Search Results (52)

Soil organic carbon (SOC) plays a critical role in climate regulation, soil fertility, and ecosystem resilience, making its accurate spatial quantification essential for sustainable land management and greenhouse gas (GHG) reporting. However, mapping SOC in heterogeneous agroforestry systems remains challenging due to vegetation cover and landscape complexity. In this study, we develop and evaluate a hybrid bare soil modelling- Digital Soil Mapping supported by ML framework to generate high-resolution soil properties predictions in Mediterranean agroforestry systems (Extremadura, Spain). A dual modelling approach was implemented, combining (i) Bare Soil modelling using Sentinel-2 multi-temporal reflectance composites and (ii) Digital Soil Mapping (DSM) supported by environmental covariates (climate, terrain, vegetation) following the SCORPAN framework. Machine learning models, namely Quantile Regression Forests (QRF) and Extreme Gradient Boosting (XGBoost), were applied and optimised using automated hyperparameter tuning (FLAML). A total of 107 LUCAS topsoil samples and 36 complementary points from the Forest ICP Level I were used for calibration and validation, with a 70/30 train–test split. Results show that Sentinel-2-based modelling can effectively capture SOC spatial variability in bare soil conditions, while DSM improves predictions in vegetated areas. Model performance reached R² values up to 0.76 (QRF, pH) and RMSE as low as 0.03 (XGBoost, N), with uncertainty quantified using the Prediction Interval Ratio (PIR) and performance further supported by RPIQ values up to 3.15. However, prediction accuracy remains sensitive to vegetation structure and sample density. The proposed framework provides a scalable and uncertainty-aware approach for SOC mapping, supporting Tier-3 GHG inventories and emerging Monitoring, Reporting, and Verification (MRV) systems. The results highlight the importance of integrating multi-source datasets and hybrid modelling strategies for reliable SOC estimation in complex landscapes. Full article

High-spatial-resolution soil data collected on behalf of the private sector (e.g., farmers) represents a largely untapped resource for EU soil monitoring initiatives. Georeferenced soil samples also raise privacy issues since sampling locations can be linked to individual farm parcels and their respective activities. This work presents a Privacy-Preserving Digital Soil Mapping Framework (PP-DSM) that enables integration of private georeferenced soil datasets while releasing only aggregated spatial outputs, minimising risks of individual farm identification. The framework consists of three components: a secure soil data processing environment that keeps private point data under institutional control, a Quantile Regression Forest (QRF) engine that produces spatially explicit predictions and uncertainty estimates, and spatial aggregation of raster outputs to the EU LUCAS 2 × 2 km monitoring grid, releasing only anonymised polygon-level statistics based on polygons centred on the grid points. This methodology is demonstrated using a case study of a published georeferenced soil dataset of organic carbon (403 topsoil samples) from the Kastoria region, Northern Greece. Aggregated predictions preserved regional soil patterns while eliminating farm-level identifiability. Across six independently validated LUCAS polygons, QRF polygon statistics differed from independent test set means by an average difference of 0.070% SOC, consistent with the expected spatial smoothing. This study suggests that privately held soil datasets can support EU monitoring infrastructures within the existing regulatory environment, contributing to Soil Monitoring Law objectives and Carbon Removals and Carbon Farming initiatives. Full article

