ABSTRACT The concept of soil is increasingly being examined within a planetary context, prompting renewed attention to the physical conditions that enable surface materials to support life. In parallel, planetary habitability has emerged as a framework for defining the minimum resources and conditions required for biological activity, independent of whether life is currently present. Here, we apply habitability principles to soils not to assert the presence of life, but to distinguish clearly between soil formation, biological viability, and functional performance. Habitability formalises a binary set of physical and chemical requirements—including accessible energy, liquid water, essential elements, and viable environmental conditions—that define whether an environment can support life in principle. By treating habitability as a state that applies to already formed soils, rather than as a defining criterion for soil existence, we resolve long‐standing ambiguity around the role of biology in soil definition. This separation leads to a tightened planetary definition of soil as ‘an organised planetary surface system, comprising generally loose mineral and/or organic material, formed through sustained genesis via water‐mediated surface coupling that produces internal physical, chemical, and/or biological organisation and enables system evolution through time, irrespective of the presence of extant life.’ Within this framework, soil health, soil quality, and soil security operate as continuous descriptors of function only within habitable or potentially habitable soils, while degradation is conceptualised as a contraction of habitable state space. Habitability thus provides a physically grounded foundation for interpreting soil function, selecting indicators, and differentiating soils from sediments and regolith across Earth and other planetary environments.

Visible and near-infrared (vis-NIR) spectroscopy enables rapid and cost-effective soil characterization, supporting the estimation of soil chemical, physical, and biological properties. Most applications, however, focus on using spectroscopy to predict individual soil properties and augment conventional laboratory analyses, while the potential of spectral information to study soil variation in space remains largely unexplored. Latent variables derived from the dimensionality reduction of spectral data offer a compact representation of soil variability, capturing multiple soil properties simultaneously. When spatially predicted, these latent variables can provide a means to investigate soil-landscape relationships in a spatially explicit context.This study aimed to (i) model the spatial variation of latent variables derived from compressed topsoil vis-NIR spectral data across Denmark at 10 m resolution using a digital soil mapping approach; (ii) identify the drivers, in relation to SCORPAN factors, controlling their spatial variation; and (iii) examine how the predicted latent variables represent the soil-landscape relationships across Denmark. An earlier study compressed the data using principal component analysis, kernel principal component analysis, shallow autoencoders, and convolutional autoencoders. To predict the latent variables in space, we used 32 predictors comprising harmonized environmental layers representing climate, relief, and parent material. A weighted machine learning algorithm was used to model each latent variable, with bootstrap resampling providing pixel-level uncertainty estimates. Overall, climate was the main driver across all methods, followed by topography and parent material, mainly the extent of clay till deposits.By integrating compressed spectral data with diverse spatial predictors, we captured complex, non-linear soil-landscape relationships. The resulting high-resolution maps offer a spatial description of soil variation in relation to soil-forming factors and spectral signature. Our maps can be used to improve point-based predictions of soil properties or provide enhanced covariates to support digital soil mapping of soil properties and classes.

Soil reflectance spectroscopy in the vis–NIR domain is a high-throughput technique for estimation of soil properties, but soil spectra are high-dimensional and collinear, often requiring compression into a lower-dimensional space for modeling and digital soil mapping. Principal component analysis (PCA) is commonly used for this purpose. With increasing data and computational resources, nonlinear methods such as kernel PCA (KPCA), autoencoders (AE), and convolutional autoencoders (CNN-AE) are increasingly applied. However, how different compression methods affect the preservation of soil-spectral information across land uses and wavelength regions is less explored. In this study, we compared PCA, KPCA, AE, and CNN-AE using 6,238 soil samples representing diverse land uses and soil types across Denmark. Nonlinear methods consistently outperformed PCA in reconstruction error, which is a measure of information loss during compression. At low dimensionalities (5–10 variables), AE and CNN-AE reduced reconstruction error by up to 60% relative to PCA, and by about 54% at 20 latent variables. The largest performance gap between PCA and other methods was for wetland and grassland samples, where PCA smoothed and attenuated certain absorption peaks in chemically informative wavelength regions. The results show PCA’s explained variance may not guarantee preservation of chemically important narrow absorption features, particularly when they contribute little to total variance. Consequently, ”explained variance” alone is not a sufficient criterion for selecting the number of latent variables. Together, our findings emphasize the importance of compression choices for preserving soil-relevant spectral information as vis-NIR continues to support large-scale soil characterization and digital soil mapping.

