Breaking wind-waves or whitecaps are a highly reflective feature on the sea surface that change the spectral properties of the ocean surface in both magnitude and spectral shape. As we move into higher spectral and spatial resolution satellites, new methods are needed to better characterize the contribution of whitecaps. New methods are needed to account for dips and troughs in whitecap reflectance moving into the near infrared to short wave infrared parts of the spectrum. Moreover, higher spatial resolution satellites can resolve individual wave facets. Here, hyperspectral reflectance measurements of whitecaps were taken in Long Island Sound both of manufactured and natural breaking waves. The data reveal that the spectral shape of a standard breaking wave can be modeled with a simple polynomial fit to the absorption by liquid water. Troughs in the reflectance spectra correspond to local peaks in absorption centered at 600, 756, 970, 1198, 1448 and 1932 nm. Additionally, this average whitecap spectrum was used to estimate the effective fractional whitecap coverage in 89 different spectra from mixed pixels ranging from 0 to 1 in whitecap fraction. Using these data, correlation matrices were constructed to assess how combinations of different wavelengths could be used to predict fractional whitecap coverage using different metrics. Regions of highest correlation were found for bands at 800 nm, 900-980 nm, and 1100-1300 nm. In addition, the amount of reflectance in SWIR bands from 1500-1800 nm was found to be highly correlated to the whitecap fraction. Various algorithms are presented to retrieve the contribution of whitecaps contributions to radiance from the visible to the short wave infrared without the use of ancillary measurements of wind speed, as is commonly done. Considerable field measurements over the last decade have shown that wind speed is only an average climatological predictor of whitecap coverage and cannot be used to accurately assess instantaneous coverage or used for high spatial resolution satellites can resolve individual wave facets. This research provides a way forward that is independent of ancillary measurements and provides a simple framework for estimating the contribution of whitecaps based on liquid water absorption features.

Area A - Board 1

Observing the amounts and distribution of the three phases of water is crucial, especially in the context of climate change and growing world population. By the quantification of liquid water, e.g., canopy water content (CWC), critical vegetation conditions like drought stress can be estimated. Detecting local and global trends in water vapor is important for climate research, since it is strongly connected to climate change feedback mechanisms. Furthermore, the knowledge about snow and ice extents is of interest for the understanding of melt processes and the resulting availability of fresh water.

Imaging spectroscopy is a suitable tool to improve the understanding of local and global water processes. It allows to identify and quantify water in different states, which is unique with respect to other remote sensing techniques and is based upon the discriminative absorption lines of the three phases in the solar spectrum.

In this contribution, we developed a coupled retrieval of the three phases of water to infer their amounts as part of the atmospheric correction of spaceborne hyperspectral measurements like the future German EnMAP mission. We linked the MODTRAN radiative transfer code to a surface reflectance model based on the Beer-Lambert attenuation law and apply a spectral fitting algorithm to iteratively adjust the model parameters to match the modeled and measured spectra. This model inversion reports the derived amounts of vapor, liquid water and ice in units of gcm-².

We first tested the algorithm on airborne AVIRIS Classic data acquired over mountainous vegetated and snow-covered areas. To validate the retrieval, we simulated EnMAP-like top-of-atmosphere radiance spectra for vegetated surfaces using a coupling of the leaf optical model PROSPECT, the 3-D HySimCaR model for canopy reflectance spectra and the EnMAP end-to-end simulation tool (EeteS). By varying input CWC and columnar water vapor (CWV) we evaluated the performance of the retrieval.

Both parameters are accurately retrieved showing an R² of 0.97 for CWV and 0.96 for CWC. For the derivation of CWV, this indicates less sensitivity to background reflectance in comparison to conventional methods like band ratio approaches. Consequentially, this leads to an accuracy improvement of the atmospheric correction, since biases in the CWV retrieval over vegetated and snow-covered surfaces are reduced and the surface reflectance spectra get smoother in water absorption regions. Despite the high R², we observe a systematic overestimation of CWC by a factor of 3.5 on average. This could be addressed to volume scattering processes within the canopy and different radiation penetration depths. For a precise quantification of CWC, a correction scheme must be developed, which accounts for scattering processes within the canopy and simultaneously incorporates the influences of different soil backgrounds and soil moisture. Nevertheless, promising results with respect to future spaceborne hyperspectral missions are presented in this study and indicate the potential of imaging spectroscopy to support the understanding of water’s role and importance in the Earth system.

Area A - Board 2

Analytical modeling gives the support one needs where there are scarce in-situ observations. That holds for the development of algorithms for the retrieval of water constituents over aquatic ecosystems within the scope of current and upcoming orbital hyperspectral missions. In this context, the deployment of hyperspectral sensors should contribute to the quantification, for instance, of narrowband and trans-spectral radiative transfer processes (e.g. fluorescence), that are required for accurate and operational retrieval of optical constituents of water bodies (e.g. phytoplankton functional types). In this work, we used the coupled atmosphere-ocean radiative transfer model SCIATRAN (Rozanov et al. 2017) to construct a database of top-of-atmosphere radiance, at-surface radiance, and remote sensing reflectance simulated over water surfaces. The SCIATRAN model incorporates scalar and vector ordinate techniques to solve the radiative transfer equation, including inelastic radiative processes (e.g. Raman scattering and fluorescence). Previous work on the identification of water types that was based on remote sensing reflectance (Hieronymi et al. 2017) is extended to include the variability of atmospheric conditions that can be associated to each optical satellite retrieval over water. We investigate, for instance, how changes in the atmosphere interfere on the spectral water properties as measured by the sensor at the satellite. We present results for different ranges of chlorophyll and dissolved matter and for different atmospheric profiles, including aerosols. We also discuss approximations that are commonly used and need special attention, like the assumption of homogeneous columns in contrast to vertical distributions of atmospheric and of water constituents.

