A prototype processor for water quality exploiting PRISMA satellite hyperspectral images has been developed in the framework of the ASI project “Sviluppo di Prodotti Iperspettrali Prototipali Evoluti” (Contract ASI N. 2021-7-I.0). The main objective of the project is the prototyping of a subset of Level 3 / Level 4 value-added products to be retrieved by processing Level 2 hyperspectral data. The Water Quality Prototype is a combination of state-of-the-art techniques for the retrieval of the following parameters, useful for the characterization of both inland and coastal waters: Phytoplankton, Total Suspended Matter (TSM) and Bottom Substrate. The prototype processor ingests at-surface reflectance product and implements adaptive semi empirical, semi-analytical and analytical methods for parameters retrieval.
An adaptive band ratio algorithm was developed for the retrieval of the concentration of phytoplankton primary photosynthetic pigment (Chlorophyll-a (Chl-a)) and the accessory pigments (e.g. phycocyanin). The processor exploits the diagnostic reflectance spectral feature of Chl-a and phycocyanin. Chl-a is correlated with both the height and position of the red-edge scattering signal near 700 nm, which shifts towards increasing wavelengths as biomass increases. Thanks to the spectral resolution of the PRISMA sensor, the relative maximum and minimum diagnostic features can be performed pixel-based and adaptively identified in the image scene. Moreover, the prototype processor implements dedicated algorithms to retrieve TSM concentration and water turbidity exploiting different wavelengths in the visible or near-infrared range. A bio-optical model inversion allows the retrieval of chlorophyll, coloured dissolved organic matter and non-algal particulate matter concentration for optical deep water, or bottom substrate coverage abundances (e.g. macrophytes, sand, rocks) for optically shallow waters. Model parameters consider the Inherent Optical Properties specific for the case studies and different bottom spectral properties. The developed approach assumes a relative linear mixed distribution of up to three different substrates and a relaxed constraint hypothesis for modelling the contribution of the substrates in bottom reflectance.
The proposed techniques have been tested on PRISMA data, acquired over different Italian lakes (Lake Garda, Mantua, Varese and Trasimeno) and coastal areas (northern Adriatic Sea). The preliminary results of the prototype products validation show a good agreement respect to the in-situ data.

The rich spectral information captured by hyperspectral satellite sensors make them useful for a number of real-world applications. Detection of a target with known spectral signature, when this target may occupy only a fraction of the pixel, is an important issue in hyperspectral applications. In this work, we describe the algorithm for material detection developed and validated in the framework of the ASI Contract no. 2021-7-I.0, whose main objective is the prototyping of Level 3 / Level 4 value-added products based on hyperspectral satellite data.
A classical approach to the problem of sub-pixel material detection from hyperspectral data is based on the generalized likelihood ratio test (GLRT). However, this approach needs some assumptions about the background distribution to make the math work, thus the performance of the detector strongly depends on how well the background has been statistically characterized. The background is usually modelled using a multivariate Gaussian distribution, and the mean vector and the covariance matrix needed to compute the GLRT are replaced with their sample estimates, retrieved using a subset of image pixel as the background. In this work, several approaches for computing the background statistics have been explored. The global approach uses the whole image to compute the background statistics. In the local approach, the image was divided into tiles to estimate the local background statistics. Finally, a cluster-based approach was developed, using several clustering options (e.g. K-means and Expectation Maximization), as well as several choices of cluster selection for modelling the background. In all cases, some exclusion techniques were tested to remove pixels that have a spectral signature similar to the target, therefore increasing its contrast to the background.
The output of the developed prototypal detector includes the GLRT values for the image, called soft detection map or “heat map”, and the binary map created applying a threshold on the GLRT to select pixels that likely contain the target material, named hard detection map.
The performance as well as the robustness of these detectors have been evaluated, with very promising results, on real hyperspectral data acquired by the ASI PRISMA Italian mission. The proposed solutions is worthy of interest for real applications.