The soil-gas diffusivity (ratio of gas diffusion coefficients in soil and pure air), Dp/D0, controls the mobility of gases in variably saturated soils, including aeration, emission and uptake of greenhouse gases. Previous soil-gas diffusivity studies have focused on lower-organic soils. Here, Dp/D0 was measured on intact soil cores at six different soil-water matric potentials between -30 and -1000 cm H2O (pF between 1.5 to 3) on 127 Danish peat top soils (12-52% SOC). The SOC level was not found to be main control of Dp/D0 for peat soils. Dp/D0 versus soil-air content (ε) curves varied as much within a narrow SOC interval (e.g., 30-36% SOC) as for the whole range. In contrast to previous observations for lower-organic soils, it was found that at each pF level, Dp/D0 increased linearly with ε, the slope, the Penman pore continuity index P, increased linearly with pF. Previous Dp/D0 models developed for lower-organic soils (typically < 4% SOC) failed to describe Dp/D0 for peat soils. A soil-water characteristic curve-dependent model (Buckingham-Burdine-Campbell, BBC) well predicted Dp/D0 for peat soils. A modified WLR model with inputs of actual and effective air-filled porosity (ε*, taken as ε around pF3) was developed. The BBC and modified WLR models were successfully validated against independent data for high-organic soils representing different climate zones and organic matter quality. This work provides a better understanding of and models for gas diffusion in high-organic soils, setting a platform for including soil-air phase properties when evaluating soil functions, security, and capital.

Soil physics models have long relied on simplifying assumptions to represent complex processes, yet such assumptions can strongly bias model predictions. Here, we propose differentiable hybrid modeling (DHM) as a paradigm-shifting framework that learns unobservable intrinsic processes from data and physical constraints, rather than simplifying them. As a proof of concept, we apply the DHM approach to the challenge of partitioning the soil water retention curve (SWRC) into capillary and adsorbed water components, a problem where traditional assumptions have led to divergent results. The hybrid framework derives this partitioning directly from data while remaining guided by simple physical constraints. Using basic soil physical properties as inputs, the DHM couples an analytical formula for the dry end of the SWRC with data-driven physics-informed neural networks that learn the wet end, the transition between the two ends, and key soil-specific parameters. The model was trained on a SWRC dataset from 482 undisturbed soil samples, spanning a broad range of texture classes and organic carbon contents. The hybrid model successfully learned both the overall shape and the capillary and adsorbed components of the SWRC. Notably, the learned patterns were consistent with pore-scale thermodynamic saturation behavior in angular pores, without relying on explicit assumptions about soil pore geometry or its distribution. Moreover, the model revealed a distinctly nonlinear transition between capillary and adsorbed domains, challenging the linear assumptions invoked in previous studies. The methodology introduced here provides a blueprint for learning other soil processes where high-quality datasets are available but mechanistic understanding is incomplete.