Hieronymi et al. 2017. The OLCI Neural Network Swarm (ONNS): A Bio-geo-optical Algorithm for Open Ocean and Coastal Waters. Frontiers in Marine Science, 4, 140.

Rozanov et al. 2017. Radiative transfer modeling through terrestrial atmosphere and ocean accounting for inelastic scattering processes: Software package SCIATRAN. Journal of Quantitative Spectroscopy & Radiative Transfer, 194, 65-85.

Area A - Board 3

Hyperspectral imaging data (whether space borne, airborne, field-based or laboratory) are excellent for providing an abundance of analysis options for the phenomena under study. In the realm of vegetation analysis, vegetation indicies (VI’s) such as Normalized Difference Vegetation Index (NDVI), Leaf Area Index (LAI) and Green Sum Index (GSI) have been standard indicies in spectral data analysis throughout the last several decades. While VI’s use is ubiquitous in the multi- and hyperspectral literature, the effective power of their combination (fusion) has not been well determined nor well presented. The fundamental reason for the lack of fused data use is simply that a straight-forward-approach to data fusion and interpretation of the results has yet to be presented. While there certainly are very sophisticated and useful data fusion products in use, they are usually very costly and require an expert level user in several disciplines unrelated to remote sensing (such as spaces-related mathematical analysis and advanced coding).
In the work presented here, we show the results of the application of a very simple yet powerful adaptive data fusion algorithm to airborne hyperspectral imagery – and by extension to space borne spectral imagery. In this work, we use the idea of a pre-determined objective to select the indicies of interest, and from that we apply consistent colour intensity assignments to the VI’s with positive correlation to the feature of interest and inverse colour intensities to the VI’s with negative correlation to the feature of interest. From there we apply a user-defined weighting scheme (linear or non-linear) to the result of each VI, then sum the results on a pixel-by-pixel basis. The results produce not a VI but rather a quantitative representative of the fusion of the VI’s themselves. The data used for this work was acquired over the Mer Bleue wetlands outside of Ottawa, Ontario, Canada (one of three supersites in Canada for the Sentinel 2 program). The results will show the power of applying this data fusion method to the problems of seasonality variability on water stress in northern wetlands.

Area A - Board 4

Abstract:
In this paper, we aim at tackling a multi-modality feature learning question in remote sensing community, that is can a limited amount of high-quality training data (e.g., hyperspectral data) improve the performance of a classification task using a large amount of low-quality data (e.g., multispectral data? Traditional approaches, such as manifold alignment, fail to perform sufficiently well for such problems, since the high-quality data is hardly available on a large scale. To this end, we propose a semi-supervised multimodality learning framework (SSML). SSML learns a joint graph structure directly from multi-modal remotely sensed data instead of artificially computing a graph by a Gaussian kernel function. The unlabeled samples can be effectively propagated on the learned graph, which enables finding a more accurate decision boundary. Extensive experiments on hyperspectral-multispectral datasets demonstrate the superiority and effectiveness of the proposed method in comparison with state-of-the-art methods.

Key-words: Classification, graph structure, manifold alignment, multi-modality, remote sensing.

With the advent of big data era, it is now easier to obtain a large amount of data than ever before. Unfortunately, most of them are usually unlabeled and low-quality. In general, a reliable classifier needs to be trained on a large amount of labeled, discriminative, and high-quality samples. Such training samples are capable of better capturing the distribution of different categories in order to find an accurate decision boundary. Although high-quality training samples are expensive to collect, we may be able to expect a small amount of high-quality training data available. In land cover classification tasks, we can easily acquire multispectral data at a global scale from Sentinel-2 or Landsat-8. However, the corresponding hyperspectral data can only be collected in very small areas due to various limitations of satellite sensors. This motivates us to ask the following interesting and challenging question: can a limited number of high-quality training data contribute to the recognition or classification task of large-scale low-quality data?

For this purpose, we align the multi-modal data in a latent subspace by jointly learning subspace projections and a linear classifier as well as adaptively estimating a joint Laplacian matrix, which can be formulated as Y=PTX with the manifold regularization tr(TXL(TX)?). Y is the label matrix represented by one-hot encoding, X denotes the labeled data from source and target domains as well as the unlabeled data and L is a joint Laplacian matrix, T and P are the subspace projections and the linear classifier.

References:
[1]. D. Hong, N. Yokoya, J. Chanussot, X. X. Zhu. CoSpace: Common Subspace Learning from Hyperspectral-Multispectral Correspondences. IEEE Transactions on Geoscience and Remote Sensing, 2018.
[2]. D. Hong, N. Yokoya, N. Ge, J. Chanussot, X. X. Zhu. Learnable Manifold Alignment (LeMA) : A Semi-supervised Cross-modality Learning Framework for Land Cover and Land Use Classification. ISPRS Journal of Photogrammetry and Remote Sensing, 2018.
[3]. D. Hong, N. Yokoya, J. Xu, X. X. Zhu. Joint & Progressive Learning from Highdimensional Data for Multi-label Classification. European Conference on Computer Vision (ECCV), 2018, pp. 478-493.

Area A - Board 5

Environmental agencies show a increasing interest to remote sensing observations for the monitoring of coastal and inland waters. Indeed, the high spatial and temporal coverage provided by satellite observations is extremely important for monitoring dynamic phenomenons involving phytoplankton such as eutrophication or harmful algal blooms. However, monitoring phytoplankton from space is still challenging. Whereas in clear and open waters, chlorophyll-a concentration, a commonly used proxy for phytoplankton biomass and eutrophication status, is well estimated, obtaining good chlorophyll-a products in coastal and inland waters is still an issue. In fact, many regional algorithms have been developed but most of them have restrictions in their domain of application preventing their utilisation at large scale. In addition, environmental agencies generally need information about phytoplankton community composition. They also need to monitor specific species which are toxic or have negative consequences for the environment.

