This study presents a methodological framework that predicts topsoil organic carbon (SOC) using spectral recordings from a low-cost near-infrared (NIR, 1350 - 2550 nm) device and geospatial covariates related to landform, climate, and vegetation. Initial attempts using the extreme gradient boosting algorithm yielded moderate accuracy metrics (RMSE 0.36 log-SOC), emphasizing the necessity for synergistic integration with geo-covariates to address existing limitations. To address this, a dual-input deep learning framework was utilized, where diverse data inputs are fed into two distinct streams. One stream utilizes a convolutional neural network with NeoSpectra sensor spectral signatures, while the other employs 214 raster spatial layers in a fully connected neural network. These streams are combined into a unified feature vector for a dense layer, generating the final prediction. Preliminary results indicate that this dual-input approach holds significant potential in enhancing predictive accuracy for handheld NIR sensors, resulting in a reduced RMSE of 30.56%.

This study evaluates the potential of EnMAP spaceborne imaging spectroscopy for mapping and monitoring topsoil properties in two diverse Mediterranean agricultural regions. The research employs state of the art linear and nonlinear machine learning techniques, spectral transformations, and pre-processing methods to analyze EnMAP data quality in comparison to in-situ measurements. The findings demonstrate promising accuracy levels in estimating key soil properties, highlighting the potential of EnMAP imagery for global soil property mapping and monitoring applications.

Soil organic carbon (SOC) content is a key indicator of soil health informing about sustainable land management practices, but parcel-wide SOC mapping is challenging as it requires high-resolution data. Unoccupied Aerial Vehicles (UAVs) can collect data with cm-resolution but are not yet fully ready to be practically implemented. The aim of this study is to provide more insights in the explanatory capabilities of UAV-derived spectral and topographical variables. To this end, mixed models were employed to estimate the SOC content of three agricultural parcels with different crop types in Greece. Results showed variations in SOC content among parcels, with a vineyard and a kiwi orchard having higher values compared to a peach orchard. All models, containing topographical and/or spectral variables, explained 81% of SOC content variation of the training dataset. Besides crop type, other topographical and spectral variables were identified as significant predictors. The study emphasizes the feasibility of UAV data and specific modeling techniques for accurate SOC estimation at the parcel level, providing valuable insights for precision agriculture. The findings recommend further exploration, including machine-learning approaches in future studies.

Soil organic carbon (SOC) constitutes a fundamental component of terrestrial ecosystem functionality, playing a pivotal role in nutrient cycling, hydrological balance, and erosion mitigation. Precise mapping of SOC distribution is imperative for the quantification of ecosystem services, notably carbon sequestration and soil fertility enhancement. Digital soil mapping (DSM) leverages statistical models and advanced technologies, including machine learning (ML), to accurately map soil properties, such as SOC, utilizing diverse data sources like satellite imagery, topography, remote sensing indices, and climate series. Within the domain of ML, self-supervised learning (SSL), which exploits unlabeled data, has gained prominence in recent years. This study introduces a novel approach that aims to learn the geographical link between multimodal features via self-supervised contrastive learning, employing pretrained Vision Transformers (ViT) for image inputs and Transformers for climate data, before fine-tuning the model with ground reference samples. The proposed approach has undergone rigorous testing on two distinct large-scale datasets, with results indicating its superiority over traditional supervised learning models, which depends solely on labeled data. Furthermore, through the utilization of various evaluation metrics (e.g., root-mean-square error (RMSE), mean absolute error (MAE), concordance correlation coefficient (CCC), etc.), the proposed model exhibits higher accuracy when compared to other conventional ML algorithms like random forest and gradient boosting. This model is a robust tool for predicting SOC and contributes to the advancement of DSM techniques, thereby facilitating land management and decision-making processes based on accurate information.

