Mapping of acid sulfate soils (ASS) has in the past focused on ASS probability maps, which are very useful to avoid environmental damage caused by these soils. However, these maps do not indicate the ASS subtypes, which may have different environmental impacts depending on whether they are actively releasing acidity and metals (sulfuric soils) or have the potential to do so (hypersulfidic soils) if the sulfidic material within them is disturbed (oxidized). Additionally, there is a particular type of soil that is close to being classified as an ASS, but where the pH criterion is not fulfilled. This soil is referred as para-ASS and may have a similar negative environmental impact as ASS. In the risk assessment of ASS, it is therefore crucial to know the location of ASS subtypes as well as para-ASS. In this study, we have created for the first time a multiclass map of ASS subtypes. Furthermore, four probability maps have been generated, one for each class. For this, the suitability of two machine learning methods for multiclass classification of different ASS subtypes has been evaluated. The methods are Random Forest (RF) and Gradient Boosting (GB), which showed very high capabilities for the classification of ASS in binary classification [1-3]. RF has given the best results with F1-score values between 71% and 80% for the four classes. An accurate and realistic multiclass map of the ASS subtypes has been created using the RF model.
[1] V. Estévez et al. 2022. “Machine learning techniques for acid sulfate soil mapping in southeastern Finland”. Geoderma 406 (2022) 115446.
[2] V. Estévez et al. 2023. “Improving prediction accuracy for acid sulfate soil mapping by means of variable selection”. Front. Environ. Sci. 11:1213069 (2023).
[3] V. Estévez et al. 2024. “Acid sulfate soil mapping in western Finland: How to work with imbalanced datasets and machine learning”. Geoderma 447 (2024) 116916.
How to cite: Estévez, V., Mattbäck, S., and Boman, A.: Mapping of acid sulfate soil types in Laihianjoki River catchment: A multiclass classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1150, https://doi.org/10.5194/egusphere-egu25-1150, 2025.
Understanding soil water management properties is crucial for agricultural, hydrological, and environmental modelling. To enhance the description of soil hydraulic processes, we developed national 3D soil hydraulic maps for Hungary at 100 m resolution, covering six soil layers down to 2 m depth (HU-SoilHydroGrids). This dataset includes continuous values of calculated soil hydraulic parameters, but aggregating this information is necessary to facilitate its use in national large-scale hydrological models with significant computational demands.
In Hungary, the methodology of the Várallyay soil water management categories map has been used for the hydrological classification of soils before the availability of HU-SoilHydroGrids. This nationwide map supports agricultural water management planning and includes nine soil water management categories and seventeen variants, established through expert rules based on field capacity, wilting point, available water content, infiltration rate, saturated hydraulic conductivity, and soil texture variations.
The newly available HU-SoilHydroGrids maps allow statistically based classification of soil hydraulic properties. In our study, we classified Hungarian soils using both national and international studies. Our methodology began with clustering via the k-means method on the HU-SoilHydroGrids database, considering eight soil hydraulic parameters across six soil depths, including van Genuchten parameters, water content at saturation, field capacity, wilting point, available water content, and hydraulic conductivity. This analysis identified twelve statistically distinct soil classes.
To ensure the inclusion of underrepresented soil groups with significant differences in water management, we refined these clusters with expert-based rules. Consequently, we further subdivided the twelve groups by soil profile depth, genetic soil type, electrical conductivity, and exchangeable sodium content. Combining statistical methods with expert-based rules, we established 68 categories. These soil hydrological groups provide a possible solution to aggregate the soil hydraulic data in environmental modelling applications.
The preparation of the HU-SoilHydroGrids dataset was carried out within the framework of the Széchenyi Plan Plus program with the support of the RRF 2.3.1 21 2022 00008 project. The derivation of the soil hydrological groups was funded by the Sustainable Development and Technologies National Programme of the Hungarian Academy of Sciences (FFT NP FTA). The statistical computations were performed in the HUN-REN Cloud (https://science-cloud.hu/) e-infrastructure.
How to cite: Szabó, B., Kolcsár, R., Mészáros, J., Laborczi, A., Takács, K., Szatmári, G., Makó, A., Bakacsi, Z., Rajkai, K., and Pásztor, L.: Aggregating 3D soil hydraulic properties for large-scale environmental modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4071, https://doi.org/10.5194/egusphere-egu25-4071, 2025.
Black soils play crucial roles in maintaining global environmental and social systems, contributing significantly to world food production and balancing carbon in the earth-atmosphere system. Monitoring productivity and land cover changes in relation to other environmental variables is essential for understanding global processes and implementing timely actions.
In our study, we analysed environmental changes of Eurasian black soils from 2001 to 2021 using time series remote sensing-based datasets. The Eurasian region is vast and exhibits highly diverse environmental conditions across its different areas; therefore, we conducted our analysis by dividing the region into distinct bioregions. Understanding the factors influencing Gross Primary Productivity (GPP) is crucial for evaluating ecosystem health and productivity under changing environmental conditions. This study investigates the relationship between GPP and various environmental variables across multiple regions, focusing on spatial and temporal dynamics. We examined the following key variables: Fraction of Photosynthetically Active Radiation (FAPAR), Solar Radiation, Soil Water Content, Temperature, Evaporation, Precipitation, and Vegetation Period.
Results show productivity increases in Chinese and Mongolian black soils, contrasting with significant decreases in large areas of Kazakh black soils. Notably, among countries with extensive black soil coverage, Russia and Ukraine exhibit areas with both declining and increasing productivity trends, reflecting the complex interplay of environmental and agricultural factors within these regions
Our findings indicate that climatic factors predominantly influence both negative and positive productivity trends, while cultivation technology levels also contribute significantly in specific regions. Climate change emerges as the primary driver of land cover change on black soils, with the net loss of croplands being the most alarming trend. This loss displays a scattered spatial pattern across Eurasia but is most prominent in the drying regions of Kazakhstan and Russia.
This research provides valuable insights into the dynamic nature of black soils and emphasizes their relevance to achieving the United Nations Sustainable Development Goals (SDGs). Ensuring sustainable management of black soils is crucial for addressing food security, mitigating climate change, and promoting sustainable land use practices in the face of ongoing environmental challenges.
How to cite: Csikós, N., Mészáros, J., Takács, K., Szabó, B., Hermann, T., Ivits, É., and Tóth, G.: Black Soils of Eurasia: two-decade environmental analysis (2001-2021), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6230, https://doi.org/10.5194/egusphere-egu25-6230, 2025.
Digital soil maps provide important information into the spatial distribution of soil properties, supporting land management decisions and Earth system modeling. This study introduces a novel methodology for creating 30-m digital soil maps across the conterminous United States (CONUS). This approach reduces uncertainties in estimating soil property distribution. The predicted soil properties include soil texture, bulk density, soil hydraulic properties (pedotransfer function-derived), pH, and organic matter - through six standard depth intervals from surface to 2-m depth. Comparative analysis demonstrates improved performance over existing soil products over CONUS.
Our methodology uses a two-step process. First, we have developed a pruned hierarchical Random Forest (pHRF) method to generate prior distributions of each soil property. Key highlights of the pHRF method include: (1) efficient selection of soil covariates, such as Sentinel 1 and 2 satellites and GOES land surface temperature; (2) implementation of a 'moving polygon' algorithm that preserves natural landscape boundaries; (3) incorporation of point-based soil measurements; and (4) development of a pruned hierarchical Random Forest algorithm that reduces uncertainties in estimating soil properties and addressing the inherent imbalance in soil survey data (uneven distribution of soil observation and underrepresented soil classes).
To further enhance the performance of the soil dataset, we also incorporate bias correction as a post-processing process. This process incorporates additional soil profile data to iteratively correct histograms of soil properties at each location. This process continues until the residual variations of soil properties between iterations fall below a predetermined threshold, indicating convergence. Our method leverages the most probable predicted soil property values to correct their distributions while accounting for spatial correlations between different soil property layers. This data-driven approach makes no assumptions about the underlying distribution of soil properties and relies on non-parametric statistical models. The resulting posterior distributions show reduced prediction uncertainties. It also demonstrates reproducibility in the final soil maps. These soil maps provide valuable input for various land management and modeling applications.