Accurate estimation of soil total nitrogen (TN) and soil organic matter (SOM) is important for sustainable soil fertility assessment and precision nutrient management. Visible–near-infrared hyperspectral sensing provides a rapid and non-destructive solution, but its inversion accuracy is strongly affected by spectral preprocessing, especially under small-sample conditions. To reduce dependence on a manually selected preprocessing operator, this study proposes a homologous preprocessing representation fusion framework based on greedy concatenation (HPRF–GC). The framework constructs multiple homologous spectral views from the same raw spectrum, selects informative views through cross-validation-guided greedy forward selection, and concatenates the selected views before random forest or support vector regression. A self-built in situ hyperspectral dataset was collected from two representative black calcareous Mollisol farms in Heilongjiang Province, China, including 200 composite samples measured with a GaiaField Pro V10 imager at 5 m height under midday illumination using white reference calibration. On this dataset, HPRF–GC reduced RMSE by 3.61% for TN–RF, 9.94% for TN–SVR, 0.87% for SOM–RF, and 7.15% for SOM–SVR compared with the strongest single-preprocessing baseline, while introducing only a modest training-time overhead. On the public LUCAS 2015 dataset, HPRF–GC achieved competitive TN prediction performance, with an $R^2$ of 0.890 and an RMSE of 1.191 under RF. These results indicate that HPRF–GC provides a lightweight, interpretable and reproducible strategy for reducing preprocessing selection sensitivity in small-sample soil hyperspectral inversion. Full article

Capillary imbibition is the process by which liquids are absorbed into porous materials as a result of capillary pressure differences at the pore scale. Accurate characterization of imbibition dynamics, particularly in the presence of gravitational potential, is essential for understanding fluid transport in diverse systems such as soil, fractured rocks, filtration media, and plant roots. This study presents systematic imbibition experiments using filter papers with pore sizes of 2.5 µm, 11 µm, and 20 µm, each inclined at 80° to quantify the influence of gravitational potential on imbibition behavior. For horizontally positioned samples, the imbibition front propagated radially and symmetrically, exhibiting a power law dependence on time. The measured temporal exponents ranged from 0.386 to 0.403, consistently lower than the theoretical value of 1/2 predicted by the Lucas–Washburn law. With increasing permeability, the temporal exponent approached the Washburn limit, indicating a marked dependence of imbibition dynamics on pore structure. For the inclined configuration at an 80° angle, the imbibition fronts remained nearly circular but exhibited a pronounced displacement of the front center toward gravity. This displacement increased with permeability, from approximately 0.497 cm for the 11 µm filter paper to 3545 cm for the 20 µm filter paper, highlighting the combined effects of permeability and gravitational potential on fluid movement. Furthermore, the advance of the imbibition front was significantly slower in the smallest pores (2.5 µm) compared to the larger ones. Experimental results were evaluated against a theoretical model proposed by Medina, demonstrating moderate quantitative agreement at early times, when gravitational potential effects are less significant. These findings confirm that both the temporal scaling exponent and the spatial evolution of the imbibition front are governed by the porous medium’s permeability and inclination angle, providing experimental evidence of deviations from ideal Washburn behavior in real porous systems. Full article

Accurate mapping of soil total nitrogen (STN) is fundamental for advancing sustainable and precision soil management. While digital soil mapping (DSM) has increasingly relied on Earth observation (EO) data, the potential of various synthetic aperture radar (SAR) features, particularly interferometric coherence and texture, remains underexplored for large-scale STN prediction. This study aimed to systematically evaluate the potential of multiple Sentinel-1 SAR-derived features, including backscatter coefficients, interferometric coherence, and texture metrics, for modeling and mapping STN across the Iberian Peninsula. We integrated 4296 soil samples from the 2018 LUCAS dataset with multi-source environmental covariates processed via the Google Earth Engine (GEE) platform. Nine modeling scenarios were designed to compare individual and combined contributions of Sentinel-1, Sentinel-2, topographic, and climatic variables using random forest (RF) and extreme gradient boosting (XGBoost) algorithms. The results indicated that the selection of SAR-derived features significantly influences prediction accuracy. Among individual Sentinel-1 feature groups, texture metrics and interferometric coherence outperformed the traditionally used backscatter coefficients, emphasizing their effectiveness in STN mapping. Specifically, texture-based and coherence-based models achieved R² values of 0.34 to 0.35 and 0.33, respectively, whereas backscatter-only models yielded the lowest accuracy (R² = 0.29 to 0.30). The integration of all three radar categories substantially improved performance (R² = 0.39 to 0.42), surpassing the performance of models based solely on Sentinel-2 optical data (R² = 0.33 to 0.34). The most comprehensive model, which combined multi-source EO data with topographic and climatic variables, achieved the highest overall accuracy with R² values of 0.51 for RF and 0.52 for XGBoost. Variable importance analysis confirmed that satellite-derived variables were the most influential group. Spatial predictions successfully captured the heterogeneity of STN across the peninsula, with higher concentrations in humid, mountainous regions and lower values in arid central plateaus and southern regions. This study demonstrates that integrating diverse Sentinel-1 radar information, particularly coherence and texture, provides a robust alternative or complement to optical data, offering a powerful tool for large-scale soil property mapping. Full article

