Conference Agenda

In current imaging spectroscopy missions for Earth surface monitoring, data is widely available and represents a valuable resource. However, effectively handling, storing, and transmitting to ground the extensive datasets generated by modern mission hosting hyperspectral payloads requires specialized onboard infrastructure and significant computational resources. As a result, there is a risk of reducing the potential to fully exploit the real satellite acquisition capabilities.

These considerations open an opportunity for the improvement of new missions and form the basis of the innovative COME-ON-BOARD-PSG! project, supported by the Italian Space Agency. This project proposes enhancing the platform with onboard computing capabilities, breaking the traditional division between the space segment, solely responsible for data acquisition, and the ground segment, responsible for data processing.

In detail, a conceptual design for the upcoming PRISMA Second Generation (PSG) mission is proposed, with a focus on augmenting the satellite's decision autonomy through onboard edge computing with machine learning (ML). This is achieved through two possible distinct technological solutions. One low-impact solution involves equipping the primary hyperspectral payload with a computing unit featuring ML algorithms for real-time onboard cloud detection of acquired images to exclude those with a cloud coverage percentage exceeding a predefined threshold. The option with a greater impact on the system involves integrating a secondary RGB payload, forward-looking and oriented between the nadir direction and the satellite velocity. Acquisitions from this camera, once analyzed by the computing unit equipped with ML, could enable the real-time evaluation of cloud coverage potentially impacting future hyperspectral acquisitions thus making informed decisions about off-nadir maneuvers maximizing the probability of capturing cloud-free images.

Furthermore, the project aims to address an issue of international interest by exploring the feasibility of equipping the satellite's main payload with ML-based real-time fire detection capabilities. This would enable the satellite to detect wildfires and high-temperature phenomena, providing timely warnings to support ground operations during natural disasters.

Data from additional and main cameras are processed by dedicated devices such as graphic processing units or field programmable gate arrays. Preliminary on-ground results show promising fire detection with PRISMA hyperspectral imagery, using convolutional neural networks trained on datasets from Australia and Oregon. Edge implementation on an Nvidia Jetson TX2 demonstrates suitability for space missions, providing 98% accuracy, 3.0 ms inference time, and 4.8 W average power consumption. The cloud detection component is still pending implementation but leverages the experience from previous successful missions such as Phisat-1.

Imaging spectroscopy in the VNIR/SWIR spectral range has demonstrated great potential for characterizing the chemical and physical properties of the Earth’s surface. Following the Hyperion space mission, other satellites have recently become operational (PRISMA, EnMap), and global missions are under study (CHIME, SBG). Most of these have a ground sampling distance (GSD) of 30 m, which limits the characterization of heterogeneous ecosystems due to a large number of mixed pixels. The BIODIVERSITY mission (ex-HYPXIM) aims to complement these space missions with a unique combination of characteristics, including a GSD of 10 m, a revisit time of up to 5 days and a spectral range from 0.4 to 2.4 µm.

To support BIODIVERSITY[1], the French scientific community has focused on identifying requirements for spectral and radiometric resolution, and absolute calibration, assessed on the basis of a set of applications covering the above-mentioned topics. For each topic, illustration and performance on the estimated variables are presented and the best configurations are deduced. These applications include:

• Classification of temperate forest species and estimation of vegetation traits at tree-level in Mediterranean forests (photosynthetic pigments such as chlorophyll ab and carotenoids, leaf water content and leaf mass per area),
• Estimation of the shallow water biodiversity and shallow water bathymetry in coastal and inland waters,
• Characterization of soil surface properties (mineral discrimination, soil moisture content) to assess soil pollution and soil quality at fine spatial resolution, providing information on the influence of soil management practices on environmental processes such as soil carbon sequestration, infiltration and retention, runoff and soil erosion,
• Monitoring cities and industrial pollution to assess urban sprawl or the quality of our environment. Among the methods tested, the benefit of the hyper-pansharpening method is proposed,
• Snow characterization: specific surface area, black carbon concentration,
• Atmosphere characterization: water vapor, carbon dioxide and aerosols.

[1] X. Briottet, et al., End-to-end simulations to optimize imaging spectroscopy mission requirements for seven scientific applications. ISPRS Open Journal of Photogrammetry and Remote Sensing 12 (2024) 100060. DOI:10.1016/j.ophoto.2024.100060

Atmospheric correction is one of the main steps when processing optical data from satellites. Its main task is the conversion of the top-of-the-atmosphere (TOA) radiance signal measured by a satellite instrument into surface reflectance. After compensation for atmospheric effects, the derived surface reflectance data can be used to retrieve geophysical properties for applications such as vegetation monitoring or water quality. In recent years, atmospheric algorithms based on optimal estimation have been proposed, with the potential to obtain highly accurate results and uncertainty propagation based on the implementation of physical principles across the data processing chain. In this context, we present MAGAC (Magellium Atmospheric Correction), a generic toolbox for processing hyperspectral satellite data. MAGAC is divided into two main modules: (1) atmospheric characterization (water vapor and aerosols) and (2) surface reflectance retrieval. The characterization of water vapor is based on the use of H2O absorption characteristics through differential absorption techniques. Aerosols are recovered with an optimal estimation algorithm using a priori information (from climatological and spectral databases) and efficient machine learning emulators of atmospheric radiative transfer models (MODTRAN). Surface reflectance recovery includes compensation for adjacency and topography effects. In this presentation, we will provide an overview of MAGAC and the validation results over PRISMA and EnMAP data in the context of the ESA/NASA ACIX-III Land activity.

NASA's Surface Biology and Geology (SBG) Mission Visible to Shortwave Infrared (VSWIR) terrestrial vegetation algorithm team is working to establish algorithmic strategies for global trait estimates. This effort involves leveraging both existing data and future campaigns that link in situ data with airborne spectra, as well as testing strategies for scaling algorithms with orbital precursors, such as NASA’s EMIT mission. Here, we provide an update on the team’s strategy in the context of the workshop’s six themes.

This update will include an overview of our efforts to create a robust and open-source database, SBG PLANTS. This initiative integrates airborne and in situ datasets starting at the radiance level, aiming to address the inconsistencies caused by varying atmospheric correction strategies across different platforms. By leveraging investments from the SBG mission, the US National Science Foundation, and leadership by university PIs, this project not only enhances the development of trait prediction algorithms but also provides a framework for researchers and the global community to contribute data. Initially, the SBG High Frequency Timeseries (SHIFT) campaign data has been leveraged to shape the database's structure. Planned contributions from the EnSpec lab and the National Ecological Observatory Network, will help ensure compatibility across well-established collection frameworks and facilitate the incorporation of a wide range of existing datasets.

An important component of this effort is to include early community engagement that supports those new to the field in their own collection efforts. We also highlight a future campaign planned by the Rocky Mountain Biological Laboratory which will serve as a beta test for this community-driven data collection strategy. This approach aims to ease the integration of different data sources, enable the community, and provide transparency as the mission progresses, ultimately enhancing the capability and accuracy of SBG VSWIR’s terrestrial vegetation algorithms.

The classification of forest cover in homogeneous units is essential for describing the forest resources of a given area. These units can be delineated by the predominant tree species or groups of species, evaluated in terms of volume, biomass, or canopy cover. Traditional classification methods, which rely heavily on field surveys and expert knowledge, are often labor-intensive and time-consuming, highlighting the need for automatic classification methods [1].
This work introduces a three-branch deep learning methodology for the classification of forest areas by using PRISMA hyperspectral images. The proposed architecture is composed by a three-branch feature extraction part, each branch employing convolutional layers with skip connections [2] tailored to the specific type of data. The balanced extracted features from all branches are then concatenated and classified by a fully convolutional network. The study areas refer to three distinct scenarios in northern, central and southern Italy. For these areas, ground-truth has been defined through a combination of photo-interpretation and in-situ surveys. The hyperspectral PRISMA images have 30-meter spatial resolution across 239 spectral bands. Additionally, the proposed architecture exploits the PRISMA panchromatic representation at higher resolution of 5-meter and digital elevation models.
For the experiments, the ground-truth data were subdivided into training and validation sets using a spatial cross-validation technique [3] based on agglomerative clustering, resulting in 10 different folds. These folds comprised 5719/3810/1499 training samples and 2451/1633/643 validation samples for the northern, central, and southern scenario, with 11, 10, and 4 forest classes, respectively. The performance of the model was assessed averaging the F1-score across folds and classes, yielding scores of 84.60%, 89.60%, and 95.10% for the three scenarios, respectively. A first qualitative assessment of the classification maps confirm the good quality of the results and the generalization capability of the architecture on a regional scale.

Acknowledgements

This activity has been supported by Italian Space Agency inthe AFORISMA project (PRISMA SCIENZA call DC-UOT-2019-061).

References
[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.
[2] Ava Vali, Sara Comai, and Matteo Matteucci. Deep learning for land use and land cover classification based on hyperspectral and multispectral earth observation data: A review. Remote Sensing, 12(15):2495, 2020.
[3] Yanwen Wang, Mahdi Khodadadzadeh, and Ra´ul Zurita-Milla. Spatial+: A new cross-validation method to evaluate geospatial machine learning models. International Journal of Applied Earth Observation and Geoinformation, 121:103364, 2023.

Over the past few decades, spaceborne hyperspectral imagery has evolved significantly, opening up new possibilities for agricultural research and monitoring. More precisely, the retrieval of crop biophysical variables appears to be of increasing interest to the scientific community as hyperspectral data become more widely available. Amongst all variables of interest, crop nitrogen concentration remains probably the most critical. In order to guarantee an optimum level of food production while minimizing their environmental impact, farmers need reliable tools to monitor the nitrogen status of their crop fields. Although numerous studies tend to demonstrate the potential of hyperspectral imagery in the field of agriculture, the spatial and temporal resolution of the most recent sensors are still the subject of debate as to their limitations.

With the same temporal and spatial resolution, hyperspectral sensors would theoretically be better suited to quantifying biophysical plant variables such as chlorophyll or nitrogen content. This study aims to provide insights into how much a better spectral resolution tends to overcome a broader spatial and temporal resolution of the current hyperspectral missions in the framework of developing biophysical variable retrieval applications.

In this study, we compare the performance of a robust model calibrated with 4 years of Sentinel-2 data with the performance of different approaches based on EnMAP data. The model based on Sentinel-2 data is based on the use of the PROSAIL radiative transfer model. This model is then inverted using a neural network to retrieve the nitrogen content of wheat. On the other hand, several nitrogen estimation methods were applied to the EnMAP data to evaluate their predictive power. Normalised Difference Index (NDI) (Band 1 - Band 2)/(Band 1+Band 2) and Simple Ratio Index (SRI) (Band 1)/(Band 2) were calculated for all possible combinations of spectral bands. The indices with the highest correlation with the CNC are then used as explanatory variables in a simple parametric regression. The other two methods are based on two different machine learning algorithms: Random Forest Regression (RFR) and Partial Least Square Regression (PLSR). The results will be compared with measurements from a large field campaign carried out from March to June 2024.

Vegetation canopies are complex three-dimensional structures where, especially in the near infrared part of the spectrum, multiple scattering dominates. These structural effects make it difficult to compare in a meaningful way data acquisitions made under different view and illumination conditions, such as by satellites belonging to different constellations with varying overpass times or spatial and spectral resolution. This also applies to simple vegetation-related remote sensing products such as vegetation indices. A common solution is the use of a physically-based vegetation reflectance model with an inversion scheme. These models, based on the theory of radiative transfer, truthfully convey the spectral scattering of vegetation based on its biochemical composition and some assumed, generally simplified structure. While they provide a more robust approach to estimate some key vegetation properties such as the leaf area index (LAI), they are limited by the assumptions made in the modeling stage, e.g. the spatial distribution of basic scattering elements (leaves or needles) which are too small to be detected by satellite sensors.

An alternative approach, based on the same fundamental equation of radiative transfer, is to use the theory of spectral invariants. The original invariants, related to the eigenvalues of the radiative transfer operator, are part of the successful MODIS LAI algorithm. Later, the first invariant (p) has been shown to be the photon recollision probabability and the other, the escape probability rho, is related to the fraction of sunlit foliage in the field-of-view of the sensor. This simple yet powerful approach has been successfully used to describe chlorophyll fluorescence and within-leaf radiative transfer. Recently, the theory has been expanded to include more parameters, such as the fraction of non-photosynthetic material and demonstrated to be able to separate within- and between-leaf scattering in a hyperspectral image.

With these developments, the spectral invariants provide a new tool to retrieve basic physical properties of the vegetation such as the albedo of the average leaf or recollision probability (a strong function of LAI, depending also on canopy clumping). The theory also provides a physical method to estimate vegetation biochemical content in a manner agnostic of canopy structure and instrument observation conditions. In this presentation, we provide the theory and a few suggestions for spectral invariant based vegetation products, which can be derived from existing and future imaging spectroscopy data.

NASA’s PACE (Plankton, Aerosol, Cloud, ocean Ecosystem) satellite mission is composed of three instruments: the OCI (Ocean Color Instrument), HARP2 (Hyper Angular Rainbow Polarimeter #2), and SPEXone (Spectro-polarimeter for Planetary Exploration one). With a mission focus of exploring Earth’s atmosphere and oceans, PACE’s novel hyperspectral and polarimetric instruments also observe across terrestrial and inland aquatic domains. The Ocean Color Instrument collects global, hyperspectral observations every 1 to 2 days, providing a platform to advance science and applications capabilities for land surface, coastal, and inland water researchers and decision makers. For example, OCI’s proposed terrestrial product suite will include several hyperspectral vegetation indices, providing land managers with detailed information on vegetation pigments linked to crop health and drought status. Similarly, OCI’s inland water suite is anticipated to include products from the Cyanobacteria Assessment Network (CyAN), extending the capacity of water quality and harmful algal bloom research. New hyperspectral algorithms may also be included to tease out specific characteristics relating to harmful algal blooms. PACE’s two polarimeters supplement OCI’s observations, seeing the Earth in polarized light and at many different angles. These instruments may assist the PACE mission with atmospheric topics, as well as terrestrial ones like the Bidirectional Reflectance Distribution Function (BRDF).

The first set of high-level mission requirements for the future French high-resolution hyperspectral mission is inherited from previous mission studies such as Hypxim, and includes requirements from a national group of end-users. Along with the strong need for a high spatial resolution and these stringent requirements, the resulting instrumental design was heavily constrained and the balance between all contributors to the image quality was still to be done.

Based on its legacy on mission engineering and mission performance, CNES has developed an End-to-End Simulator (E2ES) built to assess mission requirements and concepts as well as to help the definition of the system requirements. In addition, it was designed to aid the development and verification of detailed Level-1 and Level-2 algorithms in future mission phases. The E2ES generates hyperspectral images, taking into account the instrument transfer function and major acquisition perturbations. These images can be analyzed by expert end-users to evaluate the performance of their applications. CNES has also included representative end-user scenarios in the simulator. This approach paves the way for a massive evaluation of various instrumental configurations, to speed up the instrumental design process and focus end-user expertise on the main challenges.

