Sustainable land management requires reliable information about soil hydraulic properties. Among these properties, available water-holding capacity (AWC) is a key attribute, as it quantifies the amount of water available for plants that the soil can hold. Since direct measurements of AWC are costly, pedotransfer functions (PTF) are often used to estimate AWC, leveraging statistical relationships with properties that are easier to measure, such as texture, bulk density, and organic carbon content. This study evaluates visible near-infrared spectroscopy (vis-NIR) as an alternative approach to predict volumetric water content at field capacity (FC) and permanent wilting point (PWP) — AWC being the difference between PWP and FC. A suite of 970 vis-NIR soil spectra, recorded from air-dried, 2-mm, sieved soil samples, were associated with FC and PWP analytical data obtained from New Zealand’s National Soils Database. Partial least squares (PLS) regression and support vector machines on PLS latent variables (PLS-SVM) were used for spectroscopic modelling. With root mean squared errors below 7% and 5% for FC and PWP, respectively, our results indicate that vis-NIR spectroscopy can be used to quantitatively predict volumetric water content at FC and PWP.

Precise knowledge about the soil organic carbon (SOC) content in cropland soils is one requirement to design and execute effective climate and food policies. In digital soil mapping (DSM), machine learning algorithms are used to predict soil properties from covariates derived from traditional soil mapping, digital elevation models, land use, and Earth observation (EO). However, such DSM models are trained for a specific dataset and region and have so far only allowed limited general statements to be made that would enable the models to be transferred to different regions. In this study, we test the transferability of SOC models for cropland soils using five different covariate groups: multispectral soil reflectance composites (satellite), soil legacy data (soil), digital elevation model derivatives (terrain), climate parameters (climate), and combined models (combined). The transferability was analyzed using data from two federal states in southern Germany: Bavaria and Baden-Wuerttemberg. First, baseline models were trained for each state with combined models performing best in both cases (R² = 0.68/0.48). Next, the models were transferred and tested with soil samples from the other state whose data were not used during model calibration. Only satellite and combined models were transferable, but accuracy declined in both cases. In the final step, models were trained with samples from both states (mixed-data models) and applied to each state separately. This process significantly improved the accuracies of satellite, terrain, and combined models, while it showed no effect on climate models and decreased the models based on soil covariates. The experiment underlines the importance of EO for the transfer and extrapolation of DSM models.

Artificial neural networks (ANNs) are valuable tools for predicting soil properties using large datasets. However, a common challenge in soil sciences is the uneven distribution of soil samples, which often results from past sampling projects that heavily sample certain areas while leaving similar yet geographically distant regions under-sampled. One potential solution to this problem is to transfer an already trained model to other similar regions. Robust spatial uncertainty quantification is crucial for this purpose, yet this is often overlooked in current research. We address this issue by using a Bayesian deep learning technique, Laplace approximations, to quantify spatial uncertainty. This produces a probability measure encoding where the model’s prediction is deemed to be reliable and where a lack of data should lead to a high uncertainty. We train such an ANN on a soil landscape dataset from a specific region in southern Germany and then transfer the trained model to another unseen but, to some extent, similar region without any further model training. The model effectively generalized alluvial patterns, demonstrating its ability to recognize repetitive features of river systems. However, the model showed a tendency to favour overrepresented soil units, underscoring the importance of balancing training datasets to reduce overconfidence in dominant classes. Quantifying uncertainty in this way allows stakeholders to better identify regions and settings in need of further data collection, enhancing decision-making and prioritizing efforts in data collection. Our approach is computationally lightweight and can be added post hoc to existing deep learning solutions for soil prediction, thus offering a practical tool to improve soil property predictions in under-sampled areas, as well as optimizing future sampling strategies, ensuring that resources are allocated efficiently for maximum data coverage and accuracy.

