Digital Soil Mapping (DSM) is a demonstrated viable approach to generate spatial predictions of soil properties but Digital Soil Assessment (DSA) methods are not widely applied as the translational step from DSM to DSA remains challenging. The purpose, in generic terms, of DSA is the conversion of quantitative data on soil properties obtained through DSM to a spatial assessment of the capacity of a soil to fulfil a particular function. However, the interpretation and value of soil information to users for effective decision making is often qualitative rather than quantitative, expressing the capacity and capability of soil to deliver particular services or perform particular functions. We identify three challenges to implementing DSA for decision making and illustrate these with several case studies: 1) Bridging the gap between quantitative DSM to qualitative DSA.  Soil Quality and Health (SQH) are general terms for indicators that are associated with soil security which are neither easy to define, nor easy to quantify. Through combining the UK national soils datasets, and the SQH Bayesian inference, we were then able to predict SQH for soils across Great Britain.  We show that we are able to describe both aleatoric uncertainty and, equally important, epistemic uncertainty through a description of the experts confidence and through using multiple experts. 2) Cascading the uncertainty generated from DSM into DSA. We demonstrate using a stochastic simulation technique and specified threshold values for soil constraints for crop growth that the uncertainty can be incorporated into the resulting value assessments. We show that DSA can be used to quantify the potential contributions of soil constraints versus socio-economic, farm management and other factors, and the importance of allowing for uncertainties and having appropriate constraint criteria is illustrated by the sensitivity of our constraint estimates to the various criteria we tested. 3) Incorporating temporally dynamic environmental data. We developed a DSA that integrated a dynamic modelling approach to determine land suitability under future climatic variability.  The DSA outputs highlighted where best to grow food in the future based on soil and climate interactions, however decision making needs to address potential trade-offs in other soil services before deciding when and where to protect the best quality land. Although we illustrate, through these examples, that informative and useful spatial data about the soil can be obtained through DSA, in each case the process is elaborate and complex, with significant modelling challenges. Unlike DSM, DSA introduces a value judgement on the soil which can be difficult to capture through quantitative modelling processes.

How to cite: Corstanje, R. and Hannam, J.: Assessing the value of Digital Soil Assessment: a bridge too far?, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3921, https://doi.org/10.5194/egusphere-egu23-3921, 2023.

Sustainable agriculture is seriously threatened by severe soil erosion, which is also occurred in the Neszmély Wine Region, in the northern part of the Gerecse Hills in Hungary. In the region, three vineyards with visible signs of erosion were chosen to quantify the amount of eroded soil. The empirically based Universal Soil Loss Equation (USLE) model was utilized first to determine the soil loss. The study sites were monitored with an unmanned aerial vehicle (UAV) to create high-resolution models of seasonal and annual soil loss. After the empirically based, spatially detailed quantification of erosion, we have tested the applicability of machine learning methods to predict soil erosion for the selected parcels during the same time period. The primary concept was to use the empirically inferred erosion values as observation data to construct parcel-specific prediction models and test them on the remaining two parcels. In the model we have used (i) Sentinel 2 satellite data in the form of both native spectral bands and its derived spectral indices; (ii) terrain features derived from digital surface model created and aggregated from the UAV flights and (iii) formerly elaborated digital soil property maps as auxiliary data. Various machine learning methods (ranger, ridge, xgbLinear, enet, pls, brnn) have been tested to find the best performing predictions. Observation data were generated in the form of random points, in 100 representations. Model performances have been tested by proper measures to evaluate the applicability of the applied machine learning techniques for soil erosion mapping.

Acknowledgment: Our research has been supported by the Hungarian National Research, Development and Innovation Office (NRDI; Grant No: K 131820).

How to cite: Takáts, T., Mészáros, J., Albert, G., and Pásztor, L.: Spatial modelling of vineyard erosion using machine learning methods, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2312, https://doi.org/10.5194/egusphere-egu23-2312, 2023.

In the context of increasing soil degradation worldwide, spatially explicit soil information is urgently needed to support decision-making for sustaining limited soil resources. Digital soil mapping (DSM) has been proven as an efficient way to deliver soil information from local to global scales. The number of environmental covariates used for DSM has rapidly increased due to the growing volume of remote sensing data, therefore variable selection is necessary to deal with multicollinearity and improve model parsimony. Compared with Boruta, recursive feature elimination (RFE), and variance inflation factor (VIF) analysis, we proposed the use of modified greedy feature selection, namely forward recursive feature selection (FRFS), for DSM regression. For this purpose, using quantile regression forest, 402 soil samples and 392 environmental covariates were used to map the spatial distribution of soil organic carbon density (SOCD) in Northeast and North China. The result showed that FRFS selected the most parsimonious model with only 9 covariates (e.g., brightness index, mean annual temperature), much lower than RFE (22 covariates), VIF (30 covariates), and Boruta (76 covariates). The repeated validation (50 times) showed that the FRFS derived model performed better than these using full covariates, Boruta, RFE and VIF. Despite the similar performance of the uncertainty estimate (PICP), the model using FRFS and RFE had the lowest global uncertainty (0.86) as indicated by the uncertainty index. In addition, FRFS had the best computation efficiency when considering the steps of variable selection and map prediction. Given these advantages over Boruta, RFE and VIF, FRFS has a high potential in fine-resolution soil mapping practices, especially for these studies at a broad scale involving heavy computation on millions or billions of pixels.