The simulator as well as the main instrumental requirements obtained from these analyses will be discussed. For now, the available end-user scenarios cover the following applications: bathymetry, mineralogy and vegetation analysis. The result of this study is to assess the impact of image quality on the final application performances, not to challenge the end-user algorithms themselves. So, it is assumed that the end-user scenarios are based on state-of-the-art algorithms with representative input dataset. End-users are always improving their approaches and compensating for instrumental defects. To overcome the limitation of the implemented scenarios, all performance results are analysed by comparison to the ones obtained with the initial set of requirements.

Overall, the end-to-end simulation of hyperspectral images is an effective tool for evaluating instrument specifications and their impact on end-user applications. This study has helped defining the requirements for the current hyperspectral mission and provides useful insights for future research endeavours.

Estimation of the water-leaving surface reflectance is a critical processing step for remote spectrosocpy of aquatic environments. A future generation of VSWIR imaging spectrometers, like SBG-VSWIR, will enable new studies of water quality and benthic ecosystems globally, but doing so will necessitate reliable atmospheric correction that can account for latitudinal variability in sunglint. This talk describes an innovative new instrument-agnostic approach, based on prior work by Gege and Groetsch (2016), that models sunglint and skyglint separately. This enables imaging spectrometers at fine spatial resolution of 30m and smaller to account for the different distribution of wave facet geometries in each pixel. We use a simultaneous estimation of upwelling backscattered light, along with sky and sunglint levels, and atmospheric terms like aerosol loading that can be difficult to disentangle from sunglint effects. This enables separation and retrieval of these processes, all of which are superficially similar, but which can be distinguished using their spectrally-distinctive impacts on water leaving reflectance in the broad VSWIR interval from 380-2500. A new formalism, based on the spatial constrained retrievals of Eckert et al. (2024) enables retrieval of sunglint independently in each pixel while accounting for spatial constraints across pixels such as the local smoothness of atmospheric fields. We describe the formalism and show its implimentation and performance on a diverse set of EMIT orbital imaging spectroscopy images. Because the approach is based on physics-based models at arbitrary spectral resolution, it is intrinsically instrument agnostic and can be applied across sensors equally regardless of spectral calibration or ground sampling. This makes it a compelling option for the next generation of global imaging spectrometer missions.

A corner stone to mapping photosynthetic dynamics efficiently over large areas of land is the retrieval of sun-induced fluorescence (SIF) from passive remote sensing data. In this contribution we present a novel method to retrieve SIF from hyperspectral imagery that tightly integrates radiative transfer simulations and self-supervised neural network training. Differently to other physically constrained retrieval methods that optimize the parameters to a radiative transfer model (RTM), it reduces the prohibitive computational cost of a physical model deployed to continuous data streams. To achieve this, it couples an emulator of large-scale radiative transfer simulations with a lightweight encoder-decoder neural network architecture and is trained by optimizing a constraint based loss formulation.
This method was developed and tested in the spectral region around the O₂-A absorption band on high-quality data acquired by the HyPlant sensor, the airborne demonstrator sensor for ESA’s upcoming Earth Explorer satellite mission FLEX that aims to provide global hyperspectral imagery for SIF retrieval. In a validation study with in-situ SIF measurements we find better performance than the traditional Spectral Fitting Method (Cogliati et al. 2019). Furthermore, an adapted version of our approach yields consistent SIF estimates on hyperspectral data of the spaceborne DESIS sensor onboard the International Space Station (ISS). This result is encouraging since DESIS only provides spectrally low-resolved imagery (2.55 nm SSD, 3.5 nm FWHM) compared to HyPlant (0.11 nm SSD, 0.25 nm FWHM). In a unique data set consisting of quasi-simultaneous, spatially matching DESIS and HyPlant acquisitions, the DESIS SIF estimates achieve a mean absolute difference of less than 0.5 mW nm^-1 sr^-1 m^-2 with respect to HyPlant derived estimates. Furthermore, the method yields SIF estimates that align well with the equally ISS-based OCO-3 SIF product.
The proposed methodology could benefit research in computationally efficient full-spectrum SIF prediction from FLEX data. While our method has been tested only in the O₂-A absorption band of HyPlant and DESIS acquisitions, principally it can be adapted in a straightforward fashion for retrieval in other spectral regions and in data from different sensors. Future work will thus include recently published simulated FLEX imagery and our simulation tool developed for DESIS SIF prediction to gauge the method’s applicability in FLEX-like data.

Snow plays a key role in Earth’s radiation budget by having the highest albedo of all natural surfaces, and therefore, introducing potential cooling effects to Earth’s climate. This cooling potential is a direct function of the amount of absorbed solar radiation. However, as snow warms and melts, its effective grain size increases, lowering the albedo in the near-infrared. Likewise, small light-absorbing particles (LAPs), such as mineral dust, black or brown carbon, and algae, lead to enhanced absorption in the visible wavelengths, causing intensified melting and retreat of snow cover. Exposing the underlying darker surface contributes to climate warming and thus represents a large component of climate change. The retreat of polar glaciers and ice sheets amplifies global sea level rise, and reduced snow cover in mid-latitude mountains directly affects freshwater resources. It is therefore of particular importance to monitor the spatial and temporal changes in snow albedo and LAP absorption on a global scale.

The U.S. National Academies’ decadal survey clarifies a specific need for measurement of “spectral albedo of subpixel snow and glaciers at weekly intervals to an accuracy to estimate absorption of solar radiation to 10%.” This objective can be achieved by the upcoming orbital imaging spectroscopy missions SBG and CHIME, which will give approximately weekly combined observations of reflected solar radiation from almost all locations on Earth’s land surface. The measurements will feature both spectral and spatial resolutions of ~10 nm and 30 m, respectively, enabling the detection and quantification of grain size and LAP by resolving their subtle absorption characteristics.

However, an accurate snow albedo product requires independent and robust retrieval algorithms that account for varying illumination conditions, fractional snow-covered area, surface roughness and topography, background reflectance, as well as for optical properties of LAPs and the snow itself. In addition, we need to combine albedo products from different sensors with different spatial, temporal, and spectral resolutions to close gaps in time series due to lack of observation or too much cloud coverage. We will present the current status of the SBG snow physics and hydrology focus area and how we plan to achieve the objectives of the decadal survey. This includes the presentation of selected candidate retrieval algorithms, ATBD strategies, and community contribution plans. In addition, we will outline the needs of water resources applications for remotely sensed snow albedo and the paths forward to meet these needs with SBG and CHIME.

Among the properties of the soil, the spatial distribution of the soil texture plays a crucial role in the agricultural sector as it affects the degree of water penetration and retention, germination and the rooting of agricultural crops, susceptibility to erosion and nutrient absorption. Knowledge of soil texture variability is therefore a determining factor in addressing global environmental challenges, allowing the promotion of sustainable agricultural practices that favor a more efficient use of fertilizers and water.

The estimation and mapping of soil texture, understood as the relative proportion of clay, silt and sand, is one of the main user requirements for the Agriculture and Food Security application domain within the European Union (EU) Copernicus Earth Observation (EO) and monitoring program.

The study explores the value added by existing hyperspectral data of similar characteristics to CHIME, namely AVIRIS-NG and PRISMA, for detecting topsoil texture properties considering the USDA (United Stated Agriculture Department) classification model and the linear spectral mixture analysis (LSMA). The topsoil texture retrieval was tested on the bare soil units of two cropland areas in Central Italy. The purpose of testing LSMA is to open new windows on soil texture detection having consequent optimization and improvement of retrieval schemes and fulfilling the objectives of the CHIME mission by tuning different retrieval models including secondary algorithms.

The CHIME-like configuration allowed the recovery of textural classes in terms of fine (clay and silt) or coarse (sand) abundance with an accuracy of 40% in pixels with clay and/or silt greater than 45%, corresponding to USDA Loam -Clay Loam classes. The same behavior is found in the results of the PRISMA configuration which shows a precision of approximately 40% where the silt is high, further highlighting the superior spectral performance of the two hyperspectral differences (CHIME-like and PRISMA) compared to the multispectral configuration. The exploitation of hyperspectral mixture models for soil texture detection will be important also in the near future in the frame of the upcoming Italian IRIDE constellation that will include hyperspectral sensors and for the setup of the potential services to be delivered by the IRIDE programme first to the national users and second to commercial users.

Soil Organic Carbon (SOC) plays a key role in mitigating climate change and is crucial in the global carbon cycle, accounting for 50 to 80% of total terrestrial carbon. There is a recognized need to improve the SOC monitoring process to prevent further soil degradation and to develop adaptive agricultural management strategies. Traditional methods for SOC mapping are usually time-consuming; however, the rapid advancement of remote sensing technology offers novel approaches to SOC estimation.

This research describes the initial results of a study that aims to combine Sentinel-2 multispectral data, characterized by a high revisit time, with PRISMA hyperspectral data, characterized by high spectral resolution, to assess the current status of SOC monitoring. Both types of data underwent extensive preprocessing, and specific remote sensing indices were considered to extract bare soil areas, on which the analysis is conducted. Then, using ground truth data and remote sensing data, a dataset was created and used to train machine learning regression algorithms. Different regression models were tested.

Our results demonstrate that Sentinel-2 and PRISMA data are adequate for estimating SOC, showing that the proposed method performs effectively. These findings underscore the significant potential of remote sensing data, both multispectral and hyperspectral, for SOC mapping and highlight the need for further research in this field.

The main advantage of hyperspectral over multispectral remote sensing for coastal waters is the potential to detect different phytoplankton groups. This potential is limited in turbid waters where phytoplankton pigments may represent only a small part of the total backscatter and absorption. Algorithms that exploit spectral curvature/derivatives/anomalies are promising there. Many hyperspectral algorithms have been proposed for phytoplankton groups and tested using modelled or in situ measured reflectance, but successful validation for hyperspectral satellite data is rare, especially in turbid waters.

There are 3 possible reasons for this: 1. Hyperspectral algorithms exploiting spectral curvature/derivatives/anomalies are very sensitive to random noise, interband calibration and systematic wiggles near atmospheric absorption wavelengths. 2. Most hyperspectral satellites were not designed for this application, which has stringent signal:noise and interband calibration requirements. 3. Atmospheric correction algorithms have been designed and validated to provide accurate water reflectance but may give very inaccurate second derivative of water reflectance for wavelengths near atmospheric absorption features.

This presentation will provide a mathematical framework for the impact of random noise, interband calibration and systematic wiggles on the second derivative of reflectance. This will then be used to limit the impact of such measurement errors on phytoplankton group algorithms, using examples from simulated, in situ and satellite measurements. The question of how to deal with hyperspectral wiggles, and particularly the importance of not smoothing spectra in post-processing, should be addressed in an international framework.

Spectral indices use band division to target specific absorption features or reflectance changes (e.g., the red-edge in vegetation spectra). The resultant intensities help detect targeted objects across a scene. However, most standardized spectral indices formulas are designed for multispectral sensors (e.g., Landsat/ASTER), with hyperspectral data typically downsampled spectrally to fit these formulas.

Band averaging can work well for heterogeneous scenes, but we have found that it can suppress subtle variations over relatively homogeneous scenes, such as dense vegetation or extensive carbonate-rich zones. We postulate that this is because the higher spectral resolution is not being fully leveraged. For example, with NDVI, a slight shift in the red-edge can indicate the state of plant health. However, averaging bands in the NIR and Red range to fit the standard NDVI equation will easily overlook this indication. Similarly, in carbonate-rich zones, where calcite and dolomite may be present in close mixtures, averaging many spectral bands may also miss the necessary shifts to differentiate between these minerals.

In this contribution, we propose a Tree-structured Parzen Estimator (TPE) algorithm that can help to optimize spectral band selection for spectral index analyses with hyperspectral sensors, while retaining compatibility with well-established multispectral ones. TPE, a Bayesian hyperparameter optimization technique, improves selection based on previous trials. It treats the continuous range of hyperspectral band wavelengths as a search space and evaluates initial samples with a Gaussian mixture model. The algorithm iteratively generates new candidate band combinations by exploring areas of the search space that yield better performance.

We tested this approach for NDVI, focusing on the NIR and Red bands at the Hohes Holz, (Germany) research site, and for calcite with a signature absorption at 2337 nm in the Marinkas carbonatite region (Namibia). These applications assessed the suitability for both VNIR and SWIR spectral regions of operational hyperspectral sensors - PRISMA, EnMAP, and EMIT. The results show that for the NDVI index (NIR-RedNIR+Red), the maximally correlated band equations for PRISMA and EMIT are 913.45-664.89/913.45+664.89 and 902.37-671.09902.37+671.09 respectively. For the calcite spectra index (2190:2224/2293:2345 * 2375:2430/2293:2345), the optimized band equations are 2199.45/2322.13*2400.00/2322.13 for EnMAP and 2204.50/2330.33*2396.88/2330.33 for EMIT. The resultant hyperspectral indices highlight more subtle variations than their multispectral counterparts, facilitating comparison across sensors.

Forest ecosystems are critical for the provision of essential ecosystem services to human well-being. Remote sensing data have been used with varying degrees of success to monitor changes in forest functional traits and biodiversity indicators during the last few decades. The advent of a new generation of satellites and advanced retrieval techniques offers the possibility of obtaining accurate quantitative estimates of forest traits from spaceborne observations at different scales.

Here we investigate the potential of ASI-PRISMA hypersepctral data in estimating key plant functional traits and capturing information on plant diversity in a mixed forest ecosystem. In particular, we exploit PRISMA data to develop and test machine learning regression models as well as hybrid approaches for forest trait mapping. Emerging methods based on remotely sensed data that aim at the estimation of α and β spectral diversity, are also applied using the R package “biodivMapR”.

We conducted an intensive field campaign in the Ticino Park (Italy) in the summer of 2022 in conjunction with four PRISMA overpasses to collect species composition data to derive biodiversity indices, as well as trait samples for calibration and validation of the retrieval schemes.

A positive and statistically significant correlation was found for all the investigated traits. Among the leaf level traits, the best results were obtained for Leaf Water Content (LWC) (r²=0.97, nRMSE=4.5%) and Leaf Mass per Area (LMA) (r²=0.96, nRMSE=5.3%), while slightly worse results were obtained for Leaf Chlorophyll Content (LCC) (r²=0.51, nRMSE=16.5%) and Leaf Nitrogen Content (LNC) (r²=0.67, nRMSE=13.1%). Leaf Area Index (LAI) was estimated accurately (r²=0.89, nRMSE=9.1%). The comparison of the trait values in June and early September revealed a significant decline in both leaf biochemistry and LAI, which can be traced back to the stress induced in the Ticino Park by the severe drought that hit Europe in the summer of 2022.

We finally investigate the effectiveness of the spectral species concept for producing useful β diversity maps using the 5–30 m range of pixel spatial resolution exploiting the PRISMA panchromatic band. Preliminary results show that in the PRISMA β diversity map, the spatial patterns of plant species composition were predictably more similar within forest associations than between. Those results from PRISMA images are similar to those obtained using in situ data.

Further studies are needed to better understand the link between taxonomic, functional and spectral diversity, extending the current approach to larger areas and exploring the combination of data from other sensor types.

