Earth Observation is creating new possibilities for mapping and monitoring soil health, yet the value of these products depends on how well they respond to the needs of the people who manage, advise on, regulate, or make decisions about soils. This potential will expand further with the advent of the Copernicus Sentinel Expansion missions CHIME and LSTM. CHIME will provide hyperspectral observations across the VIS–NIR–SWIR domain, while LSTM will add thermal infrared information related to land surface temperature and emissivity. Their synergistic use is expected to broaden the range and quality of EO-derived soil health information, including indicators linked to soil organic carbon, texture, moisture, mineralogy, degradation risk, and sustainable land management.

At the same time, more advanced EO products do not automatically lead to better uptake. In agricultural landscapes, soil information is interpreted through local knowledge, advisory networks, policy requirements, certification schemes, farm constraints, and trust in the data source. NextSoils+ addresses this challenge by placing stakeholder engagement and co-creation at the centre of EO-based service development.

The project brings together farmers, agronomists, cooperatives, regional authorities, researchers, certification bodies, NGOs, and policy actors across Mediterranean pilot regions. Engagement activities include stakeholder identification, power–interest mapping, influence mapping, targeted consultations, interactive meetings, and structured feedback. These activities help clarify which soil-related decisions matter most to different actors, what information is useful in practice, and how EO-derived indicators should be communicated to support interpretation and uptake.

Early service outputs, such as soil organic carbon maps, digital interfaces, and indicator-based visual products, are used as shared reference points for dialogue. They allow stakeholders and technical teams to discuss local relevance, usability, uncertainty, trust, and integration into existing farming, advisory, certification, or reporting workflows.

By linking the emerging capabilities of CHIME and LSTM with iterative stakeholder engagement, NextSoils+ contributes practical insights into how EO-based soil health products can evolve into services that are scientifically robust, understandable, and relevant for real soil management contexts.

Extreme weather events, including droughts, heatwaves, intense rainfall and episodic non-seasonal frosts, are increasingly disrupting soil functionality and agricultural production stability. In Mediterranean agricultural landscapes, these pressures interact with long-standing soil degradation processes such as declining organic matter, erosion, salinisation, compaction and reduced biological activity, creating new risks for food security and land management. There is therefore a growing need for integrated, spatially explicit tools that can assess not only the current condition of soils, but also their capacity to buffer climatic stress, recover after disturbance and support stable crop production over time.

The Soil Resilience and Food Security Index (SRFSI) is proposed as a composite framework for evaluating the resilience of agricultural landscapes under climate-extreme conditions. The index integrates four complementary dimensions: Soil Resilience Capacity, representing the ability of soils to sustain carbon storage, water regulation, nutrient cycling, structural stability and biological functioning; Biological and Agroecological Management, capturing the contribution of practices such as compost and manure application, biofertilisers, microbial inoculants, cover crops, crop-residue retention, reduced tillage, diversified rotations and landscape elements that enhance biodiversity and water regulation; Climate-Extreme Resilience, assessing exposure and response to drought, heatwaves, heavy rainfall, erosion risk, salinity stress and crop-relevant non-seasonal frosts; and Food Security Stability, reflecting yield stability, stress-year yield retention, production reliability and post-event recovery.

Through SRFSI, a framework is proposed that can function both as a baseline diagnostic approach and as a scenario-based tool for assessing soil and production resilience. By comparing index scores under current conditions, climate-extreme scenarios and post-intervention conditions, resilience loss and adaptation gain can be quantified across different crops, soil types and management systems. Remote-sensing proxies, open soil and climate datasets, field measurements and farm-management information are combined to support implementation at field, farm, municipal and regional scales. This approach is intended to support the identification of vulnerable agricultural areas, the evaluation of biological and agroecological soil-management interventions, and the design of strategies that help maintain productive, resilient and food-secure agricultural systems under increasing climate uncertainty.

Accurate, scalable estimation of soil organic carbon (SOC) and clay content from remote sensing data is crucial for monitoring soil health at large continental scales. Standard baseline methods such as PLSR, SVR, Random Forest, Gradient Boosting etc. struggle to exploit the inherent relationship between spectral signatures and spatial neighborhoods of the corresponding samples. In order to overcome this limitation, we propose MVHGCN, a Multi-View Hypergraph Convolutional Network developed for joint SOC and clay content prediction from Sentinel-2 imagery.