How to cite: Xu, C. (. and Chaney, N.: Development of a new 30-m soil properties map over the CONUS using pruned hierarchical Random Forests and iterative bias-correction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7117, https://doi.org/10.5194/egusphere-egu25-7117, 2025.
In countries of northern Europe, formerly covered by continental glaciers, soil textural maps are a fundamental source of information for authorities, researchers and non-governmental organizations in the land use sector, environmental conservation and land use planning. Yet, the small-scale mapping is still incomplete. Digital Soil Mapping (DSM) may provide ways to subjectively and efficiently generalize a low number of field observations into regional or country-wide soil textural maps by utilizing existing DEMs, remote sensing and airborne geophysical data.
We followed a commonly applied national soil textural classification, called RT (9 classes: 1 unsorted, 4 sorted, bedrock outcrops, stones and boulders, peat). Two mapping depths were selected: surface sediment (40‒90 cm) and base sediment (> 90 m). Three study areas (9 km to 18 km in width) covering wide geological variation were selected across the country based on the availability of Surficial deposits maps at 1:20/50 000 scale, and a 2-m-DEM and airborne geophysical datasets with 50‒75 m line spacing and four frequencies of electromagnetic data. The latter raster data were complemented by DEM derivatives, canopy height model (CHM), optical Sentinel-2 and ALOS PalSAR data resulting in a dataset of 253 explanatory features. Because the number field observations of soil texture were low for training and testing of machine learning models, the study areas were combined into one dataset (i.e. field data, surface sediment n=5133, base sediment n=4009). In addition, a complementary ‘pseudo’ reference dataset was extracted from the existing maps with GIS operations (i.e. map data, surface sediment n=9817, base sediment n=10120). We applied supervised classification with random forest (RF), and performed a priori feature selection with genetic algorithm and feature importance evaluation with permutation feature importance.
The overall classification accuracies for the surface sediment classification based on field data was 85.5%, and based on map data 78.1%. For the base sediment classification the respective overall accuracies were 74.6% and 75.6%. Class-specific accuracies were highest for the most common classes with abundant training data, while classes with fewer samples were poorly classified. For example, field data-based analysis revealed that bedrock outcrops and peat achieved the highest accuracies, while sandy till, fine sand, coarse silt, fine silt, and clay, represented by the lowest number of training data, consistently showed lower performance. The feature importance results indicate that DEM and its textural derivatives were the most significant features. However, optical and SAR satellite and radiometric airborne data would also be required for best separation of the soil textural classes if the classification was to be applied across wider areas.
Although these predictive mapping results indicate moderately successful surface sediment and base sediment textural classification, the small size of the study areas and consequent highly unbalanced training sets, poses a limitation for generalization of this study. However, it indicates a potential for successful surface sediment and base sediment textural classification with DSM if applied at regional scale or country-wide. Future studies should explore the applicability of supervised deep learning algorithms to avoid calculating a high number of textural derivatives of the DEM.
How to cite: Middleton, M., Hamedianfar, A., Pohjankukka, J., Väänänen, T., Lerssi, J., Laatikainen, M., Sallasmaa, O., Räisänen, J., Valkama, M., Pirttijärvi, M., Palmu, J.-P., and Kananoja, T.: Predictive mapping of soil textural classes – Digital soil mapping case study across formerly glaciated terrains, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8659, https://doi.org/10.5194/egusphere-egu25-8659, 2025.
Digital soil mapping (DSM) relies on a broad pool of statistical methods, yet determining the optimal method for a given context remains challenging. Large benchmarking studies are needed to reveal strengths and limitations of commonly used methods. Existing DSM benchmarking studies usually rely on a single dataset with restricted access, leading to incomplete and potentially biased conclusions. To address these issues, we introduce an open-access dataset collection called Precision Liming Soil Datasets (LimeSoDa). LimeSoDa consists of 31 field- and farm-scale datasets. Each dataset has three target soil properties: soil organic matter (SOM) or -carbon (SOC), clay and pH, alongside a set of features. Features are dataset-specific and were derived from spectroscopy, proximal soil sensors and remote sensing. All datasets were processed into a tabular format and are “ready-to-go” for modeling. We demonstrated the use of LimeSoDa for benchmarking by comparing four learning algorithms: multiple linear regression (MLR), support vector regression (SVR), categorical boosting (CatBoost) and random forest (RF) on their predictive power across all datasets of LimeSoDa. The results showed that no learning algorithm was generally superior. MLR and SVR proved to be better for high-dimensional spectral datasets due to better compatibility with principal components. In contrast, CatBoost and RF had considerably stronger performances for all other datasets. These benchmarking results illustrate that the performance of a method can be very context-dependent. Therefore, LimeSoDa provides a crucial data resource for improving the development and evaluation of machine learning methods in DSM and pedoemtrics.
How to cite: Schmidinger, J., Vogel, S., and Atzmueller, M.: LimeSoDa: A Dataset Collection for Benchmarking of Machine Learning Regressors in Digital Soil Mapping , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9905, https://doi.org/10.5194/egusphere-egu25-9905, 2025.
Smart Farming (SF) practices are essential for reducing agricultural impacts on ecosystems while maintaining food security. However, the implementation of SF is often hindered by the lack of high-resolution soil property data. This study addresses this challenge by developing a cloud-based approach to predict soil texture (clay, silt, and sand) using a random forest machine learning model within Google Earth Engine (GEE) at three spatial scales: farm, regional, and national.
The analysis was conducted at four farm sites in the German states of Brandenburg and Mecklenburg-Vorpommern (with 355, 321, 392, and 151 topsoil samples), across the eastern part of Brandenburg at the regional scale (1,080 samples), and nationwide across Germany (2,199 samples). Soil data were sourced from smart farming projects and the LUCAS soil database. The datasets were split into 70% for model training and 30% for validation.
The input earth observation (EO) data included optical and radar remote sensing information from Sentinel-1 (S1) and Sentinel-2 (S2) satellites. Vegetation indices, soil indices, and bare soil pixels were calculated from S2 data, while S1 provided radar backscatter values (VV and VH polarizations). Temporal patterns were captured through statistical metrics such as mean, standard deviation, and coefficient of variation. Finally, the 71 raster datasets at farm scale and 55 raster datasets at regional and national scale were extracted at soil sampling locations and used as covariates in the random forest models. Model performance was evaluated using root mean square error (RMSE). At the farm scale, RMSE values ranged from 4.1% to 5.8% (R2 0.36 to 0.76) for clay, 5.3% to 8.7% (R2 0.35 to 0.51) for silt and 8.9% to 10.9% (R2 0.4 to 0.72) for sand. At the regional scale, RMSE values were 6.4% (R2 0.58) for clay, 6.5% (R2 0.4) for silt, and 10.7% (R2 0.46) for sand. At the national scale, clay predictions remained consistent with an RMSE of 6.9% (R2 0.49), while RMSE values for silt and sand increased to 11.1% (R2 0.51) and 14.8% (R2 0.56), respectively.
Key predictors across scales were S2 bands 11 and 12 (under bare soil conditions), S1 VV and VH backscatter, the VV-VH ratio, and elevation data from the Copernicus Digital Elevation Model. The influence of EO data was highest at farm and regional scales but diminished at the national level.
The developed models, implemented in GEE, can predict topsoil texture (0–30 cm depth) at a resolution of 10 m × 10 m for any arable field in Germany. This approach could help to increase the availability of high-resolution soil data for smart farming applications.
How to cite: Meyer, S. and Marzahn, P.: Cloud-Based Prediction of Soil Properties Using Remote Sensing Data and Machine Learning for Smart Farming Applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13053, https://doi.org/10.5194/egusphere-egu25-13053, 2025.
