There is a trend towards decentralized source separation (DSS) for wastewater treatment and resource recovery. An assessment framework is required to assess whether implementing a DSS treatment over a conventional centralized one is advantageous. This framework needs to account for the performance of the wastewater treatment plant (WWTP) and the effect that resource recovery has on closely-linked sectors such as food and energy production. A framework is lacking that covers the economic dimension, the circularity, the nature reciprocity of resource recovery and that can be applied to real-life cases. A novel WFE framework has been developed here to compare a conventional centralized and a DSS-based WWTP. This novel WFE framework contains assessment methods that are reproducible, and applicable to real-life cases. It also accounts for the local climatic conditions that determine irrigation water requirements. The comparison results revealed that the need to construct new DSS infrastructure leads to a lower economic efficiency of water treatment. Further, chemical-intensive treatment reduces the DSS's material resource circularity and efficiency. Using heat pumps increases the energy use of the DSS WWTP, causing a reduction in water treatment energy efficiency. However, the advantages of DSS show up in the freshwater and nutrient efficiency of food production as well as in the energy self-sufficiency of the WWTP. The novel WFE framework contains indicators specific to water treatment and the food production sectors to improve inter-sectoral communication. Also, including the nature reciprocity assessment can help demonstrate the issue with treated wastewater discharge, especially in arid regions with low stream flows. It can potentially help improve the acceptance of treated wastewater-based reuse. To conclude, the novel framework helps to assess real-life case studies in a more integrated and holistic way. It can help make decisions related to decentralization and source separation by simultaneously considering the water treatment, energy production, and food production sectors.

The illicit connections between sewage and stormwater pipes result in the discharge of untreated sewage into receiving rivers, posing significant odor and health hazards. While thioethers are recognized as key odorants in sewage systems, their distribution in stormwater systems remain poorly characterized. This study analyzed 12 types of thioethers in stormwater pipes sampled at 21 sites in China. Advanced analytical techniques, including Mantel analysis and Structural Equation Modeling, were employed to examine the relationships between overlying water properties, sediment microbial characteristics, and thioether concentrations. Results showed that sediment thioether loads (36.77 ± 50.14 μg S/m; range: 7.24–99.96 μg S/m) were substantially higher than those in the overlying water (12.02 ± 42.52 μg S/m; range: 0.03–92.76 μg S/m), highlighting sediment as a critical pollution reservoir. Dissolved oxygen, NH₃-N, and terrestrial-derived dissolved organic nitrogen were identified as key factors shaping sediment microbiome composition, particularly fermentative, sulfate-reducing, and denitrifying bacteria, which in turn drives thioether formation. Specifically, dominant compounds like dimethyl disulfide and dimethyl trisulfide were found to be produced through the anaerobic fermentation of methionine and redox conversion of methanethiol, as well as the anaerobic fermentation of cysteine and methylation of polysulfides. Humic substances could facilitate methanethiol redox conversion and polysulfide methylation by serving as methyl donors and enhancing electron transfer efficiency. Additionally, NH₃-N may promote microbial metabolism by providing amino groups essential for the synthesis of metabolic precursors. Therefore, effective mitigation of odorous thioethers in stormwater systems necessitates integrated strategies targeting both sulfur-containing organic precursors and nitrogen-rich pollutants.

This review explores recent advancements in modeling the flow behavior of Herschel-Bulkley (HB) fluids in pipes, discussing theoretical, semi-empirical, computational, and experimental methods. While the laminar flow of non-Newtonian HB fluids can be effectively modeled using first-principle physics, significant challenges remain in turbulent and transitional flow regimes. Existing turbulence models, though widely used, may not always fully align with experimental data, often requiring further validation or complex mathematical tuning, leading to higher computational costs. Further, the transition to turbulence in HB fluids is influenced by shear-thinning and yield stress, yet current models often fail to account for this delayed transition. Consequently, stability and Reynolds number-based transition models can exhibit inconsistencies, limiting their broader applicability. Progress is further hindered by limited experimental studies, constrained by resolution, attenuation, cost, and material combinations. Inaccuracies in rheological modeling—due to inappropriate shear rate ranges, curve-fitting techniques, or simplifying assumptions such as homogeneity and non-elasticity—further complicate flow predictions. Through this review, we delve deeper into the state-of-the-art modeling of HB fluids, highlighting progress and these challenges. Addressing these limitations requires advanced experimental and numerical studies, particularly for near-wall measurements, to better capture flow complexities and improve model predictions. This could also facilitate the development of data-driven approaches and operational envelopes that define their validity thresholds. Future research should also prioritize the independent effects of yield stress and shear-thinning properties while considering material attributes and settling phenomena in non-Newtonian suspensions. Ultimately, these advancements will enable more accurate flow predictions and practical solutions for industrial applications.