Artificial Neural Networks (ANNs) are valuable tools for predicting soil properties using large datasets. However, a common challenge in soil sciences is the uneven distribution of soil samples, which often results from past sampling projects that heavily sample certain areas while leaving similar yet geographically distant regions under-sampled. One potential solution to this problem is to transfer an already trained model to other similar regions. Robust spatial uncertainty quantification is crucial for this purpose, yet often overlooked in current research. We address this issue by using a Bayesian deep learning technique, Laplace Approximations, to quantify spatial uncertainty. This produces a probability measure encoding where the model’s prediction is deemed reliable, and where a lack of data should lead to a high uncertainty. We train such an ANN on a soil landscape dataset from a specific region in southern Germany and then transfer the trained model to another unseen but to some extend similar region, without any further model training. The model effectively generalized alluvial patterns, demonstrating its ability to recognize repetitive features of river systems. However, the model showed a tendency to favor overrepresented soil units, underscoring the importance of balancing training datasets to reduce overconfidence in dominant classes. Quantifying uncertainty in this way allows stakeholders to better identify regions and settings in need of further data collection, enhancing decision-making and prioritizing efforts in data collection. Our approach is computationally lightweight and can be added post-hoc to existing deep learning solutions for soil prediction, thus offering a practical tool to improve soil property predictions in under-sampled areas, as well as optimizing future sampling strategies, ensuring resources are allocated efficiently for maximum data coverage and accuracy.

Artificial Neural Networks (ANNs) are valuable tools for predicting soil properties using large datasets. However, a common challenge in soil sciences is the uneven distribution of soil samples, which often results from past sampling projects that heavily sample certain areas while leaving similar yet geographically distant regions under-sampled. One potential solution to this problem is to transfer an already trained model to other similar regions. Robust spatial uncertainty quantification is crucial for this purpose, yet often overlooked in current research. We address this issue by using a Bayesian deep learning technique, Laplace Approximations, to quantify spatial uncertainty. This produces a probability measure encoding where the model’s prediction is deemed reliable, and where a lack of data should lead to a high uncertainty. We train such an ANN on a soil landscape dataset from a specific region in southern Germany and then transfer the trained model to another unseen but to some extend similar region, without any further model training. The model effectively generalized alluvial patterns, demonstrating its ability to recognize repetitive features of river systems. However, the model showed a tendency to favor overrepresented soil units, underscoring the importance of balancing training datasets to reduce overconfidence in dominant classes. Quantifying uncertainty in this way allows stakeholders to better identify regions and settings in need of further data collection, enhancing decision-making and prioritizing efforts in data collection. Our approach is computationally lightweight and can be added post-hoc to existing deep learning solutions for soil prediction, thus offering a practical tool to improve soil property predictions in under-sampled areas, as well as optimizing future sampling strategies, ensuring resources are allocated efficiently for maximum data coverage and accuracy.

Radon (222Rn) is a radioactive gas of the uranium-radium decay chain and occurs naturally in soil air. Primarily by diffusion, radon migrates to the surface and can accumulate in buildings, where it is harmful to human health because of its radioactivity. To tackle this particular health hazard, radon mitigation was recently introduced into national law. Among other things, the law demanded the designation of radon priority areas by the end of 2020. These areas are defined as administrative areas with a large proportion of land affected by high radon concentrations inside buildings. In the state of Baden-Wuerttemberg these areas were selected at municipality level based on a 10x10 km national map of geogenic radon potential (GRP) provided by the German Federal Office for Radiation Protection (BfS) accompanied by a state-specific map of uranium concentrations. With the prospect of future designations, this work aims at replacing the national GRP map used in 2020 by a more sophisticated state-specific version. In doing so, we improve spatial resolution, focus on covariates that only represent the state-relevant features of geology and soil and move away from a kriging-based mapping approach towards machine learning.This ongoing study is currently based on 580 radon measurements in soil gas at 1 m depth from different surveys spread irregularly over Baden-Wuerttemberg in southeast Germany. This point dataset is combined with a set of covariates from factors that influence radon concentrations such as soil parameters, geology, relief, climate and a map of uranium concentrations created from over 4000 heavy metal measurements. Modelling is done using a random forest (RF) approach as implemented in the R packages ranger and mlr3. Preliminary results indicate that the new GRP map with a spatial resolution of 250 m is highly useful in classifying communities as vulnerable areas, which were previously not called due to uncertain underlying data. In addition, model output confirms up to 80 % of already identified radon priority areas.Machine learning algorithms such as RF with its precise learning progressions can be used to create GRP maps at regional scale at high resolution. Besides further improving the RF model, next steps will focus on explainable machine learning, i.e. to produce features that support policy makers in finding acceptance by the public in sight of a sensitive topic. This includes variable importance plots, uncertainty measures and maps representing areas of applicability. The latter will also be used to help guiding the ongoing radon measurement programme of Baden-Wuerttemberg, which currently comprises approximately 100 new locations per year.