Conference Agenda

.

Mapping and especially monitoring of soil properties is of great interest which is expressed in an increasing number of political recommendations and policies (e.g. SDGs, 4p1000). This importance is based on many ecosystem services provided by soils such as provision of food, carbon sequestration and water purification. These services are partially endangered due to progressing soil degradation in many regions caused by e.g. urbanization or deforestation. In order to investigate the status of our soils spatially and on large scales, remote sensing especially using hyperspectral sensors provides a valuable tool. This has already been proven using airborne sensors and is currently investigated by applying data of the recently launched second generation spaceborne sensors. These first studies based on spaceborne imaging spectroscopy missions show promising accuracies. A limiting factor is the availability of bare soils at the surface which show a variability in agricultural areas depending on the season. To overcome this challenge, multitemporal composite maps of soil properties can be generated which are based on a set of available multitemporal hyperspectral images recorded at different dates. Each image will show different bare fields and their combination increases the spatial coverage of the final soil map. Currently, datasets with multitemporal hyperspectral images and matching soil ground reference samples are sparse but the availability will increase with future missions like CHIME and SBG. In this study a set of multiple hyperspectral PRISMA and EnMAP images recorded over the same study site at different dates, is investigated and merged into final soil properties maps. Potential study sites with suitable datasets are located in Northern Germany (acquisitions between August 2020 and August 2021) and Central Spain (images recorded between May 2021 and September 2023). Thereby, questions such as the required number of images and general limitations are evaluated. A central aim is to evaluate different workflows of merging the multitemporal images. Possibilities are to combine all images into a synthetical image with e.g. median spectra or to compile separate soil property maps for each image and merge them at the end. The creation of up-to-date and spatially more complete maps using short time-series of a few years will be one step towards global mapping and, given the availability of longer time-series data, also monitoring of top-soil properties.

The NASA Surface Biology and Geology Decadal Survey mission is in development and includes global Visible to Short Wavelength Infrared (VSWIR) imaging spectroscopy observations with 16 day revisit and 30 m spatial sampling. The core VSWIR science addresses a range of most important Decadal Survey objectives spanning the domains of terrestrial ecology, hydrology, coastal and inland waters, and geology. A companion set of important new applications are identified and addressed. In support of these science and applications objectives, the VSWIR Project Science Team, supported by many colleagues, has developed a comprehensive set of measurement requirements that have led to the current mature VSWIR Project design. These requirements have in turn been used to establish the baseline measurement processing architecture from instrument-recorded signal, to top-of-atmosphere radiance, to surface reflectance, fractional cover, and suites of products addressing the core science and applications VSWIR objectives. At all product levels, uncertainty estimates are determined and reported. Special attention has been fucused to assure a comprehensive set of product characterization information is available at every level. This characterization information is intended to enable a broad set of additional “instrument-agnostic” algorithms that build upon the core products delivered by the Project. All Project algorithms are open. Work is ongoing to coordinate the product characterization information with other imaging spectroscopy projects and missions, to enable broad interoperability and harmonization of VSWIR derived products. We present the current status of the SBG VSWIR observation requirements and product processing plans with a focus on international cooperation in space imaging spectroscopy.

Remote sensing is emerging as a potential biodiversity monitoring tool, pushed by recent advances and the emergency of biodiversity decline. In particular, the arrival of the new hyperspectral missions, such as CHIME, has brought new perspectives for monitoring vegetation biodiversity from space since they are expected to capture an extensive range of plant functional traits and species-specific features describing different aspects of biodiversity. Nonetheless, the quantification of plant biodiversity, and in particular, plant functional diversity, is more challenging than merely estimating plant biophysical properties (many of them considered plant functional traits). Among others, the lack of adequate field datasets sampled with the aim of validating remote sensing biodiversity estimates hampers the development of robust methods and products.

We have developed BOSSE, a Biodiversity Observing System Simulation Experiment, to overcome the most fundamental issues in this emerging field. BOSSE simulates dynamic vegetation scenes that evolve over time as a function of their sensibility of meteorological conditions and represents the associated hyperspectral features in different spectral regions. BOSSE can simulate observations of varying spectral configurations and spatial and temporal resolutions, mimicking current and future remote sensing missions. Overall, BOSSE is a tool for testing hypotheses and methods and assessing the limitations and capabilities of remote sensing in capturing plant functional biodiversity.

We used BOSSE to test various hypotheses regarding the capability of different approaches and remote sensing missions to capture plant functional diversity from the simulated scenes varying over time. In particular, we focused on estimating plant functional diversity over large areas and combining hyperspectral reflectance imagery with other remote sensing products. The results determined which approaches and products can or cannot robustly infer plant functional diversity and revealed that while hyperspectral reflectance synthesizes most of the biodiversity spectral information, its combination with information from other spectral regions can increase the robustness of the estimates. The results are coherent and prove the capability of our tool to support research and new findings in the emerging field of remote sensing of vegetation biodiversity.