Alpine grasslands have experienced significant changes over the last decades due to socio-economic shifts and a changing climate, often resulting in intensification of management or abandonment. The Habitats Directive aims to preserve these habitats, but effective conservation measures require up-to-date habitat maps, the availability of which often limited by resource constraints. Remote sensing can help to mitigate this issue, assisting in the development of habitat maps and supporting the management of protected areas in the Alps. In this study, we used hyperspectral satellite data from the PRISMA mission (ASI), along with complementary data from Sentinel-1 and Sentinel-2, to achieve a two-step classification process in the Sciliar-Catinaccio Nature Park (South Tyrol, Italy). We exploited Random Forest (RF) and Support Vector Machine (SVM) classifiers to (i) classify land cover types within the park, and (ii) conduct a more detailed classification of grassland habitats, which we validated using field data. In the first step, the best results were achieved with RF and PRISMA data in combination with Sentinel-1 backscatter data (overall accuracy of 86%). SAR data played an important role, increasing sensitivity for shrubs (+15%) and woods (+9%), and improving discrimination between these two last classes. In the second step, preliminary results show that a combination of PRISMA and SAR data yielded the most accurate grassland classification, while with Sentinel-2 and SAR data we achieved lower accuracies. Our results show that a multi-sensor approach can lead to improved classification of Alpine habitats. However, significant challenges are still present. These are mostly related to grassland habitat fragmentation relative to sensor resolution, the need to collect substantial amounts of field data in areas which can be difficult to access, and frequent cloud cover. We also believe that a multitemporal approach should be tested to achieve better results whenever possible. However, low cloud-free image availability in alpine areas imposes strong limitations on this type of analyses and requires specific approaches to deal with partial cloud cover. Considering the limitations imposed by frequent cloud cover, a synergy among the missions (PRISMA, EnMAP, CHIME, SBG) could be a good way to improve the monitoring capabilities of hyperspectral satellite sensors in Alpine areas.

The spectral responses captured by hyperspectral satellite sensors can be adapted for a wide range of questions and applications relevant to Land Use/Land Cover (LULC) dynamic analysis as the estimation of fragmentation and the connectivity from natural to built-up units and the related changes in ecosystem services provision and supply in rapidly different changing landscapes.

In this study, libraries of spectra acquired in the field and containing the reflectance fingerprint of the most abundant land cover types in hyperspectral PRISMA scenes have been used to overcome the spectral variability and the composition of mixed pixels. Thus, the field spectral signatures, in addition with signatures derived by Principal Component Analysis, have been used as input in the Linear Spectral Mixing Model (LSMM), decomposing hyperspectral images and giving back their fractional abundance proportions for each pixel.

To investigate the LULC pattern, the satellite observations are integrated with field measurements and synthesized through consolidated algorithms to examine land cover macro-classes such as water surfaces, natural and seminatural vegetation, urban and artificial areas and soil-sediments presence areas. In addition, the spectral behavior of field signatures has been compared to the PRISMA spectral configuration. The results indicate that integration of images satellite images and field spectral libraries combined in Linear Mixture Models are suitable for LULC applications for diverse characterization of the landscape coverage and dynamics.

Resulting thematic mapping are presented not only in terms accuracy of LULC classes detected and investigated target zones, but in the framework for assessing food provision, recreational and biodiversity conservation considering thereby impacts of LULC and the landscape fragmentation. This work contributes to the development of LULC operational products Copernicus Land Monitoring Service portfolio considering the upcoming hyperspectral CHIME mission.

Leaf Area Index (LAI) is an important biophysical trait for modeling the energy and mass exchange characteristics between the land surface and the atmosphere of terrestrial ecosystems. Additionally, LAI serves as a valuable indicator of forest growth, biomass, and net primary production. Traditionally, LAI is estimated through destructive and non-destructive ground-based methods. However, these methods, including the use of handheld optical instruments, are time-consuming, labor-intensive, subjective, and costly. Advancements in multispectral and hyperspectral remote sensing technologies, data processing, and analytical approaches now enable explicit and accurate estimation of forest LAI.

The integration of hyperspectral/multispectral data and RTM (PROSAIL) together with Neural network (NN) markedly enhances the predictive efficacy of models for LAI estimation. In this study, we compared the use of multispectral ESA Sentinel-2 data and hyperspectral airborne CASI-1500 sensor (350-1050) data for estimating temporal variations in LAI. We conducted four campaigns on different days of the season, DOS 115 (April 24-25, 2019), 208 (July 23-26, 2020), 254 (September 4-11, 2019) and 290 (October 17-24, 2019) in the Lanzhot forest, Czech Republic. To collect the field observations, LAI measurements were collected using a Licor LAI-2000 canopy analyser instrument.

The results demonstrated the acceptable accuracy of estimation, as determined by airborne CASI-RTM, with an R² of 0.92, and Sentinel-2-RTM, with an R² of 0.95, during DOY 115. Meanwhile, the utilisation of NN resulted in the highest accuracy, as indicated by the R² values of 0.86 and 0.87 for Sentinel-2 from DOY 115 and 290. This results indicate that hyperspectral airborne and multispectral satellite remote sensing data are useful for estimating LAI.

Europe’s forests, which are naturally valuable areas due to their high biodiversity and well-preserved natural characteristics, are experiencing major alterations, so an important component of monitoring is obtaining up-to-date information concerning species composition, extent, and location. While ground-based surveys provide detailed information, satellite imagery offers a cost-effective way to map and monitor tree species over large and inaccessible areas. The availability of new-generation hyperspectral (HS) sensors offers a large volume of hyperspectral data, although at a relatively coarse spatial resolution compared to airborne sensors. In this light, the present study investigates the potential and challenges of obtaining spectral information from near-same-day HS images. We have analyzed HS images from PRISMA & EMIT satellites along with HySPEX (airborne) images. The study site is located in the Hohes Holz forest, Magdeburg, Saxony, Germany.

We performed the qualitative and quantitative comparisons of these images across the full spectral range of PRISMA sensor (lowest spectral resolution amongst all). This comparison was made while resampling and co-registering (using AROSICS python package) all sensors to 50 cm (of HySPEX), 30 m (of PRISMA) and 60 m (of EMIT). Furthermore, we extracted endmembers at all three scales using N-FINDR algorithm. These endmembers are matched with the ECOSTRESS, John Hopkins University, and USGS spectral library using spectral angle mapper (SAM), spectral information divergence (SID) and normalized spectral similarity score (NS3) metrics. Based on the matched species, we used unconstrained least squares (UCLS), non-negative constrained least squares (NNLS), and fully constrained least squares (FCLS) for species abundance estimations. The results indicate that sensors can be well combined with spectral consistency; however, issues arise when very low spatial resolution HS data (such as EMIT) is resampled to 50 cm spatial resolution. We also present the multiscale unmixing and how it becomes increasingly harder to distinguish tree species at coarser resolutions. This standardized comparison can also help in utilization of multiresolution hyperspectral sensors for spectral unmixing and target detection in other domains as well.

Due to the advantages provided by the valuable imagery they capture, hyperspectral imagers have been integrated into various airborne platforms to perform a wide range of remote sensing applications, such as environmental monitoring, precision agriculture, and surveillance missions. One of these platforms is medium-altitude fixed-wing unmanned aerial vehicles capable of long-endurance missions. While these aerial platforms offer distinct advantages, integrating hyperspectral imagers into these systems for prolonged operational capabilities presents specific systemic challenges.

One of the main challenges is the lack of a flight operator to control the system during long missions. Another difficulty comes from the need to properly adjust settings, such as the frame capture rate of hyperspectral imagers, especially those using push-broom imaging design, to match the relevant altitude when performing missions at different altitudes within the same flight.
Additionally, it is crucial to ensure the safety of the gyro-stabilizer device and other equipment carrying the imaging system during rigorous maneuvers outside the mission area of the aerial platform. This involves autonomously activating or deactivating these components to maintain the overall system health.

This study addresses these critical points and presents details of a system designed for long-duration autonomous hyperspectral imaging missions. The developed autonomous system includes two hyperspectral imagers operating in the 400-1000 nm and 900-2500 nm wavelength ranges, a FODIS sensor to perform radiance-reflectance conversion of the captured images without dependency on any specific calibration target located at a geographic point, particularly in high-risk missions, a high-performance GPS-IMU device, and a gyro-stabilizer that houses all the equipment.

The mission control software has been developed to autonomously manage the entire system based on predefined mission areas and altitudes. It ensures the optimal operation of the hyperspectral imagers and control of the gyro-stabilizer. The system has successfully conducted test mission flights at altitudes ranging from 500 m to 6000 m above ground level.

Since start of routine phase in November 2022, the Environmental Mapping and Analysis Program (EnMAP) has attracted more than 2800 registered users from over 85 countries. Users frequently access the EnMAP data archive downloading approximately 5400 tiles per month. Users can request observations through the EnMAP Instrument Planning Portal (https://planning.enmap.org). The demand for future observations however varies by region, with Europe showing the highest demand, leading to issues like overlapping orders in the same orbit. To address these challenges and improve mission efficiency, EnMAP has adjusted its acquisition strategies and planning. The goal is to improve access to EnMAP data for a larger user community and increase data acquisition in areas with very high demand. In March 2024, EnMAP introduced the "Foreground Mission," focusing on extended 990-kilometer flightlines over Europe. Initially, a set of 10 flightlines over Germany has been identified in collaboration with the Science User community. These are scheduled regularly (about eight per month) to enhance data coverage and create systematic time series during the vegetation period (March to October). The acquisition plan is updated on a monthly basis and adjusted based on other mission priorities such as ground activities or weather conditions. Users can receive advanced information about future priority observations at https://www.enmap.org/data_tools/foreground_mission/. Successfully acquired data are accessible to all registered users approximately six days after acquisition via the EnMAP Data Archive at the EOWEB Geoportal. It is also anticipated to expand the Foreground Mission to other European countries. This presentation emphasizes the outcomes of the new Foreground Mission initiative and outlines the efforts of the mission consortium to support concurrent field campaigns and mission synergies.

Surface reflectance is a key intermediate product serving as input for a number of higher level products. Therefore, assessment of its uncertainty is of high priority.

Uncertainties in surface reflectance mostly stem from the atmosphere correction process. These uncertainties are dependent on surface cover and atmospheric conditions. Therefore, surface reflectance has to be validated for actual surface covers and atmospheric conditions.

Stationary tower mounted and mobile ground based only instruments are only usable for non-vegetated or low vegetation sites. For taller vegetation like trees, airborne instruments are required. Campaigns using sensors on manned aircraft are expensive. Consequently, they are being done very infrequently - making it difficult to cover the full range of conditions. Additionally, these measurements usually require some level of atmosphere correction due to the flight height, i.e. surface reflectance cannot be measured directly. This again leads to larger, difficult to quantify uncertainties of the validation data themselves.

Sensors mounted on remotely piloted aircraft systems (RPAS) are cheaper to operate, provide more flexibility and require less logistical effort than manned aircraft systems. Their ability to fly at very low altitudes negates the requirement for atmosphere correction, i.e. they measure surface reflectance directly.

This presentation will introduce a spectrometer system mounted on a very small (2 kg) RPAS. Results from data captures at the North Australian Satellite Validation Facility will be shown. Additionally, results from a campaign comparing different RPAS systems for surface reflectance validation will be presented. Based on these experiences advantages and shortcomings of such a small system will be discussed.

Finally, the presentation will map out a pathway to a fully autonomous system that can automatically and regularly capture surface reflectance at a dedicated validation site.

In recent years, the quality and agreement between different Earth’s ground products retrieved by the past and present successful remote sensing missions have contributed significantly to a better understanding of Earth and Climate changes.

The remote sensing acquisitions, corrected for the radiative effects by the Earth’s atmosphere constituents, of the operational multi- and hyper-spectral missions, have allowed their inter-validation and the study of the possible remaining discrepancies for further harmonization of products.

The Ground Segment L2A processor for DESIS and EnMAP missions corrects the at-sensor received terrestrial reflection of the incident solar radiation and generates the Bottom-Of-Atmosphere (BOA) ground reflectance spectral image cube, along with pixel-classification masks, Aerosol Optical Thickness (AOT at 550 nm) and Water Vapor (WV) maps.

The correction of the Earth’s atmosphere using the same atmospheric correction software for multi-spectral missions like Landsat 8/9 and Sentinel-2, as well as for hyperspectral missions like DESIS and EnMAP, helps to minimize the differences between sensors due to the atmosphere contribution since the same correction steps are consistently performed.

This minimization of the uncertainty contribution across sensors after the same atmospheric correction allows the assessment of systematic effects present at L2 coming already from the L1 products. With a typical uncertainty at L2 of around 5 % the BOA reflectance at-sensor, significant biases with respect to their references can be estimated using data from the missions themselves for global overpasses, in addition to independent in-situ measurements.

In this contribution we summarize the lessons learned from the validation studies of the uncertainty of the hyperspectral L2A products and their consistency with the multi-spectral results, as well as the importance of traceable uncertainty budgets of L2A products. Statistical assessments for further validation procedures in future missions like CHIME and their reference data will be proposed for broader discussion.

This study assesses the potential of the German hyperspectral satellite sensor Environmental Mapping and Analysis Program (EnMAP) for deriving coastal water quality parameters which is compared to results from the MultiSpectral Instrument (MSI) on-board Sentinel-2. The study focused on the area around the Integrated Marine Observing System (IMOS) Lucinda Jetty Coastal Observatory (LJCO, Australia, http://lucinda.it.csiro.au/), the EnMAP validation site with the highest number of image acquisitions in Australia. A total of nine EnMAP and Sentinel-2 images acquired between June 2022 and October 2023 were used. These images were atmospherically corrected before retrieving water quality parameters. For EnMAP, the operational L2A product based on the Modular Inversion Program (MIP) and the ACWater module from the EnMAP-Box based on Polymer were used, while for MSI, Polymer and the Case 2 Regional Coast Colour (C2RCC) processor were applied. The quality of the normalized water-leaving reflectance ([ρ_w]N) retrieved from EnMAP and MSI were assessed using in situ hyperspectral (Hyper-OCR, CSIRO) and multispectral (AERONET-OC) reflectance measurements at this site. MSI and EnMAP [ρ_w]N data were processed to the water constituents including chlorophyll-a (Chl-a), coloured dissolved organic matter absorption (aCDOM), and total suspended matter (TSM) with the Water Color Simulator software (WASI). The results indicate that EnMAP-MIP processing combination in terms of [ρ_w]N has higher accuracy and less bias (Median Absolute Percentage Error (MdAPE) of 15.71% and Median Percentage Error (MdPE) of -1.57% ) compared to EnMAP-Polymer, MSI-C2RCC, and MSI-Polymer combinations. For the water constituent retrieved using WASI, results from the EnMAP-MIP processing are closer to the in situ measurements than those from EnMAP-Polymer, MSI-C2RCC, and MSI-Polymer. These results highlight the potential of EnMAP for enhancing water quality assessments. First results are presented.

