Conference Agenda

The Earth is facing unprecedented climatic, geomorphologic, environmental and anthropogenic changes, which require global scale observation and monitoring. The interest is in understanding involving Earth Observations (EO) of large extended areas, and long periods of time, with a broad variety of satellite sensors. The collected EO data volumes are thus increasing immensely with a rate of many Terabytes of data a day. With the current EO technologies these figure will be soon amplified, the horizons are beyond Zettabytes of data.

1）“Pre-trained and fine-tuning" is one of main paradigms that pre-train a fundamental model with large-scale unlabelled data in a unsupervised learning way and then retrain it with a small amount of labeled data for downstream tasks. Pre-trained models are demonstrated to be of strong generalization and adaptation to multi-tasks. To address the challenges of the difficulty and high cost of manual ground truth labeling, a three-dimensional masked autoencoder (MAE) self-supervised learning method is designed based on an improved masked autoencoder (MAE) self-supervised framework for SAR and optical image joint self-supervised learning to enhance the feature extraction ability in the vertical direction along modal channels. Experimental results show that the proposed method surpasses the state-of-the-art comparative learning and MAE-based models in land cover classification tasks and reduces data input through vertical masking to achieve a more efficient model. Furthermore, additional experiments show that the proposed model has good generalization and can maintain good representation learning capabilities on small-scale data.

2）A remote sensing image self-supervised learning method based on SimMIM is pre-trained , and a MIM-SwinUNet is fine-tuned for land cover classification model supervisedly. The experiment shows that the self-supervised pre-trained model can effectively extract generalizable image features, and when transferred to downstream land cover classification tasks, it can achieve similar classification performance with significantly reduced labeled training sample size. Based on self-supervised labeling learning methods, multi-temporal remote sensing image land cover classification and land use change analysis are carried out in the case of Shanghai area using Sentinel 1 and 2 data.

3）The challenge is the exploration of these data and the timely delivery of focused information and knowledge in a simple understandable format. In this context we envisage the monitoring of Danube Delta and Black see costal areal. The study is directed to the modeling and understanding of climate change effects, particularly droughts and maritime currents. Droughts are studied using multispectral Satellite Image Time Series (SITS) of Sentinel 2. The study case is focused on the ensemble of lakes between the Black Sea coast and Danube Delta for the period 2019 to 2022. The Sentinel 2 SITS are analyzed to quantitively measure the lakes water surface, as the case of lake “Nuntasi”. During the 2020 drought the lake was completely dearth, a channel was built connecting it to the neighboring larger lake and refilling it. The SITS characterizes both the water level and quality variation. The Black Sea surface current of in the coastline limitrophe area are analyzed using SAR SITS from Sentinel 2. The maritime surface currents are characterized estimating the Doppler frequency of the SAR images. The SITS data are used to predict current patterns

As the most common form of ocean movement, mesoscale eddies promote the redistribution of marine variables, such as temperature, salinity and nutrients, through the transport of material and energy. They have an important influence on the marine biogeochemistry cycle, marine ecosystem and marine heat balance etc. Through the 2D/3D structural visualization of multiple variables of mesoscale eddies, the motion patterns of mesoscale eddies are directly visual through graphics and images, greatly contributing to studying mesoscale eddies.

Under the Euler coordinate, various methods for extracting mesoscale eddies have been proposed based on their basic features, among which the sea level anomaly (SLA)-based methods have performed better because these methods are able to avoid extra noise and excess eddy detections. A previous SLA-based method has been provided to identify and track global eddies. This highly effective orthogonal parallel algorithm greatly improves the efficiency of recognition without reducing the accuracy of mesoscale eddy recognition. The global eddy identification and trajectory dataset is built with a total time span between 1993 and 2020, which provides a data foundation for the subsequent study of mesoscale eddies. Affected by the modulation of various physical mechanisms and the complex marine environments, there are also complex dynamic processes such as eddy splitting, eddy merging, and dipoles. In this case, an automatic recognition method of global eddy dipoles is developed in terms of the mesoscale eddy dataset and the transmission modes as well with the characteristics of dipoles are simultaneously analyzed. In addition, an algorithm named EddyGraph for tracking mesoscale eddy splitting and merging events is come up with based on multi-level topological relationships, which helps to analyze the statistical characterization of global eddy splitting and merging events.

Under the Lagrangian coordinate, eddies are the cumulative results of the state of the fluid within a given time scale, which can maintain material coherence over the specified time intervals. By using the elliptic Lagrangian coherent structures, a typical black-hole eddy is extracted based on the data of the geostrophic flow velocity field. Combined with multi-source satellite remote sensing data and in-situ data, it shows that the black-hole eddy boundary can describe material transport more objectively than the Euler eddy boundary on a longer time scale. On the regional scale, Lagrangian eddies in the Western Pacific are successfully extracted and their spatial and temporal variations are analyzed. Through normalized chlorophyll data, it is observed that Lagrangian eddies can cause chlorophyll aggregation and hole effects. These findings demonstrate the important role of Lagrangian eddies in material transport. Nevertheless, the high calculation cost during the integration process has become a bottleneck, especially when the data resolution is improved or the study area is enlarged. Therefore, SLA-based orthogonal parallel detection of global rotationally coherent Lagrangian eddies is built, whose runtime is much faster than the previous nonparallel method. Finally, a dataset of long-term global Lagrangian eddies is established.

Based on objective reference framework and criteria, the extraction and visualization of the mesoscale eddy coreline, an ocean three-dimensional structure, are achieved by extracting the valley line of the obtained from objective flow field calculations as the eddy coreline. At the same time, equipped with an integrated visualization system, named i4Ocean, a standard morphological model of the transfer function for ocean thermohaline anomaly data and pressure anomaly data is designed from the number of feature points, feature color mapping and the line shape. Volume rendering technology and spherical ray casting algorithm are utilized to more clearly and completely display the large-scale ocean 3D eddies under the condition of ensuring the rendering quality. Based on 2D and 3D flow field vector data, the spatio-temporal continuity of ocean flow field visualization is enhanced under the whole spatio-temporal continuous framework of pathline-pathline. The geometry-based visualization animation of trace becomes smoother and more stable after solving the problem of aliasing in previous visualized ocean flow fields. Applying region-based eddy detection techniques (ow method, Q method, and Ω method) to ocean flow fields, the extracted mesoscale eddies are more comprehensive. Based on the ow criterion, Q criterion, and Ω criterion, standard transfer functions are constructed to optimize the extraction effect of ocean mesoscale eddies, reduce the difficulty of analyzing ocean mesoscale eddies through user interaction transfer functions, and improve the efficiency of user interaction analysis of ocean mesoscale eddies.