Rapid advancements in hyperspectral Earth observation (EO) are opening novel possibilities for environmental monitoring. EnMAP, PRISMA, and EMIT are pioneering globally sampled hyperspectral time series, paving the way for next-generation global satellite missions like CHIME and SBG. Hyperspectral time series promise high precision in retrieving key vegetation parameters essential for better understanding ecosystem state and dynamics. Non-photosynthetic vegetation (NPV) is one such parameter, expected to be quantified globally with improved accuracy. However, developing a universal NPV retrieval scheme applicable across diverse ecosystems remains a significant challenge.

Here, we introduce a universal scheme for retrieving NPV fractional cover time series across diverse ecosystems. We utilize time series of EnMAP data acquired monthly in 2023 and 2024, covering study sites located in California and Namibia. These sites encompass diverse grasslands, shrublands, woodlands, and forests, exhibiting high spatial and temporal diversity. Our approach relies on (i) a spectral library comprising representative NPV, green vegetation, and non-vegetation spectral signatures, (ii) synthetic mixing of these library spectra, (iii) establishing a regression model based on the synthetic data, and (iv) applying the model for NPV retrieval. In situations where field/laboratory spectral measurements are unavailable, we propose a universal solution for image-based spectral library development. Our results reveal spatial-temporal patterns of NPV cover fractions across all study sites, enabling detailed assessment of vegetation condition and seasonality. Reference data confirms the accuracy of our approach in estimating NPV fractional cover across ecosystems.

Our NPV retrieval scheme across ecosystems is designed for simplicity, reproducibility, and automation to support operationalization. The EnMAP time series serves as an ideal database for developing universally applicable algorithms and showcasing the application potential of future spaceborne hyperspectral satellite missions for high-level vegetation product generation, including NPV.

Climate change and land use change re-distribute organisms likely leading to novel trait syndromes (i.e. trait co-occurrence bundles) to emerge and thus affecting the ecosystem service potential. However, we have been limited in the study of such processes to limited data availability to understand trait syndrome dynamics. Imaging spectroscopy has shown many options to measure, map and monitor plant traits, enabling the study of trait syndrome dynamics. Here we examine the extent to which trait syndromes change over time in an inland delta effected by multiple global change drivers. We utilize imaging spectroscopy data over the large and diverse landscape of the Sacramento-San Joaquin river delta in California to estimate plant traits in aquatic and terrestrial plant communities (emergent aquatic plants and riparian plants), examine their associations to identify trait syndromes and examine how these trait syndromes change over time.

To do so we examined 88 traits retrieved from imaging spectroscopy data over two periods of five years each, 2004-2008 and again 2014-2019 and describe the trait syndromes for both terrestrial and aquatic plant communities. Trait syndromes include a combination of traits related to pigments, leaf morphology, leaf biochemistry and water content and lignin, cellulose and salinity traits. Emergent plant communities show stronger changes of trait syndromes between the two periods of time in comparison to the riparian communities. We believe that this framework could be expandible to other ecosystems, and provide us with a better understanding of ecosystem dynamics and functioning.

Understanding soil is vital for the efficiency of agricultural systems. Remote sensing, which analyzes the "spectral signature" of materials, is used to obtain information on soil characteristics. The Italian PRISMA and German EnMAP hyperspectral satellite missions are expected to enhance our understanding of the spatial variability and mapping of agricultural soil properties.
This study focuses on the Jolanda di Savoia farm in Italy (Lat. 44.87°N, Lon. 11.97°E), where soils show variability due to buried paleo-channels from reclaimed marshlands. Multiple hyperspectral images, 16 PRISMA and 6 EnMAP, were acquired over this area. Concurrently, topsoil samples were collected from 32 agricultural fields, each with Elementary Sampling Units (ESU) of 30 by 30 meters. Wet analysis determined properties such as Soil Organic Matter (SOM), CaCO3, and texture.
To isolate bare soil pixel and exclude vegetation, various processes were employed, including spectral indices like Normalized Difference Vegetation Index (NDVI), non-photosynthetic vegetation (nCai) and maximum area under the curve for soil moisture. Three datasets were created: i) all combined unfiltered time series samples; ii) EnMAP's driest samples; iii) PRISMA's driest samples. The datasets underwent pre-treatment to reduce the impact of noisy bands and remove/reduce the scattering impact. Different machine learning regression algorithms and strategies were tested using a K-fold cross-validation approach. Results show that algorithm training on the EnMAP and PRISMA combined time series gives similar results to the ones trained only on the driest samples of the individual case datasets. All the three data set seams, anyhow, not performing well for sand abundance detection. Results show that on the PRISMA-EnMAP combined data set can’t be identified an optimal pre-treatment method for all soil variables retrieval. Gaussian Process Regression (GPR) algorithm produced a RMSE of 9.16% and an R² of 0.66 for clay; an RMSE of 7.8% and an R² of 0.62 for silt, while an RMSE of 2.4 and an R² of 0.61 was obtained for SOM prediction. Lastly, for calcium carbonate (CaCO3), GPR showed an RMSE of 2.9% and an R² of 0.52.
This study pointed out that GPR applied to the unfiltered hyperspectral time series of Jolanda test site shows a good potential for assessing topsoil properties with an accuracy comparable to the one obtained by selecting, among the time series, the driest samples.

.