The Life CYANOBLOOM project will develop a risk management system for the early identification of harmful algal blooms, using satellite-based data, high-frequent in situ optical measurements and in situ genetic samples. CYANOBLOOM will provide warnings based on spatial information from the satellite data, high-frequent measurements from a WISPstation at sensitive locations and when a bloom is detected genetic analysis will be applied to determine if there are toxin producing genes present. The CYANOBLOOM service objective is to ensure water quality safety.

The CYANOBLOOM method will be developed and tested at four pilot sites, in Spain, Sweden and the Netherlands. However, basins used for drinking water (the pilots in Spain), and lakes or lake areas used for recreation (the pilots in Sweden and the Netherlands) tend to not be very large: sometimes on the edge of what can be monitored with the Sentinel satellites. CYANOBLOOM will therefore also use high resolution data (PlanetScope, SuperView, Pleiades Neo), and especially for the purpose to trace differences in blooms also hyperspectral data (EnMap). In contrast to the clear advantages of the new hyperspectral satellite data with high spectral resolution, the (current) drawbacks are that acquisitions are not very frequent. Atmospheric correction algorithms and parameter retrieval algorithms specific for water also need to be (further) tested and developed, especially for the very high spatial resolution Planet Scope sensors.

Therefore, CYANOBLOOM uses WISPstation optical data for three purposes:

1. To generate semi-continuous monitoring data for the most sensitive location (e.g. bathing water area) for the days between satellite image acquisitions ​
2. To obtain in situ hyperspectral data to serve for validation and calibration of the atmospherically corrected surface reflectance data from the (hyperspectral) satellite data
3. To validate/calibrate parameter retrieval algorithms for local circumstances (using in situ optical and lab samples) and where possible relate remote and in situ monitoring of toxic events.

During summer 2024 satellite data acquisitions, WISPstation measurements and in situ sampling campaigns are being carried out. At SPECTRAL 2024 the first results of the WISPstation validation/calibration activities in CYANOBLOOM will be presented.

Aquatic ecosystems play a crucial role in fostering biodiversity and providing essential services like drinking water, irrigation, temperature regulation, energy and recreation. Nowadays, these ecosystems are facing severe pressures that compromise sustainability and water quality and availability; such impacts are altering phytoplankton growth rates and are giving rise to widespread harmful algal blooms, damaging tourism, fisheries, and human health. On a global scale, efforts are underway to promote sustainable water management practices, and in this regard leveraging the capabilities of spaceborne imaging spectroscopy emerges as a pivotal strategy, harnessing new spectral capabilities, paired with advanced retrieval algorithms, to provide novel information on the status of waterbodies worldwide. This study exploits a multisource dataset compiled by four spaceborne hyperspectral missions (PRISMA, DESIS, EnMAP and EMIT) that together offer similar spectral and spatial resolutions and complementary temporal resolutions, for improving the assessment of freshwater ecosystems. A total of 25 images over the last 3 years were collected for the following water bodies: Lake Peipsi (Estonia), Lake Võrtsjärv (Estonia), Curonian Lagoon (Lithuania) and Lake Trasimeno (Italy). Satellite reflectance products were converted into remote sensing reflectance while bad bands were excluded. The retrieval of phytoplankton, in terms of concentrations of chlorophyll-a (Chl-a), a proxy for phytoplankton biomass, and phycocyanin (PC), a photosynthetic auxiliary pigment abundant in cyanobacteria, was then conducted with two different approaches (namely BOMBER and Mixture Density Network) adaptable to any of the four sensors’ resolutions. Furthermore, BOMBER was applied to describe bottom properties in terms of bottom depth and type and to estimate the fractional coverage of submerged vegetation. The validation, performed by means of comparison with coincident in-situ measurements, showed mapping errors of the Chl-a and PC concentration products lower than 20% and an overall accuracy of about 85% for the coverage and classification of bottom types and depth. These results confirmed that hyperspectral spaceborne data can be employed to capture a comprehensive picture of aquatic ecosystems status and particularly of primary producers, as phytoplankton and aquatic vegetation. The present study demonstrates that the integration of the data provided by different but complementary sources allows for a complete analysis of dynamic processes and changes over time that takes advantage of the strengths of each data source, laying the foundation for routine monitoring, in view of future operative hyperspectral missions, such as CHIME.

Three-dimensional (3D) radiative transfer combined with machine learning regression modelling is highly effective for quantifying biophysical traits of forest canopies from optical airborne/drone observations, usable as reference data for complementary spaceborne products. This study demonstrates a semi-automated process for creating a biologically genuine 3D representation of a forest using terrestrial laser scanning (TLS), which is a cornerstone for a Fluspect and Discrete Anisotropic Radiative Transfer (DART) modelling framework retrieving leaf chlorophyll (Cab), carotenoids (Ccar), and anthocyanin (Cant) contents from a drone imaging spectroscopy data by a non-parametric regression. Combining state-of-the-art TLS and computing technology, we developed a workflow reconstructing a detailed 3D representation of the forest and tested it at the Tumbarumba permanent monitoring site in southeastern Australia (i.e., TERN supersite with 402 adult/tall eucalyptus trees in a 1 ha inventory plot). The 3D reconstruction involves: i) segmenting TLS data into leaf and wood components, ii) building 3D mesh of woody parts, and iii) recreating a realistic distribution of foliage elements. The reconstructed 3D trees created a virtual scene, which was used in DART to simulate remotely sensed drone-like images and to create a look-up table of top-of-the-canopy forest reflectance by varying Cab, Ccar, and Cant inputs in the coupled Fluspect leaf model. Random forest regression was subsequently trained with the modelled tree crowns’ reflectance and applied per-pixel on drone-acquired imaging spectroscopy data to quantify Cab, Ccar, and Cant of sunlit parts of individual crowns. Although this innovative approach is forest-type specific, it is highly automatized and allows for mapping forest biochemical traits at very high-resolution (spatial sampling distance of 5 cm). It results in a precise map depicting detailed spatial variability of pigment estimates that is required for calibrating satellite data interpretation algorithms and validating their corresponding quantitative products. The next step is an assessment of the method uncertainties propagation, ensuring that the accuracy of the estimates matches requirements for calibration and validation purposes.

The Bipartisan Infrastructure Law has allocated $320 million over five years to support the U.S. Geological Survey (USGS) Earth Mapping Resources Initiative (Earth MRI) with the goal of reducing our dependence on imported critical minerals that are essential to our security and economy. Earth MRI funds the Geological Earth Mapping Experiment (GEMx) which is a partnership between the USGS and the National Aeronautics and Space Administration (NASA) to collect imaging spectrometer data that will be used to enhance our understanding of the Unites States’ geologic framework and identify areas that have the potential to yield untapped critical mineral resources. GEMx utilizes NASA’s Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) along with the Moderate Resolution Imaging Spectroradiometer/Advanced Spaceborne Thermal Emission and Reflection Radiometer (MASTER) airborne simulator to simultaneously collect high altitude remote sensing data onboard their ER-2 aircraft. Mapping began in 2023 and over 420,000 km² of new visible to short wavelength infrared (VSWIR) imaging spectrometer plus mid-infrared multispectral data have been collected to date with additional data collections planned for 2025 and 2026. Data has been primarily collected in California and Arizona as well as a small swath in southwest Nevada. In order to provide the highest quality data to the public the USGS is also collecting an expansive set of ground calibration data from large, fairly homogenous sites across the arid western US that will be used to improve GEMx data but can also be used by current and future spaceborne imaging spectrometers.

Large-scale validation of fractional vegetation cover maps using high-resolution RGB-UAV videos.

Kevin Kühl¹, Peter Schwind¹, Paul Karlshöfer^1, Uta Heiden¹

¹ German Aerospace Center (DLR), Earth Observation Center (EOC), Weßling, Germany

Keywords: EnMAP, fCover, Validation, U-Net, UAV

Fractional Vegetation and Soil Cover (fCover) is an important land surface parameter, especially in agricultural systems. It provides quantitative cover of photosynthetically active vegetation (PV), non-photosynthetically active (NPV), and bare soil (BS) to serve the data needs for soil parameter modeling, soil erosion monitoring and the identification of land degradation. Further, it supports the observation of the impact of climate-friendly tillage practices on carbon stocks in agricultural systems (farming practices).

Validating fCover at a large is challenging because so far ground data/in-situ availability is very limited and fragmented. Nevertheless, for operational L3 processors, large-scale validation approaches are necessary to provide reliable accuracy and uncertainty measures for the land product. At the same time, the validation approach should be easy to implement, work universally and should be cost and time-efficient. The flexibility and simplicity of use of the proposed methodology offers the possibility to provide a comprehensive large-scale validation methodology based on conventional user friendly high-resolution RGB-UAV videos. It is suitable to validate fractional vegetation cover derived from e.g. EnMAP data on subpixel level.

The proposed framework is designed to validate individual/single EnMAP pixels according to a) segmenting each UAV input image in Vegetation and Background via two trained U-Net models. In the subsequent step, a support vector machine is trained to further distinguish the vegetation mask to PV and NPV per pixel and finally, a histogram-based threshold classification is done, taking advantage of different color spaces if segmentation fails due to too homogeneous surface structures. In the result, UAV videos/images can be autonomously classified into the three fCover components PV, NPV and BS. One of the principle ideas is to overcome the time-consuming manual labelling of ground data by introducing deep learning models. This consequently facilitates the processing of substantial amount of data and thus the validation of large areas. The proposed methodology underlines the reliability of the fCover processor develop at the Earth Observation Centre at DLR and the possibility to eliminate current gaps of small-scale and scattered validation data.

ESA’s upcoming hyperspectral satellite mission CHIME (Copernicus Hyperspectral Imaging Mission for the Environment) is scheduled for launch in 2028. CHIME L2A and L2B processors are currently under development. High Priority Prototype Products (HPPPs) at L2B level derived from surface reflectance are planned for vegetation (canopy nitrogen and water content, leaf nitrogen and water content, and leaf mass per area) and for soil (soil organic carbon content) and mineralogy (kaolinite abundance).

The aim of this contribution is to present the current status of activities related to the calibration and validation (cal/val) of L2B HPPPs in the pre-flight phase. We have screened for suitable data sources relevant for cal/val and identified potential gaps.

For vegetation L2B HPPPs, the most valuable data are synthetic (radiative transfer model simulated) data for different vegetation types in combination with campaign data (field trait measurements with hyperspectral, preferably spaceborne, data such as EnMap and Prisma with similar spatial and spectral resolution as future CHIME data). These datasets can be obtained mainly from the scientific community. Exploration of observation networks (e.g. ICOS, NEON, TERN) that provide measured data in standardised, comparable way every year revealed that, for example leaf-level trait data from ICOS are rarely linked to any airborne or spaceborne hyperspectral observations, whereas NEON provides both trait and airborne hyperspectral data openly.

For soil organic carbon (SOC) content, there are several sources of SOC data, e.g. European LUCAS topsoil sampling, national soil monitoring strategies and national-level maps of SOC obtained by digital soil mapping, local campaigns with imaging spectroscopy data.

The situation is different for kaolinite abundance compared to SOC. To date, we have not identified any relevant campaign datasets other than those used for product development and testing. For pre-flight cal/val, we suggest designing a controlled experiment with dedicated field sampling with hyperspectral measurements.

The EnMAP Processing Tool (EnPT), developed by the German Research Centre for Geosciences (GFZ), offers an alternative to the official EnMAP processing chain for generating orthorectified Level-2A bottom-of-atmosphere (BOA) reflectance from Level-1B top-of-atmosphere (TOA) radiance. EnPT provides various workflow and algorithm options, is open-source, and can be used as a standalone Python package, from the command line, or via a GUI in the EnMAP-Box QGIS plugin. It supports advanced co-registration and alignment to a user-provided spatial reference and implements three atmospheric correction methods: SICOR and ISOFIT for land, and ACwater/Polymer for water. Here, we present recent advances of the EnPT processing chain and their effect on the validation of EnPT's Level-2A BOA reflectance product in terms of geometric accuracy and surface reflectance quality across different atmospheric and surface conditions globally. Comparisons with the official EnMAP Level-2A product are also included.

Multiple EnMAP acquisitions from sites used to validate the official EnMAP Level-2A product were selected. These datasets were processed using EnPT with different parameters to account for all possible outputs. The results were validated using the EnVAL software developed at GFZ, focusing on geometric accuracy (co-registration between VNIR and SWIR detectors and between EnMAP and a spatial reference dataset) and spectral quality (comparing BOA reflectance with in-situ measurements). Geometric accuracy was assessed using the AROSICS algorithm.

The geometric validation includes spatial distribution maps of misregistration patterns and scatter plots of registration errors. Performance metrics such as RMSE and mean shifts were compared with the official EnMAP Level-2A product and mission quality requirements. The spectral validation involves comparing EnPT's BOA reflectance and normalized water-leaving reflectance spectra with in-situ data, allowing an absolute assessment of spectral quality. Cross-validation between EnPT and the official processing chain was performed to evaluate overall agreement.

EnPT is actively maintained and updated based on user feedback and state-of-the-art developments. Continuous monitoring of EnPT L2A outputs is necessary to account for updated algorithms within EnPT or changes in EnMAP Level-1B products provided by the EnMAP ground segment. Future work may include cross-validation with other hyperspectral sensors like PRISMA, EMIT, or DESIS.

The HYPERNETS network delivers hyperspectral water and land surface reflectance data for validation of satellite missions. While multispectral water reflectance data can be used to partially validate hyperspectral satellite missions and identify full spectrum problems such as poor aerosol correction, hyperspectral validation data is still needed. The new possibilities of hyperspectral satellite data, e.g. detection of phytoplankton groups from derivative spectroscopy, can only be validated by hyperspectral measurements.

Two autonomous hyperspectral radiometer systems, the PANTHYR and the HYPSTAR®, have been designed and deployed at diverse water sites around the world in the WATERHYPERNET part of HYPERNETS. This activity is currently evolving from an R&D project to an operational network with fully automated data acquisition, transmission and processing and improved monitoring and reliability of all hardware, software and procedural components.

The status of the WATERHYPERNET will be described here with a focus on quality control of data acquired in 2023-24 and examples of application to hyperspectral satellite missions including PRISMA, ENMAP and PACE. The added value of hyperspectral over multispectral validation data will be demonstrated. Future evolution of the network, including improvement of the measurement method and geographical extension, will be outlined.

The validation of masking algorithms for hyperspectral imagery requires a precise reference, often referred to as ground truth. Currently, most validation exercises for masking algorithms rely on hand-made annotations, but these demand significant qualified manpower, as each validation requires the creation of a new dataset. Additionally, misinterpretation of class definitions often causes overlaps in annotations between classes. Furthermore, other sources of information, such as in-situ measurements, are difficult to obtain on a global scale.
We propose a validation methodology that uses more sensitive and global sources of information, employing physical properties to validate masking products. The main objective is to retrieve reference data from missions specifically designed to sense a particular characteristic of the atmosphere or Earth's surface and use it to validate the masking products of other missions, particularly optical remote sensing. To achieve this, it is necessary to implement an interface that transforms the masking products into physical properties that match the format of the mission products used as a reference. This work addresses a use case for a spatial-spectral feature extraction-based neural network, trained with a pre-classification from the Python-based Atmospheric Correction (PACO) software developed at the German Aerospace Center (DLR). It validates the cloud masking of the Environmental Mapping and Analysis Program (EnMAP) that uses PACO as L2A land processor, using the European Space Agency's (ESA) Sentinel- 5P TROPOspheric Monitoring Instrument (TROPOMI) cloud products generated by DLR, such as Cloud Optical Thickness (COT) and cloud cover, as a reference. This validation involves finding matching location overpasses with short time differences between the two missions. Thus, the EnMAP cloud mask is resampled to the spatial resolution of TROPOMI products (5.5 km × 3.5 km) and then validated with the cloud cover product of TROPOMI, using robust metrics to compare the differences.