How to cite: Chen, S., Zhang, X., Xue, J., Wang, N., Xiao, Y., Shi, Z., Richer-de-Forges, A., and Arrouays, D.: Developing parsimonious model for digital soil mapping using forward recursive feature selection, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1497, https://doi.org/10.5194/egusphere-egu23-1497, 2023.

Spatial predictions for mapping soil properties are often prone to smoothing of distribution tails. As a result small values are overestimated and large values are underestimated. For many applications there might be not harm, but it is critical for map uses where soil property interpretation for small or large values have a substantial effect. For example, soil texture maps are relevant to implement irrigation strategies or to adjust irrigation soil moisture probes. Texture at the margin of the distributions have a much larger impact on probe adjustment than intermediate textures.

To investigate the effect of different statistical approaches on smoothing of soil texture prediction we analyzed four Swiss data sets originating from different surveys with different strength of response-covariate relationships (weak: arable land north of Berne, n = 1650; weak to medium: arable land of Canton of Zurich, n = 3920; medium: strongly cultivated Gleysols and Histosol in Seeland/Grosses Moos, n = 2510; strong: cultivated Histosols in Rhine Valley, n = 2590). We evaluated behavior around lower and upper tails of predicted distributions for commonly used methods fitted to clay, silt and sand and to additive log-ratio transformed responses: random forest, gradient boosted trees, support vector machines, Cubist regression, k-nearest neighbor, robust external-drift kriging and group lasso. In addition, we applied approaches that are supposed to alleviate the problem of smoothed predicted distributions: 1) post-processing transformation to match original variance, 2) SMOTER algorithm for imbalanced regression, 3) constrained kriging, 4) random forest with resampling weights inverse to histogram, 5) univariate distributional random forest with a distribution loss criteria and a 6) multivariate response variant of distributional random forest.

Validation was done by surveying a design-based dataset for Rhine Valley and Berne and by data-splitting otherwise. Besides computing validation statistics of mean model performance, we evaluated goodness-of-fit of univariate and multivariate distributions. Further, we judged the multivariate accuracy regarding HYPRES and USDA texture classes often used within irrigation applications.

The comparative analysis of the used methods showed that no approach outperformed the others on all datasets regarding mean overall accuracy and at the same time satisfactory prediction of tails. Sometimes random forest with inverse histogram weights was resulting in slightly better predictions at the tails, but it was closely followed by an unaltered random forest. Hence, producing proper prediction along the full distribution of a response remains a challenge.

How to cite: Nussbaum, M., Vogel, S., Oechslin, S., Tanner, S., and Burgos, S.: Smoothed predicted distributions in digital soil mapping – a comprehensive comparative study to predict soil texture for irrigation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5543, https://doi.org/10.5194/egusphere-egu23-5543, 2023.

Although improved and more effective approaches for predicting the spatial distribution of soils have long been developed, further study in this field is still required. This study proposes an algorithm for generating terrain attributes over several scales and automatically selecting the optimal scale for each predictor using the powerful Random Forests (RF) method, to increase the accuracy of soil property maps. The effectiveness of using optimal scaled predictors to improve the accuracy of soil maps is investigated on nine soil properties (clay, silt, and sand content within the 0-20 cm range, soil porosity within the 0-20 cm range, subsoil porosity, pH within the 0-20 cm range, edaphic volume, humus reserve, and base saturation within the 0-20 cm range). Experiments were carried out in two study areas in western Romania, located along the boundary between the Western Plain and the Western Hills. The first study area contains 96 georeferenced soil profiles, while the second has 92. The initial 12.5 m digital elevation model (DEM) was resampled to 25 m, then in 25 m increments to 1000 m, resulting in 40 coarser versions of the DEM. Each rescaled version of the DEM was used to derive 10 terrain attributes, resulting in 40 rescaled versions of each terrain attribute. Next, a Random Forest (RF) and a linear correlation model with each scaled terrain attribute were created using soil property values. The highest R-squared value and correlation coefficient, respectively, are used by the script to produce two sets of optimally scaled terrain attributes. All multiscale predictors, optimally scaled predictors based on the RF model, optimally scaled predictors based on the correlation coefficient, and original not scaled predictors were the four groups of predictors used to map each soil attribute. The results showed that when the predictors are optimally scaled compared to maps produced with the original unscaled predictors or with all multiscale predictors, more accurate and less uncertain soil property maps are obtained.