Urban drainage network models (UDNMs) have been widely used to facilitate flood management. Typically, a UDNM is developed using data from Geographic Information Systems (GIS), and hence it consists of many short pipes and connection nodes or manholes. To improve modeling efficiency, a GIS-based model is generally skeletonized by removing many elements. However, there has been surprisingly a lack of knowledge on to what extent such skeletonization can affect the model's simulation accuracy, resulting in uncertainty in flood risk estimation. This paper makes the first attempt to quantitatively evaluate multidimensional impacts of different skeletonization levels on hydraulic properties of UDNMs. This goal is achieved by a new evaluation framework comprising of eight existing and new metrics that make use of hydrographs, main pipe hydraulics and flood distribution properties. A real-life UDNM is used to illustrate the new framework under various rainfall conditions and different skeletonization levels. The new framework is also used to compare the performance of two compensation methods in mitigating impacts caused by model skeletonization. Results obtained show that: (a) model skeletonization can significantly affect the magnitude of peak flow at the outfall, with a maximum overestimation of up to 20%, (b) hydraulics in main pipes can also be affected by model skeletonization with the maximum flow increasing up to 35%, and (c) model skeletonization may significantly alter the flood distribution properties which has been largely ignored in past studies. These findings provide guidance for UDNM skeletonization, where their associated impacts should be aware in engineering practice.

As cities expand and land becomes built over, more rainwater will run off rather than infiltrate or evapo(trans)pirate, increasing the likelihood of urban pluvial flooding. Stormwater management and planning is essential to ensure that urban areas are well adapted to climate change, involving cooperation between diverse actors with their own objectives. Current tools to support decision-making have a narrow technical focus and do not incorporate the multi-actor context. In this paper, we present a serious game called Urban dRain, developed with the aim to integrate technical assessment of blue, green and grey solutions and actor negotiation. In the game, participants are challenged to develop a stormwater management strategy for a Dutch neighbourhood in multiple rounds, first within their own separate groups, and then collectively. We present results from validation and play-testing the final game prototype with 70 students and researchers. Results show that the game supports socio-technical learning by encouraging players to come up with a range of stormwater management plans and negotiate for their individual goals while achieving a collective goal. The game demonstrates potential to bring actors with varying perspectives together and co-develop solutions to pluvial flooding, overcoming limitations of existing technology-focused tools.

The analysis of the spatial distribution of drinking water temperature (DWT) in the drinking water distribution system (DWDS) can allow for the detection of hotspots and the identification of suitable mitigation interventions to enhance the climate resilience. For this purpose, a water temperature model is implemented in EPANET-MSX and coupled with the hydraulic model of the DWDS in the town of Almere (the Netherlands). This model is then used to assess the effectiveness of a range of interventions against the unwanted water warming under a climate scenario of an extreme air temperature increase in a Dutch summer. Finally, a solution scenario is suggested to comply with the Dutch legislative limit of 25 °C on DWT at the tap.