The ATCOR method has been established as a widely used tool for atmospheric and topographic correction of both airborne and spaceborne instruments. With the availability of new spaceborne instruments, the correction methods within ATCOR have been updated for optimal processing of the high spectral resolution instruments such as ENMAP and EMIT.

This contribution at first focuses on recent developments for optimal atmospheric correction for such instruments. Recent developments include adaption of the topographic correction approaches, the correction of aerosol scattering, the influence of scattered clouds, spectral recalibration, and the BRDF correction. The bottom of atmosphere reflectance and the spectral albedos retrieved from such optimized corrections are evaluated against standard reflectance products from ENMAP and EMIT for a limited number of sample data sets.

For ENMAP, the standard product for land applications is retrieved by a re-implementation of the ATCOR methods. Therefore, only small differences due to different configurations are to be expected. In the EMIT case, the standard reflectance products is based on JPL's ISOFIT algorithm. This algorithm integrates physical processing in an optimization routine, what results in significantly different reflectance products than achievable by strictly physical band-by-band inversion of radiative transfer models.

For both instruments, available evaluation results of standard products against ground reference targets have shown a good agreement to reference spectra. Therefore, the focus of this analysis is on cross-comparison of reflectance outputs with a special focus on critical wavelength ranges such as water vapor absorption bands, the CO-2 bands, the blue to UV spectral range, and the SWIR range above 2300nm. Also, the detectability of spectral features in mineralogy is compared to evaluate the impact of polishing and spectral fitting on spectral accuracy. The evaluation shall help for a better understanding of the quality of reflectance data from current spaceborne imaging spectrometers.

In June 2025, the IPERLITE IOD mission carrying the CSIMBA instrument will be launched. It will provide hyperspectral imagery with 20 m GSD, 80 km swath and 154 bands in the range 450-900nm at 5nm resolution. Using pointing, it will allow weekly acquisitions over sizeable regions (e.g. .240 km x 110 km at latitude 50°), making it a useful source of hyperspectral data complementary to Sentinel-2 data.

Most hyperspectral missions belong to one of two separate categories: large scientific missions with high quality requirements, or smallsat missions focusing on reducing costs at the expense of image quality. We intend to fill a gap between both with a smallsat hyperspectral mission with additional provisions to increase image quality.

The instrument consists of a hyperspectral sensor which has thin film interference filters directly deposited on a CMOS detector, and a wide swath TMA telescope. The setup allows to build a very compact instrument, but for high spatial resolutions, it struggles to collect enough light to achieve good SNR performance. This has been addressed by adding very powerful read out electronics and onboard computing capabilities which allows to acquire images with 12 stages of digital TDI, resulting in an important boost in SNR (x3.5). Moreover, the hosting platform can reduce the apparent forward ground speed by performing pitch compensation, allowing longer integration times, maintaining ample forward imaging capability. By combining several integration times improves the dynamic, the dynamic range can be enlarged.

The mission will be calibrated using advanced geometric and radiometric vicarious methods through the CalibrEO service, our comprehensive generic calibration solution. Repeated dedicated acquisitions over stable desert sites and lunar acquisitions will be carried out as well as concurrent acquisitions with other hyperspectral missions in order to facilitate cross-calibration. Thus, it can contribute to improve quality from small hyperspectral missions and bridge the gap with larger missions.

Airborne imaging spectroscopy continues to play an important role in the development and calibration/validation of spaceborne imaging spectroscopy. Until recently, it has been the only data source for simulating future hyperspectral satellite missions. Although experimental hyperspectral satellites are already providing vast amounts of data, and drones with hyperspectral systems are increasingly being used, aircrafts remain viable and flexible platforms for targeted hyperspectral campaigns, allowing the integration of multiple sensors to explore future sensor synergies. Here we present one of the few fully operational airborne research platforms for Earth observation and ecosystem research in European research space. The Flying Laboratory of Imaging Systems (FLIS) is operated by the Global Change Research Institute of the Czech Academy of Sciences (CzechGlobe). The system consists of three commercial imaging spectroradiometers and a laser scanner that acquire data simultaneously. One spectroradiometer covers the visible and near infrared part with the possibility of changing the spectral resolution (CASI-1500) and the other covers the shortwave infrared part of the electromagnetic spectrum (SASI-600) with fixed spectral resolution. These two provide full spectral data between 380-2450 nm, mainly for the assessment of biochemical properties of vegetation, soil and water. The third spectroradiometer covers the thermal longwave infrared part of the electromagnetic spectrum (TASI-600) and allows mapping of surface emissivity and temperature properties. The fourth instrument on board is the Riegl Full Waveform Laser Scanning System, which provides data on landscape topography and the 3D structure of objects. In addition to the four primary sensors, the aircraft is certified to carry the HyPlant instrument, an airborne demonstrator for the upcoming ESA FLuorescene EXplorer satellite mission FLEX. The FLIS infrastructure can be used by the international imaging spectroscopy community through open access to CzechGlobe's research infrastructure.

Data retrieved through satellite observation usually is pre-processed in different levels, where for example radiometric, geometric, and atmospheric corrections are applied, to enable a faithful interpretation and ensure comparability through time as well as between different missions. Each step in this process will inevitably introduce uncertainties to the resulting product, which need to be accounted for. Therefore, these uncertainties need to be identified, assessed, validated, and propagated through the processing chain. Due to the large data volumes, evaluating and propagating uncertainties is a challenge for recent (PRISMA, EnMap) and upcoming (CHIME, FLEX) hyperspectral missions.

We will present the current uncertainty analysis for the upcoming CHIME mission in form of an uncertainty tree diagram for the atmospheric correction, based on the FIDUCEO project and QA4EO guidelines. We will highlight the measurement function for each processing step, its sources of uncertainty, their propagation path, and the uncertainty propagation methodology. The different sources of uncertainty are combined into a few components with different error-correlation structures and stored in the L2A CHIME products. We will further briefly discuss how bottom-of-atmosphere measurements from HYPERNETS, a ground-based network of hyperspectral radiometers, can be used for validation of hyperspectral satellite data and their uncertainties.

Abstract: The prediction and mapping of the traits of leaf and canopy forest vegetation plays a critical role in assessing the functioning and dynamics of the ecosystem. Retrieval of plant biochemical and biophysical parameters is becoming feasible and entering a new phase as development of new Hyperspectral missions such as EnMAP, PRISMA, SBG and CHIME. Several retrieval algorithms are available, including parametric regression, nonparametric regression, and physically based approaches. The most widely used approach is PROSPECT that simulates leaf optical properties and SAIL, which has evolved over the past decades as the most widely used canopy radiative transfer models (RTMs). However, a combination of RTMs and machine learning regression techniques enables a fast and robust retrieval of vegetation traits from imaging spectroscopy data by considering physical principles and statistical analysis. In this study, the inversion hybrid model with PROSAIL that simulates the spectral reflectance of the leaf (PROSPECT) and canopy (SAIL) properties of vegetation based on set parameters was used to predict plant traits. A training database, namely lookup table (LUT), was established based on synthetic spectral reflectance and corresponding vegetation parameters generated by RTMs that describes its pigment, structure, biochemistry as well as background soil. The parameters are randomly derived from Gaussian normal and uniform distributions with approved ranges. The created LUT was resampled into EnMAP sensors band configurations using spectral response functions (SRF) and served as a training data set for ML regression analysis. The EnMAP image was acquired on 26 September 2023, over the test site, Marburg Open Forest (MOF), located north of Frankfurt, central Germany. Ground truth data (e.g. leaf samples and fisheye canopy images) were collected from the study and analyzed under laboratory conditions. Support Vector Regression (SVR) was successfully trained for the five properties of vegetation, including Anthocyanins (ANT), Chlorophyll a + b (CHL), Equivalent water thickness (EWT), Dry matter content (LMA), and Leaf area index (LAI). Validation using in situ Tree-M campaign dataset in EnMAP data showed (e.g., LAI with R² value of 0.85, RMSE of 0.17 m² /m² and LMA with R² of 0.4 and RMSE of 0.0048 g/cm²) that the vegetation parameters can be mapped overall with good accuracy from hybrid RTM and ML imaging spectroscopy.

The current and planned hyperspectral satellite missions, providing narrow and spectrally contiguous visible-to-shortwave infrared information, are offering crucial opportunities for applications in aquatic ecosystems. In this framework, we exploited the capabilities of PRISMA hyperspectral images in retrieving standard and innovative water quality parameters as well as benthic substrate types and bottom depth with enhanced accuracy. On 11^th September 2023, concurrent PRISMA and airborne (CASI) hyperspectral images were acquired over the Lagoon of Venice (Italy) to assess the performance of PRISMA products and to support new mission development (CHIME, PRISMA2GEN). This campaign was organized under the PRISCAV (Scientific CAL/VAL of PRISMA mission) and PANDA-WATER (PRISMA Products AND Applications for inland and coastal WATER) projects, funded by Italian Space Agency. To support the spaceborne and airborne campaign, synchronous fieldwork activities were carried out spanning coastal waters to lagoon optically-deep turbid and optically-shallow waters. In situ data were also used for assessing both remote sensing reflectance and image-derived products. The radiometric consistency of PRISMA and airborne atmospherically corrected imagery was evaluated through a match-up analysis with in situ reflectance. Water quality products were retrieved using different approaches: semi-analytical algorithms to obtain water turbidity and sediment concentrations and bio-optical modelling inversion techniques to simultaneously estimate water constituents, benthic substrate types, Secchi disk and water column depth. We performed consistency analyses among the water quality products derived from CASI and those from PRISMA, including also the products derived from the pansharpened PRISMA data, obtained from a novel deep learning pansharpening approach. The preliminary results show a good agreement for the cross-sensor comparison of reflectance, with better performances between PRISMA corrected with ACOLITE and CASI corrected with ATCOR. Comparative analysis shows CASI and PRISMA capability to mapping lagoon ecosystem conditions, enhancing the efficiency of CASI and pansharpened PRISMA in identifying variable patterns at small scale.

Terrestrial ecology often depends on observations at landscape to regional scales, but scaling up plant spectra and biogeochemistry ground observations provides a broader understanding of terrestrial ecosystems. However, translating small-scale data to larger spatial scales is not straightforward and relies on linking diverse and extensive plant traits, field, and airborne spectra datasets —collected through different sensors, experimental setups, and scales— through empirical, physical, and hybrid models. Thus, the terrestrial vegetation algorithm team from NASA's Surface Biology and Geology (SBG) Mission Visible to Shortwave Infrared (VSWIR) is developing the Plant Traits and Spectral (SBG PLANTS) database. The database will integrate field and airborne data along with metadata and connect users interactively with tools to facilitate the creation and testing of transferable, sensor-agnostic algorithms. These algorithms are designed to retrieve demonstration products for the SBG VSWIR terrestrial ecosystem, such as chlorophyll, nitrogen, and leaf water content.

SBG PLANTS' core is a PostGIS relational database deployed on Amazon Relational Database Services (AWS-RDS) that extends PostgreSQL's capabilities by supporting storage, indexing, and querying of spatial data. PostGIS can efficiently process spatial operations and functions posed in SQL for quickly querying spatial properties and relationships. The deployment allows for managing its expandable architecture design through PostgreSQL Database Management System tools (e.g., pgAdmin). SBG PLANTS comprises data from entities like leaf traits, sampled plots and individual plants (e.g., trees, flowering plants), sampled species, metadata, flight lines, airborne reflectance spectra, and sensor characteristics that relate to each other, based on an Entity Relationship (ER) diagram. The SQL-ER is flexible and can be redesigned under a version control system to incorporate other existing dataset structures from the SBG community and partners (e.g., EnSpec, NEON) in an open platform that enables findable, accessible, interoperable, and reusable (FAIR) data. Thus, addressing shared challenges in data collection, data formats and standards, modeling, data mining, storage, and accessibility. Among the many advantages, this structure enables the parallel working of concurrent users accessing the data through their preferred platforms and coding language(e.g., R, Jupyter, QGIS, Matlab, Python, IDL, etc.) for automatizing their data mining, modeling, and visualization workflows.

Access to global aerosol and surface property datasets is critical for Earth climate studies. Such data are typically obtained by applying retrieval algorithms to space-borne measurements. Current retrieval methods rely on 1D radiative transfer models, which represent the Earth surface and atmosphere as horizontally invariant, neglecting the 3D features of the simulated scene, among which atmospheric heterogeneities (broken clouds, horizontal variations of aerosol properties, etc.), as well as surface reflectance heterogeneity and terrain topological structure. While the impact of atmospheric heterogeneities on retrieval accuracy has been studied, the effect of surface heterogeneities are much less documented. Surface heterogeneity can occur over long distances, leading to adjacency effects, but also on shorter distances, leading to local complex 3D surface-atmosphere radiative coupling which cannot be captured under the 1D assumption.

In this work, we show how Eradiate, an open-source 3D radiative transfer model, can be used to investigate the effect of surface heterogeneities on a variety of synthetic scenes. We build both abstract scenes, to provide a better understanding of long- and short-distance radiative coupling phenomena, and realistic scenes, designed to mimic well-documented Aeronet sites. We introduce our modelling approach and show the current conclusions that can be drawn on the impact of surface heterogeneities on simulated reflectance values.

This research is funded by the ESA 3DREAMS project.

In recent years, there has been an increase in cases of water bodies affected by eutrophication. Studies have shown that this is due to both human activities that increase the amount of nutrients in aquatic ecosystems and the rise in average temperatures associated with global warming. The high complexity of biophysical components significantly modifies the optical properties of aquatic environments, which alters the spectral behavior of Chlorophyll-a (Chl-a), one of the most commonly used water quality parameters in eutrophication processes. This limits the generalization and optimization of models for its estimation under various conditions. Therefore, measuring spectral signatures with high-spectral-resolution instruments is essential, not only to identify differences between optical water types (OWT) but also for their classification and the development of algorithms that can adapt to different conditions. This study estimates the spectral sensitivity evaluated with the coefficient of determination R² of the PRISMA spectral band simulation with radiometer data and Chl-a concentration for various OWTs. For the classification of OWTs, a spectral library was constructed with more than 300 samples from PRISMA images with various characteristics, such as turbidity, Chl-a concentration, and bloom events. For the sensitivity measurement, the GLORIA database (globally representative hyperspectral in situ dataset) was used, which includes samples of Chl-a and remote sensing reflectance (Rrs) with a spectral resolution of 1 nm. The Rrs data were convoluted to the PRISMA spectral response function (SRF) of approximately 10 nm and subsequently classified with the spectral library and the NS³ (Normalized Spectral Similarity Score) method to obtain the OWT of each sample. The PRISMA data was simulated, and cubic spline interpolation was implemented at 1 nm to evaluate the sensitivity of each band to Chl-a for each OWT. Preliminary results indicate that for waters dominated by Chl-a, the best response is achieved in the band ratio R₇₀₃ / R₆₉₄ with an R² of 0.61, for clear waters R₆₀₁ / R₅₅₉ with an R² of 0.64, and for turbid waters R₄₁₅ / R₆₉₄ with an R² of 0.6.