How to cite: Dornik, A., Cheţan, M. A., Drăguţ, L., Iliuţă, A., and Dicu, D. D.: Digital mapping of soil properties with optimally scaled predictors, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8450, https://doi.org/10.5194/egusphere-egu23-8450, 2023.

The European Joint Programme Cofound on Agricultural Soil Management has set the goal of harmonizing national soil databases with the continental LUCAS topsoil database for the purpose of monitoring soil carbon, fertility and land degradation. Up-to-date soil surveys have poor spatial resolution and are very cost- and labour intensive, so harmonizing the existing datasets allows us to create more accurate soil maps with better resolution. The LUCAS topsoil database can be used to complement the Hungarian Soil Information and Monitoring System (SIMS) for the creation of more detailed soil-property maps. Due to the different laboratory analysis methods and sampling strategies used by the two databases, conversion was needed, so the datasets can be directly compared to each other and used together for digital soil mapping. Two sets of topsoil data were used from 2009 and 2015 from both LUCAS and SIMS respectively, so the swiftly changing soil properties (pH, CaCO3, OC, P, K) were available for a more accurate side-by-side comparison. Map products were created for the chemical soil properties of both the LUCAS and SIMS database with the ancillary data of 28 environmental covariates using random forest kriging with 10-fold cross-validation. The spatial resolution of the maps was 100 x 100 m. The raster maps were compared directly to each other using linear regression. In conclusion the results show that the LUCAS and national soil databases can and should be harmonized, merged and used together for creating more accurate soil maps with better spatial resolution at national and continental scale.

How to cite: Benő, A., Kocsis, M., Szatmári, G., Laborczi, A., Bakacsi, Z., and Pásztor, L.: Harmonization and comparison of soil chemical properties of the LUCAS topsoil database and the Hungarian National Soil Monitoring System, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15268, https://doi.org/10.5194/egusphere-egu23-15268, 2023.

The spatial distribution of soil organic carbon (SOC) serves as critical geographic information for assessing ecosystem services and guides land management for migrating carbon emissions. Digital mapping of SOC is challenging due to the complex relationships between the soil and its environmental conditions. Except for the well-known topography and climate environmental covariates, the aboveground vegetation growth, which interacts with belowground soil carbon, influences SOC significantly over seasonal and interannual variations. Although several remote-sensing-based vegetation indices (e.g. NDVI and EVI) have been widely adopted in digital soil mapping, variables indicating long-term vegetation growth status have been less used. The vegetation phenology, an indicator of vegetation growth characteristics, can be used as a potential time series environmental covariate for SOC prediction. In this study, a CNN-RNN hybrid model was developed for SOC prediction with inputs of static and dynamic environmental variables in a study area located in Xuancheng City, China. The spatially contextual features in static variables (e.g., topographic variables) were extracted by the convolutional neural network (CNN), while the temporal features in dynamic variables (e.g., vegetation phenology over a long period) were extracted by a recurrent neural network (RNN) as represented by using a long short-term memory (LSTM) network. The ten-year phenological variables before the sampling year derived from satellite-based observations were adopted as new predictors reflecting historical temporal changes in vegetation in addition to the commonly used static variables. The random forest model was used as a reference model for comparison. Our results indicate that adding phenological variables can improve the soil carbon prediction accuracy, and demonstrate that the fine-tuned CNN-RNN model is potentially effective and can be a powerful model for SOC predictive mapping. We conclude that the hybrid deep learning models have great potential to enhance soil prediction by simultaneously extracting spatial and temporal latent features from different types of environmental variables, and highlight that using the long-term historical vegetation phenology information can serve as a useful extra input for future applications in the predictive mapping of soil carbon.

References

Zhang, L., Cai, Y., Huang, H., Li, A., Yang, L., Zhou, C., 2022. A CNN-LSTM Model for Soil Organic Carbon Content Prediction with Long Time Series of MODIS-Based Phenological Variables. Remote Sensing 14, 4441
Yang, L., Cai, Y., Zhang, L., Guo, M., Li, A., Zhou, C., 2021. A deep learning method to predict soil organic carbon content at a regional scale using satellite-based phenology variables. International Journal of Applied Earth Observation and Geoinformation 102, 102428
He, X., Yang, L., Li, A., Zhang, L., Shen, F., Cai, Y., Zhou, C., 2021. Soil organic carbon prediction using phenological parameters and remote sensing variables generated from Sentinel-2 images. CATENA 205, 105442.

