Displays

The degradation of European soils is a cause for concern. Examples are the reduction of carbon content and soil fertility. The European Commission therefore recommends further research on how to better monitor soils and their changes over time and space. Digital soil mapping (DSM) is already an established method for the use of hyperspectral information from soil samples for quantifying soil properties under laboratory conditions based on soil spectral libraries. At the remote sensing level, imaging spectroscopy has already achieved good results for the prediction of soil properties on a local scale. Major advantages of this method are that topsoil maps can be updated more frequently, spatially more accurately and with less costs.

In this study we bring together pedometric and remote sensing approaches to achieve the development of soil spectral models applicable to upcoming global hyperspectral imagery, combining DSM methods and data with Earth Observation expertise. In a first step at laboratory level, we used the EU-wide topsoil database LUCAS. We investigated whether using solely spectral data (without any covariates) and selected classification algorithms combined with PLSR could allow and improve the quantification of soil organic carbon (SOC) content. The best results were obtained for the local PLSR approach with RMSE=5.16 g kg-1, RPD=1.74 and R²=0.67. In addition, the local PLSR approach was tested with LUCAS spectral data resampled to EnMAP satellite spectral resolution, resulting in a very similar SOC prediction model accuracy.

In the next step, the local PLSR approach was applied to airborne HySpex image data and simulated satellite EnMAP data from a test area in north-eastern Germany where local soil data are available for model validation. This area is associated with one LUCAS point. A direct application of the laboratory-based SOC model to the spectra of the airborne image was not possible due to higher variability in the image data caused by different environmental conditions (solar illumination, mixed soil-vegetation pixels, surface state -roughness, wetness-) and sensor performances different from the laboratory data resulting in an overall lower signal-to-noise ratio in the airborne image. Therefore, after reducing the effect of soil moisture, green vegetation cover, residues coverage, we used a two-step approach where (i) wet chemistry SOC analyses for a set of soil samples from the test area were replaced by the local PLSR approach using the LUCAS database. Then (ii) an airborne model was calibrated using the SOC content from (i) and the corresponding image spectra to calibrate an airborne PLSR. Preliminary results show a good airborne model accuracy for HySpex imagery with RMSE=3.33 g kg-1, RPD=1.59, R²=0.63 and slightly lower but still good accuracy for simulated EnMAP imagery with RMSE=3.72 g kg-1, RPD=1.45, R²=0.55. Both models were then applied to the images to produce SOC maps for bare soils and validated with existing data and previous SOC mapping works in the area based on local datasets. This approach demonstrates the possibility to replace wet chemistry by the local PLSR approach based on large scale soil spectral libraries for SOC mapping.

How to cite: Ward, K. J., Brell, M., Spengler, D., Castaldi, F., Neumann, C., Segl, K., Foerster, S., and Chabrillat, S.: Mapping soil organic carbon based on simulated EnMAP images and the LUCAS soil spectral library, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-3013, https://doi.org/10.5194/egusphere-egu2020-3013, 2020.

Any strategy to change Carbon (C) pool would have a substantial effect on functionality of numerous ecosystem functions, detachment of Soil Organic Carbon (SOC), atmospheric carbon dioxide (CO₂) concentration, and climate change mitigation. As the largest amount of the world’s C is stored in forests soils, the importance of forest SOC management is highlighted. Total SOC in forest varies not only laterally but also vertically with depth; however, the SOC storage of lower soil horizons have not been investigated enough despite their potential to frame our understanding of soil functioning. Visible–Near Infrared (vis–NIR) reflectance spectroscopy enables rapid examinations of the horizontal distribution of forest SOC, overcoming limitations of traditional soil assessment. This study aims to evaluate the potential of vis–NIR spectroscopy for characterizing the SOC contents of organic and mineral horizons in forests. We investigated 1080 forested sites across the Czech Republic at five individual soil layers, representing the Litter (L), Fragmented (F), and Humus (H) organic horizons, and the A₁ (depth of 2–10 cm) and A₂ (depth of 10–40 cm) mineral horizons (total 5400 samples). We then used Support Vector Machine (SVM) to model the SOC contents of (i) the profile (all organic and mineral horizons together), (ii) the combined organic horizons, (iii) the combined mineral horizons, and (iv) each individual horizon separately. The models were validated using 10-repeated 10-fold cross validation. Results showed that there was at least more than seven times as much SOC in the combined organic horizons compared to the combined mineral horizons with more variation in deeper layers. All individual horizons’ SOC was successfully predicted with low error and R² values higher than 0.63; however, the prediction accuracy of F and A₁ was greater compared to others (R² > 0.70 and very low-biased spatial estimates). We have shown that modelling of SOC with vis–NIR spectra in different soil horizons of highly heterogeneous forests of the Czech Republic is practical.

How to cite: Gholizadeh, A., Viscarra Rossel, R., Saberioon, M., Boruvka, L., and Pavlu, L.: Characterizing Soil Organic Carbon Content in Forests at National Scale using Reflectance Spectroscopy, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-2589, https://doi.org/10.5194/egusphere-egu2020-2589, 2020.

The rapid growth in the global population over the past few decades has resulted in the transformation of many natural ecosystems into human-dominated ones. Land-use (LU) dynamics are accompanied by an increase in resource exploitation, often causing deteriorated environmental conditions that are reflected in the soil quality. Soil quality differences between LUs can be observed and measured using near-infrared reflectance spectroscopy (NIRS) methods. The research goal was to apply, measure, and evaluate soil properties based solely on the spectral differences between both natural and human-dominated LU practices, in the dryland environment of the central Negev Desert, Israel. This goal was achieved through the development and implementation of chemometrics techniques that were generated from soil point spectroscopy. Soil quality index (SQI) values, based on 14 physical, biological, and chemical soil properties, were quantified and compared between LUs and geographical units across the study area. Laboratory spectral measurements of soil samples were applied. Significant differences in SQI values were found between the geographical units. The statistical and mathematical methods for evaluating the soil properties’ spectral differences included principal component analysis (PCA), partial least squares-regression (PLS-R), and partial least squares-discriminant analysis (PLS-DA). Correlations between predicted spectral values and measured soil properties and SQI were calculated using PLS-R and evaluated by the coefficient of determination (R²), the Root Mean Square Error of Calibration, and Cross-Validation (RMSEC and RMSECV), and the ratio of performance to deviation (RPD). The PLS-R managed to produce “excellent” and “good” prediction values for some of the soil properties, including EC, Cl, Na, Ca + Mg, SAR, NO₃, P, and SOM. Results of the PLS-R model for SQI are R² = 0.90, RPD = 2.46, RMSEC = 0.034, and RMSECV = 0.057. The PLS-DA classification of the laboratory spectroscopy was applied and resulted in high accuracy and kappa coefficient values when comparing LUs. In contrast, comparing the sampling sites resulted in lower overall accuracy (Acc = 0.82) and kappa values (K_c = 0.80). It is concluded that differentiation between physical, biological, and chemical soil properties, based on their spectral differences, is the key feature in the successful results for recognizing and characterizing various soil processes in an integrative approach.  The results prove that soil quality and most soil properties can be successfully monitored and evaluated using NIRS in a comprehensive, non-destructive, time- and cost-efficient method.

How to cite: Levi, N., Karnieli, A., and Paz-Kagan, T.: Using reflectance spectroscopy for detecting land-use effects on soil quality in drylands, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8092, https://doi.org/10.5194/egusphere-egu2020-8092, 2020.

There is a high demand for information about soil conditions in forests stands as it is crucial to ensure sustainable management, to maintain ecosystem services, to preserve timber production and establish proper pest management. Nowadays, the main drivers for changes in soil conditions are element input, forest conversion, subsoil liming and changing climate. These drivers influence nutrients and water availability and are challenging current site mapping methods. However, for impact assessment high-resolution and up-to-date information is needed. As laboratory analysis is time consuming and expensive, alternative approaches are preferred.

The project DIGI-Humus uses methods of reflectance spectroscopy in the visual and near-infrared-region of the electromagnetic spectrum for indirect measurement and prediction of physical and chemical soil properties in forest stands. For this purpose, spectral data were collected under laboratory conditions to build a database of forest soils. We used retained samples from Saxony soil survey, measuring both Oh and Ah horizons. To ensure data quality, we developed our own protocol based on literature review and self-conducted test measurements. The data has been used to successfully calibrate regression models based on different forest types and soil horizons to predict the soil parameters C and N content, C/N ratio and pH-value.

To improve model performance and test its generalization capability, the created library has been extended with new samples from a field campaign conducted in 2019 at an additional local test site. Using this data, the impact of adding new information to the modelling process and the robustness of the models could be evaluated.

The results of this research will be used to assess forest sites regarding nutrients availability, as basis for the development of site specific management strategies and to enhance and improve current methods of periodic site mapping of forest stands.

How to cite: Thomas, F., Petzold, R., Becker, C., and Werban, U.: Usage of visual and near-infrared spectroscopy to predict soil properties in forest stands, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-9107, https://doi.org/10.5194/egusphere-egu2020-9107, 2020.

Soil is an essential component in the environment and is vital for food security. It provides ecosystem services, filters water, supplies nutrients to plants, provides us with food, stores carbon, regulates greenhouse gases emissions and it affects our climate. Traditional soil survey methodologies are complicated, expensive, and time-consuming. Visible and infrared spectroscopy can effectively characterize soil properties. Spectral measurements are rapid, precise and inexpensive. The spectra contain information about soil properties, which comprises minerals, organic compounds, and water. Today, several Soil Spectral Libraries (SSLs) are being created worldwide because these datasets have a notable potential to be used as training datasets for machine learning methods that will benefit precision agriculture activity for better management of food production. Nonetheless, as SSL's are created under laboratory conditions it is not clear if it can be used to infer field conditions in situ and/or from the sky. Thus, study the relationship between RS, field spectroscopy and the laboratory measurements of soil is very important. Accordingly, this study postulates that traditional SSLs don't simulate the real spectral signatures in the field that both, satellite and airborne sensors measure as well, because they are affected by factors that are not an integral part of the soil, such as: moisture, litter, human and animal activity, plow, grass, dung, waste, etc… However, under laboratory conditions, these factors are usually removed for the preparation of SSLs. Thus, given the several SSLs available, it is necessary to evaluate the protocols that were used in these SSLs. The objective of this study is to evaluate the gap between field and laboratory spectral measurements through the analysis of the performance of spectral based models. This procedure combined two soil spectral libraries that contain 114 samples that were measured in the laboratory as well as in the field. The nature of the dataset is varied, because these samples were collected from six different fields in three countries of the Mediterranean basin: Israel, Greece and Italy. Moreover, 63 samples are mainly sandy and 51 are mainly clayey. In order to obtain optimal spectral measurements in the field, we used a new optical apparatus that simulates the sun's radiation. Next, we generated PLSR models to estimate one of the most important hydrological parameters namely “infiltration rate” that control the runoff stage, soil erosion and water storage in the soil profile. This property is strongly affected by the surface characteristics. Finally, the field based spectral model was adapted to an UAV hyperspectral sensor in order to estimate the infiltration rate from the sky. The results were successfully validated in field, and we concluded that for the estimation of the infiltration rate, SSLs must be created using surface reflectance in field because laboratory protocols can be detrimental for the performance of the dataset in question.