Hyperspectral remote sensing can help to fill this gap and to improve phytoplankton monitoring from space. Indeed, from in situ hyperspectral reflectance measurements, algorithms have been developed to detect certain phytoplankton groups or species such as dinoflagellate, cyanobacteria, diatoms, phaeocystis and to better estimate chlorophyll-a concentration in turbid waters. The present study proposes an evaluation of some of these algorithms on hyperspectral remote sensing images acquired by CHRIS/PROBA over coastal and inland waters. CHRIS/PROBA (Compact High Resolution Imaging System on the Project for On-Board Autonomy platform) is an experimental hyperspectral satellite sensor which can provide 13kmx13km images with 62 spectral bands from 415nm to 1050nm at a spatial resolution of 36m since 2001.

To investigate the capabilities of hyperspectral remote sensing sensors for ocean colour applications and phytoplankton monitoring, CHRIS images taken in 2018 and 2019 over different sites including inland waters, coastal waters and river months and plumes were processed to retrieve water reflectance and information about phytoplankton biomass and composition. First, a water specific CHRIS/PROBA processing chain was designed to retrieve water leaving reflectance from top of atmosphere radiance. This chain includes noise reduction, radiometric calibration based on spectral signal smoothing and atmospheric correction. Then, hyperspectral phytoplankton oriented algorithms were tested on remote sensing data. In particular, hyperspectral chlorophyll-a algorithm (Ruddick et al. 2001) was compared to band ratio based algorithms and the algorithm for the detection of phaeocystis globosa (Astoreca et al., 2009), a phytoplankton species blooming in the North Sea, was tested on CHRIS images. Finally, recommendations for future ocean colour hyperspectral remote sensing missions are provided from the CHRIS/PROBA experience.

references:
Astoreca R., Rousseau V., Ruddick K., Knechciak C., Van Mol B., Parent J-Y., Lancelot C., Development and application of an algorithm for detecting Phaeocystis globosa blooms in the Case 2 Southern North Sea waters. Journal of Plankton research, 31-3, 287-300, 2009.

Ruddick K., Gons, H., Rijkeboer, M., Tilstone, G. Optical remote sensing of chlorophyll a in case 2 waters by use of an adaptive two-band algorithm with optimal error properties. Applied optics, 40(21), 3575-3585, 2001.

Area A - Board 6

Recently activities were carried out to establish a theoretical model of uncertainty associated with L1b products for an Imaging Spectrometer. Further a first version of the model was implemented in an existing L0 to L1 prototype processor.
As part of the Sentinel-3 Ocean and Land Color Instrument (OLCI), this task was found of considerable interest for the scientific communities, since it would provide more comprehensive information needed in the Level 1b data, i.e. a quantitative indication of their quality.
The uncertainty model is based on the numerical methods for computing top of atmospheric radiance (TOA) developed for data processing whereas uncertainty estimation and propagation is founded on the guide to the expression of uncertainty in measurement (GUM) /1/. Estimation and propagation of uncertainty are founded on the rules provided by the guide to the expression of uncertainty in measurement GUM.
In the paper, the theory of uncertainty estimation and propagation is discussed: basic definitions (uncertainty, error) are provided as well as the uncertainty propagation rule, which is fundamental to the whole problem. Subsequently, the measurement model of OLCI /2/ is explained; it is remarkably complex as it requires different calibration and characterisation parameters (which is often fitted using variety of algorithms) -- the model strictly follows OLCI data processor in order to precisely estimate uncertainties of each processing step and properly propagate them.
To the knowledge of the authors, it is the first time that the GUM methodology is applied on an imaging spectrometer type processing chain, therefore, the overall concept elaborated on OLCI can be adapted – in principle – for any other (also: hyperspectral) imaging spectrometer.
/1/ Joint Committee for Guides in Metrology. JCGM 100: Evaluation of Measurement Data - Guide to. s.l. : Joint Committee for Guides in Metrology, 2008
/2/ “The Sentinel-3 Ocean and Land Colour Imager Optical Payloads for Space Missions”, Chapter in ISBN 9781118945148 S-E Qian (editor) (2016)