How to cite: Zhang, L., Yang, L., and Zhou, C.: Enhancing predictive mapping of soil carbon by incorporating vegetation growth dynamic information via deep learning, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1878, https://doi.org/10.5194/egusphere-egu23-1878, 2023.

For site-specific estimation of lime requirement, high-resolution soil maps of clay, soil organic carbon (SOC) and pH are required. These can be generated using digital soil mapping (DSM), in which prediction models are fitted on covariates from proximal soil sensors. However, the quality of the maps derived may differ significantly depending on the methodology applied. Hence, we assessed effects of (i) calibration sample size (5-100), (ii) sampling design (simple random sampling (SRS), conditioned Latin Hypercube sampling (cLHS) and k-means sampling (KM)) and (iii) prediction model (linear regression (LR) and Random Forest (RF)) on the prediction performance for the above mentioned three soil properties using data from two multi-sensor platforms. The present case study is based on a geostatistical simulation using 250 soil samples from a 51 ha field in Germany. Among others, Lin’s concordance correlation coefficient (CCC) and root-mean-square error (RMSE) were used to evaluate model performances. Results show that with increasing sample size, improvements of RMSE and CCC decreased exponentially. We found best median RMSE values at 100 calibration soil samples, i.e. 1.73%, 0.3 and 0.21% for Clay, pH and SOC, respectively. However, already with 10 samples, models of moderate quality (CCC > 0.65) can be obtained for all three soil properties. Both, cLHS and KM obtained significantly better results than SRS. At smaller sample sizes, LR showed lower median RMSE values than RF for SOC and pH. Nonetheless, with at least 75-100 and 25-30 samples, RF eventually outperformed LR. For clay, median RMSE was lower with RF, regardless of sample size.

How to cite: Vogel, S., Schmidinger, J., Schröter, I., Bönecke, E., Rühlmann, J., Kramer, E., Mulder, T., Heuvelink, G., and Gebbers, R.: Effect of sample size, sampling design and calibration model on generating soil maps from proximal sensing data for precision liming, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5877, https://doi.org/10.5194/egusphere-egu23-5877, 2023.

Multinomial Logistic Regression (MNLR), which is a linear and simple classification algorithm, the probability of each pixel belonging to a class can be calculated and the most probable classes and ground realities are compared in digital soil mapping. However, Random Forest (RF) algorithm, which is a relatively complex classification algorithm that can also discover non-linear relationships, the most probable classes can be determined by measuring the proportion of votes for each class, which it calls estimates of class probabilities in each pixel. In the current study, we used the map with eight FAO-WRB second level soil classes as a result of detailed soil survey mapping in an area of approximately 10,000 ha. We determined the data set points from the mapping units with the area-weighted sampling methodology. Digital soil class maps generated using the two classification algorithms and twenty-three variables representing parent material, organism and topography generated from the digital elevation model and Landsat 7 ETM satellite images. Classification accuracies measured using the confusion matrix. Overall accuracy calculated in training and validation set for MNLR, 52% 48%;  for RF, 48% and 55%, respectively. In general, machine learning algorithms try to minimize the misclassification error and thus the error in all classes is equally important. However, in soil science, the most probable and the second most probable class probabilities produced as a result of these two classification algorithms are important. Thus, confusion index (CI), which is calculated by considering the probability values of the most probable class and the second most probable class, in the training and validation sets of each classification algorithm. Mean CI values were calculated in training and validation set as 0.73 and 0.75 for MNLR; for RF 0.36 and 0.77, respectively. As the CI approaches 0, CI informs us that the most probable class strongly belongs to the class to which it is allocated. Furthermore, there is no high difference between the two models in the training and validation sets, according to the confusion matrix results. However, in the confusion index, there is a 50% difference between the mean confusion index values of the training and validation sets for the RF algorithm. CI maps created are produced according to the model established with the training set, therefore, visual interpretation and pedologist knowledge should be integrated. Accordingly, both classification algorithms failed to digital map the Chromic Cambisol class in the study area. This soil class can be determined in the field according to a subsurface chroma value and has been difficult to capture by our environmental covariate set. We suggest that in addition to giving general accuracy values in the production of any digital soil classes map, the calculation of the confusion index values and their interpretation with pedological information.

How to cite: Kaya, F. and Başayiğit, L.: Digital mapping of WRB soil classes using linear and non-linear classification-based machine learning algorithms and integration of confusion index in knowledge discovery, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-510, https://doi.org/10.5194/egusphere-egu23-510, 2023.