Soil organic carbon (SOC) is an important soil health indicator. The SOC directly influences major soil functions such as nutrient cycling, fertility, soil structure, and water and air regulation. As a result, it is important to understand both the soil’s storage capacity and the stability of this carbon. Several methods have been developed to assess distinct SOC pools based on different physical, biological, and thermal definitions. Rock Eval 6 thermal analysis partitions SOC into four operational fractions (S1 to S4), which serve as proxies for organic compounds of increasing thermal stability. Thus is S1 the most labile and easily decomposable, S2 and S3 reflecting progressively more stable organic carbon pools, and the residual refractory carbon measured as S4. Although Rock Eval 6 provides accurate and reproducible estimates of thermally defined SOC fractions, the high initial investment cost and low analytical throughput limit its application for large scale monitoring. Despite the promising application of soil diffuse reflectance spectroscopy for predicting various soil properties, a comprehensive evaluation of this method for estimating Rock Eval 6 derived SOC fractions is lacking. In this presentation, we evaluate the feasibility of vis-NIR spectroscopy as a non-destructive and cost-effective alternative method to predict Rock Eval 6 SOC fractions (S1 to S4). A total of 131 soil samples were collected from lowland areas across Denmark under different land-use types. Partial least squares regression and interval partial least squares regression were applied to relate soil reflectance spectra measured between 400 and 2500 nm to thermally defined SOC fractions. Our results indicate that vis-NIR spectroscopy can reliably estimate thermally defined SOC fractions derived from Rock Eval 6 analysis, with R² values of 0.79, 0.81, 0.78, and 0.53 for S1, S2, S3 and S4, respectively. Among the individual fractions, S2 was estimated with the highest accuracy, while S1 and S3 showed moderate predictive performance and S4 exhibited lower accuracy. Based on these statistical parameters, we conclude that vis-NIR spectroscopy is a feasible and rapid tool for estimating thermally defined SOC fractions.

Estimating the soil particle size distribution (PSD) from visible-near infrared (vis-NIR) reflectance spectra is conventionally limited to predicting discrete soil fractions (e.g., sand, silt, and clay). This approach presents significant challenges: it requires the harmonization of data from different classification systems and, by reducing the PSD to a few values, fails to reflect the entire variation in soil texture. To address these gaps, we present a novel physics-informed neural network (PINN) for the direct estimation of the continuous PSD from vis-NIR reflectance measurements. The PINN learns a continuous, differentiable, and non-parametric representation of the cumulative PSD by integrating both measurements and physical constraints imposed during training. This approach eliminates the need for harmonization and interpolation of measurements originating from different soil texture classification systems and allows the model to be trained on datasets with varying numbers of measurements per sample. Performance evaluation on 30% of the 2777 studied samples showed that the PINN achieved an RMSE of 6.77% and an R 2 of 0.97 in predicting the cumulative PSD fraction. For the texture fractions, the model achieved RMSE values of 4.72%, 3.06%, and 2.75% for sand, silt, and clay, respectively. A comparison with a physics-agnostic (i.e., physics-uninformed) version of the model revealed that both approaches performed similarly in terms of RMSE and R 2. However, the physics-agnostic model violated physical constraints even in data-rich scenarios. In contrast, the PSD obtained from the PINN maintained its physical integrity even under data sparsity conditions and consistently produced non-negative, monotonically increasing predictions that sum to 100% at the largest particle size.

Study Region: The research is conducted in the Kærby district of Aalborg, Denmark (0.6 km2). Study Focus: Urban areas with shallow groundwater levels face increasing challenges related to groundwater flooding, infrastructure damage, and sewer system infiltration under changing anthropogenic climatic conditions. This study aims to quantify and analyze components of the water balance through a comprehensive, observation-based assessment of shallow urban groundwater variability, using high-resolution data from both citizen science initiatives and automated groundwater level monitoring. New Hydrological Insights for the Region: Over a 4.5-year period, groundwater levels are monitored, revealing seasonal fluctuations of approximately ±0.5 m, and rapid dynamic responses to individual rainfall events, with observed groundwater level increases up to 6 times the corresponding rainfall depth. Cross-correlation analysis links fluctuations to antecedent precipitation and groundwater recharge dynamics. Hydraulic characterization indicates high infiltrability and a narrow vadose zone (0–2 m), explaining rapid saturation and groundwater response. The study quantifies groundwater infiltration into combined sewers using a water budget, showing that in some subcatchments, annual infiltration matches precipitation and exceeds recharge, suggesting lateral inflow and anthropogenically transformed flow regimes. Geospatial interpolation reveals inverted gradients and lowered groundwater levels driven by infrastructure-induced drainage. The study reveals a natural water balance altered by urbanization. The findings highlight the importance of understanding natural and anthropogenic hydrological processes to support climate-resilient urban water management and inform solutions mitigating groundwater impacts.