Area A - Board 7

In recent years only few hyperspectral spaceborne missions successfully continue to collect imagery with different spectral and spatial resolution on specific test sites on the whole globe. These missions, such as Hyperion (NASA, deactivated on March 2017), the Chinese HJ-1A, the Indian HySI, Hyperspectral Imager for the Coastal Ocean (HICO) and the Compact High Resolution Imaging Spectrometer (CHRIS Proba, ESA), constitute an important heritage for the new terrestrial hyperspectral spaceborne missions that are foreseen in the very near future. For example, the Chinese Gaofen GF-5 is in orbit from April 2018, while the Hyperspectral Precursor of the Application Mission (PRISMA, ASI) and the Environmental Mapping Program (EnMAP, DLR) are expected to be launched on 2019. Further missions should be the Indian Geostationary Hyperspectral Imager Satellite (GISAT), the Japanese Hyperspectral Imager Suite (HISUI), the Earth surface Mineral dust source InvesTigation (EMIT, NASA), the HYPXIM (CNES), and Spaceborne Hyperspectral Applicative Land and Ocean Mission (SHALOM, ASI-ISA) and Copernicus Expansion Sentinel candidate (CHIME). The whole of these sensors cover the visible and near-infrared (VNIR) spectral ranges, while most of them additionally cover also the short-wave infrared (SWIR) range with a spatial resolution of about 30m and a variable swath. Consequently, an important issue is to explore the potential of a simple local based methodology to follow the sensor Key Data Parameters (KDP) drifts. In such a context, this study intents to explore the efficiency of the Brook and Ben Dor (2011) Supervised Vicarious Calibration (SVC) protocol when applied to the new hyperspectral spaceborne missions. The SVC protocols are based on the use of artificial agriculture black “net” at various densities over bright homogenous background used as a vicarious calibration (VC) target. This arrangement is designed to mimic different intensities of solar radiation and provide supervised radiometric sequence from the selected bright background. The Italian Space Agency has funded a joint Italian Israeli research project “An innovative method for Vicarious radiometric CALibration for orbital HYperspectral sensors” (VCALHY), which was aimed to explore the efficiency of the artificial “net” as a VC target when applied instead of a small scale (airborne hyperspectral sensor) to a larger scale as provided by the spaceborne hyperspectral sensors (e.g., PRISMA and EnMAP). The VC site selected for this study was Amiaz playa located in an arid zone of southern Israel—the Dead Sea Rift (-240 m a.m.s.l.), consisting of a Lisan Formation characterized by calcite and clay minerals. First, we checked the SVC method over this area using a hyperspectral airborne system (AisaFENIX) that showed promising results, and then arranged a campaign to study the Hyperion's performance. We installed two 90 x 60 m black nets with 30% and 80% densities, respectively. On May 7, 2015 an Hyperion acquisition (2015, 127 Julian date) were taken place over the Amiaz area. On the same date two Hyperion images were acquired: i) on the Maccarese experimental farm in Italy; ii) on the VC test site of Ammi’az. During the Hyperion overpass and following a ground-truth measurement protocol we measured both reflectance and radiance over each net as well as over the background. At each mode the collected 120 ground-truth spectra that were averaged to represent each net's spectral performance. The performance of the retrieved VC coefficients have been expressed in terms of the standard deviation of the Percent Difference (PD) calculated between the reference ASD soil spectral reflectance and the reflectance obtained from a time series of 13 Hyperion images acquired on the Ammi’az site. The standard deviation of PD is of about 5% in the VNIR; while from 650 nm to about 1800 nm the standard deviation is below 4%. Higher standard deviation values between 5% and 9% have been observed in the far SWIR region (2200–2300 nm). These results are in accordance to what observed by Czapla-Mayer et al. (2016) in their Hyperion vicarious calibration studies performed on a wider and more instrumented calibration site. We used several indices to assess the quality assurance of the data and found out that the radiance data needs some rectification. After correcting the observed drift the reconstructed radinace values were introduced into radiative transfer models (ATCOR and ACORN) to extract the reflectance of the entire image. The results showed improvement relative to the original data, suggesting that the supervised VC (SVC) method has operational merit for hyperspectral sensors from orbit. The Hyperion data that acquired at the same day over Italy was subjected to the correction factors extracted from the Amiaz (SVC) overpass and then the reflectance was extracted using ACORN code. It was shown that the SVC radiometric alignment enable to extract better reflectance values over Italy then before the SVC based correction. The project has demonstrated that the SVC procedure, with some precautions, is useful to monitor the performance of the new spaceborne hyperspectral sensors like PRISMA, ENMAP, GF-1 or CHIME. The pixel spatial resolution, however, has in general an important impact on the coefficients retrieval because of the limit in the net’s dimension that can be realistically enrolled on the field. Nevertheless, the SVC procedure applied to Hyperion imagery has demonstrated its feasibility and its main advantage relies on the possible use of less equipped calibration sites and on the selection of sites with reduced dimension like the Ammi’az test site.

Area A - Board 8

The objectives of the high-resolution imaging spectroscopy remote sensing mission EnMAP (Environmental Mapping and Analysis Program) are the investigations of a wide range of Earth surface parameters. The high-quality standardized products will be freely available to international scientific users for measuring and analyzing the status and evolution of various terrestrial and aquatic ecosystems. Therefore, it will cover the spectral range from 420 nm to 2450 nm with a spectral sampling distance varying between 4.8 nm and 12.0 nm. It will acquire 30 km in the across-track direction with a ground sampling distance of 30 m. The launch is scheduled for 2020 with an operational lifetime of 5 years.

The EnMAP portal (www.enmap.org) is the central entry point for all users. It provides general mission information, e.g. on the space, ground, and science segment as well as related campaigns.

Approximately 5,000 km in the along-track direction will be acquired per day, which are split to several observations. Therefore, the proposal handling and observation planning portal will be established to reasonably distribute the acquisition capacity to the user community. Registered users submit proposals based on Announcement of Opportunities placed by the science segment and mission management. Proposals specify especially the required coverage. If the proposal is accepted, the user plans the observations. The repeat cycle of 398 revolutions in 27 days has an orbit altitude of 652 km and combined with an across-track tilt capability of 30° this enables a target revisit time of less than 4 days. Each region is acquirable in the repeat cycle with an out-of-nadir angle of at most 5°. The local time of descending node is 11:00.

Proposals, observations, and associated research will be presented by an interactive map supporting the establishment of a world-wide user network. In case of tasking conflicts, issued observations are prioritized especially based on historical and current cloud coverage information taking satellite constraints such as power and storage into account.

All observations will be long-term archived. The catalog search and order service portal will allow all registered users to search and browse products based on the standardized protocols CSW (catalog service for the web) and WMS (web mapping service). Because of the necessary various processing options, each product will specifically be generated for each order and will be delivered using SFTP (secure file transfer protocol).

The fully-automatic on-ground processing chain will generate calibrated hyperspectral image products at three levels for the users. Level 1A/1B products are corrected to Top-of-Atmosphere (TOA) radiances including, e.g., defective pixel flagging, non-linearity correction, dark signal (and digital offset) correction, gain matching, straylight correction, radiometric/spectral referencing, radiometric calibration, and spectral defective pixel interpolation (using a simplified pixel-based atmospheric correction). Level 1C products are orthorectified to a user selected map projection (UTM, geographic, or European projection LAEA) and resampling model (nearest neighbor, bi-linear interpolation, or cubic convolution). By the method of direct georeferencing, the physical sensor model, with a correction of sensor interior orientation, satellite motion, light aberration and refraction, and terrain related distortions from raw imagery is applied. Level 2A products are compensated for atmospheric effects to Bottom-of-Atmosphere (BOA) reflectances with separate algorithms for land and water applications. Therefore, a classification (e.g., land-water-background, cloud), aerosol optical thickness and also columnar water vapor estimation is performed to obtain surface or underwater reflectances including an adjacency correction. All processors are complemented by independent data quality control routines. Calibration equipment, especially a doped sphere with spectral features and a full aperture diffuser for Sun calibrations allows achieving a spectral accuracy of better than 1 nm as well as a radiometric accuracy of better than 5%. A geometric accuracy of 100 m is achieved – improved by on-ground processing to 30 m with respect to a used reference image.

