Conference Agenda

Cross-calibration of field spectrometers for calibration and validation (cal val) purposes presents formidable global challenges. Attaining consistency among spectrometers across diverse geographic locations is inherently intricate, bordering on unfeasible. Ideally, all spectrometers deployed at each cal val site should undergo simultaneous calibration against identical targets, both temporally and spatially. However, numerous objective variables complicate this endeavor. Discrepancies in measurement protocols among users, alongside disparities in their skill levels, significantly amplify the complexity. Furthermore, factors such as changes in solar elevation, the Bidirectional Reflectance Distribution Function (BRDF), and atmospheric attenuations introduce formidable obstacles to surface-level measurements at each site. Consequently, reflectance outcomes may exhibit pronounced variability attributable to both nonsystematic and systematic influences. In this study, we introduce an innovative methodology to harmonize ground truth measurements for both cal val and validation processes of classification products. Our approach revolves around a newly patented assembly, dubbed SoilPRO®, engineered to uphold consistent measurement geometries and remain impervious to fluctuations in lighting conditions or atmospheric disturbances. This assembly can be meticulously calibrated to absolute reflectance standards and, through the integration of an internal standard, can synchronize data collected by disparate users across varying sites. Our experimental validation encompassed trials across six distinct sites employing six spectrometers, yielding exemplary harmonization among users. Furthermore, the incorporation of internal calibration facilitated seamless alignment with soil spectral libraries generated within laboratory settings. The versatility of the assembly allows for integration with a spectrum of field spectrometers, including ASD, Spectral Evolution, and others equipped with optical sieves. This pioneering solution is poised for commercialization under the auspices of the Spectral Evolution enterprise. We applied our methodological framework to cal val EnMAP data over the Amiaz Plain in southern Israel, showcasing its superiority over conventional measurements conducted via traditional open sieve protocols.

Since its launch on April 1, 2022, the Environmental Mapping and Analysis Program (EnMAP) has provided the remote sensing community with over 75000 high-quality spectroscopic imagery products across the globe. EnMAP Earth observation products cover 30x30 km2 on the ground at 30 m spatial resolution comprising 91 spectral channels in the visible and near infrared (VNIR, 418–993 nm) and 133 spectral channels in the short-wave infrared (SWIR, 902–2445 nm) with full width at half maximum between 5 and 12 nm. The Processor and Calibration team of the EnMAP Ground Segment holds the responsibility for the processing chain, in-flight calibration, data quality control and instrument monitoring. In this contribution, we shall present an overview of all calibration and processing activities since launch, focusing on the radiometric, spectral and geometric performance throughout the mission and on the lessons learned in operating a complex hyperspectral instrument in space.

EnMAP is equipped with extensive calibration units which enable multiple types of calibration measurements, namely relative radiometric, absolute radiometric, spectral, linearity, deep space and dark frames. The different in-orbit calibration possibilities are crucial to acquire high-quality and continuously monitored data. While the SWIR sensor has proven to be exceedingly stable, the VNIR sensor degraded between April 2022 and May 2023 by 9% on average but has since stabilized. The regular absolute radiometric calibrations (with Sun diffusor) have nevertheless ensured a radiometric stability significantly below the requirement of 2.5% since launch. In addition, the spectral calibrations show that the spectral stability of VNIR and SWIR is better than 0.4 nm across the full spectral range. The dark signal has also been very stable for both instruments. The extensive and high-accuracy nature of EnMAP calibration can help prepare for intercomparison and uncertainty studies using data from current and future space-based hyperspectral missions.

The EnMAP processing chain uses the most up-to-date calibration tables to provide L1B (top-of-atmosphere radiances), L1C (orthorectified top-of-atmosphere radiances) and L2A (orthorectified bottom-of-atmosphere reflectances) level products for Earth observations. The geometric quality is excellent with typical mean geolocation and VNIR/SWIR co-registration accuracies of about 0.1 pixels. The radiometric and spectral quality is regularly controlled and, together with frequent user feedback, has led to several improvements in processing. The L2A land products are CEOS ARD certified at threshold level and the certification of L2A water products is in progress. This is particularly important towards the harmonization of data products across multiple missions.