Soil water repellency (SWR) is a natural process and affects water dynamics from nano to ecosystem scales. However, the spatial distribution of SWR at the ecosystem scale, as well as the underlying drivers across diverse habitats, land uses and soil textures, remain underexplored. This study presents a comprehensive survey of SWR in Denmark and its predicted spatial distribution, using approximately 7,500 samples. We used digital soil mapping methods (Quantile Random Forest model) to map and identify the relationship between SWR and various environmental variables, including vegetation (via satellite imagery), soil properties (texture and soil organic carbon), and landforms (slope and wetness index). The predicted maps at 10 m resolution revealed that SWR varies across different land uses and vegetation types, with higher values in areas of natural vegetation (e.g., heathlands and coniferous forests) compared to grasslands and croplands (mostly hydrophilic). The analysis also identified soil organic carbon, Sentinel band 3 (Green band − Chlorophyll absorption) and soil texture as key drivers of spatial variation in SWR at the national extent. We found that soil texture influences SWR intensity, which generally decreases as clay content increases across most land use types, except for heathlands. While the predicted maps provided valuable insights into SWR distribution and its environmental drivers, further research is needed to explore the spatio-temporal dynamics of SWR within each habitat, particularly in relation to soil moisture changes. This study highlights the potential of combining machine learning and remote sensing to provide crucial spatial information for managing water resources and enhancing ecosystem resilience in the face of climate change.

The increasing threat of soil degradation presents significant challenges to soil health, especially within agroecosystems that are vital for food security, climate regulation, and economic stability. This growing concern arises from intricate interactions between land use practices and climatic conditions, which, if not addressed, could jeopardize sustainable development and environmental resilience. This review offers a comprehensive examination of soil degradation, including its definitions, global prevalence, underlying mechanisms, and methods of measurement. It underscores the connections between soil degradation and land use, with a focus on socio-economic consequences. Current assessment methods frequently depend on insufficient data, concentrate on singular factors, and utilize arbitrary thresholds, potentially resulting in misclassification and misguided decisions. We analyze these shortcomings and investigate emerging methodologies that provide scalable and objective evaluations, offering a more accurate representation of soil vulnerability. Additionally, the review assesses both physical and biological indicators, as well as the potential of technologies such as remote sensing, artificial intelligence, and big data analytics for enhanced monitoring and forecasting. Key factors driving soil degradation, including unsustainable agricultural practices, deforestation, industrial activities, and extreme climate events, are thoroughly examined. The review emphasizes the importance of healthy soils in achieving the United Nations Sustainable Development Goals, particularly concerning food and water security, ecosystem health, poverty alleviation, and climate action. It suggests future research directions that prioritize standardized metrics, interdisciplinary collaboration, and predictive modeling to facilitate more integrated and effective management of soil degradation in the context of global environmental changes.

Soil physics models have long relied on simplifying assumptions to represent complex processes, yet such assumptions can strongly bias model predictions. Here, we propose a paradigm-shifting differentiable hybrid modeling (DHM) framework that instead of simplifying the unknown, learns it from data. As a proof of concept, we apply the hybrid approach to the challenge of partitioning the soil water retention curve (SWRC) into capillary and adsorbed water components, a problem where traditional assumptions have led to divergent results. The hybrid framework derives this partitioning directly from data while remaining guided by a few parsimonious and universally accepted physical constraints. Using basic soil physical properties as inputs, the hybrid model couples an analytical formula for the dry end of the SWRC with data-driven physics-informed neural networks that learn the wet end, the transition between the two ends, and key soil-specific parameters. The model was trained on a SWRC dataset from 482 undisturbed soil samples from Central Europe, spanning a broad range of soil texture classes and organic carbon contents. The hybrid model successfully learned both the overall shape and the capillary and adsorbed components of the SWRC. Notably, the model revealed physically meaningful pore-scale features without relying on explicit geometrical assumptions about soil pore shape or its distribution. Moreover, the model revealed a distinctly nonlinear transition between capillary and adsorbed domains, challenging the linear assumptions invoked in previous studies. The methodology introduced here provides a blueprint for learning other soil processes where high-quality datasets are available but mechanistic understanding is incomplete.