The offered operational services are complemented by a service team offering expert advice on the exploitation of EnMAP.

In order to guarantee high consistency in particular all aspects of user interfaces are realized and operated by the Earth Observation Center (EOC) of the German Aerospace Center (DLR), which has long-lasting experiences with the airborne and spaceborne acquisition, processing, and analysis of hyperspectral data.

Area A - Board 9

The PRISMA (PRecursore IperSpettrale della Missione Applicativa) Earth Observation program is one the most important Italian investments in the field of Remote Sensing. Fully funded by the Italian Space Agency and based on the Italian Industries status of the art technology, PRISMA will provide images acquired by an innovative electro-optical instrument using an Imaging Spectrometer, able to acquire in contiguous spectral bands ranging from 400 to 2500 nm, optically integrated with a medium resolution Panchromatic Camera (PAN).
The PRISMA payload uses pushbroom scanning technique to collect PAN/HYP image strips up to 1800 km in length and along sub-satellite track. Several additional data (e.g. calibration data, etc.), as necessary, is also provided for the on ground processing. PRISMA images Swath is 30 km width and the Ground Sampling Distance is 30 m (hyperspectral) and 5 m (panchromatic) at Nadir. PRISMA has also the agility to acquire data up to +/-20.7° off-nadir.
The Ground Segment data processing provides at Sensor Radiance (Level 1 products) or at Surface Reflectance obtained by applying atmospheric correction (Level 2b product) and geolocation (Level 2c/2d products).
The combined use of hyperspectral and panchromatic is expected to give an innovative contribution to Earth Observation applications, by providing both the capabilities of recognition of the geometric characteristics of the observed target and of extraction of the chemical characteristics of the constituent’s materials. The benefits expected are mainly in the field of vegetation and forest monitoring, precision agriculture, water quality assessment, raw materials and soils properties observation.
The PRISMA primary operational mode is “user driven” for the collection of hyperspectral and panchromatic images on specific targets, as requested by users. As secondary operational mode PRISMA will perform “background” images acquisition such that the satellite and downlink resources will be fully exploited. The Users can access the PRISMA system via web, either to submit new image acquisition requests and/or to access the catalogue browser, to search for images already acquired and to require products from archived data.
The launch of the PRISMA satellite is planned on February 2019 (VEGA Launcher) for a five years operational lifetime.
This paper report an overview of the PRISMA mission status and activities.

Area A - Board 10

For a long time hyperspectral imaging from satellite platforms was limited to a few specialized missions carrying complex instruments. They have been used to image confined areas very infrequently because of strong data downlink and limited detector sizes. Recent technological progress has allowed to create much more compact and lighter hyperspectral instruments. Some of the most radically innovative designs use narrow-band thin film interference filters. Imec, the Belgian research institute for nanoelectronics and digital technologies, has developed a unique technology which allows to deposit such filters directly on image sensors. With such integrated filter-sensor chip combination, the size of a hyperspectral imager can be made as compact as a comparable panchromatic imager. Such solutions respond very well to two important trends in the space business: the demand for detailed spectral information and the need to build smaller and more cost-effective satellite platforms

Within the ESA-GSTP project CHIEM an engineering model hyperspectral camera was developed making use of a sensor with direct filter deposition.
It uses narrow-band filters arranged in a linewise fashion on the sensor. Full hyperspectral information is built up by capturing images in rapid succession while scanning over an area. This makes it operationally similar to a pushbroom camera, making it ideally suited for use on satellite systems. The individual acquisitions however have the advantage that each contains a meaningful 2D view of the area.

Building upon expertise gathered with developing smaller sensors, geared for use on RPAS platforms, several advances were targeted for CHIEM.
- Higher spatial resolution and a wider swath (25m GSD and 100 km swath for a 600 km altitude) by using a larger CMOS sensor array (12 Mpx, compared to earlier 2 Mpx)
- more detector lines per spectral band.
- A higher sensitivity by developing in parallel to the more traditional Front Side Illuminated (FSI) also a Back Side Illuminated (BSI) version of the sensor
- 2 panchromatic zones to allow broad spectral band imaging for high spatial resolution and higher geometric accuracy

The on sensor filter deposition greatly surpasses the classical Linear Variable Filter systems as it allows pixel-precise control of the spectral response and thus also allowing to have several sensor lines with the same spectral filter, increasing the sensitivity.
The CHIEM hyperspectral sensor is also providing a “smart spectral solution” allowing to extract information from a selection of spectral bands through on chip & onboard processing. The powerful on board computing power, the high frame rates and the twelve sensor lines per spectral band will also allow applying different techniques as HDR imaging and digital TDI. Special care is taken to onboard processing techniques so that enormous data amounts which HS imaging systems can produce will be limited for transmission.

At present our consortium is developing a proto-flight model instrument with the aim to have an In Orbit Demonstration mission in 2020. The instrument physical dimensions will be optimised to fit a 12U CubeSat. The next steps would then include the mechanical design of the mirrors, Focal Plane assembly and eventually the manufacturing of the optical system and the full instrument. At the same time the IOD mission definition will be executed, including the identification of application scenes, processing levels and data products.

