Tutorial Session Details

This tutorial session aims to expose participants to techniques for monitoring landcover changes through the Dynamic World dataset using Google Earth Engine (GEE). The Dynamic World (DW) dataset is a continuously updating Image Collection of globally consistent, 10m-resolution, near real-time (NRT) land use land cover (LULC) predictions created from Sentinel-2 imagery produced using a deep-learning modeling approach. Images in this dataset include ten bands: nine bands with estimated probabilities for each of the nine LULC classes. This will be a hands-on workshop that will introduce the participants to this new global dataset and techniques for working with it in GEE. The workshop will have several practical hands-on exercises for building applications for Mapping Landslides, Monitoring Deforestation, Detecting Urban Sprawl and more.

The topics will be covered are:

• Introduction to Dynamic World,
• Visualization and Creating Composites,
• Exploring LULC Time Series,
• Change Detection using Probability Bands

Selected Applications:Monitoring Deforestation, Mapping Landslides, Detecting Urban Sprawl.

• Remote Sensing Fundamentals
• Basics of Google Earth Engine and familiarity with Code Editor

Earth Observation (EO) and Deep Learning (DL) have emerged as transformative technologies with immense potential for revolutionising our understanding of the planet and addressing critical global challenges. This tutorial proposal seeks to provide a comprehensive overview of the fundamental concepts in both fields and showcase the practical applications, with a specific focus on how we at SatSure leverage this synergy. In addition to theoretical knowledge, participants will have the opportunity to engage in hands-on exercises and demonstrations, where they will apply DL techniques to EO datasets, thereby reinforcing their understanding and skillset. Furthermore, the tutorial will provide insights into the current trends, challenges, and future directions in the EO and DL domains.

Following topics will be covered:

1. 1. Introduction to EO: This session aims to familiarise the participants with fundamental concepts and the complexity of working with earth observation data. The session will also highlight the importance of EO, analytics on EO, its future potential, challenges and how SatSure and KaleidEO are solving them
   2. Python and DL for EO: This session aims to introduced to fundamental of Python, various libraries used in EO and PyTorch.
   3. Vision Transformers, self-supervised learning and foundation models: In this session, the participants will be explained recent advances in Deep Learning like Vision Transformers (ViT and SAM), Self-supervised learning (MAE) and the emerging trend of vision foundation models. This session will also highlight how these techniques can be applied to EO. The participants will also gain exposure to various DL-based products from SatSure, which we hope to inspire them to learn more.

Case Study 1 (Crop Classification) : This session will be a continuation of Session 3. In this session, the participants will engage in a hands-on session through which they will classify various crops from Sentinel 2 low-resolution data. The participants will be provided with a Google Collab notebook and access to a dataset with which they can train the model.

Case Study 2 (Land Use and Land Cover): In this session, the participants will be provided with a Google Collab notebook and access to a dataset with which they can train a vision transformer for Land Use and Land Cover mapping.

Python programming and Remote Sensing Fundamentals

Nowadays, several space borne Polarimetric Synthetic Aperture Radar (PolSAR) systems are in operation as TerraSAR-X (X-Band), RADARSAT-2 (C-Band), Sentinel-1A&B (C-band), ALOS-2 (L-band), BIOMASS (P-band), SAOCOM (L-band), RCM (C-band) or GF3 (C-Band), or are planned as NiSAR (C&S-band) or ROSE-L (L-band). All of them are designed to have parametric sensitivity.

The availability of spaceborne PolSAR data provides an unprecedented opportunity for applying advanced PolSAR information processing techniques to the important tasks of environmental monitoring and risk management. PolSAR remote sensing offers an efficient and reliable means of collecting information required to extract quantitative geophysical and biophysical parameters from Earth’s surface. This remote sensing technique has found many successful applications in crop monitoring and damage assessment, in forestry clear cut mapping, deforestation and burn mapping, in land surface structure (geology) land cover (biomass) and land use, in hydrology (soil moisture, flood delineation), in sea ice monitoring, in oceans and coastal monitoring (oil spill detection) etc. The scope of different applications is increasing nowadays thanks to the availability of mulitemporal and  polarimetric acquisitions.