Researchers and practitioners have extensively utilized supervised Deep Learning methods to quantify floating litter in rivers and canals. These methods require the availability of large amount of labeled data for training. The labeling work is expensive and laborious, resulting in small open datasets available in the field compared to the comprehensive datasets for computer vision, e.g., ImageNet. Fine-tuning models pre-trained on these larger datasets helps improve litter detection performances and reduces data requirements. Yet, the effectiveness of using features learned from generic datasets is limited in large-scale monitoring, where automated detection must adapt across different locations, environmental conditions, and sensor settings. To address this issue, we propose a two-stage semi-supervised learning method to detect floating litter based on the Swapping Assignments between multiple Views of the same image (SwAV). SwAV is a self-supervised learning approach that learns the underlying feature representation from unlabeled data. In the first stage, we used SwAV to pre-train a ResNet50 backbone architecture on about 100k unlabeled images. In the second stage, we added new layers to the pre-trained ResNet50 to create a Faster R-CNN architecture, and fine-tuned it with a limited number of labeled images (≈1.8k images with 2.6k annotated litter items). We developed and validated our semi-supervised floating litter detection methodology for images collected in canals and waterways of Delft (the Netherlands) and Jakarta (Indonesia). We tested for out-of-domain generalization performances in a zero-shot fashion using additional data from Ho Chi Minh City (Vietnam), Amsterdam and Groningen (the Netherlands). We benchmarked our results against the same Faster R-CNN architecture trained via supervised learning alone by fine-tuning ImageNet pre-trained weights. The findings indicate that the semi-supervised learning method matches or surpasses the supervised learning benchmark when tested on new images from the same training locations. We measured better performances when little data (≈200 images with about 300 annotated litter items) is available for fine-tuning and with respect to reducing false positive predictions. More importantly, the proposed approach demonstrates clear superiority for generalization on the unseen locations, with improvements in average precision of up to 12.7%. We attribute this superior performance to the more effective high-level feature extraction from SwAV pre-training from relevant unlabeled images. Our findings highlight a promising direction to leverage semi-supervised learning for developing foundational models, which have revolutionized artificial intelligence applications in most fields. By scaling our proposed approach with more data and compute, we can make significant strides in monitoring to address the global challenge of litter pollution in water bodies.

Low-lying coastal cities are vulnerable to compound floods caused by many factors including river flows, tides and local rainfall. Many previous studies focus on the impacts of rainfall and tidal levels (two driving factors) on estuaries or regions near the main single river, while research about the three influencing factors on the floods for complex urban river networks remain relatively scarce. This paper proposes a framework to simulate the coastal compound flooding process, with a series of scenarios designed to explore the main flooding drivers and their worst possible combination. The approach is used to quantify the induced flooding consequences, in which a two-dimensional model is employed to conduct a case study for a city with complex river networks. The Qianshan River basin in Zhuhai City, located in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) of China is used as the case study. Application results show that: (i) the occurrence and properties of flooding in this coastal area are jointly affected by upstream river flows, downstream tidal levels and local rainfall intensity with different contribution levels, (ii) compared to tidal levels, local rainfall intensity has a larger impact on the flooding levels of the study region, and (iii) the worst scenario where a 50-year return period upstream river flow co-occurring with high tide and a 100-year return period rainfall can produce a flooding area up to 50 km2 for the study region, with an average of 1.3 m flooded depth. The proposed framework with its findings offers an in-depth understanding and quantification of coastal urban flood characteristics within the GBA of China.

Biofilm formation in drinking water distribution systems (DWDSs) poses challenges to water quality and system integrity. Traditional measurement methods often involve intrusive techniques, disrupting the biofilm ecosystem, while non-intrusive methods offer promising alternatives. This paper explores the feasibility of using non-intrusive temperature sensing to monitor biofilm growth in PVC pipes. Through experiments using the SLIMER 2.0 setup, the biofilm accumulation’s impact on the heat transfer properties is investigated. Preliminary results show successful biofilm growth under controlled conditions, with temperature measurements revealing alterations in heat resistance, hence providing a basis for biofilm monitoring. This study contributes to advancing biofilm monitoring techniques, offering insights for improved water quality management in DWDSs.

The inflow and infiltration (I&I) is an issue for many urban sewer networks (USNs), which can significantly affect system functioning. Placing sensors within the USNs is a typical approach to detect large I&I event, but deploying a limited number of sensors while achieving maximum detection reliability is challenging. While some methods are available for sensor placement, they are generally heuristic search-based methods (HSBMs) and hence the resultant sensor placement strategies (SPSs) are variable over different algorithm runs or parameterizations. This paper develops a new deterministic two-stage clustering method for SPS optimization based on information entropy. Within the first stage, the Spectral Clustering method is applied to assign USN nodes to different clusters according to their joint entropy. In the second stage, the topology structure property is considered to enable further clustering for improving detection reliability. Average I&I detection reliability is used to select clusters and the optimal SPS is identified by maximizing joint entropy of all possible solutions where a single sensor is assigned to each selected cluster. The proposed method and two existing HSBMs are applied to a real USN and their performance is compared. The results obtained show that: (i) a strong correlation coefficient R (R > 0.95) is observed between joint entropy and SPS's detection reliability, which has not been revealed before, (ii) the proposed method consistently outperforms the other two approaches in efficiently offering SPSs with about 7–15 % higher detection reliability, and (iii) the proposed method provides the optimal SPS in a deterministic manner, which makes it attractive for engineering applications.