The Namibian Soil Profile Database contains 4960 entries, all samples with geographic coordinates. Each soil property presents a different number of observations, which decreases with depth. To perform the Digital Soil Map of Soil Organic Carbon (SOC) up to 30 cm, 1298 sample points were used. The covariates used in the model were composed from land cover, geology, terrain characteristics extracted from the digital elevation model and remote sensing data. Most of the covariates carry 30 m of spatial resolution. The Random Forest model implemented in Google Earth Engine (GEE) was applied with an external validation split of 80/20 %. As the second validation layer, samples from the Namibian tier of the Soils4Africa project were applied. The use of GEE facilitated the generation of a SOC distribution map for Namibia at a spatial resolution of 30 × 30 m. The highest amounts of SOC are stored in the central region of Namibia, with SOC values ranging from 0.1 to 1.9 %. The map is suitable for national and regional decision-making, offering a baseline for determining SOC stocks, monitoring changes in SOC, assessing the effects of bush encroachment/thickening and bush control, and for agriculture implications. Mapping of soil properties in other depths are scheduled for the near future.
How to cite: Gelsleichter, Y. A., Coetzee, M., Csorba, Á., and Micheli, E.: Digital Soil Mapping of Soil Organic Carbon in Namibia Using Google Earth Engine, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14774, https://doi.org/10.5194/egusphere-egu25-14774, 2025.
Soil maps describe spatial variability by using traditional or predictive soil mapping techniques. Conventional soil maps group soils based on their similar cartographic properties, as on the legendary soil surveys, while digital soil mapping predicts the values of various soil properties through available soil point datasets and geostatistics or other pedometrical techniques. It is expected that both types of soil mapping contain some degree of uncertainty either due to the subjectivity of conventional mapping, which requires a vast amount of pedological knowledge in the field, or due to insufficient number of soil samples and mathematical errors that are underestimated, in geostatistical and pedometrical methods.
Digital maps of top-soil properties provide global and unified coverage without gaps, especially at broad regional scales like countries or continents, which is essential for understanding large-scale processes and cross-border issues. Accurate soil datasets are critical for understanding and managing Earth's vital resources. For instance, in hydrology, these digital maps improve the accuracy of models predicting water runoff, infiltration, and groundwater recharge, while for agriculture, detailed soil information enables precision farming practices, optimizing irrigation, fertilizer application, and crop selection for increased yields and reduced environmental impact. These maps also support broader applications like climate modeling and disaster response.
This study attempts to investigate the representativeness of European-scale soil maps in relation to official national soil data and to outline the conditions for the development of detailed scale soil data that will improve the European soil cartography. Specifically, the study deals with the comparison of six pan-European soil datasets in raster format for four selected soil properties, namely those of top-soil texture, soil organic carbon, pH and CEC with point data coming from detailed soil surveys that were not used for their construction. The gridded datasets are coming from the European Soil Data Centre (ESDAC) while the detailed laboratory data are coming from the soil map of Greece repository covering the agricultural areas of Greece. The European scale soil digital maps were compared with the soil point data (augers and profiles) of the soil map of Greece initially by spatially overlaying the data and extracting the paired values (raster and point) for each soil attribute followed by comparison of the abovementioned soil attributes by using several geospatial, statistical and geostatistical techniques.
The initial results provided a mixed picture with differences between the datasets greatly varying spatially and with differences to be more profound in areas with distinct soil characteristics (e.g. fine soil types). This study highlights the importance of incorporating detailed national soil data to improve the accuracy and reliability of continental-scale digital soil maps, particularly in regions with heterogeneous soil properties. Findings from this research contribute to the development of more robust and reliable global soil datasets by demonstrating the value of multi-source data integration and providing specific recommendations for future mapping initiatives.
How to cite: Palli Gravani, S., Gerontidis, S., Kopanelis, D., Kairis, O., Soulis, K., and Kalivas, D.: Evaluation of digital maps of top-soil properties compared to large-scale laboratory soil data and synergies towards a better European soils’ delineation. , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15613, https://doi.org/10.5194/egusphere-egu25-15613, 2025.
Arsenic contamination in topsoil, primarily geogenic in origin, presents significant public health risks due to its potential accumulation in agricultural products. This study focuses on predicting geogenic arsenic concentrations across European topsoil using machine learning. The analysis integrates geochemical data from the GEMAS (Geochemical Mapping of Agricultural and Grazing Land Soil in Europe) database and 15 environmental variables (climate, geological, soil, and hydrological factors) to create a map predicting the spatial arsenic occurrence at a resolution of 1 km. A threshold of 20 mg/kg was selected based on general European guidelines and its relevance to potential phytotoxicity risks.
A Random Forest (RF) algorithm is developed and applied to model the probability of arsenic exceeding the widely recognized soil guideline value of 20 mg/kg, used in many European countries. To ensure robustness, 100 iterations are performed. Model efficiency is improved through Recursive Feature Elimination (RFE), which reduces the number of predictors from 35 to 15 features. Performance is assessed using metrics including Area Under the Curve (AUC), sensitivity, and specificity. SHapley Additive exPlanations (SHAP) analysis identifies key predictors, including distance to mineral deposits, latitude, and hydrological conditions. The model preliminarily reveals that 9.2% of European grasslands and 3% of croplands, particularly in France and Spain, exceed 20 mg/kg. In areas with elevated arsenic levels, more than 5% of each crop category, including wheat, maize, rapeseed, and fodder crops, is cultivated in potentially hazardous agricultural regions.
The study highlights the important environmental variables for mapping arsenic hotspots and emphasizes the need for regional assessments to better understand arsenic hazards. While it provides an overview of arsenic occurrence in soil across Europe, local geological variability and anthropogenic impacts require further investigation. Further efforts should aim to develop models at regional to national scales to enhance arsenic risk assessments for food safety and public health. This research will strengthen intervention effectiveness and improve the prediction and management of trace element presence in soil across broader regions.
How to cite: Li, K.-Y., Covatti, G., Podgorski, J., and Berg, M.: Spatial Prediction and Assessment of Environmental Drivers of Geogenic Arsenic in European Topsoil: A Machine Learning Approach to Food Safety Risks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16141, https://doi.org/10.5194/egusphere-egu25-16141, 2025.
Monitoring and evaluating ecosystems and their interactions are essential for the effective management of both natural and agricultural landscapes. Soil, a crucial component of terrestrial ecosystems, is globally acknowledged for its complexity and diversity. It serves as a vital medium where key processes such as nutrient cycling, water regulation, and carbon storage take place. Despite its fundamental role in sustaining biodiversity and delivering critical ecosystem services, soil often receives limited attention in ecological research and environmental policies. Soil formation is a complex process shaped by environmental factors such as climate, organisms, parent material, morphology, and time. The interplay of these factors, along with inherent properties, results in a wide variety of soils across different scales. Geographic Information Systems (GIS) tools can help analyze these interactions and “unpack the mosaic” that defines a given landscape, thereby aiding soil surveys and data collection in complex environments like mountainous regions. Mountainous areas cover more than a third of Italy’s land area and provide numerous invaluable ecosystem services, from water regulation and carbon storage to recreation, timber, and high-quality food production. They also support diverse habitats and preserve Italy’s historical and cultural heritage. However, climate change, land use changes, and hydrogeological instability present significant threats to mountain ecosystems and the adjacent hills and lowlands, with frequent landslides, floods, and wildfires damaging forests, crops, and communities. Despite their importance, mountain soils in Italy remain poorly understood and largely neglected in environmental policies. With this study, we present a digital soil mapping approach to optimize soil survey campaigns in complex environments, and a Random Forest to transition from the sampled points to the final map. After sampling and soil classification is completed, a set of remotely sensed and topographic covariates related to soil forming factors is selected. A subset of covariates is chosen by recursively eliminating the least performing layers, and is then used to perform an iterative Random Forest which yields the final map. This method not only delivers accurate results for this study area, but it also provides important information regarding the intensity with which soil forming factors affect it. It can also be used to plan new sampling campaigns in unsurveyed areas, making it a powerful tool in the whole process of soil mapping in mountainous environments, which is fundamental to provide useful information and directions for the management of these vulnerable but critically important lands.
How to cite: Trenti, W., Boettinger, J., De Feudis, M., Vianello, G., and Vittori Antisari, L.: Mapping soil diversity in mountain areas in a temperate climate: methods to analyse a complex environment. A case study in the Northern Apennines (Italy), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16820, https://doi.org/10.5194/egusphere-egu25-16820, 2025.