How to cite: Francos, N., Ben Dor, E., Romano, N., Nasta, P., Szabó, B., Mészáros, J., Maltese, A., Manfreda, S., Garcia, M., and Zeng, Y.: Soil Surface Reflectance as a Tool to Estimate Water Infiltration Rate from UAV Platforms, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-2092, https://doi.org/10.5194/egusphere-egu2020-2092, 2020.

On the 22 March 2019 the Italian Space Agency (ASI) launched the PRISMA satellite, having onboard a hyperspectral imager covering the 400-2500 nm range with 234 spectral bands and about 10 nm of bandwidth. The ground spatial resolution is 30 m, plus a panchromatic camera with 5 m spatial resolution. One of the potential application areas of this scientific mission is for precision agriculture applications, among which the mapping of field-scale variability of topsoil properties is of particular interest.

PRISMA clear-sky hyperspectral images were acquired in autumn and spring 2019 on the Maccarese farm in Central Italy, in the framework of the PRISCAV project, which is aimed at a first assessment of the PRISMA data. An intensive soil sampling campaign was performed, using a ground sampling scheme adapted to PRISMA and Sentinel-2 spatial resolutions, in the fields where bare soil was exposed at the satellite acquisition dates. Soil texture (clay, silt, sand), carbonates, pH and soil organic carbon (SOC) for the collected soil samples were then determined in the laboratory.

The dataset was then used to test calibration and validation of PLSR (Partial Least Squares Regression) and RF (Random Forest) models developed using PRISMA surface reflectance data. To this aim, several pre-treatment tests were performed, including pan-sharpening at 5 m using PRISMA panchromatic data as well as Sentinel-2 multispectral data.

The results show that the good results could be obtained in particular for clay estimation. The best-performing algorithm for topsoil properties retrieval using PRISMA hyperspectral data was RF algorithm as compared with PLSR.

How to cite: Casa, R., Pignatti, S., Pascucci, S., Ionca, V., Mzid, N., and Veretelnikova, I.: Assessment of PRISMA imaging spectrometer data for the estimation of topsoil properties of agronomic interest at the field scale, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-6728, https://doi.org/10.5194/egusphere-egu2020-6728, 2020.

Spatially resolved estimates of change in soil organic carbon (SOC) stocks are necessary for supporting national and international policies aimed at achieving land degradation neutrality and climate mitigation through better land management. In this work we report on the development, implementation and application of a data-driven, statistical space-time method for mapping SOC stocks, using Argentina as a pilot area. We used the Quantile Regression Forest machine-learning algorithm to predict SOC stock at 0-30 cm depth at 250 m resolution for Argentina between 1982 and 2017, on an annual basis. The model was calibrated using over 5,000 SOC stock values from the 36-year time period and 35 environmental covariates. Most covariates were static and could only explain the spatial SOC distribution. SOC change over time was modelled using time series maps of the AVHRR NDVI vegetation index. These NDVI time series maps were pre-processed using a temporal low-pass filter to allow the SOC stock for a given year to depend on the NDVI of the current as well as preceding years. Spatial patterns of SOC stock predictions were persistent over time and comparable to baseline SOC stock maps of Argentina. Predictions had modest temporal variation with an average decrease for the entire country from 2.55 kg C m^‑2 to 2.48 kg C m^‑2 over the 36-year period (equivalent to a decline of 210.7 Gg C, 3.0% of the total 0‑30 cm SOC stock in Argentina). The Pampa region had a larger estimated SOC stock decrease from 4.62 kg C m^‑2 to 4.34 kg C m^‑2 (5.9%) during the same period. For the 2001-2015 period, predicted temporal variation was 7-fold larger than that obtained using the Tier 1 approach of the Intergovernmental Panel on Climate Change and the United Nations Convention to Combat Desertification. Prediction uncertainties turned out to be substantial, mainly due to the limited number and poor spatial and temporal distribution of the calibration data, and the limited explanatory power of the covariates. Cross-validation confirmed that SOC stock prediction accuracy was limited, with a Mean Error of 0.03 kg C m^-2 and a Root Mean Squared Error of 2.04 kg C m^-2. The model explained 45% of the SOC stock variation. In spite of the large uncertainties, this work showed that machine learning methods can be used for space-time SOC mapping and may yield valuable information to land managers and policy makers, provided that SOC observation density in space and time is sufficiently large.

How to cite: Heuvelink, G., Angelini, M., Poggio, L., Bai, Z., Batjes, N., van den Bosch, R., Bossio, D., Estella, S., Lehmann, J., Olmedo, G., and Sanderman, J.: Space-time machine learning for modelling soil organic carbon change, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-3621, https://doi.org/10.5194/egusphere-egu2020-3621, 2020.

In order to assess the carbon and water storage capacity of agricultural soils at national scale (Germany), spatially continuous, high-resolution soil information on the particle size distribution is an essential requirement. Machine learning models are good at computing complex, composite non-linear functions. They can be trained on point data to relate soil properties (response variable) to approximations of soil forming factors (predictors). Finally, the obtained models can be used for spatial soil property predictions.

We developed models for topsoil texture regionalization using two powerful algorithms: the boosted regression trees machine learning algorithm, and the differential evolution algorithm applied for parameter tuning. Texture data (clay, silt, sand) originated from two sources: (1) the new soil database of the German Agricultural Soil Inventory (BZE), and (2) the well-known, publicly available database of the European Land Use / Cover Land Survey (LUCAS). BZE texture data results from an eight-kilometer sampling raster (2991 sampling points). LUCAS data from soils under agricultural use (Germany) comprises 1377 sampling points. The predictor datasets included DEM-based topography variables, information on the geographic position, and legacy maps of soil systematic units. In a first step, a nested five-fold cross-validation approach was used to tune and train models on the BZE data. In a second step, the amount of training data was increased by adding two-thirds of the LUCAS data. Model performance was evaluated by (1) cross-validation (R_CV²), and (2) by using the remaining LUCAS data as an independent external test set (R_external²).

Models trained on the BZE data were able to predict the nation-wide spatial distribution of clay, silt and sand (R_CV² = 0.57 – 0.76; R_external² = 0.68 – 0.83). Model performance was further enhanced by adding the LUCAS data to the training dataset.

How to cite: Gebauer, A., Sakhaee, A., Don, A., and Ließ, M.: Applying machine learning and differential evolution optimization for soil texture predictions at national scale (Germany), EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11797, https://doi.org/10.5194/egusphere-egu2020-11797, 2020.

Phosphorus (P) is one of the most important plant nutrients, and farmers regularly apply P as mineral fertilizer and with animal manures. Typically, reactions with amorphous aluminum and iron oxides or carbonates retain P in the soil. However, if P additions exceed the soil’s ability to bind them, P may leach from soil to surface waters, where it causes eutrophication. The phosphorus sorption capacity (PSC) is thus an inherent soil property that, when related to bound P, can describe the P saturation of the soil. Detailed knowledge of the spatial distribution of the PSC is therefore important information for assessing the risk of P leaching from agricultural land.

In weakly acidic soils predominant in Denmark, the PSC depends mainly on the oxalate-extractable contents of aluminum and iron. In this study, we aimed to map PSC in four depth intervals (0 – 25; 25 – 50; 50 – 75; 75 – 100 cm) for Denmark using measurements of oxalate-extractable aluminum and iron from 1,623 locations.

We mapped both elements using quantile regression forests. Predictions of oxalate-extractable aluminum had a weighted RMSE of 13.9 mmol kg^-1. For oxalate-extractable iron, weighted RMSE was 33.5 mmol kg^-1.

We included depth as a covariate and therefore trained one model for each element. For each element in each depth interval, we predicted the mean prediction value as well as 100 quantiles ranging from 0.5% to 99.5% in 1% intervals. The maps had a 30.4 m resolution. We then calculated PSC by convoluting the prediction quantiles of the two elements, using every combination of quantiles, in order to obtain the prediction uncertainty for PSC.

Oxalate-extractable aluminum was roughly normal distributed, while oxalate-extractable iron had a large positive skew. The age and origin of the parent material had a large effect on oxalate-extractable aluminum, and soil-forming processes such as weathering and podzolization had clear effects on the distribution in depth. Meanwhile, organic matter, texture and wetland processes were the main factors affecting oxalate-extractable iron, so much so that they obscured any trends with depth.

The weighted RMSE of the predicted PSC was 19.1 mmol kg^-1. PSC was highest in wetland areas and lowest in young upland deposits, such as aeolian deposits and the loamy Weichselian moraines of eastern Denmark. The sandy glaciofluvial plains and Saalian moraines of western Denmark had intermediate PSC. In most cases, PSC was highest in the top soil, but in the sandy soils of western Denmark, PSC was highest in the depth interval 25 – 50 cm due to podzolization.

How to cite: Møller, A. B., Hechrath, G. J., Hermansen, C., Nørgaard, T., Knadel, M., de Jonge, L. W., and Greve, M. H.: Mapping soil phosphorus sorption capacity in four depths with uncertainty propagation, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-19356, https://doi.org/10.5194/egusphere-egu2020-19356, 2020.