The DLR Earth Sensing Spectrometer (DESIS) aboard the International Space Station (ISS) has been delivering high-quality hyperspectral data to both the scientific community and commercial users since it began operations in September 2018. With an increasing number of spaceborne hyperspectral instruments in use, DESIS data remain highly relevant due to their superior spectral resolution (2.55 nm) and the acquisition of over 300,000 product tiles over more than five years of operation. This extensive dataset makes DESIS data particularly valuable for sensor inter-comparison, inter-calibration, and integration into time series with other instruments. In this contribution, we present key insights from the DESIS calibration and derive useful lessons for future hyperspectral missions.

DESIS calibration uses the on-board LED calibration unit to monitor the spectral response of the instrument. Over the years, the average spectral performance of DESIS shows little variability over long periods of time. However, it also shows high measurement-to-measurement variability. Part of this variability is related to temperature gradients between the optical elements and can be corrected during data processing. Other variations seem to have a random nature and are not corrected (RMS ~0.1 nm), contributing to the measurement uncertainties.

Radiometric calibration, on the other hand, is based on vicarious calibration using RadCalNet sites as reference and flat-fielding of the radiometric response of all sensor elements with uniform scenes. Our results show that the radiometric calibration of DESIS can change by 3.4% over one year above 500 nm. Below 500 nm, a significantly larger variability is observed, increasing as the wavelengths become shorter. During the first 3 years of operations, we observed a fast decrease in sensitivity below 500 nm, followed by a short period of stability and then a rapid increase that has recovered an important part of the sensitivity lost during the first years.

Finally, the geometric calibration is performed by comparing ground control points (GCPs) automatically extracted from many DESIS images and from reference images with higher geometric accuracy. This geometric calibration becomes important when an on-the-fly geometric improvement per scene is not possible because no GCPs can be found. In this case, the RMSE is typically 300 metres in the transverse direction and 500 metres in the longitudinal direction. By default, however, if enough GCPs are found per image, an improvement of the geometric sensor model is estimated, whereby a RMSE of ~21 m is achieved in the north and east directions.

The Cal/Val Park is a collaborative effort between its several stakeholders (e.g. space agencies, commercial satellite data providers, etc.) and is currently led by the European Space Agency. The goal of the project is to design, set-up and operate a new ground-based calibration and validation (cal/val) site for supporting the post-launch cal/val needs of multiple land imaging satellite sensors.

The Cal/Val Park is primarily designed for optical missions operating in the visible to shortwave infrared range (400nm – 2500nm), including both multispectral and hyperspectral sensors with GSD up to 30 meters. In the realm of imaging spectroscopy, this includes current and forthcoming missions such as PRISMA, EnMAP, CHIME, and SBG. The Cal/Val Park will be implemented in three consecutive phases: design, set-up, and operations. Presently, the project is in the design phase, scheduled to be completed by end of 2024.

This paper aims to provide an update on the current status and future activities of the Cal/Val Park project. Specifically, it outlines the consolidated design of the site, including its location, layout, proposed suite of sensors, and considered radiometric, geometric and image quality targets. The chosen location is a flat, rural area in southern Tuscany, Italy, situated away from heavily urbanized zones and characterized by a relatively low probability of cloud cover throughout the year. Within this area we have identified a primary site, where the majority of targets and sensors will be installed. The layout of the primary site includes six large radiometric targets (120 m x 120 m), made of different materials (minerals, synthetic and natural vegetation) spanning a wide brightness range. These targets enable the validation of the full processing chain, from Level 1 Top-Of-Atmosphere (TOA) to Surface Reflectance and higher-level derived land products (vegetation, raw materials). The image quality targets include a large checkerboard and a set of bar targets and draw patterns to assess the spatial performances of a wide range of optical sensors, with a focus on very high-resolution missions. The design will be scalable to accommodate future cal/val needs and expansion of the domain of applicability to other sensors, notably thermal high-resolution, atmospheric, and SAR sensors.

The final technical solution will be consolidated at the end of the design phase, providing the foundation for the implementation phase. The long-term objective is to set up an operational site, providing cal/val reference data continuously for at least 10 years.

In 2021 the Remote Sensing Laboratories of the University of Zurich carried out an airborne hyperspectral imaging mission on behalf of the European Space Agency (ESA) in preparation of the CHIME mission. With the data acquired in this campaign a test bed is created to develop processing- and higher-level algorithms, where one airborne (AVIRIS – NG) and two space-based (PRISMA and DESIS) hyperspectral images were used to acquire Earth observation (EO) data within a time window of day.

To assess the data quality, the collected EO data was first harmonised and reprojected to the Sentinel-2 GRI reference image. Secondly, the data quality analysis was performed using an updated version of the RSL in-house calibration and validation (CAL/VAL) tool, which is relying on the spectral information system SPECCHIO for a centralised database approach. Thereby, two spatially co-located reflectance spectra are compared, computing the spectral difference (SD), the spectral ratio (SR).

