Conference Agenda

Satellite image time series are crucial for mapping and monitoring ecological status and changes in forested ecosystems. Forthcoming global imaging spectroscopy missions such as CHIME and SBG will increase our capacity to better characterize key ecological information on vegetation, related to canopy chemical content and species composition at a community scale. The massive amount of Earth observation data poses a challenge for scientific applications dedicated to both research and operational use: volume and complexity of imaging spectroscopy data present significant challenges in data access, storage, processing, and analysis. Enhanced accessibility and usability of global imaging spectroscopy data through cloud-based systems needs to be anticipated.

Cloud-based systems offer scalable storage solutions and powerful computational resources that can handle the large-scale data associated with global imaging spectroscopy. By utilizing cloud infrastructure, researchers can bypass the limitations of local storage and computing capabilities. This approach facilitates seamless data sharing and collaboration among scientists worldwide, promoting more extensive and integrated research efforts. One critical aspect of this transition involves developing standardized data formats and metadata, ensuring that imaging spectroscopy data are easily accessible and interpretable. Implementing standardized protocols for data ingestion, storage, and retrieval on cloud platforms is essential for maintaining data integrity and usability. Additionally, cloud-based systems can integrate advanced data processing tools and algorithms, enabling real-time analysis and visualization.

The Ecosystem Extent Task Team of the Committee on Earth Observation Satellites (CEOS) is currently exploring the possibilities offered by the combination of cloud infrastructures and standardized protocols for producing essential biodiversity variables. A demonstrator based on free and open software is currently being developed to improve our capacity for forest diversity mapping and forest degradation and dieback monitoring based on Sentinel-2 time series processing over large scales. The aim of this demonstrator is to overcome current barriers between local scale and large scale analysis, and to improve the capacity to transfer between research development and operational applications. In the perspective of future imaging spectroscopy missions, these technological advances will enhance the scalability and flexibility of imaging spectroscopy data analysis, and the synergy with other sensors.

With upcoming missions able to deliver SI-traceable satellite observations with an anticipated radiometric accuracy of 1% or better comes the challenge of providing corresponding simulations with a matching accuracy. In particular, input data assembly practices must be adapted to comply with these accuracy requirements. In this work, we assess the impact of the method used to build the atmospheric vertical profile on hyperspectral radiative transfer simulations over bright desert pseudo-invariant calibration sites. Two methods are considered: using an AFGL U.S. Standard profile with the water vapour and ozone concentrations rescaled, and deriving a vertical profile from CAMS data. We first assess the impact of vertical profile assembly on the simulated top-of-atmosphere reflectance for multi-spectral observations in spectral regions affected by water vapour, ozone, and methane. The impact on hyperspectral observations acquired by PRISMA, covering the entire visible to SWIR spectral region. When the molecular absorption transmittance is greater than 97%, both vertical profile assembly methods provide results that agree within 1%. Between 75% and 97%, the corresponding uncertainty on the TOA BRF simulation increases from 1% up to 5%. Below 75%, the uncertainty becomes larger and the method using CAMS-derived profiles is highly recommended. Additionally, we estimate the effect of the pressure and temperature profiles on the Rayleigh optical thickness, and thus on the TOA BRF in the blue spectral region.

These findings have been used to compare PRISMA, EnMAP and EMIT observations acquired over bright desert PICS with simulations performed with the Eradiate model based on CAMS vertical profiles. Results show that observations agree with the simulated calibration reference within ±5% across most of the spectral domain. The generation of this simulated calibration reference can therefore be used for the harmonization of hyperspectral observations. This research is funded by the ESA CalibrEO and QA4EO HyperPICS projects.

The Environmental Mapping and Analysis Program (EnMAP) satellite mission was launched on 1^st April 2022, designed to make a significant contribution to the availability of space-based highly calibrated imaging spectroscopy products for the characterization and monitoring of the Earth’s environment and its changes. The EnMAP mission stands for its long-term development and strong science accompanying program, high data quality requirements and challenging concept. After a successful short commissioning phase, EnMAP entered into operation in November 2022, covering the whole globe with data request on-demand at 30 m with a high spectral resolution and high signal-to-noise ratio. The primary scientific goals of the hyperspectral EnMAP mission are to study environmental changes, investigate ecosystem responses to human activities, and monitor the management of natural resources.