Intra-field heterogeneity of soil properties is function of complex interaction between biological, physical factors and historic agricultural management. Quantifying the spatial patterns of soil properties such as soil organic carbon (SOC), nitrogen (N), phosphorous, exchangeable cations, pH, soil texture will contribute to an optimization of fertilizer application and crop yields. We tested the capacity of the multispectral Micasense rededge camera mounted on a UAV in order to map the development of winter wheat and related the Red-Edge NDVI (RENDVI) from the sensor to the Plant Area Index (PAI) measured in the field. The geo-referenced grain yield of the winter wheat was measured by a combine harvester and the top soil characteristics analysed by a grid based sampling. The spatial patterns in RENDVI at three phenological stages were mapped together with the yields. For each of these images conditional inference trees were used to derive the soil properties that significantly influenced these spatial patterns. Within-field variation in PAI (cv 41 % in March, 27% in April and 9 % in May) and yield (cv 4%) can be observed. The spatial patterns of RENDVI are rather constant and their correlation with yields is highest in March and April (r=0.64). Soil properties explain between 67 to 79 % of the variance in vegetation index throughout the growing season as well as 67 % of the variance in yield.  Legacy effects of land consolidation two years earlier reflected in the field lay-out, pH and exchangeable K are significant factors explaining around 12-18 % of the variance in crop yield each. The SOC contents were overall low (8-15 g kg-1). Hence, the N supply resulting from SOC mineralization throughout the growing season covered less than 10% of the crop needs and the yield patterns did not reflect the variation in SOC contents.

How to cite: Goffart, D., Dvorakova, K., Curnel, Y., Limbourg, Q., Crucil, G., Planchon, V., Van Oost, K., and van Wesemael, B.: Spatial patterns in winter wheat development related to soil properties and historic management. A case study from central Belgium., EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8700, https://doi.org/10.5194/egusphere-egu2020-8700, 2020.

Soil phosphorus (P) is one of the most critical elements for Earth’s ecosystem. P is a component of the complex nucleic acid structure of plants, which regulates protein synthesis, plants deficient in P are stunted in growth and lead to diseases. In practice, P is most often the element responsible for eutrophication problems in freshwater meanwhile, and it is considered the macronutrient most frequently as the element limiting eutrophication because many blue-green algae are able to use atmospheric N₂. Since the Second World War overuse application of fertilizer P has leaded to lots of serious environmental problems such as eutrophication of water body.

Soil P was affected by several factors including climate, geology, time, anthropogenic activities (irrigation, industrial emission, fertilizer application, crop planting pattern etc.) and so on. This makes soil P varied in a very complex manner on both spatial and time dimension and thus increases the difficulty of estimating spatio-temporal variation of soil P. Therefore, a flexible framework is necessary for modelling spatio-temporal variation of soil P.

To explore spatio-temporal variation of soil available P, we propose a Bayesian hierarchical spatio-temporal model using Integrated Nested Laplace Approximation with Stochastic Partial Differential Equation approach (INLA-SPDE). The study was conducted on phosphorus measured by Olsen (P Olsen) and Dyer (P Dyer) methods in Britany (western France) from 1995 to 2014 with data of more than 30,000 samples of France national soil test database (BDAT).

The INLA-SPDE method exploits the Laplace approximation in Bayesian latent-Gaussian models and does not require generating samples from the posterior distribution. Hence, it can often be used for quite large data sets at reasonable computational expense. It could provide approximate marginal (posterior) distributions over all states and parameters. In this study, the constructed model includes of several components such as spatial varying trend, space varying temporal trend, effects of covariates, and residual with space-time dependent variation.

Regardless the method of quantifying phosphorus, the results indicated that the mean content of soil available P decreased between 1995 and 2014 in Britany. Our model explained 49.5% of variance of spatio-temporal variation of P Olsen in Britany in external validation dataset. For P Dyer, our model explained 50% of variance in external validation dataset. The purely spatial effects shown that the available P in west of Britany was higher than east part. Our study could contribute to better soil management and environmental protection. Further study still needed to include more related factors into the model to improve the model performance and detected more related factors (such as soil management measures) which have important effects on spatio-temporal variation of available P in soil.

How to cite: Hu, B., Saby, N., Bourennane, H., Opitz, T., Denoroy, P., Lemercier, B., and Shi, Z.: Bayesian spatio-temporal modeling of soil phosphorus in Britany in western France (1995-2014) with INLA-SPDE, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-19135, https://doi.org/10.5194/egusphere-egu2020-19135, 2020.

In this research, static Bayesian networks (BN) is presented for predicting Soil Organic Carbon (SOC) in the complex and open soil systems. BN is a graphical computational model which provides a simple technique to define the nonlinear dependencies and, therefore, to implement a compact representation of the complex systems. Moreover, the BN is used as a simulation tool for effective processing of the complex system outcomes by probability propagation methods. This permits evaluation and potential intervention in complex soil systems and determines the dependencies between different variables. We use a BN to identify key factors in predicting England and Wales SOC. Then, we explore the relationships between different key factors such as geographical, environmental or climate and their roles individually in predicting SOC, particularly to identify those which have the highest impact. The proposed BN is also used to calculate the effectiveness of these interventions where the uncertainties associated with these casual relationships at the same time. This approach works with data from the variety of sources and handles a mix of subjective and objective data and can incorporate variables which differ across the contexts. The effectiveness of the technique is demonstrated with a case study to predict SOC.

How to cite: Sedighi, T., Hannam, J., and Corstanje, R.: Using Bayesian Network for Soil Organic Carbon Prediction despite its Incredible Complexity, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-1400, https://doi.org/10.5194/egusphere-egu2020-1400, 2020.

Maintaining soil organic carbon (SOC) content is recognized as an important strategy for a well-functioning soil ecosystem. The UN Convention to Combat Desertification (UNCCD) recognizes that reduced SOC content can lead to land degradation, and ultimately low land and agricultural productivity. SOC is almost universally proposed as the most important indicator of soil health, not only because SOC positively influences multiple soil properties that affect productivity, including cation exchange capacity and water holding capacity, but also because SOC content reflects aboveground activities, including especially agricultural land management. To be useful as an indicator, it is crucial to assess the importance of both inherent soil properties as well as external factors (climate, vegetation cover, land management, etc.) on SOC dynamics across space and time. This requires large, reliable and up-to-date soil health data sets across diverse land cover classes. The Land Degradation Surveillance Framework (LDSF), a well-established method for assessing multiple biophysical indicators at georeferenced locations, was employed in nine countries across the tropics (Burkina Faso, Cameron, Honduras, India, Indonesia, Kenya, Nicaragua, Peru, and South Africa) to assess the influence of land use, tree cover and inherent soil properties on soil organic carbon dynamics. The LDSF was designed to provide a biophysical baseline at landscape level, and monitoring and evaluation framework for assessing processes of land degradation and the effectiveness of rehabilitation measures over time. Each LDSF site has 160 – 1000 m² plots that were randomly stratified among 16 - 1 km² sampling clusters. A total of 6918 soil samples were collected (3478 topsoil (0-20 cm) and 3435 subsoil (20-50 cm)) within this study. All samples were analyzed using mid-infrared spectroscopy and 10% of the samples were analyzed using traditional wet chemistry to develop calibration prediction models.  Validation results for soil properties (soil organic carbon (SOC), sand, and total nitrogen) showed good accuracy with R² values ranging between 0.88 and 0.96. Mean organic carbon content was 21.9 g kg^-1 in topsoil and 15.2 g kg^-1 in subsoil (median was 18.3 g kg^-1  for topsoil and 10.8 g kg^-1 in subsoil). Forest and grassland had the highest and similar carbon content while bushland/shrubland had the lowest. Sand content played an important role in determining the SOC content across the land cover types. Further analysis will be conducted and shared on the role of trees, land cover and texture on the dynamics of soil organic carbon and the implications for LDN reporting, land restoration initiatives as well as sustainable land management recommendations.

How to cite: Winowiecki, L. and Vågen, T.-G.: Using a data-driven network to understand the drivers of soil organic carbon dynamics across the tropic, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-15028, https://doi.org/10.5194/egusphere-egu2020-15028, 2020.

Soil information is fundamental for many global applications, such as food security, land degradation, water resources, hydrology, climate change and ecological conservation. To address these diverse needs, it is important to provide free, consistent, easily accessible and standardized soil information. SoilGrids meets these requirements being a  global product supporting global modelling and providing complementary information for the development of regional and national products in data-poor areas. This presentation will focus on the methodological aspects for modelling and mapping of global soil information. We describe the selection of models for global mapping using quantile random forest and recursive feature elimination to obtain a parsimonious model. We also use a refined cross-validation procedure to account for bias caused by spatial differences in sampling density at different depths. SoilGrids also quantifies location-specific uncertainty at global level by computing 90% prediction interval limits. 

How to cite: Poggio, L., Moreira de Sousa, L., Heuvelink, G., Kempen, B., Bai, Z., Turdukulov, U., Ruiperez Gonzalez, M., Ribeiro, E., Batjes, N., and van den Bosch, R.: Modelling and mapping global soil information, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-12945, https://doi.org/10.5194/egusphere-egu2020-12945, 2020.

Changes in soil practices and management policies are fundamental in order to satisfy future growing food and energy demand, limiting risks of soil depletion. To this purpose remote sensed data are proving crucial for precision farming and for soil characterization and monitoring. In regions where agronomic rotations are adopted, soils experience unproductive periods between two main crops ("bare soil"), causing nutrient leaching, erosion and acceleration of organic matter consumption. Although the presence of "bare soil" period is evident and well known, there are no studies able to provide a dedicated regional framework to draw attention to this issue.

This study aims at mapping soils cultivated but unproductive during certain times of the year ("bare soil") using satellite images. Once detected, "bare soils" are deeply investigated to define their surface and the time duration of the bare soil status. Thereafter, valorization scenarios for these "bare soil" are proposed considering an optimized mix of energy and cover crops.

The applied methodology includes Sentinel-2, 5-days-return-time optical images, with 20 m ground spatial resolution acquired during 2017. The images were pre-processed using the Satellite Image Automatic Mapper™ (SIAM™) and outputs subsequently processed on the QGIS platform and validated with ground truths provided by the regional agriculture authority.

Of the total Utilised Agricultural Area (UAA), results show that up to 20% is "bare soil" from July to October and about 10% is unproductive from November to April. The size of most plots varies from 0.5-2 ha, however, about 30% of the "bare soil" fields have surface size from 3 to 50 ha, sufficient to justify their agronomic exploitation. In a basic scenario where biomass sorghum is cultivated from July to October, 50% of the bioenergy demand can be met through anaerobic digestion.

This study proposes a digital soil mapping methodology able to answer several questions: if yields can be improved, in what period of the year, in which area, how large are the plots. Therefore, the potential of "bare soils" for increasing food or energy crops and to store more carbon in soils is highlighted.

How to cite: Greggio, N., Cimatti, M., Spisni, A., Sapia, L. D., Baraldi, A., Contin, A., and Marazza, D.: "Bare soil" detection addressing agricultural production optimization throughout the year: case study in Emilia Romagna using Sentinel-2 images., EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11439, https://doi.org/10.5194/egusphere-egu2020-11439, 2020.

