Jan Diels

Nitrate contamination of water bodies driven by agricultural activities remains a widespread environmental concern. In groundwater-fed surface waters, nitrate concentrations are governed by the transport of N-excess through groundwater systems. Denitrification plays a central role in mitigating this transport by reducing nitrate to inert N₂ gas in anoxic zones. The anoxic zone in an aquifer is separated from the oxic zone by a sharp boundary, i.e., the nitrate reduction front (NRF). This NRF can be retrieved from sediment colour changes, shifting from oxidised hues (yellow, brown, or red) to reduced colours (grey, green, or black), provided there are sufficient Fe-bearing minerals and redox colours are not masked by high clay or organic matter contents. Hydrochemical data from multilevel observation wells provide a more integrated signal of redox conditions, yet precise NRF delineation remains challenging since usually, groundwater sampling is not done for consecutive depth intervals. Integrating lithological and hydrochemical information therefore offers a more robust approach.This study presents a two-stage machine learning (ML) framework to predict groundwater NRF positions across Flanders. Firstly, a logistic regression model (ML1) was developed to estimate oxidation probabilities for individual borehole layers using lithological characteristics (colour, texture, stratigraphy) and relative depth within the aquifer as predictors. The model reproduced the redox conditions (reduced or oxidized) in 72% of borehole layers, which were assessed from the hydrochemistry (dissolved oxygen, [Fe2+] and redox potential) of the associated filters for groundwater monitoring.[JV2] [ES3] The derived probabilities were used to assess the likelihood that a boundary between two borehole layers is located at the NRF leading to a likelihood-depth profile. A separate likelihood profile was developed using the hydrochemistry data from the filters of a multilevel borehole. Multiplication of these two profiles yielded the most probable NRF position, resulting in NRF estimates at 1,902 locations.Secondly, these positions were used to train a gradient-boosted regression tree model (XGBoost) to predict the NRF depth for any location across Flanders (ML2i). The training data were the NRF depths at the 1,902 multilevel observation wells.

Accurate estimation of soil water content in the root zone (e.g., 0 – 30 cm) is essential for designing irrigation schedules and requires measurements that represent the field scale. Cosmic Ray Neutron Sensing (CRNS) offers a non-invasive solution that provides integrated soil moisture measurements with a horizontal footprint of approximately 7 to14 hectares and depths ranging from 15 to 83 cm, making it suitable in an area with a homogenous land use, like agricultural fields. However, CRNS sensitivity varies with both distance and depth relative to the sensor, complicating its use for estimating soil moisture in specific layers. When soil moisture is known, it is feasible to perform a forward calculation to derive neutron counts from soil water content. In this study, such calculations were performed using COSMIC, integrated with the HYDRUS-1D model. However, backward calculations, deriving soil water content from neutron counts, are not straightforward. This is because wetting and drying processes start at the soil surface, where CRNS is most sensitive. Consequently, the integrated measurement disproportionately reflects changes in the upper layers, creating a non-unique or hysteretic relationship between neutron counts and soil moisture during wetting and drying cycles. This makes predicting the 0 – 30 cm water content from neutron counts particularly challenging.To address these limitations, we explore the application of the Long Short-Term Memory (LSTM) model to predict the average soil water content in the 0 – 30 cm layer by training the model using time series of average 0 – 30 cm soil water content and neutron counts (simulated with HYDRUS-1D COSMIC) as well as meteorological data (precipitation and reference evapotranspiration). The LSTM model is well-suited because it can learn temporal dependencies and patterns of long sequence data. The initial simulations were based on three years record of synthetic data under bare soil conditions for a region in Flanders, Belgium. While initial findings indicate a potential, further research will focus on improving the model’s robustness by training the model with more diverse variables, expanding the dataset, and integrating field measurement soil moisture records to enhance its applicability across different scenarios. This research highlights the feasibility of combining CRNS measurement, physically based modelling, and data-driven techniques to improve soil moisture estimation for irrigation management.