As global food demand increases, farming systems experience heightened pressure to enhance productivity on limited arable land. In Africa, including Morocco, smallholder farms are particularly susceptible to climate variability, soil degradation, and suboptimal farming practices, resulting in yield gaps—the disparity between actual and potential yields under optimal conditions. In Morocco, yield variability is significantly influenced by soil fertility, irrigation, and climate. Consequently, quantitative assessment and mapping of key soil fertility indicators at the field scale are essential for improving yields. Remote sensing data, particularly hyperspectral imagery, presents a cost-effective and time-efficient alternative to traditional soil mapping methods. However, its potential for providing detailed local-scale soil information in Africa remains underexplored. This study utilizes high-resolution PRISMA (PRecursore IperSpettrale della Missione Applicativa) hyperspectral imagery and laboratory-analyzed soil samples to map four key soil properties—cation exchange capacity (CEC), soil organic matter (SOM), available phosphorus (P₂O₅), and exchangeable potassium (K₂O)—in the Ain el Orma agricultural area on the Saïss plateau, Morocco. Despite the advantages of hyperspectral sensors, their high processing complexity, due to redundant or correlated spectral bands, can impede machine learning model accuracy. This study compares the performance of traditional and advanced machine learning algorithms combined with dimensionality reduction techniques—PCA, UMAP, and RFE. Six well-established algorithms (XGBoost, Gradient Boosting, PLSR, SVR, and Random Forest) were evaluated as an initial step in the artificial intelligence workflow, yielding weak to moderate results. For SOM (%), the utilization of RFE resulted in the optimal performance with a substantial improvement in R² from 0.30 (PCA) to 0.36, while the Root Mean Squared Error (RMSE) decreased from 0.52 to 0.39%. Furthermore, the Ratio of Performance to Interquartile Range (RPIQ) for SOM (%) also increased from 1.58 (PCA) to 2.10. In the case of P₂O₅ (mg/kg), PCA emerged as the superior method, yielding an R² of 0.38 compared to 0.37 for RFE and -0.01 for UMAP. The RMSE decreased from 11.92 (RFE) to 11.82. For K₂O (mg/kg), PCA again proved to be the optimal method, with an R² improving to 0.13 from -0.29 with RFE and remaining superior to UMAP's 0.19. The RMSE decreased from 107.33 (RFE) to 88.51%, and the RPIQ increased from 1.50 to 1.82. Lastly, for CEC (meq/100g), PCA delivered the most accurate predictions, improving the R² to 0.68 from 0.60 (RFE) and 0.21 (UMAP). The RMSE was reduced significantly from 2.08 (RFE) to 1.88%, while the RPIQ increased from 2.47 to 2.73. These initial findings underscore the importance of feature selection and dimensionality reduction for developing robust models for soil property estimation using hyperspectral data.
Additionally, this study aims to propose advanced innovative AI models capable of enhancing the accuracy of soil maps. In conclusion, the anticipated results are expected to support the creation of accurate soil maps, necessary for spatialized analysis of wheat yield variability using hyperspectral remote sensing imagery, thus contributing to food security and sustainable agricultural practices.
How to cite: Elbouanani, N., Laamrani, A., El-Battay, A., Hajji, H., Bourriz, M., Bourzeix, F., Ait Abdelali, H., Amazirh, A., and Chehbouni, A.: Enhancing Soil Fertility Mapping with Hyperspectral Remote Sensing and Advanced AI: A Comparative Study of Dimensionality Reduction Techniques in Morocco, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18418, https://doi.org/10.5194/egusphere-egu25-18418, 2025.
Soils are the largest terrestrial carbon store on the planet. However, it is estimated that soils have lost 8% of their carbon content since human farming began. Around the world, agricultural soils are highly degraded and are a significant source of greenhouse gas emissions.
Soil carbon content is a good proxy for soil health. Healthy soils are more resilient to climate shocks, achieve high yield to input ratios, produce nutrient rich food, as well as making the soil a net carbon sink. From a climate change perspective as well as a food security perspective, there is growing interest in rebuilding soil health, requiring effective measurement and monitoring.
Traditionally, measuring soil carbon relies on in-situ sampling. This is expensive and labour intensive and so encourages low density sampling and infrequent repeat measurements. However, through a remote digital soil mapping (DSM) approach combining globally available sampling, environmental data - aligned with SCORPAN and remote sensing data we have developed a methodology which enables monitoring soil health in response to interventions and practices for example, biomineralization applications, or cropping rotations.
Using the Downforce digital twin approach, which has been calibrated globally, and validated within the case study areas, we will present a set of case studies in which we remotely monitor variable biomineralization applications over more than 10,000ha, under varying farming practices (livestock and cropping) in Victoria, Australia. The case studies provide evidence that the applied biofertilizer, which is designed to catalyse non-labile minerals and nutrients present in the soil into labile forms, not only improve soil health but also increase nutrient density in crops and livestock. The potential of this DSM approach to provide near-real time insights at hyperlocal to landscape scales is unique and is enabling transformative and adaptive management in conservation and agriculture, at a fraction of the cost of in-situ sampling.
How to cite: Knox, N., McGlade, J., Lakey, C., Kruse, K., and Sharp, N.: Remote monitoring of Soil Health: Insights into the impacts of biomineralization applications to build soil health, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19331, https://doi.org/10.5194/egusphere-egu25-19331, 2025.
Rising temperatures and drier conditions due to climate change will force Alpine forests towards their ecological limits. For an informed decision on climate-smart tree species composition, we need area-wide information about the current and future moisture regime in forest areas. For this task, soil-vegetation-atmosphere transport (SVAT) models in combination with digital mapping techniques have proven useful tools. However, depending on the selected mapping algorithm (e.g. random forest - RF, generalized additive models - GAM, neural networks - NN) and the selected train-/test-split of the input data, resulting maps can vary considerably. The method of splitting the dataset into training and test subsets can significantly impact model performance and spatial predictions, particularly when imbalanced data is present. Such imbalances in specific training/test splits can lead to inconsistencies in model development and their resulting spatial predictions. For generating maps that represent the moisture regime under current and future climate conditions, spatial consistency and reproducibility are crucial. Models must produce robust spatial patterns that are not merely artifacts of a single training/test split but reflect reliable and consistent predictions.
First, we use a lumped, physically-based SVAT model (LWF-Brook90) for reproducing the moisture regime at 2009 mapped forest sites in Tyrol and Vorarlberg (Austria). We parameterized the model with the individual soil characteristics at the sites, while considering a generic beech forest stand for sake of comparability between the sites. Based on interpolated meteorological observations and bias-corrected climate projections, we derived components of the water balance under current (1991-2020) and future conditions (2036-2065, 2071-2100) on a daily resolution. As an indicator for the moisture regime, the transpiration deficit (Tdef; i.e. the difference between potential and actual transpiration) was identified.
Using digital soil mapping techniques, we generate maps of the mean annual Tdef sum for the selected periods, incorporating geomorphometric and climate-related covariates. Feature selection is conducted using RF (based on feature importance) across multiple training/test splits, selecting the most commonly chosen features to build RF and NN models. GAM, by contrast, employs a smaller, expert-based set of covariates for improved interpretability. To ensure robustness, multiple runs are performed for each algorithm. Forming ensemble means prevents random biases from imbalanced training data, while deviation maps help identifying uncertainties in the mapping process.
Statistical metrics (e.g., R², RMSE) on an independent validation set reveal greater variation within a single algorithm than between different algorithms, complicating the identification of a "better" approach. To address this, we propose a weighted ensemble approach that accounts for performance on independent validation data, enabling reliable and spatially consistent outcomes. The resulting maps will aid in identifying suitable tree species under future climate conditions at the slope scale.
This work was carried out within the WINALP21 project, funded by the INTERREG VI-A program (grant number BA0100020).
How to cite: Huber, T., Zieher, T., Simon, A., Gadermaier, J., and Klebinder, K.: Derivation of robust moisture regime indicator maps in Alpine forests considering climate change – balancing uncertainty by multi-algorithm ensembles, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19804, https://doi.org/10.5194/egusphere-egu25-19804, 2025.