Salinization and alkalinization are worldwide among the soil degradation threats in irrigated schemes affecting soil productivity. Niger River basin irrigated schemes in the Sahel arid zone are no exception (ONAHA, 2011). The use of remote sensing for identifying and evaluating the level of these phenomena is an interesting tool. The launching of the Sentinel2 satellite constellation (2015) brings new perspectives with high spectral and temporal resolutions images. The aim of this study was to develop a methodology for detection of salt-affected soils in this climatic condition.

To achieve our goal, we used two types of data: remote sensing and ground truth data.

Two complementary approaches were used: one by observing salinity on bare soil by the use of salinity index (SI) and the other by observing the indirect effects of salinity on the vegetation during eight (8) rice growth phases  using vegetation index NDVI.

Remote sensing data were acquired from multi temporal sentinel2 images over 4 years (from 11/12/2015 to 30/11/2019). One hundred and fifty seven (157) images were downloaded (one image each 5 days) and corrected from atmospheric effects and some bands resampled to 5 m using python software. The salinity and vegetation indices were calculated. NDVI index was calculated and NDVI integral between NDVI curve and the threshold of 0.21 NDVI calculated for the eight growing cycles.

Ground truth data were collected in 2019 during the dry growing season (January – may 2019) from 24 calibration plots and 40 validation plots. One hundred and twenty (120) soil samples collected and analyzed for pH and electrical conductivity and finally forty six (46) biomass samples were collected, air dried and weighed for biomass yield and 46 grains samples collected for grain yield.

NDVI integral proved to be good indicator for yield variations and could distinguish crops behavior according to the growing period. It also makes it possible to distinguish plots which were not cultivated or with weak growth due to strong constraints of which the main one is salinity. It showed also that the effect of salinity on growth differs according to the growing season and the possibility of managing irrigation. Bare soil analysis distinguishes fields with different salinity indexes despite the low number of dates for which bare soil can be observed.

Ascending Hierarchical Classification (AHC) enabled to identify four classes of NDVI dynamics over time and bare soil salinity index. High saline soils according to direct soil measurements were related to the class characterized by high frequency of no-cultivation during the dry season and low NDVI integral during the wet season. Multi-temporal Sentinel2 images analysis enabled therefore to detect rice crop fields affected by salinity through its influence on crop behavior. This approach will be tested over the whole paddy schemes of the Niger River valley.

How to cite: Issaka, M., Christian, W., Didier, M., Pascal, P., Hervé, N., and Guéro, Y.: Soil salinity assessment using temporal series of Sentinel2 satellite images in irrigated paddy fields of Western Africa , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11855, https://doi.org/10.5194/egusphere-egu2020-11855, 2020.

In this study, we predicted and mapped soil salinity using machine learning (ML) and digital soil mapping (DSM) approaches. Support vector regression (SVR) and the hybrid of SVR with wavelet transformation (W-SVR) where applied to correlate soil salinity of the upper 200 cm of soil to a wide range of environmental covariates derived from a digital elevation model (DEM), remote sensing (RS) and climatic data. Results indicate that W-SVR performed better in predicting soil salinity at all depth intervals with scattered index ranging from 1.45 to 1.68 compared to the standalone SVR. This is particularly true at the lowest soil depth when W-SVR indicated ~1.5 times higher accuracy compared to the SVR. At this soil depth topographic features are the main covariates in the models. For topsoil salinity, land use represented by RS features controls the spatial distribution of the salinity widely. Independent from soil depth, climatic features are the most important predictors for soil salinity in all ML models. The predicted salinity maps show the highest salinity for soils in the eastern parts of central Iran. Furthermore, the importance of topographic features for all ML algorithms coincides with most landform characteristics in central Iran and confirms a close relation of soil salinity not only to land use practices like irrigation but also to soil-landscape relationships in this dry region.

Keywords: Soil salinity, machine learning, spatial variation, central Iran, support vector regression, wavelet transformation

Acknowledgments

Ruhollah Taghizadeh-Mehrjardi has been supported by the Alexander von Humboldt Foundation under the grant number: Ref 3.4 - 1164573 - IRN - GFHERMES-P.

How to cite: Taghizadeh-Mehrjardi, R., Toomanian, N., Shamshirband, S., Mosavi, A., Behrens, T., Schmidt, K., and Scholten, T.: Predicting and mapping of soil salinity using machine learning algorithms in central arid regions of Iran, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-18516, https://doi.org/10.5194/egusphere-egu2020-18516, 2020.

Area-wide high resolution information of organic layer properties is required for assessing the current nutrient availability in forest stands. Together with climate, location, parent material and terrain predictors, vegetation is known to have a direct impact on the characteristics of the organic surface layer of forest soils and therefore plays an important role in predicting its chemical properties.

Here, we use multi-temporal Sentinel-2 (S2) Earth Observation (EO) data as a proxy for various vegetation characteristics to spatially predict forest organic layer properties at 10 m spatial resolution in Saxony/Germany. To ensure full data coverage of the study area, we used multi-temporal statistical measures of S2 data generated by the FORCE algorithm. Complementary to ancillary predictors (climate, location, parent material, terrain), we compared three different sets of vegetation related data as predictors: (1) categorical tree species groups from a field survey, (2) multi-temporal statistical measures derived from S2 data and (3) S2 multi-temporal statistical measures and additional S2 multi-temporal spectral indices.

We used random forest regression models to estimate pH value, base saturation, C/N ratio and effective cation exchange capacity of the organic surface layer and the upper 0-5 cm of the first mineral soil horizon. For model evaluation 5-times 10-fold cross-validation was applied. For predictor evaluation and selection, we used recursive feature elimination.

The results indicate that S2 data can serve as a vegetation proxy when predicting forest organic layer properties. For example, the cross-validation estimate of the prediction error in scenario (3) for C/N ratio (organic surface layer) is about 7.3 % lower than in scenario (1). In some cases the explanatory power is higher compared to the field survey data of the tree species groups, probably due to the high local variability of EO based data. This may help to reveal short range variability of chemical properties. We conclude that the three scenarios show comparable results and thus multi-temporal S2 data can be used as a vegetation proxy to spatially predict chemical properties of the organic surface layer of forest soils.

How to cite: Becker, C., Frantz, D., Petzold, R., Schmidt, K., Behrens, T., and Scholten, T.: Using multi-temporal Sentinel-2 data to predict chemical properties of the organic surface layer of forest soils, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-21992, https://doi.org/10.5194/egusphere-egu2020-21992, 2020.

The study presents the analysis of effect of changes of the open surface of arable soils occuring due to the influence of agricultural practices or natural factors (mainly, precipitation) on the possibility of assessment of organic matter content in the arable layer with optical remote sensing data.

The object of the research was gray forest arable soil of a test field located in the Yasnogorsky district of the Tula region. In 2019, the field was complete fallow.

During field work conducted on the test field on 15.08.2019, the spectral reflectance of the surface of arable soils and a wetter subsurface horizon was measured at 30 points. At the same points, 30 mixed samples of the arable horizon were collected for laboratory estimation of organic matter content.

Spectral reflectance was measured using a HandHeld-2 field spectroradiometer, which operates in the range 325–1050 nm with a step of 1 nm.

Proximal sensing data were smoothed with Savitzky-Golley function and recalculated into Sentinel-2 bands using Gaussian function.

We also chose seven Sentinel-2 scenes for 2019 for the studied region: 2.04.2019, 17.04.2019, 20.04.2019, 5.05.2019; 6.06.2019, 19.06.2019, 28.08.2019. Atmospheric correction for chosen scenes was performed with Sen2Cor model in SNAP. Aftewords we extracted reflectance values at points, where we collected spectral data and soil samples in the field.

Then we calculated a number of spectral indices and ratios for both proximal and Sentinel-2 data which were further used in regression modelling. Models were cross-validated by bootstrapping.

At field scale, difference in moisture content did not significantly affect the accuracy and quality of the models. R²adjcv of model for dry surface layer was a bit higher than in case of model for wet subsurface layer (0.77 vs. 0.72). RMSEPcv and RPIQ for both cases were very close (0.71 and 0.71; 2.09 and 2.12).

When we used models developed based on proximal sensing data to calculate OM content with Sentinel-2 data at different acquisition dates, we found that the accuracy of OM prediction varied. In some cases RMSE was higher than 7 % and predicted OM content was two times higher than actual.

Models developed based only on Sentinel-2 data for different acquisition dates, varied in accuracy, quality and informative bands. R²adjcv of most models was about 0.72-0.83, RPIQ was 2.09-2.07, and RMSEPcv was in the range of 0.56-0.77 %.

Therefore changes in surface state of arable soils result in a situation when for each state we have different model. That imposes restrictions on further use of such models for remote evaluation and monitoring of organic matter content in arable soils. To deal with this problem, it is necessary to account for soil surface state when developing models for properties of arable soils based on optical remote sensing data.

The research was funded by the Ministry of Science and Higher Education of Russia (contract № 05.607.21.0302). 

How to cite: Prudnikova, E. and Savin, I.: Modelling of soil organic matter of arable soils with optical remote sensing data: the impact of soil surface state, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20185, https://doi.org/10.5194/egusphere-egu2020-20185, 2020.

Since the onset of agriculture, soils have lost their organic carbon to such an extent that the soil functions of many croplands are threatened and there is therefore a strong demand for mapping and monitoring critical soil properties and in particular soil organic carbon (SOC). Pilot studies have demonstrated the potential for remote sensing techniques for SOC mapping in croplands, given their large spatial coverage and high temporal resolution. It has however been shown that the assessment of SOC may be hampered by crop residues. In this study we tested the effect of the threshold for the cellulose absorption index (CAI), on the performance of SOC prediction models for bare cropland soils. Airborne Prism Experiment (APEX) hyperspectral images covering an area of 230 km² in the Belgian Loam Belt were used together with a local soil dataset. We used the partial least square regression (PLSR) model to estimate the SOC content based on 104 georeferenced calibration samples, firstly without setting a CAI threshold, and obtained a satisfactory result (R²=0.49, RPD=1.4 and RMSE=2.14 g kgC^-1 for cross-validation). However, a cross comparison of the estimated SOC values to grid-based measurements of SOC content within three fields revealed a systematic overestimation for fields with high residue cover. We then tested different thresholds of CAI in order to mask pixels with high residue cover, by eliminating calibration samples used in the PLSR model based on this threshold. The best model was obtained for CAI threshold of 0.8 (R²=0.59, RPD=1.5 and RMSE=1.76 g kgC^-1 for cross-validation). These results reveal that the purity of the pixels needs to be assessed aforehand in order to produce reliable SOC maps. Preliminary results indicate that an index based on the SWIR bands of the MSI Sentinel 2 sensor is also capable of detecting crop residues. However, the application under moist conditions and for different types of residues needs to be confirmed.