The PRIS4VEG project aims at the development and optimization of techniques and algorithms for an innovative and quantitative monitoring of vegetation in agricultural and forest ecosystems using PRISMA data. In particular, the PRIS4VEG project focuses on the generation of functional plant trait (i.e. biophysical, biochemical and ecophysiological) products from PRISMA reflectance data and their use for the generation of higher level products related to ecosystem functional diversity and ecosystem heterogeneity.
In this contribution, we exploit hyperspectral data cubes collected by the PRISMA satellite to develop and test a hybrid retrieval workflow for forest trait mapping and to estimate spatial patterns of plant functional diversity of mixed forest ecosystems. The hybrid retrieval scheme consisted in the use of physically based radiative transfer models for the forward simulation of a set of spectral responses as a function of the model input variables, and in the use of machine learning regression algorithms to learn the relationships between the simulated spectra and the model input variables. The model trained on the simulated dataset was then applied to the real remotely sensed spectra for estimating the traits of interest. Finally, the accuracy of the proposed retrieval scheme was evaluated against ground data collected in correspondence of PRISMA overpasses. Based on their ecological importance in terms of plant functioning, the plant traits investigated at leaf and canopy level were: leaf chlorophyll content (LCC), leaf water content (LWC), leaf mass area (LMA) and leaf area index (LAI).
Strong correlations were found between measured and predicted values of the LCC and LAI. Slightly worst results were achieved for LMA and LWC. Overall, the estimated LCC and LAI showed value distributions within the ranges expected based on the field measurements. The LCC and LAI maps showed some similarities, but they were not totally correlated. Based on the tree species functional composition obtained from a previous classification performed using airborne data, we observed that LAI showed a larger inter-species variability. Also, the LCC did not differ significantly in regeneration and mature stands, while LAI was higher in the regeneration stands.
Finally, emerging methods aimed at the estimation of ecosystem heterogeneity (representing the degree of non-uniformity in land cover, vegetation and physical factors) through spectral diversity metrics were applied to hyperspectral reflectance and vegetation indices. Information theory metrics were also computed on vegetation trait maps to characterise the ecosystem heterogeneity. The spatio-temporal patterns in the ecosystem heterogeneity maps were discussed in relation to several factors/processes expected to drive biodiversity changes in the study areas.
The results obtained in this study demonstrated that the retrieval of a broad set of leaf and canopy traits from space using hybrid retrieval schemes is feasible, paving the way for future operational algorithms for the routine mapping of vegetation traits from spaceborne sensors. Also, the use of hyperspectral reflectance can improve our ability in biodiversity mapping compared to state-of-the-art measures based on broad band vegetation indices available from current platforms.

The cryosphere has a relevant role for understanding the changes on our planet. Because of its importance, cryosphere has been investigated by addressing different remote sensing instruments and techniques. Even though many products and studies exist from multispectral and radar images, the exploitation of hyperspectral images is still in an early phase because of the limited availability of these sensors up to now. The study of the cryosphere by means of remote sensing hyperspectral data in the domain of reflected solar radiation (400-2500 nm) can contribute to determine key surface parameters such as albedo, grain size, liquid water content, concentration of organic and inorganic impurities, distribution of debris cover on glaciers, proglacial lakes and their surface characteristics. In this context, PRISMA data represent a unique opportunity for the development and optimization of algorithms for the estimation of physical parameters of snow and ice and are particularly well suited for investigations in complex morphologies such as alpine areas.
In this contribution, we present the main activities undertaken within the SCIA project founded by the Italian Space Agency. The main goal of the project is the development and optimization of algorithms to generate products useful for monitoring the cryosphere in different geographic and climatic context, with particular focus on mountainous alpine areas. The overall methodology is based on the jointly exploitation of satellite images, in-situ measurements, and radiative transfer modeling.
Particular attention will be initially paid on the generation of surface reflectance corrected for topography and including adjacency effects. We will present simulations performed by using radiative transfer models and the influences of the snow parameters on surface reflectance. Preliminary algorithms for detecting snow parameters, such as albedo, grain size and light absorbing impurities will be also presented at different scales. For assessing processes due to glacial-lakes interaction we will show some examples of deglaciation processes revealed by the amount of suspended solids into the lake. By exploiting the state-of-the-art solutions of hypersharpening (fusion of PAN and hyperspectral images) we map lakes in terms of number, size and shape and to compute the chromaticity coordinates and the dominant wavelengths to support the analysis of lake water properties. Finally, some examples of debris covered glacier are also addressed. Although the high level of accuracy and automation achieved to map ice and snow by satellite sensors, the recognition of supra-glacial debris is still an issue when the glacier snout is debris covered, and more in general, for those glaciers that are partially or totally debris-covered, the exploitation of spectroscopic method help the detection of such conditions.
In summary, imaging spectroscopy is promising for the detection of all these parameters and the possibilities offered by PRISMA to detect subtle spectral features can open new perspectives in the remote sensing of the cryosphere.