Plant invasion is a fast-occurring phenomenon that causes loss of biodiversity and depletion of natural resources and negatively impacts ecosystem functioning, agriculture production, economy, and human health. Understanding the causes and effects of species invasions is a priority in ecology and conservation biology. One crucial step in evaluating invasive species' impact is to map their spread and relative abundance across a vast region. In this context, the present study investigates the potential of spectral unmixing to derive sub-pixel abundance of Acacia saligna in the coastal dune areas of Israel.

The availability of new-generation hyperspectral sensors offers unprecedented opportunities for understanding the spectral properties of vegetation. For this purpose, we acquired DESIS (April 04, 2023), EnMAP (April 04, 2023), and EMIT (April 16, 2023) images over Nitzanim national park (Near Ashdod) in Southern Israel. We extracted endmembers using three widely used methods, i.e., automatic target generation process (ATGP), pixel purity index (PPI), and N-FINDR. These endmembers were matched with the ECOSTRESS spectral library and Acacia saligna (collected in field locations at Givatolga and Sdotyam in Israel) reflectance spectra using a spectral angle mapper (SAM) with a threshold of 0.10. Based on the matched endmembers, we used non-negative constrained least squares (NNLS) for abundance estimations. These abundance estimation results are validated with reference proxy abundance maps from very high-resolution Vision-1 multispectral images (March 31, 2023, and April 17, 2023) and airborne UAV images (April 02, 2023).

Studying invasive plants at the ecosystem level is a complex and resource-intensive task, where remote sensing plays a crucial role in data acquisition and environmental evaluation. Our study offers practical insights by providing a comprehensive understanding of the spectral properties of Acacia saligna. Ecologists can use these insights to manage the actual and potential distribution and relative abundance of invasive species across large regions over an appropriate period, thereby aiding in their effective management and understanding of their causes and consequences.

Imaging spectroscopy (IS) data from missions like EnMAP, PRISMA and the upcoming CHIME and SBG programs holds immense potential for environmental monitoring, agriculture, and mineral exploration. With the increasing availability of IS data, users more than ever require solutions allowing to visualize and analyse images with hundreds of collinear bands. Also, appropriate tools are needed to combine IS data with data from other sources, e.g., field measurements and multi-spectral instruments, in a meaningful way. Conventional GIS and remote sensing software can often only be used to a limited extent here, e.g. due to high costs, limited access, or insufficient flexibility when integrating data from different sensors.

The EnMAP-Box is an open-source python plugin for the QGIS geoinformation system that addresses these problems. It provides functionality for the visualization and analysis of IS data and spectral libraries. The EnMAP-Box adds more than 150 algorithms to the QGIS Processing Framework, e.g. to yield classification maps or estimations of continuous biophysical variables. These algorithms can be easily integrated into other workflows and run in different environments - from single Laptops to cloud-based processing clusters. Various research applications use the EnMAP-Box as a platform to provide easy access to context-specific applications, e.g., the EnMAP preprocessing tools (EnPT), the EnMAP Geological Mapper for Mineral Classification (EnGeoMAP) or an Interactive Visualization of Vegetation Reflectance Models (IVVRM).

The EnMAP-Box has been successfully used in a variety of applications, ranging from land cover classification to mineral mapping and environmental change detection. It is used in remote sensing education, by researchers, land managers, public authorities, and private companies. Our presentation gives a short overview on the EnMAP-Box, presents the latest developments released with EnMAP-Box 3.15, and finally gives an outlook on the development steps until end of 2026.

Imaging spectroscopy has emerged as a powerful tool for characterizing soil chemical and physical parameters across various temporal and spatial resolutions, leveraging data from laboratory, field, and remote sensing sources. Soil properties such as texture, mineralogy, inorganic and organic carbon, and nitrogen content can be estimated through spectroscopic analyses, providing crucial information for assessing soil carbon storage, fertility, and degradation status. The ability to conduct large-scale soil surface mapping makes imaging spectroscopy highly attractive to meet societal requirements for monitoring and improving soil health through sustainable soil management practices. The advent of current and upcoming next-generation hyperspectral satellite sensors offers new opportunities to produce high-quality maps of soil surface properties (L3/L4 products) on a larger scale and with increased temporal resolution. However, spectral modeling faces limitations due to the significant spectral influence of dynamic surface conditions such as fractional vegetation cover, varying soil moisture and roughness, and the general lack of harmonized spectral preprocessing and reference data. This contribution presents several case studies from the initial years of EnMAP's operation and the ESA Worldsoils project, highlighting the ongoing challenges in using imaging spectroscopy for mapping soil properties and degradation. We propose an operational machine learning-based workflow for generating L3 soil products including uncertainty assessment and discuss challenges related to sensor properties, spectral processing, and spatial sampling design. Furthermore, we examine the impact of dynamic surface conditions on soil property mapping at the scale of Earth Observation, as well as the importance of scale dependency for ground reference sampling. Finally, we provide an outlook on the capabilities of transferring this approach to ESA's upcoming hyperspectral satellite mission CHIME (Copernicus Hyperspectral Imaging Mission for the Environment) for L3 soil products.

Artificial nighttime light sources show characteristic spectral behavior in the spectral range visible to human eyes (VIS). But also in the near infrared (NIR) and in the short-wave infrared (SWIR) narrow-band emission lines of artificial light sources as well as broad-range emissions from fires do exist, allowing for an identification of light sources. The identification and mapping of Earth’s night environment and lighting practices supports a range of applications such as socio-economic development (access to power and electricity, efficient illumination techniques) or environmental impacts (light pollution). These nighttime light signatures have been studied using field and laboratory spectroscopy, and were also measured and mapped using airborne hyperspectral sensors with a high signal-to-noise ratio (SNR). But from space, the mapping of these artificial light spectra is more challenging as the SNR levels of imaging spectrometers with a reasonably high spatial resolution of ~30 m GSD are low.

The first published nighttime light spectra measured by satellite are from EnMAP, which are detailed in this presentation. It is worth pointing out that neither the EnMAP sensor nor the processors were designed for these low signal levels, so no requirements are applicable regarding absolute at-sensor radiance values, SNR and linearity. First, the standard L1B and L1C products generated by the EnMAP Ground Segment are used as inputs. For comparison, also a non-operational processing for the Dark Current (DC) correction is used to improve the SNR and to account for remaining DC artefacts. Finally, in order to identify and to map the different light types, a classic spectral signature mapping approach is applied, using documented emission lines of over forty lamps by ELVIDGE et al (2010). For high-temperature emissions, a Planck curve fitting is tested, keeping in mind the expected nonlinearity of radiances at this low signal level. It is worth pointing out that the full VNIR-SWIR range was used in the identification step as certain lamps also show diagnostic emissions in the infrared.

The transfer of these methods is shown for brightly illuminated metropole areas such as Las Vegas, but also tests are conducted for areas with lower illumination levels. Based on these EnMAP results, the potential and limitations of using available hyperspectral sensors for nighttime applications is demonstrated.

Information on European soils and their chemical and physical characteristics are essential to achieve the ambitious goal to have all European soils in a healthy condition by 2050 (defined in the EU soil health law). In recent years, soil compositing techniques based on multispectral satellite archives have been developed and established to generate input data for spectral and digital soil mappings. The surface reflectance composites (SRC) select bare soil pixels from a multitemporal data stack by using spectral index thresholds. However, due to the limited spectral information of multispectral systems (e.g. Sentinel-2), residuals from non-photosynthetically active vegetation (NPV) cannot be fully excluded. This might also impact the quality of the soil parameter models.

The novel idea presented here is to use the quantitative outputs of the semi-operational fractional vegetation cover processor (fCover) to select bare soil pixels from a Sentinel-2 time series and thus, overcome threshold-based indices. fCover provides quantitative measures of photosynthetically active vegetation (PV), non-photosynthetically active vegetation (NPV) and bare soil (BS) from hyperspectral satellite images (e.g. EnMAP, PRISMA). However, the EnMAP-based outputs cover a small portion of a Sentinel-2 scene and also just provide information for selective Sentinel-2 scenes in time. In this work, a modified deep learning model Hybrid-SN was trained using S2 images as inputs and EnMAP-based fCover maps as labels to predict fCover for the complete Sentinel-2 scene.

The predicted optimized S2 fCover outputs are then used to define bare soils in each Sentinel-2 scene as input for the subsequent temporal compositing. The resulting SRCs are compared to those developed by the threshold-based Soil Composite Mapping Processor (SCMaP). SCMaP is a fully automated approach to make use of per-pixel based bare-soil compositing. The difference is quantified based on an evaluation technique developed for comparing different SRCs.

By exploiting synergies of hyperspectral derived products and the comprehensive S2 archive using an innovative Deep Learning approach, the selection of undisturbed bare soil areas can be enhanced and thus, the derivation of soil information can be improved.

The retrieval of vegetation biophysical and biochemical datasets is vital for measuring and monitoring the impact climate change has on global ecosystem health and food security. Accurate time-series of these variables provide a means for assessing the effectiveness of mitigation and adaptation efforts that help drive environmental policy development and progress toward meeting international climate commitments. Central to this is the provision of uncertainty information which can help determine a data stream’s fitness-for-purpose. Radiative transfer models (RTMs) are commonly used to estimate vegetation parameters from remote sensing observations, through model inversion of hyperspectral reflectance. To improve the robustness of the inversion, uncertainties should be reported. This study tests the ability of hyperspectral inversion methods to deal with uncertainty information in realistic crop canopy architectures where the parameter values are known. That is achieved by utilising two RTMs: Librat and PROSAIL, to test well-characterised simulated hyperspectral data, while simultaneously assessing individual sources of uncertainties and their contribution to the output uncertainty. Librat is an RTM that uses Monte Carlo ray tracing to estimate the illumination of an object within a scene and translate it into per-pixel hyperspectral reflectance. This study uses Librat to render a realistic three-dimensional canopy architecture of a heterogeneous corn field, with radiometrically accurate light conditions and known parameters. The output hyperspectral reflectance of this simulated study area is then inverted using PROSAIL, an RTM that simulates individual leaf reflectance and transmittance, as well as the optical properties of the whole canopy. The obtained vegetation parameters are compared to the known variables used to simulate the scene. A methodology for propagating the uncertainty of each retrieved parameter and its evaluation against the known input parameters is presented, identifying the spectral limitations of the inversion algorithm. The next steps involve assessing the accuracy of the model inversion procedure by measuring synchronous in-situ data of the various output parameters and surface reflectance. Refining agricultural crop reflectance with quantified uncertainties will allow better establishment of the plant sensitivity to environmental factors such as extreme temperatures and water stress, hence developing and implementing more appropriate mitigation policies. We make recommendations for the validation of biophysical and biochemical parameter time-series.

Accurate quantitative spatial measurements of canopy nitrogen content are key information for environmentally and economically optimized fertilization decisions in crop production. Although, during the vegetative growth stages of crops, strong correlations between chlorophyll and canopy nitrogen content exist, which are exploited by multispectral earth observation systems to provide guidance for fertilization measures in spring, quantification success rates of fertilization strategies, i.e. Nitrogen Use Efficiency (relation between nitrogen input via fertilizer and nitrogen uptake into the biomass), becomes challenging as soon as the chlorophyll-nitrogen-correlation is resolved during senescence.

Spectroscopy, however, provides access to the subtle and shallow absorptions of protein-based biomass constituents via contiguous high-resolution observations in the SWIR. With spectrometers now increasingly becoming available from spaceborne platforms, these measurements not only can be performed in a spatially explicit way but also become available for agricultural sites across the Globe. It is obvious that delineation of empirical models for nitrogen retrieval via in-situ measurements is limited, if the heterogeneity that can be found across the Globe, e.g. due to different management strategies which are implemented in the various agricultural systems, is targeted. Model-based retrieval methods that incorporate physical knowledge thus are the only feasible pathway towards quantitative nitrogen measurements, as the laws of physics and chemistry are independent from the socioeconomic/cultural frames that prevail in different parts of the world. The latest generation of feasibly invertible turbid-medium canopy reflectance models (PROSAIL-PRO) enables the differentiation of carbon-based and protein-based biomass constituents in the SWIR. To enable computationally efficient spatial mapping, the outputs of reflectance models are used to train machine learning regression algorithms (i.e. hybrid retrieval). The compilation of the training data base largely determines retrieval performance. Thereby, techniques of Active Learning enable selecting the most informative samples from large databases to compile trimmed and efficient training data, while avoiding redundancies. Although the training data in those cases exclusively consists of simulated data, in-situ measurements still can be used to support the sample selection process.

We apply a time-series of in-situ measurements from six consecutive growing seasons of winter wheat, acquired in the MNI test area (Southern Germany), to generate an externally optimized active learning training data base from simulated reflectances. Based on this data, different machine learning models are trained specifically targeting the SWIR domain, where the physical absorptions of proteins can be found. Validation is achieved by applying the models to data acquired during the HyperSense flight campaign 2022.

Retrieving forest variables from spaceborne imaging spectroscopy data is challenging due to natural variability in species composition, 3D canopy structure, and phenology. To develop robust, reliable, and fully operational retrievals of high-quality vegetation products from future hyperspectral satellite missions (e.g., CHIME, SBG), field or simulated forest trait data and spectral signatures that capture the potential variability of natural forests are crucial.

We present a simulated dataset, so called look-up tables (LUT), for Central European temperate broadleaf forests, demonstrating its potential for machine learning approaches. The dataset was simulated using the 3D Discrete Anisotropic Radiative Transfer (DART) model. Detailed virtual forest scenes, down to the individual leaf level, were generated from terrestrial laser scans of real trees, covering an area of 30 by 30 meters.

Leaf-level trait variations and simulations of 2000 leaf-level optical properties were performed using PROSPECT PRO. Canopy reflectance simulations for three different canopy covers, eight LAI levels, nine sun zenith angles, and twelve azimuth geometries were conducted in DART-Lux version 5.10.0, resulting in approximately 3.5M unique combinations. The resulting images were processed into two databases: one containing the reflectance of the entire forest scene and the other containing only reflectance from sunlit pixels. This dataset will be opened to the research community for testing and to support the development of high-level vegetation products from spaceborne imaging spectroscopy data.

The optimal amount of training data for machine learning models is not clearly established, but these methods generally benefit from large data volumes. A common guideline is to have at least ten times as many training data points as the number of features. For deep learning, even more data is typically required. Establishing a scalable data collection pipeline is essential. For tasks such as predicting biophysical parameters of vegetation, high-quality data representative of true vegetation conditions is crucial.