Accurate spatial soil information at the field scale is critical for sustainable land management and environmental modeling. This research investigates methods for mapping plant-available phosphorus by integrating sensor technologies, targeted sampling strategies, and geostatistical approaches.
Field-scale soil mapping in Denmark typically employs a uniform 100-meter sampling grid with interpolation to estimate soil properties. This study explores the potential of targeted sampling informed by proximal and remote sensing technologies, terrain variables, and existing national-scale soil maps. The sensor technologies evaluated include electromagnetic induction (EMI), gamma-ray sensors, and aerial imagery. Although these sensors are widely applied to assess soil texture and organic carbon content, their application in phosphorus mapping is relatively novel. The study relies on data from seven fields located in Weichselian morainic landscapes in Denmark. The fields covered 4 – 37 ha each and mainly comprised loam and sandy loam soils.
Targeted sampling strategies were designed using k-means clustering. We used measurements of Olsen P as a proxy for plant-available phosphorus, which was then mapped using Gaussian Process Regression. The performance of sensor-informed approaches was compared to methods based on spatial coverage sampling and interpolation. Each method was tested with different numbers of soil samples used for calibration.
At the field level, Olsen P was found to have a moderate to strong positive correlation with organic matter. The values were generally higher in topographic depressions and areas with darker soils in the aerial images. Variogram analyses indicated that phosphorus measurements exhibit spatial autocorrelation with effective ranges of 23 to 167 meters in different fields, highlighting opportunities to optimize sampling strategies based on site-specific spatial variability.
Mapping accuracy improved with increased sampling density in both sensor-based and conventional approaches; however, sensor-derived covariates provided significant accuracy gains. The sensor-based methods generally achieved accuracies that were unattainable by conventional approaches, irrespective of the sampling density. The sensor-based methods also stayed effective with low sampling densities (less than 0.5 samples ha-1), which was not the case without sensors.
This study highlights the potential of combining spatial geostatistics with sensor-based approaches to improve phosphorus mapping. The results demonstrate that such integration can reduce sampling density requirements while enhancing phosphorus mapping precision, offering a cost-effective and scalable solution for field-scale nutrient management.
How to cite: Møller, A. B., Greve, M. H., Pedersen, I. F., Knudsen, L., and Lemming, C.: Mapping plant-available phosphorus at the field scale using targeted sampling, sensors, and geostatistics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19828, https://doi.org/10.5194/egusphere-egu25-19828, 2025.
Agricultural Land Classification (ALC) for England provides an assessment of the quality of agricultural land by determining physical and chemical limitations to agricultural production. It distinguishes six grades of land based on climate, topography and soil characteristics of a site. The original "Provisional" ALC map was created during the 1960s-70s before the National Soil Map was created, plus new guidelines were introduced in 1988. A new "Predictive" ALC map has now been produced following the 1988 guidelines and drawing on not only the National Soil Map at 1:250,000 scale but also more recent detailed soil mapping that covers some 25% of the country. Whilst this new Predictive ALC map benefits from geospatial representations of ALC within a GIS environment, the map is produced solely by combination of reclassified parameters, it has boundary artefacts and missing nuances in soil information and climate that can affect the decision making about ALC at a local level. Digital Soil Mapping, through its capability to leverage machine learning methods, can capture the intricate nonlinear relationships between ALC and input climate, soil, and topographic variables, thereby enhancing the existing methods for directly predicting soil function for agricultural production through ALC. In this study, an alternative ALC map for England was produced by leveraging the detailed soil maps and boosted classification trees. Nine climate features, four soil features, and 17 topographic features were used as predictors and stratified random sampling technique was used to extract the training data from detailed soil maps. We achieved 76% accuracy on training and 74% on validation and testing data and applied the model to generate ALC for the whole country. Comparison to the predictive ALC map revealed some grade changes and improved continuity of ALC grades in some areas and an estimate of the uncertainty that was not available to users of the provisional map. Spatially explicit assessment of uncertainty allows for the efficient allocation of resources for additional soil surveys needed to improve portions of the predictive ALC map, which is one of the primary advantages of using digital soil mapping approaches in soil assessment.
How to cite: Keay, C., Mukherjee, K., Zawadzka, J., and Hannam, J.: Digital soil assessment – enhancing the provisional Agricultural Land Classification Map for England , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19972, https://doi.org/10.5194/egusphere-egu25-19972, 2025.
The constant and detailed monitoring of soil properties is crucial for having an up-to-date status of the health of our soils. This requires sufficient sampling points to meaningfully and accurately represent the soils of a whole country. Topsoil datasets can be very different regarding point density, spatial distribution and representativity. Soil sampling is also very cost- and labour-intensive, which is why combining existing national and international datasets is an efficient way to create larger datasets for the creation of accurate soil property maps. In the case of Hungary, these datasets were the Hungarian subset of the topsoil dataset of the Land Use/Cover Area frame Survey (LUCAS) and the Hungarian Soil Monitoring and Information System (SIMS). The purpose of this study is to investigate whether combining harmonized soil data from different soil monitoring systems improves the quality and accuracy of the predicted soil property maps.
The physical soil properties (sand-, silt-, and clay content) were harmonized by converting the SIMS dataset to a uniform 0-20 cm depth using mass preserving splines and matching the particle size limit of the LUCAS dataset (FAO/WRB) to the SIMS dataset (USDA). After the harmonization the two datasets were merged together and Additive Log Ratio transformation was used to assure that the particle fractions add up to 100%. This resulted in y1 and y2 values which were used in Random Forest Kriging to create the predicted maps. These maps were converted back to sand-, silt-, and clay content maps. The same procedure was applied to the LUCAS and SIMS datasets resulting in their respective sand-, silt-, and clay-content maps. The particle maps of the combined dataset were compared directly to the SIMS and LUCAS particle maps using linear regression. The quality of the predicted maps were measured and compared. Soil texture maps were created from the particle fractions using the USDA soil texture triangle. The soil texture map of the combined dataset was directly compared to the LUCAS and SIMS soil texture maps using the taxonomic distance between the predicted values of the map pairs. The result of the study show, that the quality and accuracy of the combined datasets’ predicted soil property maps were only slightly better than the maps predicted by LUCAS and slightly worse than the maps predicted by the SIMS dataset. This lead us to conclude that merging datasets alone won’t improve the quality of the soil property maps and that different approaches are required.
How to cite: Benő, A., Szatmári, G., Laborczi, A., Kocsis, M., Bakacsi, Z., and Pásztor, L.: Using a combined topsoil dataset from two soil monitoring systems to create predicted soil-physical property maps and comparing them with the predicted maps of the original datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19986, https://doi.org/10.5194/egusphere-egu25-19986, 2025.
Accurate soil mapping is crucial for agriculture, land, ecosystem and environmental management. Digital Soil Mapping (DSM) is one of the most conventional and widely used methods for mapping soil. This study introduces a novel strategy for DSM by incorporating the neighborhood effect of environmental covariates (ECs), aiming to enhance mapping accuracy of soil properties. The research focused on modeling organic carbon, cation exchange capacity, bulk density, and pH in southern Canada using 18 ECs derived from the Soil Landscapes of Canada dataset and satellite imagery. Two strategies were compared: a conventional approach using standard ECs, and a proposed method incorporating neighboring ECs through Inverse Distance Weighting. Both strategies employed Gaussian Process Regression for modeling. Results demonstrated significant improvements in accuracy using the proposed strategy. Mean absolute errors were reduced by 32%, 36%, 28%, and 14% for organic carbon, cation exchange capacity, bulk density, and pH, respectively. The proposed method also decreased the coverage of high-error areas and improved R² values across all soil properties. Moreover, mean uncertainty in soil property modeling decreased by 3.4% to 3.9% using the proposed strategy. This study highlights the importance of considering spatial context in DSM through neighborhood effects. The proposed strategy offers a more nuanced and accurate approach to soil property modeling, with potential applications extending beyond soil science to other environmental mapping domains. These improvements in soil mapping accuracy have significant implications for sustainable land management, precision agriculture, and environmental conservation.