How to cite: Dvorakova, K., Shi, P., Quentin, L., and van Wesemael, B.: Soil organic carbon mapping from remote sensing: The effect of crop residues, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8253, https://doi.org/10.5194/egusphere-egu2020-8253, 2020.

Rhenish opencast mines located in the central west of Germany have used about 330 km² of land so far. Of this, some 230 km² have been recultivated, including 125 km² of arable land. After recultivation, the land is cultivated for at least seven years by the mining company before let to the farmers. Where new farmland is envisaged, the stackers spread pure loess mixed with soil material of the original Luvisols (loess loam) at the top of the refilled mining areas. After a certain settling time, this layer must be at least two meters thick. In a next step, the loess is levelled in a soil-sparing fashion using caterpillars with extra-wide rawler tracks. Even if care is taken that the loess layer will not be heavily compacted during levelling, local soil compaction is one of the major problems, as leveling often is performed during unfavorable moist soil conditions. These local compactions lead to reduced crop growth during either wet or dry growing seasons and result in yield losses over periods of many years. Localizing and evaluating such compacted field zones would enable the mining company to perform a physical soil melioration before handing over the land to a farmer.

To identify local soil compaction, a field study was performed in 2019 on a selected field with known variability in crop performance within the recultivated area of the Garzweiler mine in North Rhine-Westphalia, Germany. Over the course of 5 months, the field was intensively investigated using geophysical methods such as electromagnetic induction (EMI) and electrical resistivity tomography (ERT). Additionally, soil samples were taken to determine soil water contents, bulk density, penetration resistance, and soil texture.

The geophysical maps gathered, clearly show zones of higher electrical conductivities in the soil, which were associated to conventionally measured subsoil compaction. Regression of bulk densities with EMI data yielded good results allowing to map out compacted zones within the field and also to quantify compaction. Hence, geophysical methods provide a promising approach to plan soil melioration measures.

How to cite: Weihermüller, L., Schmäck, J., Mertens, M., Endenich, M., van der Kruk, J., Vereecken, H., Welp, G., and Pätzold, S.: Detection and quantification of soil compaction in a post-mining landscape by geophysical methods, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-3271, https://doi.org/10.5194/egusphere-egu2020-3271, 2020.

Soil pH is one of the most important soil parameters, due to its importance for for soil management and food security. Spatial distribution of pH could altered by the different environmental conditions, such as geology, climate or soil-vegetation interactions. pH has an ecological function in controlling spatial distribution of plant species, conditioning absence or presence of different species due to soil pH ability or modifying mineral solubility. Hence, pH and remotely sensed land surface phenology (LSP) could be associated. The objective of this work was two-folded: i) mapping the soil pH of Andalusian soils and ii) the evaluation of new features derived from remote sensing which are related to seasonal cycles of vegetation applied to digital soil mapping

We developed a pH model using 3215 pH measurements at different locations together with three types of predictor features: terrain (elevation, slope, hydrological attributes…), climatic (annual and monthly precipitation and maximum and minimum temperatures) and phenological features extracted from remotely sensed vegetation indices time series (date of the start of spring, date of the end of senescence, growing season length, end of the growing season, length of the growing season, maximum peak, and large seasonal integral as a proxy of productivity). The LSP features were obtained from time series of NDVI that were computed from the MODIS weekly surface reflectance product (MOD09Q1 v6) at a spatial resolution of 250 for the entire study period. The performance of  multiple lineal regression (MLR) and Random Forest was evaluated within the framework of a high dimensional feature space.

The results showed that RF outperformed MLR (R²: 0.66 and 0.58; RMSE: 0.76 and 0.83). ph and feature pairwise correlations were higher for the phenological features: median of large integral (-0.55); median of maximum peak (-0.51); valley depth (0.48); median of date of start of spring (-0.47), median of value on the date of start of spring (-0.46). The most important features in RF prediction were almost the same: the median of large integral, valley depth, maximum temperatures in September and median of maximum peak, showing that LSP features were relevant in pH spatial modelling, with an better performance of RF model.

How to cite: Canero, F. M., Rodriguez-Galiano, V., Cardenas-Martinez, A., and Luque-Espinar, J. A.: Modelling and mapping soil pH in Andalusia (Spain) using phenological products as predictor features, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-5455, https://doi.org/10.5194/egusphere-egu2020-5455, 2020.

Soils have critical relevance to global issues, such as food and water security, climate regulation, sustainable energy, desertification and biodiversity protection. All these examples require accurate national soil property information and there is a need to scientific support to develop reliable baseline soil information and pathways for measuring and monitoring soils. Soil sustainable management is a global issue, but effective actions require high-resolution data about soil properties. Two projects, GlobalSoilMap and SoilGrids, aim at delivering the first generation of high-resolution soil property grids for the globe, the first one by a bottom-up approach (from country to globe), the latter by top-down (global). The GLobAl Digital SOIL MAP (GLADSOILMAP) consortium brings together world scientific leaders involved in both projects. The consortium aims at developing and transferring methods to improve the prediction accuracy of soil properties and their associated uncertainty, by using legacy soil data and ancillary spatial information. This approach brings together new technologies and methods, existing soil databases and expert knowledge. The consortium aims at transferring methods to achieve convergence between top-down and bottom-up approaches, and to generate methods for delivering maps of soil properties. These maps are essential for communities from climate and environmental modeling to decision making and sustainable resources management at a scale that is relevant to soil management. The consortium will ensure links with the numerous actors in geosciences of the world, and will contribute to improving their skills in digital mapping and their national and international legibility. The actions include 4 main Work Packages (WP) subdivided into several tasks that are summarized below:

WP0 Management of the project

WP1 Legacy and ancillary data for Digital Soil Mapping (DSM)

Test the potential of new ancillary data for DSM

Explore methodologies to merge and/or harmonize different products

Propose methods for harmonizing products to a common date

WP2 Methods for sampling, modelling and mapping soils in space and time

Testing and developing new methods/models for prediction

Testing methods for estimating complete probability distribution

WP3 Methods for estimating model and map uncertainty

Develop methods of uncertainty spatial assessment

Develop methods do deal with censored data/soft data

Solve the question of influence on the age of the rescued soil data on predictions

WP4 Scientific outreach and capacity building

Produce an exhaustive review of GlobalSoilMap initiatives and results all over the world

Revise and update the GlobalSoilMap specifications by keeping them at the state-of-the-art level

Show relevance of gridded, Global, DSM by use cases and communication to end users

The added value of the consortium is to allow a direct scientific exchange between members that should result in synthesis papers, in the identification of the major knowledge gaps, and in extending, deepening and disseminating knowledge of DSM, with the final aim to contribute to the achievement of global soil maps. Another added value of the consortium will certainly be to foster the creation of new ideas.

Acknowledgements: the Consortium GLADSOILMAP is supported by LE STUDIUM Loire Valley Institute for Advanced studies.

How to cite: Arrouays, D., Libohova, Z., Minansny, B., Mulder, V. L., Poggio, L., Roudier, P., Richer-de-Forges, A. C., Bourennane, H., nehlig, P., Martelet, G., and Lagacherie, P.: The consortium GLADSOILMAP, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8335, https://doi.org/10.5194/egusphere-egu2020-8335, 2020.

In the 21^st century, soils are at the crossroads of global issues (i.e., food security, water security, biodiversity protection, climate change, and ecosystem services) and essential to achieve some of the Sustainable Development Goals. Although soils are central to these global issues, their management requires local actions and knowledge, which requires fine-resolution soil information. With an emphasis on broad-scale studies (>10,000 km²), this review outlines recent progress in the development of GlobalSoilMap, an initiative to provide a global fine-resolution grid of soil properties with quantified uncertainties using the bottom-up approach. This review provides an overview related to the soil data source, environmental covariates, spatial prediction, modelling and mapping techniques, uncertainty qualification, and target soil properties. The main findings of this review are: (1) A great increase of publication was observed after 2012, reaching a peak in recent years; (2) Australia and China were the most active countries; (3) Geoderma was the most frequent journal that was preferred by authors to publish related studies; (4) More than a half of the studies did not report soil sampling design; (5) Data splitting was the most frequent strategy for model evaluation, and independent validation was rarely used; (6) Nonlinear predictive model (i.e., machine learning) was becoming popular than ever before; (7) Relief, organisms and climate were the top three SCORPAN factors used in modelling; (8) Soil organic carbon (or soil organic matter) was the top soil property of interest.

This review also highlights the perspectives of GlobalSoilMap for further improving the quality of soil information globally and making it practical in decision making.

How to cite: Chen, S., Mulder, V. L., Poggio, L., Roudier, P., Libohova, Z., Minasny, B., Shi, Z., Hannam, J., and Arrouays, D.: Digital mapping of soil information at a broad-scale: A review on GlobalSoilMap, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-8426, https://doi.org/10.5194/egusphere-egu2020-8426, 2020.

Soil physical properties and soil water regime have been in the focus of soil surveys and mapping in Hungary due to their importance in various environmental processes and hazards, like waterlogging and drought, which endanger extended areas. 
In the late ‘70s a category system was elaborated for the planning of water management, which was used as the legend of a nationwide map prepared at a scale of 1:500.000. Soils were characterized qualitatively (e.g.: soil with unfavorable water management was defined with low infiltration rate, very low permeability and hydraulic conductivity, and high water retention), without quantification of these features. The category system was also used for creating large-scale (1:10.000) water management maps, which are contained legally by expert’s reports prepared on the subject of drainage, irrigation, liquid manure, sewage or sewage-sludge disposal. These maps were prepared eventually, essentially for individual plots and are not managed centrally and are not available for further applications.
Recently a 3D Soil Hydraulic Database was elaborated for Europe at 250 m resolution based on specific pedotransfer functions and soil property maps of SoilGrids. The database includes spatial information on the soil water content at the most frequently used matric potential values, saturated hydraulic conductivity, Mualem-van Genuchten parameters of the moisture retention and hydraulic conductivity curves. Based on similar idea, the work has been continued to produce more accurate and spatially more detailed hydrophysical maps in Hungary by generalizing the applied pedotransfer functions and using national soil reference data and high resolution, novel, digital soil property maps.
We initiated a study in order to formalize the built-in soil-landscape model(s) of the national legacy map on water management, together with the quantification of its categories and its potential disaggregation. The relation of the legacy map with the newly elaborated 3D estimations were evaluated at two scales: nationwide with 250 m resolution and at catchment scale with 100 m resolution. Hydrological and primary soil property maps were used as predictor variables. Unsupervised classifications were performed for spatial-thematic aggregation of the soil hydraulic datasets to identify their intrinsic characteristics, which were used for the elaboration of a renewed water management classification. Hydrological interpretation of the categories provided by the optimum classifications has been carried out (i) by their spatial cross-tabulation with the categories of the legacy map and (ii) using the interval estimation of the applied soil hydraulic properties provided for the individual water management categories. Machine learning approaches were used to analyze the information content of the legacy maps’s category system, whose results were used for its disaggregation. Conditionally located random points were sequentially generated for virtual sampling of the legacy map to produce reference information. The disaggregated maps with the legend of the traditional water management classes were produced both on national and catchment level.