Soil water repellency (SWR) significantly impacts water infiltration and soil health, influencing ecological processes across various habitats. Although the mechanisms behind SWR remain partially unclear, it is influenced by both soil and biological properties. While several studies have examined SWR in agricultural soils, fewer studies have focused on natural habitats. This study examines the relationships between soil properties (electrical conductivity (EC), pH, and total carbon (TC)), prokaryotic communities, and potential SWR (measured by the molarity of ethanol droplet test, 60°C pretreatment) in 1153 soil samples spanning 33 habitat types across Denmark. Using path model analysis, we show that both biotic and abiotic factors contribute significantly to SWR. A model including pH, EC, TC, and prokaryotic community composition (β-diversity) could explain ~50% of the variation in SWR, with β-diversity and TC being the most important properties. Furthermore, we reveal distinct variations in SWR across habitat types, which cover a wide range of SWR, from not water repellent to strongly water repellent. Prokaryotic α-diversity was negatively correlated to the degree of SWR, and we found a clear gradient in β-diversity from the highest to the lowest degree of SWR. The degree of SWR was divided into five classes, and we identified 69 genera indicating one or a combination of the SWR classes, which could potentially be used as indicators of the degree of SWR. This research underscores the importance of including the microbial communities in studies examining SWR. In perspective, the observed relations between SWR and soil prokaryotic diversity and community composition also imply that SWR could become a key biophysical indicator of soil health.

Agricultural activity on drained lowlands is a common practice in Denmark and there are suggestions to rewet some of them for climate mitigation purposes. Rewetting those lowlands might result in a change in microbial community composition. This study investigates the current prokaryotic diversity and community composition in soil samples from cultivated lowlands to provide the baseline for monitoring changes after rewetting. Furthermore, variations in soil properties between sites are examined, and the properties driving differences in prokaryotic diversity and community composition are identified. In total, 116 samples were collected from field sites across Denmark that were categorized as one of four different land-use types: Crop, Grass, Fallow, and Other. Soil properties were selected to cover chemical (soil water repellency, pH, electrical conductivity), hydrological (depth to ground-water table, soil water content at field capacity (-100 hPa)), nutrient-related (total nitrogen, organic carbon, carbon-to-nitrogen-ratio, fractions of pyrolizable and residual organic matter), and structural (total porosity, pore size distribution index) functions of the soil. Soil samples exhibited significant variations in their chemical and physical properties, including pH ranging from 2.02 to 7.55, organic carbon ranging from 3 g 100g-1 to 50 g 100g-1, soil water repellency ranging from 71.27 mN m-1 (hydrophilic) to 33.85 mN m-1 (very strongly hydrophobic), and total porosity ranging from 51% to 95%. Soil samples clustered according to soil class (mineral, organo-mineral, organic, highly organic) but not according to land-use type (crop, grass, fallow, other). Prokaryotic alpha diversity, measured as Shannon’s diversity index (H), ranged from 4.16 to 5.89 across samples and could best be predicted by pH, followed by total porosity, fraction of pyrolizable carbon, and pore size distribution index. The pH alone explained 36% of the variation in H between samples. Hierarchical clustering identified three prokaryotic clusters highly correlated with pH. A weak correlation was found between differences in community composition (beta diversity) and geographic distance (r = 0.15, p < 0.001). However, pH was also the main driver of beta diversity, explaining 11% of the variation. At the same time, models including additional variables only had marginally better explanatory power. In conclusion, pH was the predominant driver of prokaryotic alpha and beta diversity across land-use types in lowland soils.