Observations of soil states obtained from in-situ or remote sensors have various sources of errors. A crude way to represent these errors is to assume that part of the error is purely random whereas another part persists and does not change over time. Since the persistent part does not cancel out when more data become available over time whereas the random part does, the partitioning of the error into a persistent and random part is important to assess the uncertainty of model parameters and model predictions that are derived from these observations. Two approaches can be followed to represent these systematic errors in model parameter estimation. The first approach represents the systematic error as an additional parameter representing the bias that is estimated using additional unbiased observations, which we assumed to have only random errors. A second approach represents the systematic error as a covariance in the error-covariance matrix. The uncertainty of the model predictions in the first approach consists of a term that represents the uncertainty of the bias estimation, which is independent of the magnitude of the bias and depends only on the uncertainty of the unbiased additional observations. When additional unbiased measurements are included in the second approach, which represents bias as error covariance, smaller model prediction uncertainty is obtained than using the first approach. This is especially the case when the covariance representing the bias is smaller than the variance of the average error of the random observations. Including prior knowledge about the bias in the error covariance, reduces the model parameter and prediction uncertainty.

The SWIM² framework (Hendrickx et al., 2025) integrates a soil water balance model with in situ sensor data and soil moisture samples through Bayesian inverse modelling. The calibrated model then generates probabilistic 10-day soil moisture forecasts, enabling real-time, site-specific irrigation advice. SWIM² was validated in a real-time setup for 18 vegetable cropping cycles on agricultural fields in Flanders, Belgium, with reliable precipitation data. Although using minimal prior knowledge and despite sensor bias, SWIM² achieves robust soil moisture predictions for a 7-day horizon, with accuracies comparable to sensor measurements. We also assessed the impact of model parameter and weather forecast uncertainty on SM prediction uncertainty, water stress prediction and irrigation advice by integrating the calibrated model ensemble with ensemble-based probabilistic weather forecasts, resulting in high detection rate and accuracy in predicting water stress triggering the irrigation threshold.Time series of vegetation indices such as NDVI and LAI from Sentinel-2 optical remote sensing as well as LST from Sentinel-3 contain much information on crop growth and crop evapotranspiration. Additionally, the new NISAR mission is promising for high-resolution surface soil moisture observations. We assess relations between in situ measurements and model outputs (crop growth curve, actual ET and SWC), and the remote sensing data, and we discuss opportunities of these data to improve soil moisture and ETa predictions.Reference: Hendrickx, M.G.A., Vanderborght, J., Janssens, P., Laloy, E., Bombeke, S., Matthyssen, E., Waverijn, A., Diels, J. (2025). Field-scale soil moisture predictions in real time using in situ sensor measurements in an inverse modeling framework: SWIM². Authorea Preprints, doi:10.22541/ESSOAR.175103915.57413983/V1.

Soft fruit such as raspberry ( Rubus idaeus L.) are commonly cultivated under protective structures to shield plants from adverse weather and extend harvest periods. Agrivoltaic systems offer an alternative to disposable plastic hoop covers by combining crop cultivation with solar energy production, yet their effects on crop microclimate remain poorly understood in temperate regions. In this study, we investigated summer canopy microclimate responses to different agrivoltaic layouts at two raspberry production sites in the Netherlands. At the first site, we compared semi-translucent portrait panels and landscape opaque panels with a conventional polyethylene tunnel. At the second site, an opaque panel system was evaluated alongside an open-field control. We monitored photosynthetically active radiation, canopy air temperature, humidity, and vapor pressure deficit from July to late August and compared both daily averages and diurnal maxima and minima. Agrivoltaic structures substantially reduced incoming radiation, with semi-translucent panels providing more uniform light distribution than opaque arrays. Average daily canopy temperature differed little across treatments (≤0.5 °C), but distinct diurnal patterns emerged: opaque arrays consistently cooled around solar noon on high-radiation days (up to 0.5°C cooler than control), while semi-translucent arrays occasionally induced midday warming of about 0.4°C under wind-calm conditions. Vapor pressure deficit under polyethylene tunnels was more similar to agrivoltaic systems than to open-air plots. Wind and rainfall modulated the microclimate changes independently of PV layout. Our findings highlight within-season design-specific trade-off for summer conditions: opaque panels favor thermal buffering but reduce light uniformity, whereas semi-translucent panels improve spatial light distribution at the cost of limited buffering and occasional warming. Microclimate contrasts depended on panel material properties, and are consistent with differences in radiation balance arising from material properties and ground coverage ratio among polyethylene films, opaque panels, and semi-translucent modules. They provide initial evidence that agrivoltaic systems can offer microclimate moderation comparable to conventional plastic polytunnels, while simultaneously generating renewable energy.