Computational models for water resources often experience slow execution times, limiting their application. Metamodels, especially those based on machine learning, offer a promising alternative. Our research extends a prior Graph Neural Network (GNN) metamodel for the Storm Water Management Model (SWMM), which efficiently learns with less data and generalizes to new UDS sections via transfer learning. We extend the metamodel’s functioning by adding flowrate prediction, crucial for assessing water quality and flooding risks. Using an Encoder–Processor–Decoder architecture, the metamodel displays high accuracy on the simulated time series. Future work is aimed at incorporating more physical principles and testing further transferability.

The concept of circular economy, aiming at increasing the sustainability of products and services in the water and other sectors, is gaining momentum worldwide. Driven by this concept, novel bio-composite materials produced by recovering resources from different parts of the water cycle are now manufactured in The Netherlands. The new materials are used for different products such as canal bank protection elements, as an alternative to similar elements made of hardwood. As much as these new materials are appealing from the sustainability point of view, they may leach toxic substances into the aquatic environment given some of their ingredients, e.g., cellulose recovered from wastewater treatment. Therefore, a methodology for the assessment of related environmental risks is needed and it does not exist currently. This paper addresses this knowledge gap by presenting a framework for this. The framework is based on European environmental risk assessment guidelines, and it includes four key steps: (i) hazard identification, (ii) dose–response modelling, (iii) exposure assessment and (iv) risk characterisation (i.e. assessment). As part of the first step, laboratory leaching tests were carried out to evaluate the potential release of specific chemical substances such as heavy metals and resin compounds into the aquatic environment. Laboratory test results were then used as input data to evaluate the risk of potential leaching from canal bank protection elements into surface water. A deterministic model was used first to identify the chemicals exceeding the guideline threshold. Subsequently, a stochastic model was applied to evaluate the environmental risks across a range of leachate concentrations and water velocities in the canal, thereby simulating a broader spectrum of possible situations. The risk analyses were conducted for four alternative bio-composite materials made of different ingredients, two different flow conditions (stagnant water and advective flow) in two types of canals (wide ditch and primary watercourse) and for two different water levels based on season conditions (summer and winter conditions). The results obtained from leaching tests identified Cu, Mn, Zn, styrene and furfuryl alcohol as potentially troublesome chemicals. In the case of stagnant water, the absence of a flow rate increases the residence time of the chemicals in the surface water, resulting in a higher PEC/PNEC (i.e. risk) value. However, under stagnant case conditions, environmental risks for all chemicals considered turned out to be below the safety threshold. In the advective case, the existence of a flow rate, even at low velocities simulating the conditions of ‘almost no flow,’ contributes to increased dilution, resulting in lower PEC/PNEC ratio values. The results presented here, even though representing real-case scenarios, are only indicative as these are based on laboratory leaching tests and a number of assumptions made. Additional field tests involving collecting and analysing water and sediment samples from the canal where the canal bank protection elements are located, over a prolonged period, are required to come up with more conclusive findings.

Intermittent water supply systems are prone to air entrapments during the pipe filling phase. This work aims to analyse and discuss the numerical results obtained by applying the recently developed AirSWMM model, an extension of SWMM incorporating air phase, to a laboratory network. Experimental data consisting of pressure-head at multiple locations and video recordings of air entrapments are collected in a single loop network with a high point, for different pipe-filling conditions, system layouts and node elevations. Experimental tests have shown that the air entrapment occurred not only at the high point but also throughout the pipe network, creating air pockets with elongated shapes and larger volumes than for single pipes. AirSWWM model with air-entrapment formation, growth and transport is tested in the pipe network, and results are compared with measurements. AirSWWM model can correctly locate large air pockets but underestimates their volume.