Sensor related uncertainties estimated by the sensor operators, were propagated to the SD and SR, following the guide to the expression of uncertainties, enriching the total uncertainty budget with an uncertainty associated to geolocation. This uncertainty is estimated via a Monte Carlo approach, randomly moving the sensor specific point spread function (PSF) within pixel distance in across and along track direction, statistically summarising the spectra deconvolved with the PSF at each randomly selected position. In addition to the SD and SR, the measurement agreement between the two measurements was computed, to statistically describe the similarity of the two measurements.

Finally, the single EO data sets are compared pixel wise to a virtual reference sensor, which was computed with a weighted mean of the three data sets and the data sets were compiled to a multi scale data set (MSDS).

Running the CAL/VAL based on the sensor specific PSF reveals not only the EO data quality, but also the characteristics of the site selected for CAL/VAL. Especially the homogeneity and the size relative to the sensor ground sampling distance prove to have a great impact on the CAL/VAL results. Including the uncertainty data in the CAL/VAL exercise, helps to interpret the results of the measurement agreement, where greater uncertainties of the single measurements lead to a better agreement between these.

With this contribution we summarise the importance of a fully traceable uncertainty budget, including the uncertainties linked to geolocation as well as their impact on target selection for CAL/VAL intercomparison exercises.

Both CHIME and SBG plan Level 2+ products related to vegetation traits, including leaf nitrogen, pigments and water content, as well as leaf mass per area. For natural vegetation, the most likely algorithms for implementation with global-scale imagery are either fully data-driven or hybrid models. Natural vegetation is complex, and foliar traits vary at seasonal time scales due to phenology and year-to-year due to climate and disturbance. This is further confounded by the diversity of species and physiognomies that could mix in a nearly infinite array of combinations at a 30 m pixel scale. Thus, there is an urgent need to develop a collaborative array of networks of measurements at appropriate spatial (especially species composition) and temporal (especially traits) scales for calibration and validation of spaceborne trait outputs, including downstream products derived from traits, such as functional diversity. Although some networks that collect compatible data exist, none currently support the most robust implementation of trait modeling methods, and none collect data that meet potential requirements to capture phenological variation. This need for comprehensive trait data is shared by the biological community. They eagerly await the launch of SBG and CHIME because of the potential to fill in the massive gaps in our knowledge of trait distribution in space and time. Any network that would provide data needed for satellite missions would also provide vast new in-situ information with great scientific value in its own right, and the network envisioned for satellite missions would need to build upon or leverage existing networks. The needs for remote sensing-based approaches and plant ecology using only in-situ data are not incompatible, even if there may be requirements unique to each. Here, we present a hierarchical framework for a network of networks that could support remote sensing-based foliar trait estimation as well as the broader (biological) scientific community. This effort extends from the 2020-2024 sTRAITS working group, funded by iDIV, the German Centre for Integrative Biodiversity Research.

Radiometric calibration is crucial for providing a physical meaning to the digital values of an image by linking them to a parameter known as radiance at the top of the atmosphere. Calibration contributes to the data quality obtained through optical remote sensing and makes possible to combine measurements from various instruments. Hyperspectral imaging instruments, due to their multiple bands and wide spectral range, present significant challenges for their radiometric calibration and validation.

The purpose of this presentation is to showcase the current and future means of radiometric calibration of hyperspectral imaging systems at CNES, along with the main current calibration results obtained from these systems.

Over the last two decades, using the SADE/MUSCLE calibration system [1], CNES has calibrated and cross-calibrated a wide range of multispectral imagers, including SPOT1 to 7, MERIS, Pleiades, Pleiades-Neo, POLDER, VENµS, Sentinel-2 MSI, Sentinel-3 OLCI & SLSTR, … Furthermore, CNES participates in the radiometric validation of several systems via its multispectral photometers [2] located at La Crau and Gobabeb as part of the RadCalNet network [3].

This presentation will begin by demonstrating how current radiometric calibration methods dedicated to multispectral imagers can be adapted to hyperspectral imagers like PRISMA or EnMAP. Additionally, results from cross-calibration on desert targets [4] of these hyperspectral imagers with widely used reference sensors will be presented. Radiometric validation results will also be shown using the in-situ photometer installed at instrumented sites of La Crau and Gobabeb.