To develop carbon farming practices, decision-makers need detailed spatial data on the soil carbon sequestration (SCS) opportunities that conventional crop and soil management creates. This study exploratorily assessed SCS capacity across agricultural land in the Valencian Community using a simple carbon balance model within a GIS framework. Within this modelling approach, maps of net primary production (NPP), land-use-derived crop harvest indices, current soil organic carbon (SOC) stocks, and NPP and SOC mineralization coefficients were combined. Results show that while NPP across Valencian croplands and grasslands ranges from 0.64 to 6.43 Mg C ha⁻¹ yr⁻¹ (mean 2.42 Mg C ha⁻¹ yr⁻¹), the actual SCS capacity is much lower, ranging from −0.04 to 1.31 Mg C ha⁻¹ yr⁻¹ (mean 0.25 Mg C ha⁻¹ yr⁻¹). Significant variation exists among land uses: rice paddies exhibit the highest SCS capacity, while olive groves present the lowest. Between 2017 and 2021, SCS in Valencian agroecosystems may have offset the sector’s greenhouse gas (GHG) emissions, primarily driven by pasture and citrus because of their large extent and moderate SCS capacity, making agriculture a net-zero emitter. However, helping achieve cross-sectoral mitigation targets will depend in part on the widespread deployment of regenerative soil management (RSM) practices. While this study identifies priority areas for RSM implementation, further research is needed to determine which specific practices are most suitable for each location to maximize SCS. Full article

Soil heavy metal contamination has attracted widespread concern globally, with nickel (Ni) posing distinct environmental and human health risks. However, high-precision prediction of soil Ni concentrations at a large scale remains inadequately explored. In this study, spectral data from 18,675 sampling points were compiled to investigate the prediction of Ni concentrations. Two widely applied and highly effective preprocessing techniques, namely first-order derivative and second-order derivative, were explored. Following spectral preprocessing, three advanced machine learning models, namely random forest, extreme gradient boosting, and light gradient boosting machine (LGBM), were constructed and compared for Ni prediction. These models exhibited robust predictive performance and excellent generalization capability. Among them, the optimal model integrating the second-order derivative and LGBM achieved a coefficient of determination (R²) of 0.582 on the training set, which was further improved to 0.613 after hyperparameter tuning. On the test set, the model yielded superior predictive results with R² = 0.585, mean squared error (MSE) = 130.284, and mean absolute error (MAE) = 6.468. Feature importance analysis identified the critical spectral bands for Ni concentration prediction, including 508–509 nm, 894–895 nm, 2214.5–2215.5 nm, and 778.5–779.5 nm. This study establishes an efficient framework for predicting soil Ni concentrations, providing valuable insights for improving predictive accuracy. It also offers theoretical support for the sustainable management of soil environments and long-term soil heavy metal risk mitigation. Full article