How to cite: Biswas, A. and Fathololoumi, S.: A new digital soil mapping approach based on the adjacency effect, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20163, https://doi.org/10.5194/egusphere-egu25-20163, 2025.
Determining soil parameters is essential for rational soil use, sustainable soil management, cost-effective monitoring, and collecting baseline data for targeted soil mapping.
The aim of our research is to perform non-destructive spectroradiometric measurements on the archived soil sample bank of the HUN-REN ATK Institute for Soil Sciences, which includes comprehensive laboratory background data. This initiative seeks to develop a nationwide soil spectrum library that spatially represents the entirety of Hungary’s land cover and soil types, encompassing thousands of data points. This digital database facilitates the identification of correlations between traditionally measured soil properties and spectral characteristics. The ultimate objective is to enable the cost-effective and rapid estimation of certain soil parameters—such as soil organic matter (SOM) content, CaCO3, and pH—that are otherwise difficult, time-consuming, or expensive to measure.
The spectral database is built on two key pillars. The first comprises 5,500 soil samples collected from agricultural lands in 2011–2012 as part of the Hungarian Soil Degradation Observation System (HSDS). The second consists of 2,000 soil samples gathered from tree plantations and control areas (including pastures, fallow lands, and agricultural plots) across the Great Hungarian Plain between 2012–2014. This approach has enabled the successful inclusion of a wide range of land cover types in Hungary, spanning multiple soil layers.
Spectral measurements were performed using an ASD Field Spec 4 spectroradiometer, focusing on the visible–near-infrared region of the electromagnetic spectrum. Reflectance values were measured across a wavelength range of 350 to 2500 nm, covering 2,151 spectral bands. The recorded reflectance values underwent consistent pre-processing, which included steps such as conversion to absorbance, splice correction, noise reduction, and smoothing. Further, additional data scenarios were generated by applying advanced processing techniques, including standard normal variate (SNV) transformation, detrending, and first- and second-order derivatives.
The relationships between soil properties and soil spectra were analyzed using various machine learning techniques—such as Generalized Linear Models (GLM), Distributed Random Forest (DRF), Gradient Boosting Machine (GBM), and Deep Learning Neural Networks (DLN)—implemented in an R programming environment using the 'h2o' package.
Initial results based on the HSDS database show that SOM and CaCO3 contents are best estimated using the DRF model with absorbance and first-derivative spectra (R² = 0.705, RMSE = 0.528 for SOM, and R² = 0.632, RMSE = 5.756 for CaCO3). For soil pH estimation, the DLN model achieved an R² of 0.677 and RMSE of 0.483 when using absorbance, second-derivative spectra, and SNV transformation.
Regarding the forestry soil database, preliminary results are presented in this poster, as investigations are still ongoing. These efforts are being supplemented with XRF data, which are expected to enhance estimation accuracy when combined with spectral data.
How to cite: Balog, K., Mészáros, J., Kovács, Z. A., Vass-Meyndt, S., Koós, S., Pirkó, B., Szabó, A., Tóth, T., Gribovszki, Z., Laborczi, A., Bakacsi, Z., László, P., and Pásztor, L.: Development of a Nationwide Soil Spectrum Library and Digital Soil Assessment Based on Archived Soil Samples Using Pedometrics and Spectral Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9429, https://doi.org/10.5194/egusphere-egu25-9429, 2025.
Soil texture plays a fundamental role in influencing water retention, nutrient dynamics, erosion susceptibility, and carbon sequestration, making it essential for sustainable agricultural practices. Accurate monitoring and mapping of soil texture components, such as clay, silt, and sand, are crucial for effective soil and water management. This study explores the potential of combining radionuclide monitoring data and Gamma-Ray Spectrometry (GRS) with quantitative modelling techniques for soil texture estimation, focusing on transferring a predictive model developed in one location to another.
The research builds on work conducted in 2023 at the Hydrological Open-Air Laboratory (HOAL) in Petzenkirchen, Lower Austria, to assess the transferability of a predictive model for soil texture to the experimental farm of the University of Natural Resources and Life Sciences (BOKU) in Raasdorf, near Vienna, Austria. Soil sampling campaigns at Petzenkirchen (2023) and Raasdorf (2024) provided input data for the model. Soil texture was analyzed using the PARIO system, which applies the Integral Suspension Pressure (ISP) method, based on Stokes' law, to determine particle size distributions. Portable gamma-ray spectrometry (GRS) was used to measure activity concentrations of radionuclides (40K, 232Th, and 238U) at multiple locations in each field, serving as predictors for soil texture components through a Python-based statistical model initially developed in Petzenkirchen.
The integration of GRS data with quantitative modelling revealed critical relationships between radionuclide concentrations and soil texture components. Moderate positive correlations of 232Th (0.59) and 238U (0.72) with silt, and moderate negative correlations with clay (-0.62 and -0.74), indicate that radionuclides preferentially associate with silt-sized particles due to their larger surface area and mineralogical properties. Additionally, a strong inverse relationship between clay and silt (-0.92) reflects their complementary distribution within the soil matrix. Strong correlations were observed between 238U and both silt (R² = 0.8, p = 4.1 × 10⁻⁵) and clay (R² = 0.78, p = 5.6 × 10⁻⁵) demonstrating its predictive potential. These strong associations formed the basis for selecting 238U as a key predictor for soil texture estimation in Raasdorf.
The predictive models from Petzenkirchen were applied to estimate silt and clay content in Raasdorf using 238U as a predictor. The model performed well for silt, achieving a mean error of 10% and an RMSE of 0.07 g, indicating strong agreement between observed and predicted values. However, predictions for clay exhibited greater variability, with a mean error of 25% and an RMSE of 0.28 g. This discrepancy highlights the need for localized calibration to address site-specific differences in soil mineralogy and radionuclide binding affinities between the two fields.
This research demonstrates how integrating radionuclide monitoring with quantitative soil texture modeling provides a scalable and cost-effective approach for digital soil mapping in agricultural landscapes. Future work will refine the model by leveraging advanced GRS data analysis, such as radionuclide ratios (e.g., 238U/232Th) and spatial variability, to improve clay predictions and assess uncertainties in soil property estimations. These efforts aim to enhance the applicability of digital soil mapping for precision agriculture and sustainable land management.
How to cite: Cheng, Y., Toloza, A., Mbaye, M., Mitchell, J., Said Ahmed, H., Trust, B., Dercon, G., Neugschwandtner, R., and Kemetter, J.: Advancing Soil Texture Estimation Across Agricultural Soil Types in Austria Using Portable Gamma-Ray Spectrometry, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10312, https://doi.org/10.5194/egusphere-egu25-10312, 2025.
The introduction of the Directive on Soil Monitoring and Resilience proposed by the European Parliament and Council is supposed to be preceded by specific preparatory works at Member State level, such as the definition of so-called soil districts together with the development of a soil monitoring system based on the elaborated zonalization. Three subsequent terms of Presidency of the Council of the European Union (Belgian, Hungarian, and Polish) aimed to finalize the concept elaboration and to legislate the Directive, so far without success. As a consequence, final delineation of soil districts could not been elaborated so far. Nevertheless, certain tests were carried out to establish a proper zonalization.
The first drafts of the text of the Directive introduced a set of criteria that seems relatively simple in the legislative formulation, however, their implementation by Member States poses several number of methodological challenges. In the present paper soil health is approached from soil degradation point of view and soil districts from the regionalization of soil degradation respectively, which latter has already been addressed from time to time in the last decades.
In the frame of Land Degradation Mapping Sub-project of PHARE MERA ’92 -, identification, delineation and description of Hungary’s major land degradation regions at 1:500,000 scale were accomplished by building and analyzing a digital land degradation geographic database in the late ‘90s. The applied GIS analysis techniques were mainly based on traditional cartographic methods and had not exploited the opportunities, which were later emerged in DSM.
The former initiative of the Commission of the European Communities by the Thematic Strategy for Soil Protection proposed a comprehensive approach to soil protection with ample freedom on how to implement its requirements on the identification of threats and specific risk areas left to Member States. In 2007, the techniques available at that time provided by DSM together with the renewed interest in spatial delineation of areas endangered by various soil threats were combined for the recompilation of land degradation regions of Hungary. Different levels of specific threats were determined in the form of categories. For the overall characterization of degradation regions, indices were introduced serving as spatial land degradation indicators.