Finally, the presentation will discuss several radiometric calibration means currently under development at CNES, including the development of a photometer prototype for in-situ hyperspectral measurements at La Crau, one of the RadCalNet sites.

References :
[1] Cabot F., O. Hagolle, C. Ruffel, P. Henry, Remote Sensing Data Repository for In-Flight Calibration of Optical Sensors over Terrestrial Targets, Proceedings of SPIE, 3750, Denver, 1999.
[2] Meygret, A.; Santer, R.; Berthelot, B. ROSAS: A robotic station for atmosphere and surface characterization dedicated to on-orbit calibration. In Proceedings of the Earth Observing Systems XVI, San Diego, CA, USA, 13 September 2011.
[3] Bouvet, et al. RadCalNet: A Radiometric Calibration Network for Earth Observing Imagers Operating in the Visible to Shortwave Infrared Spectral Range. Remote Sens. 2019, 11, 2401
[4] Lacherade, S. & al, Cross Calibration Over Desert Sites : Description, Methodology, and Operational Implementation. IEEE Trans. Geosci. Remote Sens., vol51, no. 3, pp. 1098 – 1113, 2013

Earth Observation represents an irreplaceable resource that can contribute to the pursuit of multiple strategic, social and economic goals.

The COOL project was selected in the framework of the Call for Ideas ‘Scientific Activities to Support the Development of Earth Observation Missions’, promoted by the Italian Space Agency (ASI), and it is the result of a collaboration between the Institute of Methodologies for Environmental Analysis of the National Research Council (CNR-IMAA), the National Institute of Geophysics and Volcanology (INGV) and the University of Cagliari (UNICA).

The objective of the project is to develop an advanced, modular, and scalable system, capable of supporting the consolidation of the scientific readiness level (SRL) of the PRISMA Second Generation (PRISMA-SG) mission and related L2 products, in view of the specific characteristics of its new hyperspectral sensor and with particular reference to off-nadir acquisition geometries.

The outcomes of the COOL project consists of: a) an L2-product processor, based on the most advanced methods of radiative transfer physics in the earth/atmosphere system and specialised for PRISMA-SG characteristics, including aspects related to off-nadir view; b) a top-of-atmosphere (TOA) radiance simulator for different types of soil/raw terials; c) characterization of two mineralogical/raw material sites (Sale ‘e Porcus salty pond - Oristano, and Campo Pisano – Carbonia-Iglesias, Italy) as references for simulation and validation activities.

Regarding the L2 processor, advanced algorithms are implemented to optimise the atmospheric and topographic correction procedures for the generation of L2 products, including reflectance and water vapour maps. A sensitivity study is underway to assess the impact of the of-nadir view on the at sensor signal and estimate the residual errors expected after atmospheric correction of the PRISMA-SG data.

Concerning the mineralogical/raw material sites, the general mineralogical composition of Campo Pisano is known but dedicated spectral signature surveys and samples collection will be conducted to calibrate and specify the signals. The site of Sal’e Porcus salty pond has been chosen as venue of the International Remote Sensing Summer School organised by INGV and UNICA, with the support of Italian Society of Remote Sensing (AIT) and ASI to deeply study its spectral setting.

Feasibility analyses are in progress at both locations to establish their suitability for hyperspectral CAL/VAL activities.

The Earth Surface Mineral Dust Source Investigation (EMIT) and the DLR Earth Sensing Imaging Spectrometer (DESIS) instruments operate simultaneously on the International Space Station (ISS). Being mounted on the same platform, the two hyperspectral imaging spectrometers have opportunities for coincident views of different surface types and atmospheric conditions, including those of the Radiometric Calibration Network (RadCalNet) sites. Here, we focus on EMIT and DESIS observations of a RadCalNet site and conduct comparative analyses of EMIT and DESIS. We check the intercomparison of DESIS and EMIT in light of the automated in-situ measurements from the site as a quantitative inter-calibration approach. The results of this study show the utility of RadCalNet for harmonizing scenes from hyperspectral instruments. In addition, EMIT and DESIS observe the RadCalNet sites under various sensor-sun geometries because of the precessing orbit of the ISS. The changing sun-sensor geometries of these observations provide the opportunity to study the sensitivity of the comparison results to the illumination and observation angles. As a result, we can assess the directionality dependencies of the surface reflectance of the site. The intercomparison method discussed here can be used for calibration and validation of the radiometric data products of future hyperspectral instruments for the Surface Biology and Geology (SBG) and Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) missions.