Area A - Board 11

The advent of spectral high-resolution top-of-canopy observations bears the promise to turn formerly ill-posed problems in the retrieval of soil and canopy parameters into a well-posed ones. In addition, the increased information content may be used to retrieve more parameters of more complex physical models. We demonstrate the design and inversion of such a model, aiming at the simultaneous retrieval of soil type, soil moisture saturation, snow cover fraction, leaf are index (LAI), fraction of photosynthetically active radiation (fAPAR), chlorophyll-, anthocyanin-, and carotenoid content, along with other structural parameters. The application of Automatic Differentiation (AD) allows for the efficient inversion of the system, along with a stringent treatment of uncertainties. AD in reverse (adjoint) mode produces code which provides the gradient vector, the derivatives to all free parameters at a computational cost independent of their number. The inversion is carried out as a gradient-based minimisation of a cost function. This approach allows to do the inversion for multiple observations with varying number and uncertainty at once, and for the inclusion of soft constraints. Second order derivatives at the optimum are used for the assessment of posterior uncertainties, also efficiently computed from applying AD twice to the model code.
Other attempts in this direction without AD tend to be designed as inversions of some sort of emulator, as for instance Artificial Neural Networks (ANN), mostly for performance reasons. By doing the inversion before an optional emulation, our approach allows for an observation-based sampling of the training data, close quality-control of individual inversions, and requires only minimal prior knowledge of the parameters to be retrieved. Another benefit is the convenient evolution and extension of the system, since the codes required for the inversion are automatically generated by AD. Pre-mission synthetic (simulated) EnMAP data are the primary source for an initial benchmarking of the system.

Area A - Board 12

The Białowieża Forest represents the remaining parts of the vast primeval forest that once stretched across the European Plain. This World Heritage UNESCO site, with its 141,885 ha, plays a crucial role for biodiversity conservation.
The ongoing LIFE+ ForBioSensing project has the general aim of understanding and monitoring the complex dynamics of Białowieża Forest by the combined collection and use of field and remote sensing data over the whole study area. The Białowieża Forest is composed by an area in which stands are actively managed, a forest reserve, and a national park area.
Original data collection interested >600 small size plots (circular plots with 12.6 m radius), distributed in the whole forest, in which height, diameter at breast height, and species information was recorded in 2015. Above ground biomass was computed using regional allometric models. These plots were surveyed in 2015 with Airborne Laser Scanning (ALS) and hyperspectral data in NIR-SWIR range.
Hyperspectral data were acquired in July 2015 with a HySpex VNIR-1800 (2.5 m) & SWIR-384 (5 m) sensors; data were radiometrically calibrated and geometrically corrected using ALS. Atmospheric correction was performed in ATCOR4 using MODTRAN5model, and a spectral smoothing filter (Savitzky-Golay) was also applied. Shadows were masked, and VNIR and SWIR bands were stacked, and resampled at 2.5 m spatial resolution. Hyperspectral pixels values were averaged over the plots, for those pixels with >60% of area included in the plot. Different vegetation indices were then computed using ENVI software.
ALS was collected in same month with a Riegl LMS-6800i sensor, with a FOV of 60° and a density of 7 points per square meter. Data classification was conducted in Terrascan and data processing in R statistical package, leading to a Canopy Height Model and a Digital Terrain Model raster (1m spatial resolution) generation. The subtraction of these two layers allowed the extraction of canopy height metrics from the plots, and computation of different average canopy metrics.
Although the original field plots resulted useful for management and silviculture purposes, they were of limited usefulness for ecological monitoring and for remote sensing data link, as in this old growth forest the the large tree crowns caused important edge effects, as evidenced by preliminary analysis.
In order to overcome this issue, the size of field plots was artificially increased using ALS data, following the method proposed by Sterenzack et al. (2018): the homogeneity of the areas surrounding small plots was evaluated with ALS, selecting 100 plots in very homogeneous forest stands. For these 100 plots the radius was increased to 25m and above ground biomass was proportionally recomputed.
The ALS-AGB relationship, between ALS plot mean height and field AGB, was moderately good (Coefficient of determination = 0.7) and the addition of hyperspectral metrics improved the result but not significantly. This moderate result can be explained by possible additional problems in ground survey caused by geopositioning errors and dense understory presence.
Species dominance was derived for these plots using the map by Kamińska et al. (2018), developed by means of orthophotos and ALS data analysis.
Correlation with several hyperspectral vegetation indices (HVIs) was assessed for plots having a clear dominance of spruce, pines, and deciduous species. Plots were assigned to each class when the presence of that species resulted higher than 60%. Seven plots out of 100 resulted dominated by spruce, and Pearson correlation with three different HVIs resulted very high (Caimax = 0.89; Mresrmax = 0.88; Mrendvimax = 0.87). Plots dominated by pines were 21, and correlation with three different HVIs was moderate (Ndnisd = -0.63; Ndwimin = 0.63; Ndiisd = -0.6). 50 plots resulted dominated by deciduous species, and correlation with three different HVIs was high (Mtvi1median = -0.79; Tvimedian = 0.79; Tcarimedian = -0.78).
The obtained results, that have to take into account the uncertainty associated to the map of Kaminska et al. (2018), provide interesting information on the use of hyperspectral data for ecological monitoring.
First results indicate that for dominance identification, different spectral regions are important; the availability of the shortwave region (not always present in hyperspectral sensors) and red edge bands was fundamental. The presence of dead material resulted highly and positively correlated to Spruce dominated plots, together with chlorophyll leaf content. This can be linked to the high mortality of this species in the area in the period of the survey, caused by bark beetles attacks. Pines dominated plots resulted negatively correlated to nitrogen concentration, and linked to canopy water content information, but for these plots the relationship was weaker with respect to the other two groups.
The plots dominated by deciduous species resulted highly correlated to LAI information. Finally, a Random Forest classification algorithm was used to test the capability of the nine HVIs to map dominant species in Białowieża Forest. HVIs discriminate deciduous and pines with high performances (4% of error rate) while spruce was not distinguished, probably due to the low number of plots occurring in this group.