In the last decade the Hungarian soil spatial infrastructure (HSSI) has been renewed, GSM conform digital soil maps on primary together with certain secondary, derived soil properties were elaborated in the frame of DOSoReMI@hu. The work has been continued with the modelling of certain soil functions and (degradation) processes. For the support of Soil District designation all, nationally relevant soil degradation processes have been digitally (re)mapped using specific DSM approaches based on HSSI and relevant spatial environmental ancillary data. The newly (re)complied soil degradation maps have then been submitted to spatial classification procedures to regionalize the processes. The results of the various classification scenarios have been used to produce alternatives for soil districts.
How to cite: Pásztor, L., Takács, K., Szatmári, G., Csikós, N., Laborczi, A., Benő, A., Koós, S., Farkas-Iványi, K., and Bakacsi, Z.: Regionalization of soil degradation for the support of soil district designation in Hungary, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12044, https://doi.org/10.5194/egusphere-egu25-12044, 2025.
The application of Vis-NIR spectroscopy for physico-chemical soil properties estimations, like soil organic carbon, and digital soil mapping for the scopes of enhancing precision agriculture, promote soil carbon sequestration and improve soil health is fastly developing thanks to the use of machine learning algorithms and big data handling.
With our Subterra Green device, developed by S4 Mobile laboratories, a mobile field unit equipped with a visible and near infrared (VNIR) spectrometer and a load cell for measuring probe insertion force, we are able to collect spectroscopic data until 90 cm underground, down to a 1 cm resolution.
As part of the EU founded PHENET project, among many others, one specific scope is to conduct soil surveys among various soil types, including highly fertile chernozems, to less productive gleyic or cambisols. Samples collection for training and testing of machine learning based models, takes place from the humid continental zones of Austria to the temperate oceanic climate of Portugal. Ground-truthing data is verified with laboratory biochemical analysis of the selected soil samples. The ultimate goal would be to estimate important soil parameters in-situ and provide digital soil maps on larger scales (several hectares), providing this with the highest accuracy possible by using pre-processing techniques such as external parameter orthogonalization or direct standardization to correct detrimental effects caused by varying water content, bulk density, soil texture etc.
Developing precise machine learning based models using Vis-NIR spectroscopy and subsequently generating high-resolution digital soil maps leads us to fast, non-destructive and cost-effective monitoring of soil physico-chemical properties over space and time.
How to cite: Csibi, A.-E., Sanden, H., Baykalov, P., Pereira, R., Cachada, A., Rewald, B., and Perry, D.: Developing machine learning based models for soil parameters prediction and mapping using Vis-NIR spectroscopic data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12259, https://doi.org/10.5194/egusphere-egu25-12259, 2025.
Forest soils play a critical role in the global carbon cycle and in mitigating greenhouse gas emissions. However, the storage of soil organic carbon (SOC) in European forests and its future dynamics remain unclear. This study applied a digital soil mapping (DSM) approach based on LUCAS soil datasets, using machine learning models to predict SOC’s current and future spatial distribution in European forests. The analysis incorporated sixteen key environmental variables derived from multi-source remote sensing and ecological data, identified through the Boruta feature selection method, highlighting soil properties, climate, and vegetation as the dominant factors shaping SOC distribution. The results revealed that SOC stocks are currently concentrated in northern Europe, the UK, and the Alps, with temperate forests exhibiting a notable increase in SOC stocks as forest age increases. Under the high-emission pathway (SSP585), SOC storage was projected to increase by 2100, particularly in young forests. This study provides a baseline assessment of SOC storage in European forests and insights into its future trends, offering valuable guidance for forest management and carbon sequestration strategies.
How to cite: Ji, X. and Purushothaman Vellayani, A.: Mapping soil organic carbon in European forests under future climate scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13467, https://doi.org/10.5194/egusphere-egu25-13467, 2025.
Land degradation have become critical environmental issues, leading to reduced soil and water quality and reduced yields in arable lands. Soil aggregate stability (AS) refers to the ability of soil aggregates to resist disintegration or breakdown and thus presents a measure to assess the soils susceptibility to applied forces. Despite its significance, for example in soil erosion processes, spatial data on AS is scarce and only few studies on the spatial behaviour of AS exist, frequently due to the high monetary, work and time expense needed to gain data.
Machine learning approaches are increasingly used due to their high accuracy when incorporating various co-variables and are already achieving promising results in the field of AS. However, it is often unclear how well these models perform in different environmental conditions and landscapes, particularly outside their training sites.
The aim of this study is to identify and compare the best-performing variables for the soils in two study sites in northern Germany with different environmental conditions and to evaluate if and how well a model, trained on one study site would perform on another. To accomplish this, a total of 500 topsoil samples (250 each) were collected from the two study sites. They were analysed for soil properties, including AS, soil texture, SOC, pH, electrical conductivity. Additionally, a range of topographic indices and additional data (e.g. Crop, geology) were analysed.
The preliminary results show, that SOC, topographic wetness index (TWI), slope and the fine sand fraction were deemed the best performing variables in the random forest model in study site A. The model achieved an r2 of 0.575 and RMSE 7.992. Analysis of the terrain indices in study site B show channel network base level, aspect and analytical hillshading as the best performing terrain indices. A performance gap of the model would indicate limited transferability, as the model may have overfitted to site A's specific landscape and conditions. Additional models will be tested to determine which ones transfer more effectively between different sites.
How to cite: Feldmann, D., Saggau, P., Duttmann, R., and Kuhwald, M.: Optimizing Soil Aggregate Stability Predictions with Machine Learning: A Comparative Analysis of Input Variables and spatial transferability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15662, https://doi.org/10.5194/egusphere-egu25-15662, 2025.
Soil particle size fractions (PSF) directly affect the movement and retention of water, nutrients, and air, which are critical for providing optimal conditions for crop growth. This study applies three log-ratio transformation techniques for compositional data: additive log-ratio (ALR), centered log-ratio (CLR), and isometric log-ratio (ILR). Digital soil mapping (DSM) combined environmental covariates (satellite images, terrain features, and climatic data) and machine learning (ML) models (Cubist and random forest (RF)) to create a predictive map for soil particle size distribution across Taiwan. Model accuracy was evaluated using R², root mean squared error (RMSE), Aitchison’s distance (AD), and the right ratio of the predicted soil texture types (RR). The analysis revealed that CLR transformation combined with RF (RF_CLR) had the best performance, with the highest R² values (sand: 0.59, silt: 0.29, clay: 0.51), as well as the lowest RMSE (sand: 16.51%, silt: 10.65%, clay: 8.45%) and AD (0.08). The accuracy of RR (45%) was consistent across different log-ratio transformation methods. The various sampling sizes influence the computational efficiency of the data. Therefore, different sampling sizes were tested for the best-performing combination (RF_ CLR). When the sampling size was less than 50% of the original sampling size (N = 22,000), the prediction performance of PSF showed a significant decline. Our findings can serve as a valuable reference for soil management and crop cultivation planning in Taiwan.
How to cite: Syu, C.-H. and Lo, Y.-C.: Digital mapping of soil particle size fractions across Taiwan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14247, https://doi.org/10.5194/egusphere-egu25-14247, 2025.
Spatially detailed quantitative data regarding soil physical/hydraulic properties is in high demand for a range of modeling applications. EU-SoilHydroGrids has demonstrated its utility at the European level. HU-SoilHydroGrids, has been developed for the whole area of Hungary at 100 m spatial resolution with several enhancements in its elaboration process. A further step toward larger spatial resolution is based on NATASA (Hungarian acronym for Profile-level Database of Hungarian Large-Scale Soil Mapping) initiative to produce large-scale 3D Soil Hydraulic Databases (LS-HU-SoilHydroGrids).