We explore the quality of LUT and their potential to augment or substitute in-situ measurements. We examine the data characteristics and models that yield the highest prediction accuracy, including preprocessing steps (e.g., normalization, data space transformation) and hyper-parameter selection. We evaluate three data inputs: 1) a limited (<100 data points; not scalable) set of in-situ training data, 2) a dataset closely resembling in-situ data (1000-10k data points) formed using domain expertise and similarity metrics, and 3) training on the entire simulated dataset (>3M data points). We assess the best method and provide recommendations for including LUT in a training pipeline.

Accurately mapping plant traits and biogeochemistry across vegetation types is essential for understanding ecosystem functions, monitoring environmental changes, and managing natural resources. Remote sensing (RS), in particular imaging spectroscopy, enables mapping these vegetation traits across large spatial scales. However, there is a research gap in comparative assessments of different trait retrieval methods at different phenological phases. The 2022 SBG High Frequency Timeseries (SHIFT) campaign offers a unique opportunity to evaluate the performance of various remote sensing trait retrieval methods by providing a sub-seasonal time series of imaging spectroscopy data over Mediterranean terrestrial vegetation.

This study compares three methods for retrieving key plant traits using the SHIFT dataset: Partial Least Squares Regression (PLSR), a Bayesian approach, and Spectral Feature Analysis. The traditional PLSR method, widely used in RS vegetation trait mapping, transforms spectral data into latent variables for trait estimation but lacks the analytical way to estimate the model uncertainties. In this study, we indirectly evaluate the PLSR uncertainty through 200 training permutations and take the standard deviation of the 200 predictions as the model uncertainty. Unlike PLSR, the Bayesian approach does not require latent spectral transformation/training permutations. Instead, it directly selects relevant spectral features for each trait, enhancing interpretability and computational speed while performing rigorous uncertainty quantification. Spectral Feature Analysis focuses on characterizing specific absorption features related to vegetation traits through metrics like absorption depth and area via continuum removal. This approach could lead to more sensor-agnostic trait models by leveraging direct changes in spectral features.

Comparing these methods using the SHIFT dataset is crucial for understanding their strengths and limitations. This intercomparison advances the Surface, Biology and Geology mission scoping needs and enhances our ability to monitor and manage ecosystems.

Bark beetle (Ips typographus) has recently become an important cause of forest dieback in spruce forests in Central Europe, besides other types of disturbance like windthrow and clearcutting. While monitoring forest vitality, in general, is feasible with operational multi-spectral satellite systems such as Sentinel-2, analyzing the dynamics of biochemical and structural plant traits over larger areas requires spaceborne imaging spectroscopy. This study aims to analyze time series of PRISMA, DESIS and EnMAP data to better understand how forest traits are affected during green- to red-attack stages.

We employed spectral mixture analysis (SMA) and radiative transfer model (RTM) inversion on the hyperspectral satellite data. Spectral endmembers for SMA were extracted from a 1-meter resolution airborne hyperspectral AVIRIS-NG dataset collected on July 23, 2021 as part of a European-wide CHIME campaign Hypersense organized and funded by ESA, NASA-JPL and RSL Zurich. Health (damage) status (HS) of tree crowns had been known from an intensive ground sampling campaign conducted on the overflight day jointly by Trier University, LIST, and the NP administration. SMA resulted in relative abundances of healthy, moderately damaged, and severely damaged tree crowns within 30-meter satellite pixels. Using the RTM model INFORM together with a hybrid inversion approach, the forest traits chlorophyll (Cab), water content (Cw) and leaf area index (LAI) were retrieved from the hyperspectral satellite images.

Health status and forest trait changes over time were compared with ground observations of forest disturbance. Results of the analysis revealed significant relationships between remotely sensed HS, Cab, Cw and LAI changes and ground observations. Thus, the research contributes to a better understanding of how plant traits change derived from imaging spectroscopy during forest disturbances such as bark beetle attacks and can help support the development of early warning systems.

BRDF (Bidirectional Reflection Function) normalizes the reflectance data accounting for variations due to changes in viewing-illumination geometry, which is crucial specially for reducing noise in surface reflectance time series or when comparing observations spatially.

In the multi-spectral domain, BRDF methods rely on inverting the BRDF model coefficients using several observations of a limited period of the same area from various sun-view geometries. For medium and high resolution sensors with narrow angular sampling, BRDF retrieval complicates. HABA (High resolution Adjusted BRDF Algorithm), proposed by Franch et al.,(2019) is used to normalize the Sen2like product and is based on disaggregating coarse resolution MODIS BRDF parameters inferred from the VJB method at Sen2like level.

There is limited research concerning the applicability of kernel-based BRDF correction models to hyper-spectral imaging, and is mostly focused on airborne data mosaics, and on anisotropic datasets premeasured in a laboratory environment. However, these methods are far from being operational.

Within the framework of incoming future hyper-spectral missions as CHIME, there exists a necessity for a well-established operational BRDF hyper-spectral correction. This work explores the adaptability of HABA for the retrieval of BRDF parameters at hyper-spectral domain. The first approach consists in a linear interpolation over the multi-spectral BRDF parameters retrieved by HABA to generate the hyper-spectral ones based on the hypothesis of high correlated BRDF properties over nearby wavebands, while accounting for spectral consistency and integrity, which are critical on the hyper-spectral domain.

We evaluate this approach over two sites in Gobabeb (Namibia) and Panzerwiese (Munich,Germany). After implementing HABA, the methods will be evaluated to transfer the model to the hyper-spectral domain to correct EnMAP data by studying the noise reduction and comparing the results with in situ nadir measurements.

Complementary, in this work we provide the first results of a field campaign where different set of angular measurements with ASD spectroradiometer will be held over an orange tree field in Valencia (Spain). Four main blocks of measurements are planned: first, angular measurements taken from a crane to characterize the canopy BRDF along the solar principal plane, varying only the observation angle and repeating each set in a different Solar Zenith Angle (SZA); second, measurements from the crane but pointing the spectroradiometer at nadir (fixed VZA), varying the SZA; third, angular measurements analogous to the first ones but using a goniometer to characterize the BRDF at tree level; and fourth, angular multi-spectral drone measurements using the MAIA sensor.

The generation of high-fidelity synthetic data is crucial for large dataset analysis in numerous areas, such as astrophysics, climatology and remote sensing. In the particular context of remote sensing and hyperspectral imaging, it allows observation missions to be prepared and validated upstream, and real data sets to be completed during operation. Furthermore, while modelling non-linear physical phenomena such as ocean dynamics or radiative transfer can be very accurate, it cannot be expressed analytically and generally requires very costly numerical simulations. Their application when solving classical statistical inverse problems for parameter inference such as the Metropolis-Hastings Markov Chain Monte Carlo (MCMC) algorithm is therefore prohibited, since these Bayesian inference algorithms rely on hundreds of thousands of direct model evaluations in order to estimate the posterior probabilities of the model parameters. It is therefore critical to have a way of generating data accurately, quickly and efficiently.

Recent advances on deep learning based emulators enable high precision predictions at low computational cost and are therefore a promising alternative to simulation. In hyperspectral remote sensing, current emulators are limited to the emulation of spectra, limiting their ability to capture the inherent complexities of scene-level information within hyperspectral data. In this work, we investigate the Variational Autoencoder’s (VAE) capacity to emulate Sentinel-3 Ocean Colour hyperspectral images from the associated L2 biophysical products. The emulator is built in two steps: we first train the VAE to reproduce input images from a learned low dimensional latent representation, and, secondly, we train an interpolator to map the biophysical parameters to the learned latent space. This model design allows for more modularity as each part can be (re)trained and validated on its own.

Our method outperforms traditional methods like artificial neural networks and kernel ridge regression reaching a high coefficient of determination (R²) of 0.89 while being 2.5 faster than ANN when run on GPU. Moving forward, we propose to evaluate our method based on a practical setting such as biophysical parameter retrieval. Additionally, we are planning to explore alternative deep learning methods such as Generative Adversarial Networks, while focusing on evaluating their accuracy and efficiency.

The idea of SCIA (Sviluppo di algoritmi per lo studio della Criosfera mediante Immagini PrismA) project is to exploit the hyperspectral information derived from PRISMA data, managed by the Italian Space Agency (https://www.asi.it/en/earth-science/prisma/), to estimate cryosphere parameters such as liquid water content (LWC), snow grain size (GS), snow albedo, snow impurities, glacier covers mapping and characteristics of glacial lakes.

The use of data on mountainous areas (Alpine arc) highlighted the need to create a dedicated BOA reflectance product to achieve an adequate geometric correction-both in terms of accuracy and terrain compensation-and to radiometrically correct for atmosphere using ad hoc parametrization of the MODTRAN radiative transfer code, by considering adjacency and slope effects.

The developed products yield promising results with RMSE of 3.0% in the case of LWC estimated by exploiting a new superficial snow water index, SSWI, and GS values comparable with the field collected measurements in the range of minimum and maximus values.

The effect of inorganic and organic impurities deposited on the snow surface was then investigated by parameterizing the BioSnicar radiative transfer model with observed concentrations of snow algae and mineral dust.

To map glacier covers the debris covered ice is detected by exploiting specific absorption features of ice, snow, moraines and vegetation, modelled as spectral indexes and then combined in a machine learning classification framework to recognize main glacier cover classes (clean ice, debris covered ice, moraine).

Finally, glacial and periglacial lakes are detected and characterized using standard methodologies (e.g. NDWI index and chromaticity analysis), with the advantage of using hyperspectral data. The panchromatic band is integrated into data fusion and pan sharpening techniques to achieve better spatial detail.

Good potential of PRISMA hyperspectral data in the cryosphere field can be highlighted from these results; in particular, the high spectral information enables more precise characterization of cryosphere variables, by improving the accuracy in band selection for the spectral indices defined in the algorithms compared to multispectral data. Moreover, the higher spectral resolution allows to use spectral features that otherwise would not be exploitable with multispectral data. However, what is highlighted is the need to pre-process the data to obtain a product suitable for use in mountain environments, typically characterized by high topographical complexity.

Acknowledgements: This work is carried out within Contract “SCIA” n. 2022-5-E.0 (CUP F53C22000400005), funded by ASI in the “PRISMA SCIENZA” program. PRISMA Product - © Italian Space Agency (ASI). All right reserved.

The development of high level vegetation products from emerging spaceborne imaging spectroscopy data requires high quality in situ data to support accurate product calibration and validation. In general, such data are more readily available for crop ecosystems than for forests due to their simpler canopy structure and easier accessibility of leaf samples. Leaf trait sampling in forest ecosystems is more challenging due to the complex 3D canopy structure, variability in species composition, phenology, and environmental effects. Reaching the top, sunlit parts of forest canopies and obtaining representative leaf samples to reflect the natural trait variability within the canopy can be particularly tedious.

In this study, we present a comprehensive dataset of in situ measured forest biochemical and structural traits combined with plot-averaged hyperspectral signatures derived from airborne data collected using the Flying Laboratory of Imaging Systems (FLIS; https://olc.czechglobe.cz/en/flis-2/). The airborne data include simultaneous acquisitions of hyperspectral signatures in the VNIR and SWIR ranges, spectral emissivity in the TIR range, and point cloud descriptive statistics derived from an airborne laser scanner. Measured traits include leaf chlorophyll, carotenoids, nitrogen, cellulose, protein, water content, leaf mass per area, and leaf area index. Data were acquired during several campaigns in the Czech Republic, conducted in temperate forests dominated by Norway spruce, European beech, or English oak species. Trait data are presented as plot-level means and standard deviations, calculated from samples taken from the upper, sunlit branches. The associated spectral data represents mean reflectance from of the 30 x 30 m area around the field sampling locations.

This forest trait dataset will be made available to the scientific community as open data to support further research and development of vegetation products from upcoming spaceborne hyperspectral instruments such as COPERNICUS CHIME, or SBG.

The Italian Space Agency (ASI), through its Downstream and Integrated Application Unit, fosters innovation by spearheading programs that translate space data into tangible solutions.

One significant program is PRISMA-SCIENZA, which since 2022 promotes the development of novel algorithms and products from PRISMA hyperspectral images to provide added-value to EO applications. This initiative aimed to develop the skills of national hyperspectral community, thus promoting full PRISMA data exploitation with an “open & free” data access policy as well as consolidating the national know-how in synergy and in view of both present and developing programs (e.g. EnMAP, FLEX, Sentinel, PRISMA SG, IRIDE, CHIME).

PRISMA-SCIENZA initiative funded 15 Projects, comprising research centers (35%), universities (40%), SMEs (24%) and 7 international bodies. They have developed advanced techniques and products based on PRISMA data, through the joint use of other satellite data, in situ measurements and models, spanning various application fields, e.g. topsoil properties; agriculture; vegetation and forests; urban land-cover, cryosphere, air quality, coastal and inland water; and cultural-heritage.

Upon analyzing the projects results, the following major insights emerged:

• The projects assessed PRISMA data via in-situ measurements, supported by airborne/drone campaigns, lab activities, and satellite optical/radar analyses. These diverse methodologies ensured a detailed understanding of PRISMA's performance and capabilities and led to the improvement of pre-processing techniques such as radiometric and radiometric corrections, to the detection and removal of artifacts, adjacent effects and bad bands.
• Different algorithmic solutions have been used, improved and developed, among these, Artificial-Intelligence and biophysical models have dominated.
• Cutting-edge solutions to overcome typical limitations of hyperspectral data (e.g. medium spatial resolution and high noise levels) have been promoted, including super-resolution and pansharpening techniques, as well as the use of AI for both noise reduction and the generation of synthetic PRISMA images.
• Such activities have supported high-level training for young scientists and researchers through 15 research grants, 20 fixed-term contracts, and 18 doctoral scholarships and have been translated in 52 conference proceedings and 16 papers submitted to peer-reviewed journals.

This work will provide an overview of the results, evaluate the gaps still to be addressed, highlight the most promising areas of research, and discuss potential improvements to future missions.

Earth observation (EO) provides powerful tools for evidence-based policy-making in the European Union (EU). Currently, we enter a golden age of EO science, where the ever-increasing availability of high-quality data converges with a growing demand for evidence-based policy-making. Although the European Copernicus Programme already offers a wealth of high-level products and services, current environmental policy needs still have to be fully addressed, for instance, by providing information derived from higher temporal and spectral resolutions. Therefore, we systematically reviewed current and upcoming European environmental policies, evaluating their needs for enhanced EO data capacity with a focus on the potential of state-of-the-art and future multi- and hyperspectral data.

Among existing policies, in particular, the Common Agricultural Policy (CAP), the Regulation on Deforestation-Free Products (EUDR), the proposed EU Soil Monitoring Law, and the proposed Forest Monitoring Law (FML), may greatly benefit from added-value products, which could be included in the Copernicus Land Monitoring Service. Before the end of this decade, the Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) is highly anticipated to deliver data with improved spectral resolution and will thus complement the data streams delivered by multispectral routine systems, such as Sentinel-2 (S2). Furthermore, there will be S2 Next Generation (S2NG) in space in the early 2030s, with improved spectral capabilities compared to S2 and higher spatial resolution than CHIME.