Affordable autonomous soil sensors and IoT technology enable real‐time soil moisture monitoring, which offers opportunities for real‐time model calibration and irrigation optimization. We introduce an irrigation decision support system SWIM ² (Sensor Wielded Inverse Modeling of a Soil Water Irrigation Model), a digital twin that integrates continuous sensor data and unbiased, periodic soil samples with an FAO‐based soil water balance model using a Bayesian inverse modeling algorithm, DREAM _(ZS) (DiffeRential Evolution Adaptive Metropolis). SWIM ² estimates 12 soil and crop parameters and their associated probability distributions and correlations, providing soil moisture predictions with uncertainty estimates. The SWIM ² framework is illustrated and validated in a real‐time setup for 18 vegetable cropping cycles on agricultural fields in Flanders, Belgium, with in situ precipitation data. Although using minimal prior knowledge and despite sensor bias, SWIM ² achieves robust soil moisture predictions for a 7‐day horizon, with accuracies comparable to sensor measurements. Predictions improve substantially in precision within the first 20 calibration days and maintain high predictive power throughout the growing season. The impact of in situ measurements and temporal covariance of the observational errors (“error covariance”) was assessed, indicating that good knowledge of the error covariance and independent soil moisture samples are essential to correct for sensor bias and ensure accurate model calibration, while continuous sensor data ensure accurate and precise estimates of the dynamics. This study demonstrates the use of soil moisture sensor data in a Bayesian inverse modeling framework, offering practical solutions for real‐time soil moisture prediction and irrigation decision‐making, enhancing water management across agricultural fields.

In large-scale, multi-national research projects on agricultural cropping systems such as SoilCare (Horizon 2020), ensuring consistency, comparability, and timely reporting of the (meta)data of the agricultural experiments across diverse partners has been a persistent challenge. To address these concerns, the SoilCare project developed a comprehensive data management system centered around a standardized template for the collection, organization, and sharing of experimental data and metadata from cropping systems. This template, designed to support harmonized sharing, analysis, and documentation through a common structure and terminology, meets the interdisciplinary requirements of modern agricultural research. Experimental data and metadata were structured into five core pools: 1. basic information, 2. field information, 3. experimental setup, 4. agricultural management data and 5. measured data/results. The Excel-based template was carefully structured to support integration into a relational database, enabling robust monitoring, analysis, and traceability of experimental processes and outcomes. The database schema and template, together with the (meta)data collected using this system during the SoilCare projects, were made openly available via Zenodo. The standardized approach ultimately enabled unified analyses and harmonized reporting across all experimental sites, demonstrating the system’s effectiveness in facilitating collaborative agricultural research at scale.

Affordable autonomous soil sensors and IoT technology enable real-time soil moisture monitoring, which offers opportunities for real-time model calibration and irrigation optimization. We introduce an irrigation decision support system SWIM² (Sensor Wielded Inverse Modeling of a Soil Water Irrigation Model), a digital twin that integrates continuous sensor data and unbiased, periodic soil samples with an FAO-based soil water balance model using a Bayesian inverse modeling algorithm, DREAM(ZS) (DiffeRential Evolution Adaptive Metropolis). SWIM² estimates 12 soil and crop parameters and their associated probability distributions and correlations, providing soil moisture predictions with uncertainty estimates. The SWIM² framework is illustrated and validated in a real-time setup for 18 vegetable cropping cycles on agricultural fields in Flanders, Belgium, with reliable precipitation data. Although using minimal prior knowledge and despite sensor bias, SWIM² achieves robust soil moisture predictions for a 7-day horizon, with accuracies comparable to sensor measurements. Predictions improve substantially in precision within the first 20 calibration days and maintain high predictive power throughout the growing season. The impact of in-situ measurements and temporal covariance of the observational errors (“error covariance”) was assessed, indicating that good knowledge of the error covariance and independent soil moisture samples are essential to correct for sensor bias and ensure accurate model calibration, while continuous sensor data ensure accurate and precise estimates of the dynamics. This study demonstrates the use of soil moisture sensor data in a Bayesian inverse modeling framework, offering practical solutions for real-time soil moisture prediction and irrigation decision-making, enhancing water management across agricultural fields.