Digitial processing of the soil observation records of the still available soil observation legacy data originating from large-scale surveys carried out in Hungary between the 60s and 90s was firstly finalized for the watershed of the Lake Balaton in order to support hydrological modelling studies on the catchment. The digitized soil observations are firstly used in digital mapping of primary soil properties at a scale of 25 meters, which DSM products then will be similarly adapted as the 100 m resolution DOSoReMI.hu products for the derivation of soil hydraulic property predictions down to 2 meters for six standard GSM soil depth layers, thus providing the “Balaton watershed LS-HU-SoilHydroGrids”.
Prior to step forward particle size fractions (i.e., sand, silt, and clay contents) were targeted to be mapped in a case study based since NATASA includes information on soil taxonomy and basic soil chemical and physical properties, but no direct information on sand, silt and clay content, only an indirect parameter, namely, the upper limit of soil plasticity. Since particle size distribution is not only crucial for assessing soil degradation, hydrology and fertility, but also a basic information to model the planned hydraulic properties, we developed pedotransfer functions (PTFs) to compute the particle size distribution from the soil properties available in the NATASA dataset (1,372 soil profiles). The PTFs were trained and tested on the Hungarian Detailed Soil Hydrophysical Database (3,970 soil profiles) using random forest method. For the prediction model, i) additive log-ratio transformed clay, silt and sand content were used as the dependent variables, and ii) the upper limit of soil plasticity, soil type, calcium carbonate content, organic matter content and pH were included as independent variables. The results indicate that the R² values of the PTFs are 0.69 for clay, 0.58 for silt, and 0.74 for sand content. Since the NATASA database contains soil information from different depths, we splined the data into six standard depth layers (0–5, 5–15, 15–30, 30–60, 60–100 and 100–200 cm depths). The spatial modelling was performed by random forest kriging (RFK) using environmental auxiliary variables. The R2 values of the RFK models range from 0.19 to 0.67 for clay content, from 0.49 to 0.62 for silt content and from 0.69 to 0.74 for sand content.
Acknowledgement: This work has been carried out within the framework of the Széchenyi Plan Plus program with the support of the RRF 2.3.1 21 2022 00008 project and the Sustainable Development and Technologies National Programme of the Hungarian Academy of Sciences.
How to cite: Kocsis, M., Kassai, P., Szatmári, G., Makó, A., Mészáros, J., Laborczi, A., Magyar, Z., Takács, K., Szabó, B., and Pásztor, L.: Application of machine learning-based pedotransfer functions to produce large-scale maps of particle size fractions using big legacy data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20245, https://doi.org/10.5194/egusphere-egu25-20245, 2025.
The understanding of soil characteristics in alpine regions is crucial for the comprehension of infiltration or runoff dynamics. However, the estimation of soil properties in these areas poses significant challenges, which can be attributed to the complexity of the terrain, the variability of microclimates, or the limited accessibility. The existing database on soil properties in these regions is currently insufficient, so there is an urgent need for alternative approaches to reliably predict the basic soil properties.
We employed Bayesian Regression Models (BRMS) to predict basic soil properties, including soil texture and organic carbon content. This method combines environmental covariates derived from remote sensing data and digital elevation models (DEMs) with prior knowledge about the various alpine soil types and their associated properties in order to enhance the accuracy of predictions in these heterogeneous landscapes. This approach accounts for spatial variability and uncertainty, producing robust estimations of key soil properties, even with limited field observations. The results demonstrate significant spatial variability in soil properties, influenced by factors such as altitude, slope, and vegetation cover.
This study combines traditional statistical approaches with domain expertise, thereby facilitating enhanced soil property estimation in challenging environments. The methodology provides a machine learning framework for similar applications in other remote or heterogeneous regions with limited data. It contributes to global initiatives focused on the comprehensive assessment of soil quality and the implementation of environmentally land management practices.
How to cite: Darmann, F., Kumpan, M., Schwaighofer, I., Strauss, P., and Weninger, T.: Estimation of basic soil properties in alpine areas using a Bayesian Regression Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21923, https://doi.org/10.5194/egusphere-egu25-21923, 2025.
The ability of soil to store a large amount of organic carbon (SOC) is one of its most important characteristics, making it an active and indispensable participant in the global carbon cycle. SOC influences various soil related functions and services, such as agricultural productivity, water retention and management, buffering capacity against toxic elements and compounds, which are essential to provide healthy food and clean drinking water. Furthermore, SOC is widely recognized as playing a crucial role in mitigating and addressing various environmental crises and challenges, such as climate change, land degradation, declining biodiversity, water and food security. Consequently, not only soil scientists but also researchers from other disciplines, practitioners, stakeholders, and even policymakers have shown growing interest in information on the spatial and temporal variability of SOC at various scales.
In the past few years, significant efforts have been made in Hungary to predict the spatial, and more recently, the spatiotemporal variability of SOC using various digital soil mapping techniques. Recently, a space-time model of SOC was developed using a combination of machine learning and space-time geostatistics to predict SOC change at point support and various aggregation levels (i.e., 1 × 1 km, 5 × 5 km, 10 × 10 km, 25 × 25 km, counties, and the entire country) for Hungary (Szatmári et al., 2024). This work is based on soil data derived from the Hungarian Soil Information and Monitoring System between 1992 and 2016, as well as spatially and temporally exhaustive environmental covariates. Notably, geostatistics plays a central role by accounting for the spatiotemporal correlation of errors, which is essential for reliably quantifying the uncertainty associated with the aggregated SOC change predictions. The performance of the developed model was assessed using five times repeated 10-fold cross-validation, yielding acceptable results. A series of SOC maps were compiled for the period between 1992 and 2016 for each support, along with the quantified uncertainty, representing a significant advancement in Hungary. Furthermore, the presented methodology can overcome the limitations of recent approaches in spatiotemporal SOC modelling, allowing the prediction of SOC and SOC change, with quantified uncertainty, for any year, time period and spatial scale. This capability addresses current and anticipated demands for dynamic SOC information at both national and international levels.
The aim of this presentation is to outline the methodology developed, to highlight some methodological challenges, to present the resulting maps, and finally, but importantly, to discuss these findings in a broader context.
Acknowledgements: This research was funded by the National Research, Development and Innovation Office (NKFIH; grant number: FK-146391) and the János Bolyai Research Scholarship of the Hungarian Academy of Sciences.
References:
Szatmári, G., Pásztor, L., Takács, K., Mészáros, J., Benő, A., Laborczi, A., 2024: Space-time modelling of soil organic carbon stock change at multiple scales: Case study from Hungary. Geoderma 451, 117067.
How to cite: Szatmári, G., Laborczi, A., Takács, K., Mészáros, J., Benő, A., Koós, S., Bakacsi, Z., and Pásztor, L.: Recent results in spatiotemporal modelling of soil organic carbon changes in Hungary, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6260, https://doi.org/10.5194/egusphere-egu25-6260, 2025.
Digital Soil mapping (DSM) at continental and global scale provides standardised global information layers. It is also an important tool to create soil information layers for areas for which local soil survey information is lacking. The recent availability of global and continental remote sensing derived products coupled with the ease-of-access to computational resources has made the production of such layers easier across the globe. Therefore, it is ever more important to assess the quality of DSM-derived products, in particular the type of information they can actually provide to users (i.e., fitness for intended use).
DSM studies commonly assess prediction uncertainty using various approaches, including multiple simulations or quantile random forests. However, this does not encompass all the potential elements that could be used to characterise the uncertainty of a DSM product. In this study we are going to assess maps based also on area of applicability (i.e., the area in covariate space where the model learns about relationships based on the training data) and the landscape heterogeneity both in the landscape itself and in covariate space.
We present examples of continental and global mapping products, highlighting main uncertainty-related issues and how these influence suitability for intended use by stakeholders, decision makers and users in general at the given resolution. The examples come from a range of projects with different aims and goals. The results permit some practical reflections on how to integrate all the above elements to identify regions where the confidence in the predictions is highest and the associated uncertainty lowest. We will integrate the practical reflections with information collected from a user survey on requirements and usability of continental and global DSM products.
How to cite: Poggio, L., Batjes, N., Kempen, B., Genova, G., and Rossiter, D.: DSM for global and continental applications: model applicability, spatial uncertainty and maps assessment , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11053, https://doi.org/10.5194/egusphere-egu25-11053, 2025.