Soil is a critical natural resource that requires continuous monitoring to support sustainable agriculture. Among soil properties, pH is an essential indicator because it strongly affects nutrient availability and biological activity. Visible–Near-Infrared (Vis–NIR) spectroscopy offers a rapid and cost-effective solution for soil pH prediction, but traditional machine learning models often struggle to effectively extract features from high-dimensional spectral data. To address this challenge, we propose a Multi-Scale Attention Wavelet Convolutional Neural Network (MAWC-Net), which integrates multi-scale convolutions, attention mechanisms, and a Haar Wavelet Decomposition Module (HWDM) to enhance spectral feature representation. Experiments on the LUCAS2009 topsoil dataset demonstrate that MAWC-Net achieves superior prediction accuracy compared with conventional machine learning and deep learning baselines. These findings highlight the potential of wavelet-enhanced deep neural networks to advance soil property modeling and support precision agriculture. Full article

Assessing soil fertility is a complex task as it is determined by natural and anthropogenic factors, including specific agronomic interventions (e.g., fertilization and crop rotation) and broader soil management (e.g., tillage and drainage). For agricultural management, soil represents a primary production factor whose chemical–physical characteristics and macro-elements content must be known. This work presents the maps of three macronutrients, i.e., N, K, and P, in the topsoils (0–30 cm layer) of the Emilia-Romagna (21,710.1 km²) region in NE Italy. The maps and associated uncertainty at 100 m resolution were obtained via digital soil mapping (DSM) resorting to Quantile Random Forests using topsoil data from the regional soil database (N = 34,750). As Emilia-Romagna is characterized by two distinct major landforms, i.e., the intensively cultivated alluvial plain and the extensively managed mountain range of the Northern Apennines, each representing nearly half of the region, two distinct sets of numerical and categorical covariates were used as predictors for the DSM estimation of each macronutrient. Results highlight an average N content of approximately 1.57 ± 0.83 (standard deviation) g kg⁻¹ in the alluvial plain and of 1.63 ± 0.49 g kg⁻¹ in the Apennines. For exchangeable potassium (K), concentrations were 275.90 ± 92.6 mg kg⁻¹ and 210.2 ± 86.3 mg kg⁻¹ in the plain and Apennines, respectively. A stark contrast was observed for available phosphorus (P), with mean values of 40.4 ± 11.0 mg kg⁻¹ in the alluvial plain, dropping to 15.2 ± 6.1 mg kg⁻¹ in the Apennines. Such results provide useful information for assessing the fertility of regional soils and provide a reference baseline for soil quality monitoring. The resulting macronutrient maps were eventually compared with those based on the Land Use and Cover Area frame Survey (LUCAS), which represents the reference baselines at the EU scale. Full article

Identifying the primary soil parameters, weather variables and crop management practices that influence spatial variations in crop water use is essential for strategically defining optimal agricultural management practices. In this study, soil physico-chemical, weather and crop management variables were used through random forest (RF)-based modeling to evaluate the determinants of actual evapotranspiration (ETa) in winter wheat across Slovakia. ETa was estimated using Landsat imagery and the Python implementation of the Surface Energy Balance Algorithm for Land (PySEBAL), along with information from the Land Use/Cover Area frame Survey (LUCAS) over four cropping seasons. Overall, good agreements were found between PySEBAL-derived ETa and measured values, with RMSE and R² values of 0.93 mm and 0.87, respectively. Seasonal ETa values ranged from 434.87 mm to 506.12 mm, with the highest and lowest average values found in the 2011/2012 and 2017/2018 cropping seasons, respectively. The RF model showed good performance in predicting seasonal ETa, with an RMSE of 21 mm/season for the training data and 32 mm/season for the validation data, and R² values of 0.90 and 0.72, respectively. Our analysis indicated that ETa was primarily influenced by relative humidity, wind speed, solar radiation, altitude, and pH. The study further indicated that wheat production was unsuitable above 600 m elevation, while optimal crop water use occurred below 200 m. Addressing issues such as soil erosion and acidification could improve wheat crop water use efficiency across Slovakia. This modeling approach can serve as a basis to develop a crop water use forecasting system for sustainable wheat production in the region. Full article