Acknowledgment: The research has been supported by the Hungarian National Research, Development and Innovation Office (NRDI) under grants KH124765, KH126725, the János Bolyai Research Scholarship of the Hungarian Academy of Sciences and the MTA Cloud infrastructure (https://cloud.mta.hu/).

How to cite: Szabó, B., Laborczi, A., Szatmári, G., Bakacsi, Z., Makó, A., Braun, P., and Pásztor, L.: Renewal of a national soil water management category system and legacy map by data mining methods, digital primary and hydrological soil property maps, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-9234, https://doi.org/10.5194/egusphere-egu2020-9234, 2020.

There is increasing demand for up‐to‐date spatial information on soil organic carbon (SOC). Meanwhile, Unmanned Aerial Vehicles (UAV) provide flexible technology for monitoring land surface features with high spatial resolution at plot scale. Suitably performed, airborne imagery simultaneously provides spectral and terrain based spatial auxiliary data, which can be used as predictors in DSM-type modelling of topsoil OC.

To test its applicability for spatial prediction of topsoil OC, an aerial survey was carried out on a plot situated on a gently undulating slope by a Cubert UHD-185 hyperspectral snapshot camera mounted on a Pixhawk-based octocopter. The camera is capable to record electromagnetic spectrum between 450-950 nm in 125 spectral bands on 50×50 pixels images and the panchromatic spectrum in 1 Mpx images. Because of the narrow field-of-view of the UHD-185, three consecutive flights were needed to cover the whole area (cca. 10 ha); all were happened in the hours close to noon and flown in automatic flight mode to ensure the right over- and sidelap between images to make possible the photogrammetric processing. Despite the automatic flights a surveying grade GPS unit was also used to survey 12 markers, evenly distributed on the field to orthorectify images later.

The hyperspectral and panchromatic images were pre-processed in Cubert Edelweiss to produce different versions of them depending on the used spectral information to investigate later how built-in pan-sharpening method affects the prediction accuracy. The generated datasets are the native and pan-sharpened hyperspectral mosaics. Later the photogrammetric processing was performed in Agisoft Photoscan for both hyperspectral datasets, resulting in two georeferenced outcomes: a common digital elevation model (DEM) and two hyperspectral orthomosaics of the area, each exported with 1 m spatial resolution. Further data editing steps were carried out in R, generating various versions of exported hyperspectral orthomosaics: mosaic containing all of the 125 spectral bands; filtered (where spectrally overlapping bands with high correlation were removed based on Full Width at Half Minimum information) and Principal Component Analysis transformed versions.

Based on different kind of spectral orthomosaics and DEM combinations, a custom R script using Random Forest algorithm generated 36 predicted layers for topsoil OC, which were validated by Leave-One-Out Cross-Validation, hence independent mean and RMSE errors could be calculated for each dataset combinations. The overall best performing datasets were provided by the FWHM-filtered hyperspectral orthomosaic, hence the lowest mean error is resulted by the filtered, pan-sharpened PCA-transformed combination containing the DEM and its derivatives. However, in the RMSE values there were no significant difference between the six lowest RMSE combinations, but mostly the pan-sharpened and PCA-transformed versions perform better.

How to cite: Mészáros, J., Jakab, G., Árvai, M., Szabó, J., Tóth, M., Keller, B., Szatmári, G., Szalai, Z., and Pásztor, L.: Predicting topsoil organic carbon using UAV-based hyperspectral sensor, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-9324, https://doi.org/10.5194/egusphere-egu2020-9324, 2020.

The main objective of DOSoReMI.hu (Digital, Optimized, Soil Related Maps and Information in Hungary) initiative has been to broaden the possibilities, how demands on spatial soil related information could be satisfied in Hungary, how the gaps between the available and the expected could be filled with optimized digital soil (related) maps. During our activities we have significantly extended the potential, how goal-oriented, map-based soil information could be created to fulfill the requirements. Primary and specific soil property, soil type and certain tentative functional soil maps were compiled. The set of the applied digital soil mapping techniques has been gradually broadened incorporating and eventually integrating geostatistical, machine learning and GIS tools. Soil property maps have been compiled partly according to GlobalSoilMap.net specifications, partly by slightly or more strictly changing some of their predefined parameters (depth intervals, pixel size, property etc.) according to the specific demands on the final products. The nationwide, thematic digital soil maps compiled in the frame and spin-off of our research have been utilized in a number of ways.

Soil hydraulic properties (saturated hydraulic conductivity, wilting point, field capacity, saturated water content) were mapped applying generalized pedotransfer functions on available, primary soil property maps supplemented with further environmental co-variables, which were also used in the elaboration of the specific PTF.

Spatial assessment of certain provisioning and regulating soil functions and services was carried out by the involvement of soil property maps in digital process/crop models, which properly simulate the soil-plant-water environment conditioned by various factors based on actual, predicted or presumed data. Specific outputs of the modelled processes provided adequate information on functional behavior of soils.

Programs or studies dedicated to the designation of areas suitable for irrigation; risk modelling of inland excess water hazard; mapping of potential habitats; spatial assessment and mapping of ecosystem services were heavily relied on the novel type spatial soil information. The approaches sometimes required certain modifications of the standard GSM products due to various reasons.

The paper will present various national functional applications of primary soil property maps provided by DOSoReMI.hu.

Acknowledgment: Our research was supported by the Hungarian National Research, Development and Innovation Office (NRDI; Grant No: KH126725).

How to cite: Pásztor, L., Laborczi, A., Szabó, B., Fodor, N., Koós, S., and Szatmári, G.: Functional applications of primary soil property maps provided by DOSoReMi.hu, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-9645, https://doi.org/10.5194/egusphere-egu2020-9645, 2020.

Digital terrain analysis may be a useful tool for modeling the extent of Cs-137 soil contamination patterns after the Chernobyl disaster. The test area of the  Kostica River basin (Bryansk Region, Russia)  covers an area of 19,4x11,6 km and is characterized by relatively low levels of ¹³⁷Cs contamination after the Chernobyl accident  in the range of  2.4  to 33 kBq/m². It is just 4-18 times higher than the global fallout which was equal to 1,75 kBq/m² in 1986.

The purpose of  the research was to obtain estimates of the transformation of initial ¹³⁷Cs patterns as influenced by different landscape factors (DEM attributes) with a grid resolution of 100, 50 and 25 m. Different kinds of DEM  curvatures calculations  may be done by using SAGA, Whitebox GAT and Grass for each grid size model.

In the case under study two informational layers were made use of  to evaluate processes of ¹³⁷Cs redistribution in the River Kostica basin. These are: 1) SRTM  layer with a resolution of 90 m  and  2) the data of air-gamma survey with a resolution of 100 m.  The total watershed area of the Kostica River occupies 225 km². SRTM data were resampled in a coordinates and georeference system of AG (air-gamma survey was represented in the Gauss-Kruger coordinate system) lay  with a resolution of 100 m.

The results of the air gamma survey conducted in the summer of 1993, give clear evidence that the processes of ¹³⁷Cs lateral migration took place due to nearly a fourfold increase of ¹³⁷Cs in the lower slope as compared to the surface of the watershed during a seven-year period after the Chernobyl accident.

We examine the effect of grid size of the  digital elevation model (DEM) on the erosion simulations. For resampled grid data with a resolution  of 50 and 25 m  we apply SAGA-GIS Module “Resampling” and compare the results with those of  the original method of simplicity versus fitting (SvF).  The method SvF is devoted to finding a compromise between simplicity of the model and precision of replication of experimental data. The integral in the range of squared second derivatives was used as a measure of simplicity, with usual standard deviation being applied as a measure for replication of experimental data.

The study is based on the concept of sediment and hydrological connectivity. We apply GIS-based models considering lateral soil migration to analyze sediment cascade systems. Soil erosion was evaluated based on an analysis of Cs-137 migration determined using the LS factor implemented by GRASS GIS.

The reported study was funded by RFBR according to the research project № 20-07-00701A 

How to cite: Linnik, V., Sokolov, A., Ivanitsky, O., and Saveliev, A.: Modelling the extent of Cs-137 soil contamination patterns at the Kostica River basin (Bryansk Region, Russia), EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-10173, https://doi.org/10.5194/egusphere-egu2020-10173, 2020.

Soil Organic Carbon (SOC) plays a crucial role in agricultural ecosystems. However, its abundance is spatially variable at different scales. In recent years, machine learning (ML) algorithms have become an important tool in the spatial prediction of SOC at regional to continental scales. Particularly in agricultural landscapes, the prediction of SOC is a challenging task.

In this study, our aim is to evaluate the capability of two ML algorithms (Random Forest and Boosted Regression Trees) for topsoil (0 to 30 cm) SOC prediction in soils under agricultural use at national scale for Germany. In order to build the models, 50 environmental covariates representing topography, climate factors, land use as well as soil properties were selected. The SOC data we used was from the German Agricultural Soil inventory (2947 sampling points). A nested 5-fold cross-validation was used for model tuning and evaluation. Hyperparameter tuning for both ML algorithms was done by differential evolution optimization. 

This approach allows exploring an extensive set of field data in combination with state of the art pedometric tools. With a strict validation scheme, the geospatial-model performance was assessed. Current results indicate that the spatial SOC variation is to a minor extent predictable with the considered covariate data (<30% explained variance). This may partly be explained by a non-steady state of SOC content in agricultural soils with environmental drivers. We discuss the challenges of geo-spatial modelling and the value of ML algorithms in pedometrics.

How to cite: Sakhaee, A., Gebauer, A., Ließ, M., and Don, A.: Soil Organic Carbon Prediction at National Scale (Germany), EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-10552, https://doi.org/10.5194/egusphere-egu2020-10552, 2020.