Isolation valve closures (IVCs) can effectively assist pipe maintenance and management in water distribution systems (WDSs), but they inevitably cause secondary impacts on the WDS’s performance. Previous studies have mainly focused on how to optimally operate or locate valves, but few efforts have been made on investigating the secondary impacts induced by IVCs. To this end, six quantitative metrics are proposed to comprehensively evaluate physical, hydraulic, and water quality impacts caused by IVCs. These metrics are used to explore how different network topologies, valve closing strategies, and valve placement strategies affect an IVC’s overall impact on WDS performance. Applications to three real WDSs show the following: (1) the proposed metrics can effectively reveal underlying impacts caused by IVCs, especially the associated water quality risk that has rarely been considered before; (2) in addition to their surrounding pipes, IVCs can affect the water quality in pipes that are far away from the isolated segments; (3) a highly looped WDS is more likely to have higher water quality risk (e.g., due to flow direction reversal) but a lower hydraulic influence level (e.g., low pressure) compared to a WDS with many branched structures; and (4) while closing valves near the failed pipe is an overall strategy to reduce hydraulic impacts, it may also produce high water quality risk. The proposed metrics and the assessment framework are practically meaningful as they offer not only an improved understanding of the secondary impacts caused by IVCs, but also guidance for the decision-making process regarding valve maintenance and management.

A biokinetic model based on BioWin's Activated Sludge Digestion Model (ASDM) coupled with a nitrous oxide (N₂O) model was setup and calibrated for a full-scale wastewater treatment plant (WWTP) Amsterdam West, in the Netherlands. The model was calibrated using one year of continuous data to predict the seasonal variations of N₂O emissions in the gaseous phase. This, according to our best knowledge, is the most complete full-scale data set used to date for this purpose. The results obtained suggest that the currently available biokinetic model predicted the winter, summer, and autumn N₂O emissions well but failed to satisfactorily simulate the spring peak. During the calibration process, it was found that the nitrifier denitrification pathway could explain the observed emissions during all seasons while a combination of the nitrifier denitrification and incomplete heterotrophic denitrification pathways seemed to be dominant during the emissions peak observed during the spring season. Specifically, kinetic parameters related to free nitrous acid (FNA) displayed significant sensitivity leading to increased N₂O production. The obtained values of two kinetic parameters, i.e., the FNA half-saturation during ammonia oxidising bacteria (AOB) denitrification and the FNA inhibition concentration related to heterotrophic denitrification, suggested a strong influence of the FNA bulk concentration on the N₂O emissions and the observed seasonal variations. Based on the suboptimal performance and limitations of the biokinetic model, further research is needed to better understand the biochemical processes behind the seasonal peak and the influence of FNA.

The increasing focus on sustainability and circularity is driving the global production of environmentally friendly products. The Netherlands started producing new bio-composite materials which are created by reclaiming resources from various sectors of the water industry. These materials can be used for a variety of applications including façade elements in buildings. However, their potential environmental impact, particularly with regard to leaching of potentially harmful substances into surface water, necessitates further evaluation. To address this issue, a systematic environmental risk assessment methodology combined with novel experimental data is presented here. To collect this data, façade panels made of two different bio-composite materials were first subjected to a series of laboratory tests, including analysis in both new and weathered forms, the latter subject to a cyclic UV radiation and high humidity, in order to simulate the effects of aging. Leaching tests were then conducted to determine the potential release of specific chemical substances such as heavy metals and resin compounds, under two different rainfall conditions (every day and more extreme). The data generated this way was used to perform the risk assessment using the existing European ERA framework. The results obtained reveal different leaching behaviour of the new and weathered samples, as well as between the two analysed bio-composite materials, depending on the rain intensity. To overcome the uncertainties caused by the limited input data, a sensitivity analysis was carried out whereby leaching concentrations and rainfall intensities were varied and their influence on the environmental risk was assessed. The results obtained demonstrated that, despite some variability, both materials appear safe to use, i.e., with estimated risks below the established safety threshold. While these findings provide a preliminary indication, they are based on laboratory conditions and assumptions hence further field studies are recommended to obtain more definitive conclusions.

The paper proposes a novel methodology to locate and quantify entrapped air pockets created during pipe-filling events often found in intermittent water supply systems. Different filling conditions were tested in an experimental pipe with a high point. Measurements were taken and video recordings were carried out to assess air pocket volumes for different air release conditions at the downstream end of the pipe. The stochastic nature of air pocket creation resulted in varying air volumes. A new numerical model capable of simulating the air pocket creation, dragging and entrainment has been proposed. The new model, AirSWMM, was implemented as an extension of the Stormwater Management Model (SWMM) with stochasticity of air pocket formation reproduced by simulations with different air entrainment rates. The obtained numerical results show that the proposed model, even though based on a single-phase one-dimensional flow, can accurately locate and approximately quantify the entrapped air pocket volumes.