Accurate and rapid acquisition of soil texture information is crucial to evaluating soil quality, formulating soil and water conservation strategies, and guiding agricultural resource management. Compared with traditional machine learning methods, convolutional neural networks (CNNs) demonstrate superior accuracy in soil texture prediction. To overcome the limitations of existing lightweight models in spectral modeling, such as insufficient single-scale feature representation, limited channel utilization, and branch redundancy, and to meet the demand for lightweight architectures, we propose a novel dynamic feature modeling approach: Multi-scale Routing Attention Network (MSRA-Net). MSRA-Net integrates grouped multi-scale convolutions with an intra-group Efficient Channel Attention (gECA) mechanism, combined with a multi-scale weighting strategy based on a Branch Routing Attention (BRA) mechanism, thereby enhancing inter-channel feature interaction and improving the model’s ability to capture complex spectral patterns. Furthermore, we introduce a multi-task learning variant, MSRA-MT, which employs uncertainty dynamic weighting to balance gradients magnitude across tasks, thereby improving both stability and predictive accuracy. Experimental results on the LUCAS and ICRAF datasets demonstrate that the MSRA-MT model consistently outperforms baseline models in terms of performance and robustness ($RMS{E}_{mean}$ = 9.190 and $RMS{E}_{mean}$ = 8.189 for ICRAF and LUCAS, respectively). Prior knowledge-based soft constraints may hinder optimization by amplifying intrinsic noise, rather than improving learning effectiveness. Full article

Visible near-infrared (VNIR) spectroscopy offers a cost-effective solution to quantify the spatiotemporal dynamics of soil organic carbon (SOC), especially in the context of rapid advances in spectra-based local modeling approaches using large-scale soil spectral libraries. And yet, direct temporal transferability of VNIR spectroscopic modeling (applying historical models to new spectral data) and its capability to monitor temporal changes in SOC remain underexplored. To address this gap, this study uses the LUCAS Soil dataset (2009 and 2015) from France to evaluate the effectiveness of localized spectral models in detecting SOC changes. Two local learning algorithms, memory-based learning (MBL) and GLOBAL-LOCAL algorithms, were adapted to integrate spectral and soil property similarities during local training set selection, while also incorporating LUCAS 2009 soil measurements (clay, silt, sand, CEC) as covariates. These adapted local learning algorithms were then compared against global partial least squares regression (PLSR). The results demonstrated that localized models substantially outperformed global PLSR, with MBL achieving the highest accuracy for croplands, grasslands, and woodlands (R² = 0.72–0.79, RMSE = 4.73–20.92 g/kg). Incorporating soil properties during the local learning procedure reduced spectral heterogeneity, leading to improved SOC prediction accuracy. This improvement was particularly pronounced after excluding organic soils from grasslands and woodlands, as evidenced by 13.3–21.1% decreases in the RMSE. Critically, for SOC monitoring, spectrally predicted SOC successfully identified over 70% of samples experiencing significant SOC changes (>10% loss or gain), effectively capturing the spatial patterns of SOC changes. This study demonstrated the potential of localized spectral modeling as a cost-effective tool for monitoring SOC dynamics, enabling efficient and large-scale assessments critical for sustainable soil management. Full article