The idea of the PRIMARY (PRIsma for Monitoring AiR qualitY) project, recently co-funded by the Italian Space Agency (ASI), is to exploit the potentialities provided by the ASI-PRISMA (PRecursore IperSpettrale della Missione Applicativa) mission for monitoring particulate-matter related air quality at an urban/sub urban scale. In particular, the project aims at using the information contained in the PRISMA hyperspectral data to extract information on the atmospheric particulate matter concentration and composition. At the present state of art, a satellite-based product that provides a ‘speciated’ aerosol load as the one foreseen in PRIMARY is not available, in particular in terms of spatial resolution and computational time for image processing.
Given the complexity of the hyperspectral data inversion, PRIMARY will follow a physics-based machine learning approach. More specifically, we will exploit the ability of neural networks (NNs) in recognizing even very weak and highly non-linear relationships between radiances derived from the PRISMA hyperspectral sensor (NN input) and aerosol chemical-physical quantities (NN output). For training the NNs, a novel dataset will be generated starting from a number of statistically significant atmospheric and aerosol profiles provided by the Copernicus Atmosphere Monitoring Service (CAMS), with the related aerosol optical properties obtained through a post-processing tool called FlexAOD (https://doi.org/10.5194/acp-19-181-2019). Finally, the corresponding synthetic electromagnetic measure simulating the PRISMA acquisition in terms of spectral radiance is obtained using a radiative transfer model (e.g. Libradtran). This brand new dataset will have a global coverage in order to enable the operability of the PRIMARY algorithm on a large scale.
A crucial part of PRIMARY will be the validation phase and output refinements. Within the Project, “ground truth” measurements will be collected within specific field campaigns to be held in the urban area of Rome (Italy), which has been selected as Pilot Target area of the project. In this framework, ground-based and in situ aerosol observations (e.g. from the BAQUININ super site) will be coupled to measurements with aerial systems (UAV and/or helicopters) to better characterize the aerosol variability within the atmospheric column.
Thanks to the foreseen flexibility of the neural network algorithms, in the final phase of the project, the developed methodology will also be evaluated for data fusion schemes between PRISMA and other missions focusing on atmospheric research, in particular Sentinel-5P/TROPOMI.
A multidisciplinary group, with complementary expertise, composes the PRIMARY team. It will be coordinated by the Earth Observation research group of the University of Rome “Tor Vergata” that will be also engaged in the NN algorithms design and development. The University of L’Aquila, together with the Institute of Atmospheric Pollution Research (CNR-IIA), will work on model-based synthetic data generation. The Institute of Atmospheric Sciences and Climate (CNR-ISAC) will lead the validation activities in collaboration with SERCO, that will also develop an automatic procedure for PRISMA products analysis and co-location with data collected within field campaigns. Moreover, several environmental agencies, at both regional and national level, will provide support for the final products user requirements definition and validation.

Methane emissions from fossil fuel production activities typically happen as so-called “point emissions”, namely plumes emitted from small surface elements and containing a relatively large amount of gas. The detection and elimination of these methane emissions have been identified as “low hanging fruits” to reduce the concentration of greenhouse gases in the atmosphere.

Satellites offer a unique capability for global monitoring of methane emissions. The retrieval of methane from space measurements typically relies on spectrally-resolved measurements of solar radiation reflected by the Earth's surface in the shortwave infrared (SWIR) part of the spectrum (~1600–2500 nm). The Sentinel-5P TROPOMI mission allows to monitor methane emissions around the globe at regional scales with a daily resolution, but lacks the spatial resolution needed to pinpoint single point emissions.

Imaging spectrometers, such as PRISMA, sample that 2300 nm region with tens of spectral channels and a typical spatial resolution of 30-m, which can be exploited for the detection of point emissions and the attribution to particular emitting elements. PRISMA is currently the only 400–2500 nm imaging spectrometer with potential for high resolution methane mapping currently accessible to the international science community

In this contribution, we evaluate the potential of PRISMA to map methane point emissions. Our retrieval of methane concentration enhancements is based on a matched-filter based algorithm applied to PRISMA spectra in the 2300 nm shortwave infrared spectral region. We perform a simulation-based sensitivity analysis to assess the retrieval performance for different sites. We find that surface brightness and homogeneity are major drivers for the detection and quantification of methane plumes with PRISMA, with retrieval precision errors ranging from 61 to 197 parts-per-billion in the evaluated images. The potential of PRISMA for methane mapping is further illustrated by real plume detections at different methane hotspot regions, including oil and gas extraction fields in Algeria, Turkmenistan, and the USA (Permian Basin), and coal mines in the Shanxi region in China. Our study reports several important findings regarding the potential and limitations of PRISMA for methane mapping, most of which can be extrapolated to upcoming satellite imaging spectroscopy missions.