This paper presents a novel physics‐informed neural network (PINN) approach for developing pedotransfer functions (PTFs) to predict continuous soil water retention curves (SWRCs) based on soil textural fractions, organic carbon content, and bulk density. In contrast to conventional parametric PTFs developed for specific SWRC models, the PINN learns a non‐specific form of the SWRC from both measurements and physical constraints imposed during the training process. This approach allows the estimated SWRC to maintain its physical integrity from saturation to oven‐dry conditions, even in scenarios with sparse data. The new approach is particularly effective for tackling the challenges encountered in developing PTFs on large SWRC data sets, which often have an imbalance toward the wet‐end () and include numerous samples with limited and unevenly distributed measurements, many of which do not meet the requirements to fit traditional SWRC models. We compared the performance of the PINN with that of a conventional physics‐agnostic neural network using a data set of 4,200 soil samples. While both networks performed similarly at the wet‐end where data are abundant, with RMSE values of around 0.041 m 3 m −3 , the PINN excelled at the dry‐end () where data are sparse and unevenly distributed, achieving a normalized RMSE of 0.172 (RMSE = 0.0045 m 3 m −3 ) compared to a normalized RMSE of 0.522 (RMSE = 0.0136 m 3 m −3 ) for the conventional neural network. The SWRC derived from the PINN is differentiable with respect to matric potential, making it well‐suited for integration into models of water flow in the unsaturated zone.

Soil CO2 dynamics play a crucial role in the global carbon cycle, particularly in agricultural soils where management practices influence CO2 dynamics. This study quantified soil CO2 concentrations and fluxes under different fertilization treatments (chemical vs. organic) in a soybean field and evaluated the performance of the SOILCO2 model with different gas diffusivity models. Soil moisture, temperature, and CO2 concentrations were continuously monitored at multiple depths throughout the growing season. The results showed that microbial and root-associated CO2 production were enhanced in the presence of crops and organic matter, with microbial activity significantly increasing the CO2 concentrations, particularly after rainfall events. Gas diffusivity, a critical factor in CO2 transport modeling, was evaluated using the Millington–Quirk (MQ) and Water Linear Reduction (WLR) models. The WLR model with fitted parameter provided better agreement with measured gas diffusivity and observed CO2 profiles as compared to the MQ model. Sensitivity analysis demonstrated that an accurate representation of gas diffusivity is essential for a realistic simulation of soil CO2 behavior. This study highlights the importance of high-resolution field monitoring and model calibration for improving CO2 transport predictions, and supports the use of the WLR model for agricultural soils with varying textures and moisture conditions.

Agricultural activity in drained lowlands accelerates peat decomposition and greenhouse gas emissions. Rewetting is increasingly adopted across Europe to mitigate these emissions, but its effects on soil microbial communities remain poorly understood. We examined prokaryotic communities and network structure in Danish lowland soils designated for potential rewetting to improve our understanding of these communities and their drivers. We included less-studied soil properties linked to hydrology and structure (soil water content at field capacity, van Genuchten pore-size distribution index (n), and soil water repellency (SWR)) in addition to common properties such as pH and organic carbon (OC). We analysed 113 soil samples across land-use types (grass, fallow, crop, other) spanning mineral to organic soils with gradients in pH (2.0-7.6), OC (0.025-0.499 kg kg-1), SWR (33.9-71.3 mN m-1), and soil structure (n: 1.1-1.3). Prokaryotic alpha diversity (Shannon-Wiener index: 2.9-5.7) was best predicted by pH, followed by porosity and n. Together, pH, porosity, n, and OC accounted for 24.5% of the variance in community composition. Hierarchical clustering identified three prokaryotic clusters strongly aligned with pH. Network analysis revealed marginal differences when comparing samples from fallow and grass, while complexity increased progressively across clusters. Interestingly, high-pH soils showed the highest alpha diversity but the least complex networks, while low-pH soils showed the opposite. In conclusion, soil pH emerged as the dominant driver of prokaryotic communities in Danish lowlands, but hydrological and structural properties also played important roles. Network complexity provided complementary insights into ecosystem organisation beyond diversity alone.