The Environmental Mapping and Analysis Program (EnMAP) is a new spaceborne German hyperspectral satellite mission, whose primary goal is to generate accurate information on the state and evolution of the Earth´s ecosystems. The core themes of EnMAP are monitoring environmental changes, ecosystem responses to human activities, and management of natural resources such as soils and minerals. EnMAP started on 1stApril 2022 and is now in operational phase since over six months, with strong expectations regarding data quality and impact on soil research. In this paper, we aim to demonstrate in a few case studies the observed current capabilities for EnMAP with regard to soil mapping based on different test sites and methodologies. Key soil properties could be derived and spatially mapped in agricultural test sites in semi-arid and temperate zones such as Soil Organic Carbon (SOC) content important for soil health and carbon sequestration, texture (clay content) important for soil fertility, and carbonate content. Additionally, we test different standard and state-of-the art methodologies, including new scenarios for time-series of hyperspectral remote sensing data for improved soil products.

Analysis-ready data generated through datacube approaches are essential in Earth Observation (EO) data analysis. By pre-processing and organizing large volumes of satellite imagery into spatio-temporal datacubes, analysis-ready data facilitates efficient and streamlined analysis workflows, enabling quick access and manipulation of the multi-dimensional EO data. This allows researchers, scientists, and stakeholders to focus on analyzing the data, detecting trends, monitoring changes, and deriving actionable information. In this paper, we describe the architecture of the Greek Soil Data Cube which ingests Sentinel-2 data from the entire country and generates the annual bare soil reflectance composites which in turn may be used to predict soil physicochemical properties using artificial intelligence techniques.

This paper presents a novel methodology for estimating cropland topsoil organic carbon (SOC) content using open-access Earth Observation (EO) data and deep learning techniques in Greece. We address the scarcity of ground-truth reference data using a data augmentation technique utilizing spatial neighbors. By incorporating neighboring pixel information in a 3x3 grid (corresponding to 30 x 30 m meters in Sentinel-2), we achieve robust predictions and provide uncertainty estimations for SOC distribution in Greek croplands. Experimental results demonstrate the superiority of deep learning techniques over conventional methods in terms of estimation accuracy (RMSE=3.51, R2=0.59, RPIQ=2.21). This methodology offers a cost-effective and efficient solution for directing land management practices to enhance soil health and mitigate carbon emissions.

This study introduces the development of Spatially Upscaled Soil Spectral Libraries (SUSSL) approach to assess spectral disturbances caused by variations in surface conditions in remote sensing-based soil property prediction. The SUSSL incorporates realistic cropland reflectance scenarios using spectral modelling and aggregation techniques. By convoluting the spectral database to multispectral and hyperspectral satellite sensors, the sensitivity of spectral indices in retrieving undisturbed surface reflectance is evaluated. Preliminary findings indicate that the spectral disturbance effects significantly impact the accuracy of soil organic carbon (SOC) estimations, resulting in a noticeable loss compared to bare soil spectra. However, strict filtering criteria using spectral indices exhibit promise in enhancing SOC modelling performance, particularly for multispectral sensors. Hyperspectral sensors demonstrate higher baseline accuracies even in disturbed soil cases. This research highlights the importance of accounting for surface condition variations for reliable soil property mapping. Future work involves leveraging machine learning techniques on SUSSL data to improve prediction accuracy and spatial coverage of soil properties using Earth Observation data.

Novel computational algorithms along with cloud computing services present a great potential to revolutionize the processing of Earth Observation (EO) data for topsoil mapping. This work presents the first insights of the WORLDSOILS project, where open-access Copernicus Sentinel-2 data and auxiliary terrain attributes were synergistically utilized to derive regression models. Building on key results from well-studied data mining approaches, the current study compares various approaches of temporal mosaicking to generate a multi-year median composite dataset of exposed bare soil pixels, over cropland areas in European Union. Finally, we utilized a Random Forest model based on soil samples (calibration = 80% and validation 20%) of the recently released LUCAS 2015 database to predict soil clay content and organic carbon. Following a masking approach to generate a composite of multi-date bare soil pixels, a promising prediction performance (R2 = 0.52, n = 7605) was achieved for clay content, while the predictive performance for soil organic carbon was significantly lower (R2 = 0.30).