SAR Polarimetry represents today a very active area of research in Radar Remote Sensing, and for instance operational polarimetric applications start to be operational in the frame of the Sentinel-1. Consequently, it becomes important to train and to prepare the future generation of researchers to this very important topic.

Following topics will be covered:

1. Overview of Imaging Radar Polarimetry
   • 1. Brief History of Imaging Radar Polarimetry
     2. Spaceborne Polarimetric SAR Systems
2. Electromagnetic Vector Wave and Polarization Descriptors
   • 1. Monochromatic Electromagnetic Plane Wave, Polarization Ellipse
     2. The Jones and Stokes Vectors, the Wave Covariance Matrix
     3. Polarisation basis and Elliptical basis transformation SU(2) and O(4) Unitary groups
     4. Partially polarised waves and Wave Polarisation Decomposition (Born & Wolf) Synthesis
3. Electromagnetic Vector Scattering Operators
   • 1. The Polarimetric Backscattering Sinclair Scattering S Matrix
     2. The Scattering Target Vectors k and W
     3. The Polarimetric Coherency T and Covariance C Matrices
     4. The Polarimetric Mueller M and Kennaugh K matrices
     5. Change of Polarimetric Basis
     6. Target Polarimetric Characterization
4. Polarimetric SAR Speckle Statistics and Filtering
   • 1. Introduction
     2. One-Dimensional SAR Systems
     3. Polarimetric SAR Data Models
     4. Dual and Fully Polarimetric SAR Data Speckle Noise Models
     5. Polarimetric SAR Data Filtering for Single and Multitemporal Acquisitions
5. Polarimetric Target Decomposition Concept
   • 1. Introduction
     2. Eigenvector-Based Decomposition
     3. Model-Based Decompositions
     4. Coherent Decompositions
     5. Dual-pol polarimetric decomposition
6. PolSAR Terrain and Land-Use Classification
   • 1. Maximum Likelihood Classifier Based on Complex Wishart Distribution
     2. Classification Based on Scattering Mechanisms and Wishart Distribution
     3. Quantitative Comparison of Classification Capability: Fully Polarimetric SAR versus Dual- and Single-Polarization SAR
7. Polarimetric SAR Applications
   • 1. Polarimetric Temporal Series Analysis. Change Detection and Temporal Analysis
     2. Polarimetric Differential SAR Interferometry. Subsidence and Landslide Monitoring
     3. Pol-InSAR Forest Mapping and Classification
     4. Snow Monitoring
     5. Ship Detection and Analysis

This lecture is intended to scientists, engineers and students engaged in the fields of Radar Remote Sensing and interested in Polarimetric SAR image analysis and applications. Some background in SAR processing techniques and microwave scattering would be an advantage and familiarity in matrix algebra is required.

A major strength of geospatial and location technologies is their ability to integrate and analyze data from diverse providers concerning many different phenomena so as to better understand or predict what is happening in a given area. However, this diversity of data means that preparing the acquired data for integration and analysis remains a time-consuming task. Furthermore, many geospatial data users lack the expertise, infrastructure, and internet bandwidth to efficiently and effectively access, preprocess, and utilize the growing volume of geospatial data available for local, regional, and national decision-making.

CEOS defines ARD as “satellite data that have been processed to a minimum set of requirements and organized into a form that allows immediate analysis with a minimum of additional user effort and interoperability both through time and with other datasets.” However, formal Standardization of the concepts relating to ARD are not widely understood or specified.

OGC ARD SWG along with the ISO/TC 211 ARD Standard project team are working to define a multi-part Standard that specifies a set of minimum requirements that a geospatial data product shall meet in order for the product to qualify as an ARD product. This tutorial aims to provide an overview of the ARD standardization process and highlights the challenges as well as the opportunities posed by such development. The flow and utilisation of standardized ARD will be demonstrated during the tutorial.

The tutorial is self-contained and participants will be provided with the necessary information and support to run all exercises. Basic familiarity with EO data, OGC and ISO standards will be helpful.