Soils serve as critical carbon reservoirs, playing an essential role in climate change mitigation and agricultural sustainability. Accurate soil property determination relies on soil spectral reflectance data from Earth observation (EO), but current vegetation models often oversimplify soil conditions. This study introduces a novel approach that combines radiative transfer models (RTMs) with open-access soil spectral libraries to address this challenge. Focusing on conditions of low soil moisture content (SMC), photosynthetic vegetation (PV), and non-photosynthetic vegetation (NPV), the coupled Marmit–Leaf–Canopy (MLC) model is used to simulate early crop growth stages. The MLC model, which integrates MARMIT and PRO4SAIL2, enables the generation of mixed soil–vegetation scenarios. A simulated EO disturbed soil spectral library (DSSL) was created, significantly expanding the EU LUCAS cropland soil spectral library. A 1D convolutional neural network (1D-CNN) was trained on this database to predict Soil Organic Carbon (SOC) content. The results demonstrated relatively high SOC prediction accuracy compared to previous approaches that rely only on RTMs and/or machine learning approaches. Incorporating soil moisture content significantly improved performance over bare soil alone, yielding an R² of 0.86 and RMSE of 4.05 g/kg, compared to R² = 0.71 and RMSE = 6.01 g/kg for bare soil. Adding PV slightly reduced accuracy (R² = 0.71, RMSE = 6.31 g/kg), while the inclusion of NPV alongside moisture led to modest improvement (R² = 0.74, RMSE = 5.84 g/kg). The most comprehensive model, incorporating bare soil, SMC, PV, and NPV, achieved a balanced performance (R² = 0.76, RMSE = 5.49 g/kg), highlighting the importance of accounting for all surface components in SOC estimation. While further validation with additional scenarios and SOC prediction methods is needed, these findings demonstrate, for the first time, using radiative-transfer simulations of mixed vegetation-soil-water environments, that an EO-DSSL approach enhances machine learning-based SOC modeling from EO data, improving SOC mapping accuracy. This innovative framework could significantly improve global-scale SOC predictions, supporting the design of next-generation EO products for more accurate carbon monitoring. Full article

Accurate prediction of soil lead (Pb) content in small sample scenarios is often limited by data scarcity and variability in soil properties, with traditional spectral modeling methods yielding suboptimal precision. To address this, we propose a transfer learning-based framework integrated with SHAP analysis for predicting soil Pb content in the Gejiu mining area, Yunnan. Using pH data from the European LUCAS soil database as the source domain, spectral features were extracted via a 1D-ResNet model and transferred to the target domain (130 soil samples from Gejiu) for Pb prediction. SHAP analysis was applied to clarify the role of spectral characteristics in cross-component transfer learning, uncovering shared and adaptive features between pH and Pb predictions. The transfer learning model (ResNet-pH-Pb) significantly outperformed direct modeling methods (PLS-Pb, SVM-Pb, and ResNet-Pb), with an R² of 0.77, demonstrating superior accuracy. SHAP analysis showed that the model retained key pH-related wavelengths (550–750 nm and 1600–1700 nm) while optimizing Pb-related wavelengths (e.g., 919 nm and 959 nm). This study offers a novel approach for soil heavy metal prediction under small sample constraints and provides a theoretical basis for understanding spectral prediction mechanisms through interpretability analysis. Full article

Visible and near-infrared (Vis–NIR) spectroscopy enables the rapid prediction of soil properties but faces three limitations with conventional machine learning: information loss and overfitting from high-dimensional spectral features; inadequate modeling of nonlinear soil–spectra relationships; and failure to integrate multi-scale spatial features. To address these challenges, we propose ReSE-AP Net, a multi-scale attention residual network with spatial pyramid pooling. Built on convolutional residual blocks, the model incorporates a squeeze-and-excitation channel attention mechanism to recalibrate feature weights and an atrous spatial pyramid pooling (ASPP) module to extract multi-resolution spectral features. This architecture synergistically represents weak absorption peaks (400–1000 nm) and broad spectral bands (1000–2500 nm), overcoming single-scale modeling limitations. Validation on the LUCAS2009 dataset demonstrated that ReSE-AP Net outperformed conventional machine learning by improving the R² by 2.8–36.5% and reducing the RMSE by 14.2–69.2%. Compared with existing deep learning methods, it increased the R² by 0.4–25.5% for clay, silt, sand, organic carbon, calcium carbonate, and phosphorus predictions, and decreased the RMSE by 0.7–39.0%. Our contributions include statistical analysis of LUCAS2009 spectra, identification of conventional method limitations, development of the ReSE-AP Net model, ablation studies, and comprehensive comparisons with alternative approaches. Full article