Satellite multi-spectral remote sensing has been used extensively in mapping the nature and characteristics of the terrestrial land surface including vegetation, rock, soil and landforms across global through to local scales. However, with the exception of hyper-arid regions mapping rock and soil from space has been problematic due to vegetation that either masks the underlying substrate or confuses the spectral signatures of geological materials (i.e. diagnostic mineral spectral features) making them difficult to resolve. A barest earth multi-spectral algorithm operating on time series satellite archives can now significantly reduce the influence of vegetation and provide enhanced mapping of soil and exposed rock from space.

The methodology firstly applies a high-dimensional statistic called a ‘weighted geometric median’ which is robust to outliers or contamination (such as cloud cover, shadows, detector saturation, and pixel corruption) by removing sub-populations in the data. The weighted geometric median also maintains the relationship between all the spectral wavelengths which is important for the later implementation of image enhancement techniques based on the spectral signatures of minerals. The second component of the methodology applies a weighting scheme that preferences the bareness of pixels from those pixels that exhibit a vegetation influence.  After considerable experimentation a single model weighting scheme using a loss function that minimises NDVI was found to be the most robust for application at the continent scale. Customised calibration and weighting schemes can also be developed for local study areas. The result of this process for a given time series is an estimation of the barest state relating to either soil or exposed rock. The approach does not require local calibration and can be applied to other satellite archives globally.

We have applied the barest earth algorithm on Landsat and Sentinel-2 multispectral datasets to develop a suite of enhanced image products over the Australian continent to support digital soil mapping, geochemical modelling and mineral exploration. Image enhancements include individual band composites, ratio bands and selected principal component analysis. These enhanced mineral products provide new and improved inputs for machine learning and more broadly geo-spatial modelling/mapping.  The bare earth products significantly reduce the effects of fire scars in semi-arid areas of the continent and seasonal variations in vegetation cover that to-date have limited the use of satellite remote sensing in mapping soils in agricultural landscapes. The bare earth algorithm can be applied across different time intervals (e.g. annually, deeper time since mid-1980’s) and has the potential to establish environmental baselines for understanding and responding to food security, climate change, environmental degradation, water scarcity, and threatened biodiversity.

How to cite: Wilford, J. and Roberts, D.: Bare Earth – enhanced multi-spectral satellite imagery for mapping soil and exposed rock., EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-12371, https://doi.org/10.5194/egusphere-egu2020-12371, 2020.

Earth Observation (EO) has an immense potential as an enabling tool for mapping the spatial variation of the topsoil layer. Additionally, machine learning based algorithms deployed on cloud computing infrastructures have a great potential to revolutionize the processing of EO data. This paper aims to present a multi-dimensional Sentinel-based Soil Monitoring Scheme (S2MoS) based on open-access Copernicus Sentinel data and the Google Earth Engine platform to map soil properties. Building on key results from existing data mining approaches to extract bare soil reflectance values the current study presents i) preliminary insights on the synergistic use of open access SAR and optical images obtained from Sentinel-1 and Sentinel-2 sensors; and ii) evaluate the efficiency of machine learning algorithms to predict soil attributes based on multi-temporal analysis. In that regard, this study evaluated, based on Sentinel images extending over a 3 years period (2017-2019), the performance of two state of the art machine learning approaches, namely random forest and neural networks. Spatial thresholds values of 0.25 and 0.075 for Normalized Difference Vegetation Index and Normalized Burn Ratio 2 indices respectively were applied to mask bare soil pixels. In this study, we used 5000 soil data belonging to cropland land use from the European LUCAS topsoil database. We calibrated the models based on 4000 soil samples and then validated this approach with the rest 1000 samples  predict soil clay content. A higher prediction performance (R²=0.53) was achieved by the inclusion of both types (SAR and optical) of observations using the neural network model, demonstrating an improvement of about 5% in overall accuracy compared to the R² using the multi-year median optical composite.

How to cite: Tziolas, N., Tsakiridis, N., Ben Dor, E., Theocharis, J., and Zalidis, G.: A multi-dimensional Sentinel-based Soil Monitoring Scheme (S2MoS) for soil clay content estimation, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13549, https://doi.org/10.5194/egusphere-egu2020-13549, 2020.

Czech soil data is a mess. Modelling infiltration, or its probably most watched companion - runoff, has been quite a painful process for any researcher or practitioner studying any site larger or more heterogeneous than a few parcels of arable land. There are at least three main national soil databases in the Czech Republic, each of different age, scope, classification system and - most unfortunately - different administrator. So far Research Institute for Soil and Water Conservation has taken good care of data for agricultural land, while The Forest Management Institute did his job considering forest soils. A few other research institutes manage their own specific databases. There has been no service available providing consistent data for the whole country, nor methodology giving some guidelines on how to cope with differences in existing datasets, though a few large-scale applications and studies do exist. This contribution presents preliminary results of a running project TJ02000234 - Physical and hydropedological soil properties of the Czech Republic. It aims at harmonizing and combining available datasets and deriving layers of soil texture and hydropedological properties. Next the project aims at gathering available measurements of hydraulic properties of Czech soil types and their partial validation and extending with field measurements in the scope limited by the 2-years of project duration. The derived database and data products will be published in the form of a certified map as well as offered to professionals through an online GIS portal. Design planners in the Land consolidation, flood and soil erosion mitigation projects as well as professionals in public administration and researchers in environmental disciplines will benefit from the publication of this consistent data.

How to cite: Strouhal, L., Kavka, P., Beitlerová, H., and Žížala, D.: Harmonizing, merging and publishing hydropedological data for the Czech Republic, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-17487, https://doi.org/10.5194/egusphere-egu2020-17487, 2020.

The minimum set of indicators recommended for tracking progress towards LDN against a baseline are: land cover, land productivity and carbon stocks above and below ground. While land cover and its change can be and actually is operatively monitored by Earth Observation in a relatively straightforward manner, spatio-temporal assessment of the two other, soil related indicators poses challenges.

Soil organic carbon (SOC) stock in Hungary was first mapped in the frame of Global Soil Organic Carbon Map initiative. The Hungarian Soil Information and Monitoring System was used to create the GSOC product with quantile regression forest, which made the assessment of local uncertainty possible.  The map was produced with 500 meter spatial resolution and aggregated for the predefined 1 km grid. Since it used data collected in the first field campaign, in 1994, consequently its estimates represent that year’s state.

In 2018 a national report was expected by UNCCD on LDN firstly quantifying trends in carbon stocks above and below the ground. Based on global databases (ESA Climate Change Initiative Land Cover Dataset, SoilGrids250) default values were assigned to countries, which were asked about its acceptance or providing more accurate estimations based on national datasets. Similarly to the global initiative, SOC change estimation was not based on soil reference data dating from two distinct dates, but on the only available spatial prediction and changes of SOC were exclusively attributed to changes in land cover. Corine Land Cover Change maps were used to derive the GSOC estimations for the base year (2000) as well as for the target year (2012) from the original SOC map (representing 1994) according to Trends.Earth tool guidelines. SOC change between 2000 and 2012 was estimated by the difference of the two predictions.

In the next step, the SOM measurements on the samples collected in 2010 in the frame of Hungarian Soil Information and Monitoring System became available to map soil organic carbon stock in the topsoils (0-30 cm) of Hungary for the year 2010. New modelling was carried out based on the experiences of GSOC estimations, the map was produced with 100 m resolution using quantile regression forest for both years. 10-fold cross-validation was used for checking the accuracy of the spatial predictions and uncertainty quantifications. The performance of the spatial predictions and uncertainty quantifications was appropriate, which was verified by the computed biases, the root mean square errors, accuracy plots and the G statistics. Based on the compiled SOC stock maps, we assessed the spatial and temporal changes of SOC stocks on the whole area of Hungary except artificial surfaces and water bodies. The total SOC stock in the topsoil increased by 27.18 Tg over the respective period. We compared our estimate with others provided by global and continental SOC stock inventories. The comparison pointed out that a SOC stock map compiled by a given country can provide more accurate estimates at national level. We recommend applying the SOC stock map of 1992 as baseline to track and assess SOC stock change in Hungary.

How to cite: Pásztor, L., Laborczi, A., and Szatmári, G.: Spatio-temporal modelling of soil organic carbon stock for the support of national level assessment of land degradation neutrality in Hungary, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-17642, https://doi.org/10.5194/egusphere-egu2020-17642, 2020.

Legacy soil data arising from traditional soil surveys are an important resource for digital soil mapping. In the Czech Republic, a large-scale (1:10 000) mapping of agricultural land was completed in 1970 after a decade of field investigation mapping. It represents a worldwide unique database of soil samples by its national extent and detail. This study aimed to create a detailed map of soil properties (organic carbon, ph, texture, soil unit) by using state-of-the-art digital soil mapping (DSM) methods. For this purpose we chose four geomorphologically different areas (2440 km² in total). A selected ensemble machine learning techniques based on bagging, boosting and stacking with random hyperparameters tuning were used to model each soil property. In addition to soil sample data, a DEM and its derivatives were used as common covariate layers. The models were evaluated using both internal repeated cross-validation and external validation. The best model was used for prediction of soil properties. The accuracy of prediction models is comparable with other studies. The resulting maps were also compared with the available original soil maps of the Czech Republic. The new maps reveal more spatial detail and natural variability of soil properties resulting from the use of DEM. This combination of high detailed legacy data with DSM results in the production of more spatially detailed and accurate maps, which may be particularly beneficial in supporting the decision-making of stakeholders.

The research has been supported by the project no. QK1820389 " Production of actual detailed maps of soil properties in the Czech Republic based on database of Large-scale Mapping of Agricultural Soils in Czechoslovakia and application of digital soil mapping" funding by Ministry of Agriculture of the Czech Republic.

How to cite: Minařík, R., Žížala, D., and Juřicová, A.: Creation of detailed soil properties maps of the Czech Republic based on national legacy data and digital soil mapping, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-18480, https://doi.org/10.5194/egusphere-egu2020-18480, 2020.

Previous studies have shown that remote sensing data can be very useful input into soil prediction models. This input usually represents reflectance from bare soils, which, however, make up only a small part of the total area in a given part of the year. For eliminating masking effect of vegetation time series of individual images (Žížala et al. 2019; Shabou et al. 2015; Demattê et al. 2016; Blasch et al. 2015a) or multitemporal composites of spectral data can be used. Exposed Soil Composite Mapping Processor (SCMaP) (Rogge et al. 2018), Geospatial Soil Sensing System (GEOS3) (Demattê et al. 2018), Bare Soil Composite Image (Gallo et al. 2018), and Barest Pixel Composite for Agricultural Areas (Diek et al. 2017), all developed from Landsat time series, multitemporal bare soil image developed from RapidEye time series (Blasch et al. 2015b), or bare soil mosaic (Loiseau et al. 2019) derived from Sentinel-2 data can serve as examples of such composites. However, only some of the composite products have been used yet to predict soil properties. Promising results were achieved; however, the potential of these spectral composites has not yet been tested in a relevant number of studies. Further research is needed for its evaluation.