Within EnMAP scientific exploitation and support program in the operational phase, priority is placed on the development of demonstration of scientific achievements for various cutting-edge and environmental applications of benefit to society today, and on new developments in open source algorithms and educational tools to support the increased use of the EnMAP and in general of imaging spectroscopy spaceborne missions through developing free and open-source data processing tools and an education program to foster an expert user community. In this presentation, we will focus on showing selected examples of new geo- and bio- EnMAP products, and new achievements and plans to integrate science algorithms and to release new resources within EnMAP-Box toolbox and HYPERedu education program. Further, these advances shall support the establishment of synergies and collaboration with concordant missions (e.g. PRISMA, EMIT), and support the development of open test bench datasets and algorithms as key long-term objectives, which can serve as precursor for future missions like CHIME and SBG.

Atmospheric radiative transfer models (RTMs) play a critical role in satellite data processing by modeling the scattering and absorption effects caused by aerosols and gas molecules in the Earth's atmosphere. As the complexity of RTMs increases and the demands of future Earth Observation missions escalate, conventional Look-Up Table (LUT) interpolation approaches encounter significant challenges. Statistical methods called emulators have been proposed as an alternative to LUT interpolation, yet they often fall short in operational satellite data processing due to their slow performance. Moreover, their black-box nature hampers our understanding of how the predictions are made and limits the physical interpretability of these models.

Our research introduces a physics-aware solution that leverages multi-fidelity methods to enhance the accuracy and runtime efficiency of Gaussian Process (GP) emulators. By investigating the impact of the number of fidelity layers, dimensionality reduction, and training dataset size, we demonstrate that multi-fidelity emulators can achieve relative errors in surface reflectance below 0.5% and perform atmospheric correction of hyperspectral satellite data in just a few minutes. Furthermore, we propose a methodological framework that improves model emulation and interpretability through machine learning feature selection. Our wrapper-forward feature selection method integrates physics knowledge into model emulation, striking a balance between accuracy and interpretability. Applied to global sensitivity analysis (GSA) and emulation, our approach identifies critical features in the RTM input space, providing a significant advancement in the field of physics-aware machine learning-based emulation, enhancing both the performance and understanding of atmospheric RTM emulation.

In this presentation we will describe the current and on-going activities for physics-aware emulation and we will share with the community a suite of functions and tools for automating the creation and generation of atmospheric RTM emulators.

FLEX (FLuorescence EXplorer) is ESA's 8th Earth Explorer mission. The mission aims to provide an insight of the photosynthetic activity of vegetation by characterising its full energy balance (i.e., incoming radiation, reflected light, surface temperature, and fluorescence). Flying in tandem with Sentinel-3, FLEX will provide advanced biophysical products to understand photosynthetic activity with potential applications into stress detection and food productivity. In the last 4 years, several activities have been carried out spanning the development of FLEX core Level-1B (L1B) and Level-2 (L2) mission products, their validation with simulated data generated by an end-to-end mission performance simulator (E2ES), FLEX-related field campaigns, and processing of in-situ data. We recently started the activities for the implementation of FLEX Ground Segment within the so-called Data Innovation and Science Cluster (DISC), a team of scientists and technical experts in various domains related to the FLEX mission that will ensure efficient mission operations to provide the best data quality and will involve potential users of the FLEX data through outreach activities. The activities to be carried out by the FLEX DISC will encompass the following: (1) consolidating the prototype of L2 processor and industrialising it into an Instrument Processing Facility (IPF), (2) developing a collaborative platform that allows bringing FLEX users and algorithms to data, (3) developing tools to monitor the quality of the FLEX data products (Level-0 to L2), (4) design and implement a calibration/validation during commissioning phase and regular operation, (5) monitoring the FLEX data quality and maintaining calibration/validation, and (6) implementing evolutions of the processing algorithms to ensure state-of-the-art data products.

The goal of this presentation is twofold: (1) giving an overview of ESA's FLEX mission and products; and (2) describing the DISC project from its objectives to the consortium and on-going activities.

.