Area A - Board 13

Technological advancements of the last decade, have extended our imaging capabilities to various scientific domains, such as geomorphological monitoring and planetary exploration. The identification of minerals on the composition of igneous volcanic rocks with the use of Hyperspectral Imaging (HSI), is extremely significant for the characterization of the geochemical systematics on planets. Earth’s volcanic activity deposits massive and voluminous products which are spectrally convoluted, therefore mapping the latter with HSI is both challenging and intriguing. Hence the use of refined hyperspectral data, provides information on absorption spectral features that indicate the presence of minerals and an early estimation of basaltic Lava Flow (LF) alteration levels, mainly focusing on the Visible (VIS) and Near Infrared (NIR) part of the spectrum [1]. Spectral Unmixing (SU) techniques are adopted in order to identify: i. pure components (endmembers) present in the image and ii. their fractional proportions (abundances) within the pixel [2].
In the specific research, we focus on the Sicilian volcano, Mt. Etna, for an accurate estimation of the abundances of deposited volcanic products on older lava fields. The case study is intriguingly diverse, since it is an incubator of volcanological phenomena. For this purpose, a NASA EO-1 Hyperion HSI image of Etna is selected and for the sub-pixel analysis we propose four signal transformations that are assimilated in Linear Least Squares Unmixing (LLSU). After extensive pre-processing a total of 140 bands were retained. Our approach is narrowed to the part of the Historical Eruptions (1536–1669 AD) on the sensor's FOV, where deposited products present lower correlation coefficients than those of younger lava flows. The latest geological map of Etna (Fig.1c) is used as ground truth [3] for an extensive comparison of the LFs delineation and spatial distribution of the respective abundances. The resulting abundance maps from each method, are quantitatively evaluated via the Structural Similarity Index (SSIM) [4] and ultimately an efficiency assessment is given.
Mt. Etna is a large basaltic complex stratovolcano with a broad base diameter of 1.178 Km2. Its volcanism started approximately 500 ka ago, from a submarine fissure of the Gela-Catania basin and is characterized by two major components: i) the central summit craters and ii) the flank eruptions. Summit events are consistent for many years, while flank eruptions occur every few years and are correlated with tectonic faults and shallow seismic activity [5]. Etna has the ability to change its land field rapidly, vigorously and continuously. It is capable of producing both brief paroxysmal episodes and long-standing, relatively milder eruptions. During the last 100 years Mt. Etna has produced on average 107 m3 of new lava per year [6], both from its summit craters and from its flanks. The youngest active bulks of Etna are Ellitico and Mongibello. We focus our research on the persistent activity of the Mongibello volcano, which entrails the major volcanic formation of Torre del Filosofo. It contains voluminous products younger than 122 BC, with continuous temporal succession and similar lithologies. The formation is divided in three sub-formations formed in different time periods [3]. From the latter, the Mongibello Formation 1 (MF1) is ideal for testing unmixing processes, since it contains mostly altered lavas with high reflectance. The respective lava fields are distinct, voluminous and often do not mingle with younger lavas. Considering the Hyperion's limited swath range (Fig.1a) and spatial resolution over Etna, lava fields from 1536 – 1669 AD are contained in the data cube. Also, the initial mage is masked beyond the borders of the MF1 formation (Fig.1c).
Appropriate Regions of Interest (ROIs) are selected within denser lava fields and close to the eruption vents, via spectral profile examination. The endmembers were extracted from the mean pixel values for each ROI and in total 13 endmembers corresponding to 8 LFs (1536, 1537, 1566, 1610, 1614-24, 1634-36, 1646-47 & 1669 AD), two scoria cones (1610 & 1646-47 AD) and three artificial materials (industrial, semi-urban and tile rooftop areas) are selected. The determination of the number of endmembers is critical, since underestimation may result in poor representation of the mixed HSI pixels, whereas overestimation in an overly segregated area. We assume that measured pixel spectra are linear or bilinear combinations of the 13 endmembers and implement four signal transformations [7], which are integrated on a Linear Least Squares Unmixing (LLSU) non-negativity constrained method. The resulting abundance maps indicate the fraction of each endmember on the respective lava field and they are compared against the LF’s spatial distribution as shown in the geological map. Indicatively, we present the abundance maps of the first method (Fig.2) and an RGB color-composite of spatially correlated LFs, along with the extracted endmember spectral signatures (Fig.3).


Figure 1. Location of the study area: (a) Hyperion image over Etna, (b) MF1 masked image (c) superposition of MF1 in the geological map.
Figure 2. Top row: Abundance maps resulting from method 1. Bottom row: Respective region in the geological map. (a) From left of right: LFs 1536, 1614-24, 1634-36, 1669 with similar spectral trends, (b) From left to right: scoria cones and industrial areas.
Figure 3.(a) RGB color-composite of the 1536, 1614-24, 1669 correlated LF abundance maps, (b) Spectral signatures of the selected 13 endmembers

Keywords: hyperspectral data, Etna, lava flow characterization, linear spectral unmixing, bilinear spectral unmixing, fast fourier transform, structural similarity index
REFERENCES
[1] Keshava, N.; Mustard, J.F. Spectral Unmixing. IEEE SPM, 2002, vol. 19, pg. 44-57, doi: 10.1109/79.974727
[2] Dobigeon N.; Altmann Y.; Brun N.; Moussaoui S. Linear and Nonlinear Unmixing in Hyperspectral Imaging. Chpt. 6, Data Handling in Science and Technology Elsevier, 2016, Vol. 30., http://dx.doi.org/10.1016/B978-0-444-63638-6.00006-1
[3] Branca, S.; Coltelli, M. and Groppelli, G. Geological map of Etna volcano, 1:50,000 scale. Ital.J.Geosci., 2011, Vol. 130, No. 3, pp.265-291, doi: 10.3301/IJG.2011.15
[4] Zhou, W.; Bovik, A. C.; Sheikh, H. R. and Simoncelli, E. P. Image Quality Assessment: From Error Visibility to Structural Similarity. IEEE TIP, 2004, Vol. 13, pp. 600–612, doi: 10.1109/TIP.2003.819861
[5] Buongiorno, M. F. et al. ETNA 2003 field campaign: Calibration and validation of spaceborne and airborne instruments for volcanic applications. ASI Projects: I/R/157/02, I/R/203/02.
[6] Allard, P.; Behncke, B.; D'Amico, S.; Neri, M. and Gambino, S . Mount Etna 1993-2005: Anatomy of an evolving eruptive cycle. Elsevier Earth-Science Reviews, 2006, Vol. 78, pp. 85-114. doi: 10.1016/j.earscirev.2006.04.002
[7] Singh, K.D.; Ramakrishnan, D.A. Comparative Study of signal transformation techniques in automated spectral unmixing of infrared spectra for remote sensing applications. IJRS, 2017, 1235-1257, doi: 10.1080/01431161.2017.1280625