Aims of this study are to analyze and to compare the prediction ability of models using different types of multitemporal bare soil composites derived from Sentinel-2 images and their applicability for mapping soil properties in large areas. The study was conducted on a regional scale in the soil heterogeneous region of central Czechia with dissected relief and variable soil properties, where data from 100 soil profiles with soil analytics were available. Sentinel-2 images from 2016-2019 were used for composite formation in the python numpy environment. Different methods of cloud masking, bare soil identification and data aggregation (both already used in previous studies and newly derived) have been tested to compare which is the most suitable for prediction of soil properties. The principles of digital soil mapping and machine learning algorithms (random forest and support vector machine multivariate methods) were used for prediction.

Results reveal that Sentinel-2 multitemporal bare soil composites can be successfully applied in the prediction of soil properties. The setting of basic parameters of composite creation is very complex and challenging and it requires to use exact algorithms for masking clouds and bare soil. Soil moisture and surface roughness also greatly affect spectral characteristics of bare soil and thus a very important aspect of compositing is finding appropriate statistics to derive final pixel values of reflectance (minimum, mean, median, ...). One possible way to minimize the effect of moisture and surface roughness may be incorporation radar backscatter information from Sentinel-1. However, it further complicates the processing of data and makes the composite creation more complex.

The research has been supported by the project no. QK1820389 " Production of actual detailed maps of soil properties in the Czech Republic based on database of Large-scale Mapping of Agricultural Soils in Czechoslovakia and application of digital soil mapping" funding by Ministry of Agriculture.

How to cite: Zizala, D.: Sentinel-2 Multitemporal Bare Soil Composites for predicting soil properties using machine learning methods, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-18564, https://doi.org/10.5194/egusphere-egu2020-18564, 2020.

The status of the Cstock at any position in the landscape is subject to a complex interplay of soil-state factors operating at different scale and regulating conflicting processes resulting either in soils acting as sink or source of carbon.  Since spatial variability is characteristic of large landscape, key drivers of C stock might be specific for subareas compared to those influencing the whole landscape. Consequently, calibrating separately models for subareas (local models) that collectively cover a target area can result in different prediction accuracy and Cstock drivers compared to a single model (global model) that covers the whole area. The goal of this study was therefore to (1) assess how global and local models differ in predicting the litter, mineral soil and total Cstock in Sweden boreal forest, (2) identify  the key variables  in  forest  Cstock prediction and their scale of influence. We here use the Swedish National Forest Soil Inventory (NFSI) database and the digital soil mapping approach to evaluate the prediction performance of the random forest  that is calibrated locally for the northern (N-SE), central (C-SE) and southern (S-SE) Sweden and for the whole Sweden (global model). Models were built by considering (1) only site characteristics which are direct record on plot during NFSI, (2) remotely sensed variables and (3) both site characteristics and remotely sensed variables. Local models are generally more effective for predicting Cstock after testing on independent validation data. Using remotely sensed variables with soil inventory indicates that such covariates have limited predictive strength but that site specific covariates show better explanatory strength for C stocks. The latter also were the main drivers for Cstock both locally and globally. Investment could focus in mapping these influential site covariates which have potential for future Cstock prediction  models.

How to cite: Hounkpatin, O., Stendahl, J., Lundblad, M., and Karltun, E.: Predicting the spatial distribution of C stock in Swedish boreal forest using remotely sensed and site-specific variables , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-22026, https://doi.org/10.5194/egusphere-egu2020-22026, 2020.

The application of digital soil mapping in precision agriculture is extremely important, since an assessment of the spatial variability of soil properties within cultivated fields is essential in order to optimize agronomic practices such as fertilization, sowing, irrigation and tillage. In this context, it is necessary to develop methods which rely on information that can be obtained rapidly and at low cost. In the present work, an assessment is carried out of what are the most useful covariates to include in the digital soil mapping of field-scale properties of agronomic interest such as texture (clay, sand, silt), soil organic matter and pH in different farms of the Umbria Region in Central Italy. In each farm a proximal sensing-based mapping of the apparent soil electrical resistivity was carried out using the EMAS (Electro-Magnetic Agro Scanner) sensor. Soil sampling and subsequent analysis in the laboratory were carried out in each field. Different covariates were then used in the development of digital soil maps: apparent resistivity, high resolution Digital Elevation Model (DEM) from Lidar data, and bare soil and/or vegetation indices derived from Sentinel-2 images of the experimental fields. The approach followed two steps: (i) estimation of the variables using a Multiple Linear Regression (MLR) model, (ii) spatial interpolation via prediction models (including regression kriging and block kriging). The validity of the digital soil maps results was assessed both in terms of the accuracy in the estimation of soil properties and in terms of their impact on the fertilization prescription maps for nitrogen (N), phosphorus (P) and potassium (K).

How to cite: Mzid, N., Pignatti, S., Veretelnikova, I., and Casa, R.: Covariates selection assessment for field scale digital soil mapping in the context of precision fertilization management, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-4904, https://doi.org/10.5194/egusphere-egu2020-4904, 2020.

     Soil organic carbon (SOC) is a key property that affects soil quality and the assessment of soil resources. However, the spatial distribution of SOC is very heterogeneous and existing soil maps have considerable uncertainty. Traditional polygon-based soil maps are less useful for fine-resolution soil maps modeling and monitoring because they do not adequately characterize and quantify the spatial variation of continuous soil properties. And recently, digital soil mapping of organic carbon is the main source of information to be used in natural resource assessment and soil management. In this study, we collected 100 soil samples on a 50 m grid to conduct soil maps of topsoil (0-20 cm) organic carbon in a 500×500m field and evaluate the uncertainty by spatial stochastic simulation. The map of soil organic carbon generated by inverse distance weighting interpolation indicated that the average topsoil SOC is 11.59±0.61g/kg with averaged standard deviation error is 0.61. In order to evaluate the uncertainties, numbers were defined as 50, 100, 200, 500, 1000, 5000, 10000 with interval of 2×2 m to conduct conditional simulation. The standard deviation error gradually declined from 0.74 to 0.51 g/kg. Then, the uncertainty of SOC was expressed as the range of the 95% confidence intervals of the standard deviation error. Maps of uncertainty showed fine spatial heterogeneity even the numbers of simulations reached 10000. Compared with inverse distance weighting interpolation method, conditional simulation approach can improve the fine-resolution SOC maps. For some points, the simulated values deviated from the averaged values while closed to the observed values. On the whole, the maps of uncertainty showed larger waves in the field-edge and different SOC contour border. Consideration of the sample distribution and sampling strategy, the uncertainty map provides a guide for decision-making in additional sampling.

Key words: Soil organic carbon (SOC); uncertainty assessment; conditional simulation; digital soil mapping

Acknowledgements

This material is based upon work funded by National Natural Science Foundation of China (No. 41601213), Major science and technology projects of Henan (171100110600), the Key Science and Technology Program of Henan (182102410024).

How to cite: Guo, Y., Liu, T., Shi, Z., and Wang, L.: Digital soil mapping of organic carbon and it’s spatial distribution uncertainty in field scale, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13202, https://doi.org/10.5194/egusphere-egu2020-13202, 2020.

One of the biggest challenges for digital soil mapping is the limited of field soil information (e.g., soil profile descriptions, soil sample analysis) for representing soil variability across scales. Global initiatives such as the Global Soil Partnership (GSP) and the development of a Global Soil Information System (GloSIS), World Soil Information Service (WoSis) or SoilGrids250m for global pedometric mapping highlight new opportunities but the crescent need of new and better soil datasets across the world. Soil datasets are increasingly required for the development of soil monitoring baselines, soil protection and sustainable land use strategies, and to better understand the response of soils to global environmental change.  However, soil surveys are a very challenging task due to their high acquisition costs such data and operational complexity. The use of legacy soil data can reduce these sampling efforts.

The main objective of this research was the rescue, synthesis and harmonization of legacy soil profile information collected between 2009 and 2015 for different purposes (e.g., soil or natural resources inventory) across Ecuador. This project will support the creation of a soil information system at the national scale following international standards for archiving and sharing soil information (e.g., GPS or the GlobalSoilMap.net project). This new information could be useful to increase the accuracy of current digital soil information across the country and the future development of digital soil properties maps.

We provided an integrated framework combining multiple data analytic tools (e.g., python libraries, pandas, openpyxl or pdftools) for the automatic conversion of text in paper format (e.g., pdf, jpg) legacy soil information, as much the qualitative soil description as analytical data,  to usable digital soil mapping inputs (e.g., spatial datasets) across Ecuador. For the conversion, we used text data mining techniques to automatically extract the information. We based on regular expressions using consecutive sequences algorithms of common patterns not only to search for terms, but also relationships between terms. Following this approach, we rescued information of 13.696 profiles in .pdf, .jpg format and compiled a database consisting of 10 soil-related variables.

The new database includes historical soil information that automatically converted a generic tabular database form (e.g., .csv) information.

As a result, we substantially improved the representation of soil information in Ecuador that can be used to support current soil information initiatives such as the WoSis, Batjes et al. 2019, with only 94 pedons available for Ecuador, the Latin American Soil Information System (SISLAC, http://54.229.242.119/sislac/es),  and the United Nations goals  towards increasing soil carbon sequestration areas or decreasing land desertification trends.  In our database there are almost 13.696 soil profiles at the national scale, with soil-related (e.g., depth, organic carbon, salinity, texture) with positive implications for digital soil properties mapping. 

With this work we increased opportunities for digital soil mapping across Ecuador. This contribution could be used to generate spatial indicators of land degradation at a national scale (e.g., salinity, erosion).

This dataset could support new knowledge for more accurate environmental modelling and to support land use management decisions at the national scale.

How to cite: Armas, D., Guevara, M., Bezares, F., Vargas, R., Durante, P., and Oyonarte, C.: Digital soil mapping: the challenge to obtain the best soil dataset and create a precise environmental model to support land use management at a national level (Ecuador). , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20873, https://doi.org/10.5194/egusphere-egu2020-20873, 2020.