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.

Different spectroscopy methods such as visible near-infrared (Vis-NIR) spectroscopy have proven to provide useful information on soil physical and chemical properties. The majority of previous studies have focused on multivariate regression methods such as partial least squares regression (PLSR) to predict soil characteristic features from soil spectra. The objective of these efforts was to provide precision data for agricultural and land management practices where the knowledge about the soil differences in each part of the field can be used for variable-rate irrigation, seeding, liming, fertilising and pesticide application. Since the currently available machines can only apply few (less than 10) variable classes of seed, fertiliser, etc., using soil classification seems to be a more appropriate option compared to regression. As a part of Danish National Soil Survey in 1980s, 2460 soil samples were collected from 789 soil profiles in 4 depths (0-30, 30-60, 60-100 and 100-200cm) throughout Denmark (some profiles did not have samples from all 4 depths) and tested for several soil characteristics including complete soil texture, organic carbon content and calcium carbonate in the lab. Based on these soil characteristics, all soil samples were classified into 8 soil types or 12 soil classes in the Danish Soil Classification System (JB system). Later the Vis-NIR spectra of samples were measured using a FOSS DS2500 spectrometer in the range of 350-2500nm. Partial least squares and support vector machines discriminant analyses (PLS-DA and SVM-DA, respectively) were used to calibrate and cross-validate classification models where soil Vis-NIR spectra were used to classify each sample from each depths into its corresponding JB soil type and soil class. The results show excellent classification accuracy and specificity (>80% and >90%, respectively) for samples from the same depths and when samples from all depths were combined. We found that the high false negative rate (low sensitivity) was mainly due to the models classifying samples in the neighbouring classes of the actual class (e.g. classifying a sample in JBC 2 as belonging to JBC 1 or JBC 3). This clearly indicates that calibrating the classification model on the uncertain hydrometer data (with 1.4-2% reproducibility error) was the main reason for classification of samples in the neighbourhood of the actual class. In conclusion, given the highly uncertain reference methods for soil classification, using Vis-NIR spectroscopy with PLS-DA provides a very rapid, inexpensive, and highly reliable method for soil texture classification on a national scale.

Groundwater-dependent terrestrial ecosystems (GWDTE) have been increasingly under threat due to groundwater depletion globally. Within the past 200 years, there has been severe artificial drainage of low-lying areas in Denmark, leading to a gradual loss of GWDTE nature habitat areas. This study explores the spatial-temporal loss of Danish GWDTE using historically vectorized topographical maps. We carry out geographic information systems (GIS) overlap analysis between different historical topographic maps with signatures of GWDTE starting from the 19 th century up to a current river valley bottom map as a reference period. This is because farmworkers and monks have practiced drainage by ditching since the early middle ages (1100-1200). We then examine the changes in two protected GWDTE habitat types in different periods and different hydrologic spatial locations. Results reveal a decrease in the area of GWDTE over the last 200 years. We attribute this to different human interventions that through e.g., drainage, have impacted the low-lying landscape throughout history. We further conclude that downstream parts of the river network have been exposed to less GWDTE habitat loss than upstream ones. This indicates that upstream river valleys are more vulnerable to GWDTE decline. Therefore, as a management measure, caution should be exercised when designing these areas for agriculture activities using artificial drainage and groundwater abstraction since this may lead to further decline. In contrast, there is a higher potential for establishing constructed wetlands or rewetting peatlands to restore balance.