ESTIMATING THE SOIL AGGREGATES SIZE FRACTIONS IN ARID AND SEMI-ARID ENVIRONMENTS USING PEDOTRNASFER FUNCTIONS AND SPECTRAL ANALYSIS

This paper investigates the relationship of six aggregation fractions (2-1.4mm F1; 1.4-1.0 mm F2; 1.0-0.5 mm F3; 0.5-0.25 mm ; F4; 0.25-0.1mm ; F5, 0.1mm > F6) and their average (AVG) with relation to 74 soil attributes.  The database used in this study is based on the heritage soil spectral library (SSL) of Israel over both semi-arid and arid regions representing seven global soil orders from the  USDA list. The database is composed of chemical, physical and spectral measurements. The spectral data consisted of reflectance measurement using the IEEE  standard protocol and across the VIS-NIR-SWIR spectral region as well as XRF analyses of microelements and XRD determination, wet chemistry analysis and quantitative assessment of the soil mineralogy. The correlation matrix between all attributes enables to isolate four cementation agents (CAs) of the soil micro aggregation namely: clay content, clay mineral specious (smectite), organic matter, and free iron oxides contents. Generating a Pedo Transfer Function (PTF) using these CAs revealed equations that well predict the aggregate size fraction of F1,F2,F3, and the average aggregation stage (AVG ) of all soils. A separate spectral-based analysis to evaluate directly the fraction sizes from reflectance measurements without any prior information regarding the CAs, was also generated and revealed high statistic performance with spectral assignments that belonged to the four selected CAs and their derivatives. The final conclusion was that the micro aggregation stage of soil can be assessed directly or indirectly via PTF and also by using spectral analysis and data mining approaches. Assuming that the reflectance information from hyperspectral remote sensing means such as EMIT (NASA initiative) PRISMA (ASI) and EnMAP (DLR) are available today (2022) this approach may add a great deal to the NASA  mission aims at investigating the potential of soil to act as a dust source.

How to cite: Ben-Dor, E., Francos, N., and Ogen, Y.: Estimating of the  Soil Aggregate Size Fraction in Arid and Semi-Arid Soils using Reflectance Spectroscopy, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-79, https://doi.org/10.5194/egusphere-egu23-79, 2023.

Improved understanding of vegetation impacts on soil strength could improve erosion and landslide assessment, stabilization and resilience efforts, and off-road vehicle mobility efforts while minimizing damage to vegetation or soil surfaces. It has been shown that belowground biomass, along with soil moisture and type, influences soil strength. Standard handheld geotechnical instruments, such as cone penetrometers, vane and torvane, typically used for measurements of soil strength were developed for civil engineering projects. These devices have been applied to studies in natural areas to determine shear strength profiles, namely wetlands, stream banks, slopes, and coastal dunes. However, in cases where soils are soft or complex, can be saturated, or where vegetation exists, such as coastal areas or wetlands, traditional geotechnical instruments provide uncertain and highly variable results, can miss vegetation contributions to soil strength, and are difficult to compare to one another. A few efforts exist to address these issues. In the case of wetlands, Sasser et al (2018) developed a Wetland Soil Strength Tester (WSST), which measures the torque required to shear the combined soil-vegetation wetland matrix using a four-pin design inserted into the wetland soil 15 cm. In our ongoing work to use remote sensing to map soil and vegetation properties, even when soil is obscured by vegetation, we found that the geotechnical measurements insufficiently capture soil strength in the presence of vegetation. We adapted the WSST design and systematically tested against traditional geotechnical measurements broadly (variety of vegetated terrain, mostly grasses and shrubs, ranging from low to high heights), not necessarily just wetlands, to investigate the utility of the measurements in applications where soil strength is a primary parameter. Soil types included sand, loam, and clay, and each were tested in wet and dry conditions. Measurements of peak torque were recorded at intervals of 90 degrees up to four revolutions with and without surface vegetation present to account for more soil properties than shear failure. Results of the analyses showed that the method can account for a greater variation in initial shear strength than the cone penetrometer-based cone index in soft to medium hard soils, such as wet and dry sands with vegetation and moist to wet clays. Analyses of torque-based measurements beyond initial shear, greater than 90 degrees, revealed a non-uniform and non-linear change in force required as sheared vegetation-soil matrix is further manipulated. Hard soils, especially dry clays were beyond the maximum limits of torque measurements and did not add valuable information beyond traditional geotechnical measurements. Based on these results, a preliminary model of the relationship between the torque-based measurement and the cone index was developed, terrain maps were enhanced, and further applications are being analyzed.  The torque weighted index soil strength test techniques have potential application in stability, erosion, and mobility studies as well as our ongoing research in using remote sensing for indirect inferences of soil properties in vegetated areas.

