic variation in gain between internal calibration cycles of the antenna, which introduces a significant bias when we try to estimate line-of-sight surface velocities from the estimated Doppler frequency shifts. We investigate the impact of the gain problems on the derived surface velocity product and propose a simple correction of the raw data.

This paper discusses problems found when applying current techniques for estimating sea ice motion using synthetic aperture radar (SAR) to a non-trivial dataset covering the Barents Sea. We review two commonly used similarity measures and their weaknesses when the ice rotates and deforms. We then extend a current algorithm to include rotation estimation and give some considerations on parameter selection. Initial impressions from the use of HV polarization to complement use of HH polarization is also discussed.