Sectorization of a water distribution network (WDN) into district meter areas (DMAs) is considered a key strategy for an efficient management of WDNs. Typically, it involves a two-stage procedure: a clustering stage, in which the division of the WDN into clusters is performed, and a dividing stage, which involves the placement of valves and flow meters on the cluster’s boundary pipes to define the DMAs. While recently published methods attempt to enhance both the clustering and dividing stages, they fail to provide decision-making flexibility. They also neglect to consider the presence of existing valves in the WDN, which can significantly affect the evaluated implementation cost, often considered the primary decision-driving factor. This paper presents improvements to the previously introduced DeNSE method for sectorization of WDNs, aiming to address these deficiencies. The methodology consists of a clustering stage, based on the network uniformity index, and a dividing stage, in which the originally used heuristic procedure is replaced with Genetic Algorithm (GA) optimization, minimizing implementation cost. Consideration of existing valves in WDN and criteria for water supply security are also included in the dividing stage to offer a better estimate of implementation costs and post-sectorized operational efficiency of the WDN. Finally, GIS visualization is implemented, and a hydraulic model of the sectorized WDN (EPANET file) is generated, providing practitioners with valuable insights and decision-making flexibility. The methodology is tested on a part of the Amsterdam WDN in the Netherlands, serving as a pilot for methodology evaluation. A range of feasible sectorization solutions is generated and compared based on implementation cost and three performance indicators (PIs). The paper provides an in-depth discussion on the selection of preferable sectorization solution. The reported results demonstrate the method’s efficiency in optimizing sectorization solutions with minimum implementation cost whilst preserving the WDN operational efficiency and meeting the local design criteria.

Sewage pipe defects can significantly affect the urban water environment, like leakage of pollutants through pipe cracks to groundwater. Currently, sewage pipe defects are detected mainly through closed-circuit television inspection, which is conducted manually and is time-consuming. This study proposes an integrated deep-learning-based algorithm to detect and quantify pipe cracks from images, namely the Crack Detection and Characterization (CDC) method. The method is based on models created in two steps (i) crack detection by semantic segmentation, and (ii) crack length quantification using an innovative algorithm. The CDC algorithm is verified by images both artificially created in the laboratory and from actual inspection. For both laboratory and field cases, the CDC method is verified precisely. From the results, the CDC method exhibited a higher accuracy in crack identification and length quantification than other existing models. The results also show that the deblurring process can greatly improve accuracy. This study can contribute to decision-making in sewage pipe maintenance and water environment management by providing an innovative way of more efficient and accurate pipe defect assessment compared with traditional labor-intensive work.

Nitrous oxide (N₂O) emissions from wastewater treatment plants (WWTPs) exhibit significant seasonal variability, making accurate predictions with conventional biokinetic models difficult due to complex and poorly understood biochemical processes. This study addresses these challenges by exploring data-driven alternatives, using long short-term memory (LSTM) based encoder-decoder models as basis. The models were developed for future integration into a model predictive control framework, aiming to reduce N₂O emissions by forecasting these over varying prediction horizons. The models were trained on 12 months and tested on 3 months of data from a full-scale WWTP in Amsterdam West, the Netherlands. The dataset encompasses seasonal peaks in N₂O emissions typical for winter and spring months. The best performing model, featuring a 256–256 LSTM architecture, achieved the highest accuracy with test R² values up to 0.98 across prediction horizons spanning 0.5 to 6.0 h ahead. Feature importance analysis identified past N₂O emissions, influent flowrate, NH₄⁺, NO_x, and dissolved oxygen (DO) in the aerobic tank as most significant inputs. The observed decreasing influence of historical N₂O emissions over extended prediction horizons highlights the importance and significance of process variables for the model's performance. The model's ability to accurately forecast short-term N₂O emissions and capture immediate trends highlights its potential for operational use in controlling emissions in WWTPs. Further research incorporating diverse datasets and biochemical process inputs related to microbial activities in the N₂O production pathways could improve the model's accuracy for longer forecasting horizons. These findings advocate for hybridising deep learning models with biokinetic and mechanistic insights to enhance prediction accuracy and interpretability.