Enhanced spectral capabilities will offer significant improvements. In particular, this is expected for assessing forest degradation (for instance due to drought), or post-fire recovery. Specific nutrients, like nitrogen content, can be obtained from hyperspectral signatures at improved accuracies and used for the optimization of fertilization strategies. Moreover, hyperspectral data is a valuable tool for the assessment of soil organic carbon content and monitoring non-photosynthetic vegetation. Also, spectral discrimination of plant (crop, tree) species and improved biodiversity mapping are facilitated. From our investigations, we recommend exploiting:

(1) EO time series (continuous data) for more robust retrieval assessments and to better understand vegetation dynamics;

(2) scientific pre-cursor missions (EnMAP, PRISMA) to bring products to higher technology readiness levels;

(3) improve the in situ component, required for robust validation;

(4) synergies of future sensor systems (CHIME, S2NG), complementing each other in spectral, temporal and spatial dimensions.

We conclude that upcoming spaceborne imaging spectroscopy and S2NG data streams may emerge as a game-changer and indispensable tool for environmental policy implementation in the EU, providing enhanced traceability of key processes and changes.

With the launch of hyperspectral satellites like PRISMA and EnMAP, the availability of imaging spectroscopy data is continuously increasing and the interest in hyperspectral data analyses is growing. However, training courses and educational resources on imaging spectroscopy are still scarce. Therefore, HYPERedu, an online learning initiative for hyperspectral remote sensing, was conceptualized and is implemented as part of the EnMAP science program since 2019.

HYPERedu provides online learning resources on principles, methods and applications of imaging spectroscopy, addressing students at master level, as well as professionals in research, business, and public authorities. The resources comprise annotated slide collections, practical hands-on tutorials (based on the EnMAP-Box software), a number of educational videos (YouTube, www.youtube.com/@HYPERedu_GFZ) and a series of Massive Open Online Courses (MOOCs). The first MOOC on the fundamentals of imaging spectroscopy (2021) was followed by shorter MOOCs on selected hyperspectral application fields, including agricultural applications (2022), EnMAP data access and image preprocessing techniques (2023) and soil applications (2024).

The first MOOC, “Beyond the Visible: Introduction to Hyperspectral Remote Sensing”, teaches the principles and basics of imaging spectroscopy, using state-of-the-art interactive eLearning approaches and is designed to take 5-8 hours to be completed at one’s own pace, whereas the following mini-MOOCs can be completed within 2-4 hours. Upon successful completion, participants receive a certificate, as well as a listing of all course content in a diploma supplement.

Further MOOCs on forestry, geology as well as inland/ coastal waters are under preparation. All MOOCs are available as interactive online versions as well as offline documents (PDF format), allowing participants to take the course even when a stable internet connection is lacking. Moreover, all resources are continuously revised and extended and increasingly used in training courses, university teaching and individual learning.

Outlook for the future

All materials and courses are hosted on the EO College platform (eo-college.org) and are provided free of charge under a CC-BY License. Even though HYPERedu was initiated as part of the EnMAP science program, it is regarded as an initiative by and for the hyperspectral community. Particularly in view of upcoming missions with regular global coverage like CHIME and SBG and for user uptake such as by public authorities, HYPERedu has the potential to greatly enhance the training and education of (future) hyperspectral data users.

This contribution aims to present and discuss the concept, implementation status and cooperation opportunities of HYPERedu with the hyperspectral community.

EnMAP Hyperspectral data is delivered with a set of quality layers including cloud and cloud shadow masks. These are primarily created via band thresholding which relies on physical principles. For Kenya, some uncertainties have been identified in this cloud mask which may affect further processing such as land use classifications. Predominant issues are related to the missed detection of small, isolated clouds, cloud shadows, and cloud borders being underrepresented. These circumstances can have negative impacts on automated processing routines where not every image is visually inspected by an expert user. To address this issue, a three-step improvement strategy has been developed. In the first step, the delivered cloud and cloud shadow masks are refined via band indices and thresholding. Additionally, areas known to often be faulty, such as coastal regions with shallow water, are removed. Thereby, the masks, which are largely accurate, are improved on the basis of physically based reasoning and optimized to act as machine learning responses. In the second step, the refined mask layers are used for training a machine-learning model. We employ a combination of Nonlinear Iterative Partial Least Squares (NIPALS, or PLS) for dimensionality reduction and XGBoost for classification. This allows for introducing the entire spectral range of EnMAP Data, reducing its dimensionality while maximizing the covariance between predictors and the response, and finally using XGBoost for classification. The latter of which can handle potential non-linear relationships better than standard PLS Discriminant Analysis (PLS-DA). In the last step, the prediction is postprocessed with image morphological operations to smooth the output. This is done in such a way that the underestimation of clouds and cloud shadows is prevented at the cost of a slightly increased risk of overestimation along the cloud/cloud shadow edges. This processing chain allows for fast, accurate refinement of EnMAP L2A quality layers. As training and prediction are applied to each tile or swath individually, no further data is required. This approach is modular and customizable and can be integrated and adapted to any automatic EnMAP processing pipeline through a simple Python package.

Coastal and inland aquatic ecosystems are of fundamental interest to society and economy, given their tight link to urbanization and economic value creation. These ecosystems, which are continuously impacted by natural processes and human activities, play a significant role in the carbon cycle, and they comprise critical habitats for biodiversity. Systematic, high-quality and global observations, such as those provided by satellite remote sensing techniques, are key to understand complex aquatic systems. While multitudes of remote sensing missions have been specifically designed for studying ocean biology and biogeochemistry as well as for evaluating terrestrial environments, missions dedicated to studying critical coastal and inland aquatic ecosystems at global scale are non-existent. Thus, these ecosystems remain among the most understudied habitats on the Earth’s surface. A satellite mission called Global Assessment of Limnological, Estuarine and Neritic Ecosystems (GALENE), is proposed to ESA’s Earth Explorer Mission Idea call to respond to current and future challenges linked to coastal and inland ecosystems. The mission concept consists of a synergy of three innovative instruments, namely a hyperspectral sensor, a panchromatic camera and a polarimeter. GALENE will then provide optimized measurements of these aquatic ecosystems by enabling an adaptive spectral, spatial, multidirectional and polarimetric sampling of properties and processes in water column, benthic habitats and associated wetlands. GALENE will substantially contribute to solving global water challenges, including water pollution and ensuring clean drinking water supply for all and protecting coastal areas and populations. The GALENE science objectives and the main innovative features will be presented.

The Project for OnBoard Autonomy-1 (Proba-1) mission launched in 2001 and carried the Compact High Resolution Imaging Spectrometer (CHRIS) alongside a high-resolution camera and instrument payloads focused on debris and space radiation. CHRIS stopped acquiring data in December 2022, due to the availability of more modern hyperspectral missions, and previously provided up to 62 bands over the 400-1050 nm spectral range, with reduced specification in the 400-450 nm region, operating in five different acquisition modes (Barnsley et al. 2004, https://doi.org/10.1109/TGRS.2004.827260).

The CHRIS dataset was processed using an SSTL/Airbus processing chain that has not significantly changed since the early 2000s when radiometric calibration campaigns were undertaken within a few years of launch. Code (developed in C) was used to process image, dark current and calibration data to create the Level 1 (L1) product. Surrounding this were packages/processes to support the generation of the final format of the product: corrects GPS dropouts, calculate observation angles, collate metadata, and reformat the data into the HDF format. However, work is needed to address specific issues and make the datasets more usable, as currently, the CHRIS data is supplied as L1 data, with users expected to perform the processing steps to Level 2 (L2).

The reprocessing steps under development include atmospheric correction, cloud detection, and geometric correction. There is also the option to perform noise reduction, including destriping. Calibration analysis results (Lavender et al. 2022, https://doi.org/10.1109/IGARSS46834.2022.9884457) showed a relatively close radiometric alignment compared to reference data over land, and CHRIS's performance was not significantly worse compared to more modern hyperspectral missions. However, improvements are under consideration, and the calibration of aquatic sites needs to be reviewed.

A topic that needs significant work is the geometric correction, as the Proba-1 mission had limitations with its pointing accuracy; the overlap of multiple views varies. Also, the processing chain didn't attempt to apply georeferencing or orthorectification that would have allowed the L1 or L2 products to reach sub-pixel accuracy. The current analysis builds on parallel activities and experience with other optical missions such as Landsat (Saunier et al. 2017, https://doi.org/10.1109/Multi-Temp.2017.8035252) and PROBA-V (Toté et al. 2018, https://doi.org/10.3390/rs10091375). In addition, critical for hyperspectral data is retaining spectral integrity.

Ongoing analysis is focused on specifying the reprocessing that aims to generate CEOS complaint Analysis Ready Data products, i.e., surface and aquatic reflectance (http://ceos.org/ard/). The activities are being presented to gather user feedback and support the optimum setup for the reprocessing.

The ever-increasing panorama of earth observation missions and thus an ever-increasing availability of data makes it more important to build new generations of researchers trained in their use and to develop and update algorithms dedicated to the recognition of surface parameters. At the same time, it is important to reduce the distance in terms of technological and scientific preparation between generations of researchers. For these reasons the Department of Chemical and Geological Sciences of the University of Cagliari (DSCG-UNICA), and the Remote Sensing Lab of the National Institute of Geophysics and Volcanology are engaged in the promotion and execution of Summer Schools dedicated to remote sensing topics. These institutes jointly organise annual summer schools dedicated to specific topics in the disciplines of Science, Technology, Engineering and Mathematics (STEM). These events are designed to attract those professionals who wish to deepen in an immersive and sharing context, advanced scientific research topics both on specific applications and on more general and transversal ones such as CAL/VAL. The INGV and the DSCG-UNICA, jointly with the Italian Association of Remote Sensing (AIT), in cooperation with the Italian Space Agency (ASI) organised several edition of the International Remote Sensing Summer School: Experiencing Remote Sensing: Advanced summer school on instruments and methodology for a CAL/VAL site for Optical data. This School has been held on July 2023 and it has been replicated on September 2024. In July 2024 with The Montagna Sicura Foundation together INGV and the Department of Chemical and Geological Sciences of the University of Cagliari (DSCG-UNICA), organised a second edition of the International Remote Sensing Summer School "Experiencing Remote Sensing, dedicated to the Mont Blanc (Italian Western Alps, Courmayeur. These Summer School are oriented towards recent graduates on PhD courses and young researchers, but also to personnel from the world of industry. The planning for the next three years foresees two events per year, one of those (September) is constant ant it will be dedicated to the CAL/VAL activities also to support the institution of an EU CAL/VAL site for hyperspectral data, and one (July) dedicated to specific argument such as coastal and inland areas (2025 July, Italy), remote sensing for desert/arid areas (2026 July Italy or Africa TBD) and a School dedicated to Thermal Sensors Remote Sensing in collaboration with JPL-NASA (2027).

The Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) is one of the High-Priority Candidate Missions (HPCM) endorsed by ESA for the expansion of the Copernicus Sentinel missions. CHIME will provide routine hyperspectral sampling of Earth surface reflectance over the solar spectral range (400-2500 nm) at a 30 m spatial resolution with a revisit of 22(11) days with one (two) satellite(s). CHIME observations will support EU- and related policies for the management of natural resources and assets providing a major contribution in the domains of raw materials and sustainable agricultural management with a focus on soil properties, sustainable raw materials development and agricultural services, including food security and biodiversity.

Currently under phase C the development of the CHIME mission is performed by a consortium led by Thalès Alenia Space in France (as prime contractor) and OHB System AG in Germany (for the instrument). The Observation Performance Simulator (OPSI), is a software tool being developed by ACRI-ST under the management of the above partners as ATBD providers, to support the development and verification of the space segment as well as the development of the ground segment.

The OPSI is devoted to simulate the instrument acquisition and its different acquisition modes (along with the platform behaviour), to prototype the corroborating ground segment processors which calibrate the payload measurements to TOA radiance (at L1b) and orthorectified TOA reflectance (at L1c) and to assess the instrument performance by comparing true and estimated parameters generated at different stages. In order to accomplish the above objectives OPSI is composed of an Instrument Performance Simulator (IPS), a Ground Processor Prototype (GPP) and a Performance Assessment Module (PAM).

This presentation is dedicated to the status of the OPSI Software System at its Critical Design Review, at which it offers to the leading consortium and to ESA a software tool able to simulate the radiometric and geometrical aspects as well as to emulate onboard compression and to provide mission performance figures to be expected from the instrument design.

The Earth Surface Mineral Dust Source Investigation (EMIT) imaging spectroscopy mission was launched to the international space station (ISS) on the 14^th of July 2022 to characterize the mineral composition of the Earth’s arid land regions, provide new constraints on the radiative forcing impacts of mineral dust aerosols in the Earth System today, and assess potential changes in the future. EMIT has measured more than 100 billion spectra across six continents, which include arid lands, adjacent regions, calibration sites, and other areas. These measurements have been calibrated to at-instrument radiance, atmospherically corrected to surface reflectance, and made freely available from the NASA Land Process Distributed Active Archive Center (LP DAAC). Early in the mission, EMIT demonstrated an important capability to measure localized sources of methane and carbon dioxide greenhouse gases (GHGs). All EMIT measurements and products include uncertainty estimates. In 2024, EMIT entered an extended mission phase with new observation objectives to support expanded science objectives spanning the domains of terrestrial ecology, coastal oceans and inland water, snow and ice hydrology, geology, and more. EMIT extended mission observations are anticipated to support a broad set of new applications related to agriculture, forestry, critical minerals, water quality, wildfire fuels, water resources, surface plastics, etc. We present the EMIT measurements, calibration/validation, products, current science results along with uncertainty estimates, and plans for the extended mission. EMIT measurements are openly available and algorithms are open source. Efforts are ongoing to support interoperability, instrument-agnostic algorithms, and high-level product harmonization within uncertainty bounds. These EMIT related efforts support preparatory activities for the Surface Geology and Biology Decadal Survey mission with a next-generation VSWIR imaging spectrometer that is part of the NASA Earth System Observatory.

Shallow coral reef systems are of the most diverse and economically valuable aquatic ecosystems on the planet. The global decline of these ecosystem underscores the need for advancements in the observations of shallow water ecosystems. The launch of recent and upcoming imaging spectrometers (i.e., PRISMA, EMIT, CHIME, SBG-VSWIR) provides a pathway for the detection and discrimination of dominant shallow benthic community structures (i.e., coral, sand, algae, seagrass). Thus, this study evaluate NASA’s EMIT Imaging Spectrometer shallow water retrievals for coral reef systems across three of the main Hawaiian Islands: 1) Kāneʻohe Bay on Oʻahu 2) Olowalu Reef, Maui, and 3) South Shore of Molokaʻi. Based on Thompson et al. 2017 methodology, bathymetry, water column properties, and benthic fractional cover are derived for each site. Preliminary findings suggest that refinements in bathymetric and water column properties retrievals are needed for improved downline estimates of benthic fractional coverage. Overall, simultaneous retrievals of bathymetry, water column properties, and dominant benthic coverage from EMIT offers new approaches mapping the structure and composition of global reef ecosystems.