How to cite: Abelev, A., Merrick, T., Liang, R., Vermillion, M., Li, R.-R., Buma, W., and Swanson, C.: Torque-Weighted Index Soil Strength Test (TWISST) for vegetated terrain, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8418, https://doi.org/10.5194/egusphere-egu23-8418, 2023.

Soil scientists, farmers, and policy makers are seeking cost-effective alternatives for measuring, monitoring, and verifying changes in soil carbon as a result of land use change and adoption of sustainable management practices. The increasing interest is in line with the need to establish protocols for accurate representation of changes and variability in soil carbon, especially for sustainability metrics and carbon markets. As conventional soil sampling and laboratory analysis schemes are costly and labour-intensive at large scales, remote sensing-based approaches are promising. Digital soil mapping, which uses satellite images as explanatory variables in regression models, is increasingly used primarily to estimate and monitor changes in soil carbon over time and spatially. This study is intended to demonstrate how satellite images coupled with other environmental covariates and field measurements of soil carbon can be used to build models to explain soil carbon variability and change. The study was conducted at the North Wyke Farm Platform, an experimental farm located in Devon, in which different land-uses and management systems including permanent pasture, high-grass sugar and arable crops have been monitoring since 2012. Soil samples were collected in 2012 in a regular grid scheme and analysed for soil carbon, bulk density and other parameters. Subsequent measurements were also carried out in 2016, 2018, 2019, 2020, and 2021. Soil carbon values were then related to the spectral reflectance of Landsat-8 and terrain variables to build a baseline prediction model. A gradient boosting algorithm, which was parameterised to find the best parameters that minimise the square error of the prediction, was used. After the model was trained and tested for the baseline year, it was applied to satellite images and terrain attributes of the subsequent years and maps of soil carbon was obtained for each year. The predicted values for each year were compared with measured values in each field. Our results demonstrated that Landsat-8 images coupled with terrain attributes were able to explain 33% of the changes in soil carbon observed between 2012 and 2021. Future research is needed to improve these estimates and take full advantage of remote sensing and machine learning models in monitoring, measuring, reporting and verification of soil carbon stocks in agricultural systems.

How to cite: Valadares Galdos, M. and E. Q. Silvero, N.: Are remote sensing images accurate enough to detect changes in soil carbon over time? An example from the North Wyke Farm Platform, UK, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8597, https://doi.org/10.5194/egusphere-egu23-8597, 2023.

Soil organic carbon (SOC) is a crucial component of soil and is used as a proxy for soil health and fertility. SOC influences water and nutrient holding capacity, nutrient cycling and stability, and water infiltration and aeration properties of soil. Proximal Sensing and Remote sensing are two powerful tools well covered in literature used for quantitative analysis of SOC. It is difficult to examine the nutrients of the soil due to insufficient time for collecting soil samples from agricultural fields during the crop rotation process in countries like INDIA, where extensive agriculture is practiced. In this scenario, linking soil spectral libraries (SSL) developed from proximal sensing with RS image data would enable instantaneous estimation of SOC over a large command area.

A model (calibration using multivariate techniques) which contains the variability of the target site soils should be constructed in this process to extract useful information. However, many times this criterion is not easy to fulfil. To solve this problem, we propose building a soil spectral library using soil samples synthesised in lab conditions and further constructed SSL to estimate SOC and essential soil nutrients. In this regard, spectrum data of collected Alluvial soil samples were generated in lab conditions using ASD FieldSpec 4, while simultaneously analysing soil samples based on standard methods in soil science. A soil sample collected from the field was selected as the master sample, and sub-samples were prepared by combining soil with organic fertiliser and chemicals (spatial structure similar to soil compounds) of various compositions. The emissivity spectra of soil sub-samples were used to construct a spectral library and later used in the machine learning model to estimate the spatial variation of SOC utilizing space-based hyperspectral image (PRISMA). The results revealed that the proposed model for SOC estimation using spectral library is significant for the instant estimation of SOC. Future scope includes testing the approaches' capability for estimating essential soil nutrients in various soil origins.

How to cite: Bonthu, S. R. and Hassan Rangaswamy, S.: Soil Organic Carbon Estimation in Croplands by Integrating Soil Spectral Library and PRISMA Data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11445, https://doi.org/10.5194/egusphere-egu23-11445, 2023.

Hyperspectral images of new generation satellites, such as the PRISMA system of the Italian Space Agency (ASI), offer an important advantage for monitoring topsoil properties from the field to the regional scale. Although numerous studies about the prediction of soil features by proximal soil spectroscopy have been carried out, the hyperspectral remote sensing of soil features still shows several limits and difficulties. Disturbance of soil surface, namely grass or crop cover, surface stoniness, roughness, and soil moisture, influence the results and need corrections. In addition, the resolution of satellite hyperspectral images is rather low (PRISMA = 30 m) for many objectives.