Water vapour sorption is essential to understand the hygric behaviour of building materials. This study introduces an experimental method for quantifying vapour sorption dynamics. We applied the automated, non‑equilibrium Dynamic Dewpoint Isotherm (DDI) method to generate detailed ad- and desorption isotherms for 12 building materials. An overall sorption response surface (SRS; moisture storage as function of water activity or relative humidity) was generated for each material by completing a cyclic run of 14 ad- and desorption isotherms. DDI-measured moisture storage capacity at apparent equilibrium agreed well with an equilibrium method. The wideness (magnitude of hysteresis) and shape (surface and pore-network controlled) of the SRS varied greatly in regard to material composition, porosity, and density. This was used to group the materials in regard to level of moisture dynamics with high-porosity and bio-based materials showing the highest level of sorption dynamics. The DDI-SRS concept seems useful to illustrate and quantify dynamic moisture storage behaviour of building materials and elements under changing relative humidity.

A coil probe (CP) for time-domain reflectometry (TDR) with a sensor length of 40 mm (CP40) was developed for long-term field soil moisture measurements in a thin surface soil layer (0–3 cm depth). In laboratory, soil moisture measurements of CP40 were nearly identical to those of a traditional two-rod type TDR (2RTDR) probe (rod length: 15 cm, diameter: 0.3 cm, spacing: 3 cm). The CP40 measurement accuracy was between 0.01 m3/m3 and 0.03 m3/m3. For long-term field soil moisture measurements, five CP40 units were installed in the highly wetted soil at the Sanzai site (SS), which is in the permafrost area of the Taiga, and dry soil at the Mandalgobi site (MGS) in the semi-arid area of Mongolia. Four units accurately measured soil moisture at both sites over six years (2002–2007and 2008–2009). Three units succeeded in conducting precisely continuous soil moisture measurements between 2008 and 2022 at the MGS. Two units successfully measured soil moisture over 21 years at both the sites. The representativeness of the CP40 soil moisture measurements in highly wetted soils was low because of heavy rainfall and soil heterogeneity. The accuracy of CP40 soil moisture measurements in highly wetted soils was slightly lower than that of the traditional 2RTDR probe. However, the accuracy of the CP40 soil-moisture measurements in the dry soil was comparable to that of the traditional 2RTDR probe. CP40 units are durable and their field soil moisture measurements demonstrate stable and precise performance (bias, RMSE), even under severe environmental conditions.

The soil water retention curve (SWRC) is essential for describing water and energy exchange processes at the interface between the solid earth and the atmosphere. Despite its importance, measuring the SWRC using standard laboratory methods is challenging and time-consuming. This paper presents a novel physics-informed neural network (PINN) approach for developing pedotransfer functions (PTFs) to predict continuous SWRCs based on soil texture, organic carbon content, and dry bulk density. In contrast to conventional parametric PTFs developed for specific SWRC models, the PINN learns a non-specific form of the SWRC by effectively integrating both measurements and physical constraints into the training process. This approach allows the estimated SWRC to maintain its physical integrity from saturation to oven-dry conditions, even in scenarios with sparse data. The new approach is particularly effective for tackling the challenges encountered in developing PTFs on large SWRC datasets, which often have an imbalance towards the wet-end and include numerous samples with limited and unevenly distributed measurements. We compared the performance of the PINN with that of a conventional physics-agnostic neural network using a dataset of 4200 soil samples. While both networks performed similarly at the wet-end where data are abundant, the PINN excelled at the dry-end where data are sparse and unevenly distributed, achieving a normalized RMSE of 0.172 compared to 0.522 for the conventional neural network. The SWRC derived from the PINN is differentiable with respect to the matric potential and can be seamlessly integrated into the governing equations of water flow in the unsaturated zone.