The proposed architecture addresses three main drawbacks of standard approaches via the following contributions. First, hyperedges can capture higher-order relationships between soil samples, connecting pixels through both spectral or geographic similarity within a single cluster, instead of decomposing those relationships into pairwise links. Second, a multi-view formulation processes six spectral transforms of each sample (continuum-removal, absorbance, Savitzky-Golay derivatives, SNV-normalized variants), encoded via a 1D-CNN, as parallel views fused through a learned View-Based Attention Fusion module that adaptively weights each transform via an attention mechanism. Third, the model operates on two parallel branches, a global cosine-similarity branch and local multi-scale spatial branch (within-pixel and within-tile), which are fused at every block via a Cross-Talk blending unit, so local geographic proximity and global spectral similarity can be simultaneously exploited. Finally, we introduce Adaptive Hypergraph Learning (AHL), an additional attention mechanism that refines incidence matrices during training, and a multi-task head that predicts SOC and clay jointly under an uncertainty-weighted loss with learnable per-task variance.

We evaluate our approach on data from the LUCAS 2015 Topsoil survey matched to satellite-derived reflectance composites, generated from multi-temporal Sentinel-2 imagery. The proposed model is compared against a series of 1) Standard machine learning methods (PLSR, SVR, Random Forest, Gradient Boosted Trees), a state-of-the-art 1D-CNN developed specifically for estimation of soil-properties from remote sensing data, and several baseline graph and hypergraph networks. The results are reported on the original target scale, after inverting the log1p and z-score transforms. The proposed MVHGCN is consistently found to outperform all standard baselines, the respective graph and hypergraph base models, as well as the state-of-the-art 1D-CNN model.

Soil spectroscopy provides a rapid and cost-effective approach for estimating soil properties relevant to soil health assessment. However, the transfer of spectroscopic models calibrated on large soil spectral libraries to new sites remains challenging because spectral responses are affected by soil type, mineralogy, organic matter composition, sample preparation, sensor characteristics and the size of local calibration datasets. Large soil spectral libraries, such as LUCAS, offer broad European-scale spectral coverage, but their use as a single calibration population may mask important spectral heterogeneity, which can affect modelling accuracy for sites with distinct soil conditions and limited local samples.

Within AI4SoilHealth, multi-sensor spectral datasets were collected from European pilot sites and combined with wet-laboratory analyses for key soil health indicators. This work evaluated whether cluster-informed calibration strategies can improve the transferability of soil spectroscopic models from LUCAS to AI4SoilHealth pilot-site samples. The analysis included visible–near infrared spectroscopy, mid-infrared spectroscopy and laser-induced breakdown spectroscopy. For each spectral technique, LUCAS spectra were clustered into spectrally coherent calibration domains, and pilot-site spectra were assigned to the most relevant spectral domains using the same clustering framework. Random Forest models were then developed using continental, cluster-specific and locally spiked calibration strategies. Model performance was evaluated for soil organic carbon, total nitrogen, pH, clay, silt, sand, phosphorus and electrical conductivity.

The results indicate that structuring the LUCAS spectral library into more homogeneous calibration domains improved prediction accuracy across pilot sites. For Vis–NIR spectroscopy, locally spiked cluster-based models improved predictions for soil organic carbon, total nitrogen, clay and electrical conductivity in most pilot cases. MIR spectroscopy showed the most consistent improvements for soil organic carbon, total nitrogen and pH, reflecting the sensitivity of the mid-infrared region to organic matter composition, mineral constituents and chemically specific absorption features. LIBS also showed positive median gains across the evaluated properties, with particular relevance for pH, texture fractions and phosphorus, consistent with its sensitivity to elemental composition.

Overall, the results demonstrate that cluster-informed calibration can improve the transferability of continental-scale soil spectral libraries to local monitoring contexts, supporting the scalable prediction of soil health indicators across diverse European environments.