The aim of this work is to test geostatistical techniques to group proximal electromagnetic induction and remote hyperspectral data to increase the resolution and the accuracy of topsoil texture prediction. Two types of croplands have been used for this study: i) JOL- an arable field in Northern Italy (Jolanda di Savoia, Ferrara) of about 15 ha; ii) BRO- seven vineyards of a winery in central Italy (Brolio castle, Siena), for a total surface of about 30 ha.

In JOL, three dates of PRISMA images were selected: BARE- during the bare soil period (14/2/2021), VEG1- during the summer crop, namely corn (4/6/2021), and VEG2- during the winter crop, namely wheat (30/4/2022), to obtain information about the soil surface and the response of vegetation. In BRO, only two dates of PRISMA images with scarce cloud cover were available, 18/12/2020 and 01/12/2022. In both the dates, the grapevines have no leaves and the interrow was tilled by chisel plow 2-3 weeks earlier. The weeds partially covered the grapevine rows and inter-rows.

To reduce the dimensionality of the hyperspectral data and to preserve as much as possible their information content, a principal component analysis (PCA) of the spectra extrapolated from each image pixel was carried out. The first PCs (PC1) explained most of the variance of the images, therefore, it was selected for analysis. Regarding proximal electromagnetic induction, apparent electrical conductivity of shallower depth (ECa₁, about 0-50 cm) has been used. To predict topsoil spatial variation, two geostatistical methods have been tested: i) Regression Kriging (RK), forward stepwise for p < 0.5, and ii) Multiple Geographically Weighted Regression (GWR) with gaussian weighting function.

In the study field of arable land (JOL), 36 samples were used for model calibration and cross-validation. The prediction of clay and sand was unsuitable because the spatial correlation with ECa₁ or PRISMA images was lacking. On the other hand, SOC showed spatial correlation with BARE-PC1, whereas pH with VEG2-PC1.

In the vineyards, a calibration dataset of 70 points and a validation dataset of 20 points have been used. Clay prediction showed the best results using RK with ECa₁ and PC1_2020, providing R² of 0.68 and RMSEP of 6.43 g·100g^-1. Sand prediction showed slightly better results using GWR (R² of 0.78 and RMSEP of 8.74 g·100g^-1), although PC1 of hyperspectral data did not show clear improvements in prediction. Further analyses will include additional PRISMA images acquired during the growing season.

How to cite: Priori, S., Marrone, L., and Casa, R.: Hyperspectral PRISMA images and geophysical proximal sensing data fusion to map topsoil features: vineyard and arable land case studies, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11567, https://doi.org/10.5194/egusphere-egu23-11567, 2023.

Monitoring of soil quality and its degradation is essential to face the big challenges of food security and climate change. One of the main soil parameters to observe is the content of soil organic carbon (SOC) which is linked to soil fertility and other ecosystem services. We investigate the potential of spaceborne hyperspectral images to estimate the content of SOC in the uppermost soil layer. Therefore, we use the spectral information of bare soil pixels in multiple PRISMA images together with chemically analyzed SOC contents of a range of local soil samples and a large-scale soil spectral library. The study site is located in the North-East of Germany within the long-term observatory of TERENO-NE and near the village of Demmin. We compare different machine learning and regression algorithms (Partial Least Squares Regression, Random Forest, Gaussian Process Regression, spectral SOC indices) for each of the images separately and for a synthetic multitemporal image. The best performing models are applied to all bare soil pixels to produce SOC content maps. The preliminary results show medium to high quality models for most cases. With an increasing number of hyperspectral satellites in orbit the outcomes of this case study can provide valuable information for future SOC mapping and monitoring.

How to cite: Ward, K., Foerster, S., and Chabrillat, S.: Mapping soil organic carbon with hyperspectral spaceborne images, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12145, https://doi.org/10.5194/egusphere-egu23-12145, 2023.