Area A - Board 14

In agricultural soils, due to repeated tillage operations, soil properties, e.g. texture and soil organic content, can be considered as uniform over the cultivated layer. Therefore, they can be potentially estimated from the bare soil surface reflectance. Soil texture, which originates through weathering and pedogenetic processes, is not usually changed by land management practices, but it may be altered by erosion, deposition, truncation, landfill, etc.
Hyperspectral remote sensing offers the potential of increasing the accuracy of the retrieval of topsoil properties, thereby reducing the cost of collecting and interpreting soil samples, with great application potential, e.g. in the context of precision agriculture.
The VNIR-SWIR reflectance spectra of soil shows patterns for different wavelength bands responding to different chemical and/or mineralogical compositions or molecular groups. In particular, soil reflectance is generally lower in the visible range and higher in the near infrared (700-2500 nm) region, with three major specific bands around at 1400, 1900 and 2200 nm. These features can be associated with clay minerals, OH features of free water at 1400 and 1900 nm, and lattice OH features at 1400 and 2200 nm. In addition, the spectra showed a small reflectance peak around 2200 nm, which may be due to organic molecules (e.g., CH2, CH3, and NH3), SiOH bonds, cation OH bonds in phyllosilicate minerals (e.g., kaolinite, montmorillonite). Analysis of spectral curves showed that variations in reflectance intensity and shape of the spectral curves can be due to differences in soil texture due to the correlation between the shape of NIR spectra and soil texture.
In general, it is well documented, e.g. from studies carried out using spectral libraries in the lab, that topsoil variables retrieval by EO data is particularly sensitive to spectral resolution. The capability of accurately characterizinge some soil components (e.g. SOC) depends on bandwidth, position and signal characteristics in relation to chemical absorption features (e.g. overtones of hydroxyl group, etc.), and to other characteristics not directly connected to chromophores. Spatial resolution is also expected to affect the accuracy of toposoil properties retrieval, as a function of the spatial structure of the properties, but it more difficult to assess, given the support scale issue, deriving from the punctual soil sampling that is generally carried out.
The forthcoming hyperspectral satellite missions, such as PRISMA, EnMAP and ESA Copernicus CHIME mission, will allow a considerable improvement of the current soil mapping capability in the near future.
The present study, carried out in the context of an ESA supported study for CHIME Mission Requirements Consolidation, explores the sensitivity of the topsoil properties estimation accuracy to spectral and spatial resolution. Data were acquired by the airborne imaging spectrometer AVIRIS-NG, in the spectral range 380 nm to 2510 nm with 5 nm bandwith, with a pixel size of 3.7 m, in the context of a field campaign carried out near Grosseto (Central Italy) in the summer of 2018. Bare soil fields were imaged and soil samples were collected and subsequently analyzed in the lab. Semivariogram analysis allowed to explore the spatial structure of the soil properties from ground data and ordinary kriging was used to produce maps of the soil properties and of their uncertainty. Two different multivariate methods were then used for the calibration and validation of clay and soil organic carbon data estimation models: partial least square regression (PLSR) and random forest (RF). These techniques were applied to the original AVIRIS-NG data, as well as to data resampled to lower resolution spatially or spectrally or both. The results allow the assessment of the advantage of the highest spatial and spectral resolution available, also for retrieval accuracy of properties with a semivarigram range exceeding the image pixel size.

Area A - Board 15

There is great interest in the use of hyperspectral data for the retrieval of top soil properties, enhanced by the growth of the use of Earth Observation products in the framework of crop management and precision agriculture. This thematic domain is one of the priorities in the framework of the Copernicus Space Component (CSC) in the mid-2020s, in order to meet user needs and to reinforce existing services by increasing monitoring capacity and improving variable retrieval efficiency, also by the introduction of an hyperspectral mission, namely CHIME.
During summer 2018 an ESA aerial multi-sensor hyperspectral campaign has been conducted in Italy in the Grosseto area (Tuscany) for the twofold purpose of contributing to the FLEX and CHIME missions’ studies. Contemporary field surveys have been performed to acquire reference spectra and more generally ground truth. In this framework one hundred soil samples on a varied field (around 70 ha wide), have been also collected. The scope was to record a dependable data set to be fully exploited for the evaluation of retrieval capabilities as related to soil properties using hyperspectral data. In particular, thanks to the cardinality and distribution of samples it is possible to study the spatial variability of some principal soil variables (soil texture, soil organic carbon, salinity) of topsoil layer, for which the samples have been analysed in the lab (also spectrally).
Several aerial hyperspectral overfligths have been performed with different sensors in the period June 5th - July 10th allowing to sense the target field in different soil conditions as related to crop practices and development (i.e. successively after harvesting, ploughing, secondary tillage and seedbed preparation, crop emergence). The contribution introduces the soil data set collected in the field, the hyperspectral data available and discusses the impact of different field conditions on the remote signal as a first step in the study of topsoil properties retrieval capabilities for the CHIME mission.

Area A - Board 16