Variable rate irrigation (VRI) shows attractive cost/benefit ratios, as compared to drip irrigation, when implemented in large orchards with suitable field and planting geometries. To evaluate the effectiveness of apparent electrical conductivity (ECa), elevation (Z), topographic wetness index (TWI) and time-series of Sentinel-2 NDVI imagery for delimiting VRI zones in olive groves, a study was conducted in a 40-ha commercial plot in southern Spain (Ecija, Seville), equipped with a linear move sprinkler irrigation system and with trees on a 4 × 7 m grid. Soil samples were collected at 140 points on a regular grid with depth intervals of 0.3 m down to 1.2 m and analyzed for soil texture. Relationships between soil texture, topography, ECa and NDVI were analyzed using correlation analysis and regression trees. Time series of correlation between ECa and NDVI showed a seasonal pattern because of the growth-decline pattern of the grass soil cover. The regression tree analysis showed that ECa and elevation were most relevant for classifying NDVI (R²=0.70). Fuzzy k-means classification using ECa+Z yielded 4 classes while for ECa, ECa+Z+TWI and ECa+Z+TWI+NDVI 2 classes were obtained. The zoning based on ECa+Z classified clay content and the 0.95 percentile NDVI successfully. This classification was adopted for VRI since the involved variables can be related to soil water availability. Confounding effects of coarse fragments and soil water content on the clay-ECa relationship could be resolved in future studies by measuring theses variables to improve further the classification.

Acknowledgement

This work is funded by the Spanish State Agency for Research through grants PID2019-104136RR-C21 and PID2019-104136RR-C22/AEI/10.13039/501100011033 and by IFAPA/FEDER through grant AVA2019.018.

How to cite: Vanderlinden, K., Martínez García, G., Ramos Rodríguez, M., and Mateos Iñiguez, L.: Effectiveness of Sentinel-2 imagery, apparent electrical conductivity and topography for delineating site-specific management zones in an olive grove in southern Spain., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12368, https://doi.org/10.5194/egusphere-egu23-12368, 2023.

Cropland soil organic carbon (SOC) is key to maintain soil fertility for plant growth and mitigating climate change by storing considerable amount of organic carbon. Accurate mapping of cropland SOC is essential for improving soil management in agriculture and assessing the potential of different strategies aiming at enhancing regional carbon sequestration. Digital Soil Mapping represents an intermediate approach between labor-intensive soil measurement survey and uncertain SOC modelling. However, most of the widely-used environmental predictors employed in current cropland SOC mapping describe the natural conditions. Indeed, anthropogenic activities, particularly agricultural management practices have profound impacts on agricultural soils, but have rarely been considered in previous research on SOC digital mapping.
Here, we filled that gap by incorporating within the Extreme Gradient Boosting (XGBoost) model several key cropland management practices including carbon input, length of cultivation, and irrigation as management covariates, together with natural variables in order to predict the spatial distribution of cropland SOC in a traditional agricultural area in the Tuojiang River Basin, China. This approach revealed the dominant role of carbon input in explaining SOC variation in this intensively cultivated areas, followed by elevation and soil pH. Adding cropland management practices to natural variables improved prediction accuracy, with the coefficient of determination (R²), the root mean squared error (RMSE) and Lin’s Concordance Correlation Coefficient (LCCC) improving by 16.67%, 17.75% and 5.62%, respectively. Our research highlights the necessity of considering cropland management practices alongside environmental predictors in order to provide more reliable prediction of cropland SOC. We conclude that the construction of spatio-temporal database of agricultural management practices is a research priority as it has a very strong potential, not only to provide accurate digital SOC maps when incorporated within XGBoost model, but also to better initialize the SOC stocks in process-oriented model, such as Dynamic Vegetation Models and Earth System Models.

How to cite: Wang, Q., Le Noë, J., Li, Q., Lan, T., Gao, X., Deng, O., and Li, Y.: Incorporating agricultural practices in digital mapping improves prediction of cropland soil organic carbon content: the case of the Tuojiang River Basin, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2225, https://doi.org/10.5194/egusphere-egu23-2225, 2023.

A methodology of quantitative assessment of soil biomass productivity with 100 m spatial resolution on a country coverage is presented. The traditional land evaluation approach - where crop yield is the dependent variable - was followed using measured yield and net primary productivity data derived from satellite images, together with digital soil and climate maps. Further to characterization of soil biomass productivity based on measured data, the weight of soil properties in productivity was also quantified, thus providing an information basis to design sustainable land management practices. To produce these results, we used the Random Forest method for our calculations. The study considers high-input agriculture, which is predominant in the country.  Biomass productivity indices for major crops (wheat, maize and sunflower) as well as general productivity indices were computed for the whole agricultural area of Hungary. The assessment can be repeated for monitoring purposes to support general monitoring goals as well as for reporting in relation to the Sustainable Development Goals of the United Nations. Nevertheless, based on the findings we also propose a method that enables the periodical update of the evaluation, that can also be used as monitoring biomass productivity in the context of climate change, land degradation and the advancement of cultivation technology.

How to cite: Csikós, N., Szabó, B., Hermann, T., Laborczi, A., Matus, J., Pásztor, L., Szatmári, G., Takács, K., and Tóth, G.: Cropland productivity evaluation based on earth observation and direct measurements, a 100m resolution country assessment, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4318, https://doi.org/10.5194/egusphere-egu23-4318, 2023.