Worst-case Conditions Research Articles

Hyperparameter Learning Under Data Poisoning: Analysis of the Influence of Regularization via Multiobjective Bilevel Optimization.

Machine learning (ML) algorithms are vulnerable to poisoning attacks, where a fraction of the training data is manipulated to deliberately degrade the algorithms' performance. Optimal attacks can be formulated as bilevel optimization problems and help to assess their robustness in worst case scenarios. We show that current approaches, which typically assume that hyperparameters remain constant, lead to an overly pessimistic view of the algorithms' robustness and of the impact of regularization. We propose a novel optimal attack formulation that considers the effect of the attack on the hyperparameters and models the attack as a multiobjective bilevel optimization problem. This allows us to formulate optimal attacks, learn hyperparameters, and evaluate robustness under worst case conditions. We apply this attack formulation to several ML classifiers using L2 and L1 regularization. Our evaluation on multiple datasets shows that choosing an "a priori" constant value for the regularization hyperparameter can be detrimental to the performance of the algorithms. This confirms the limitations of previous strategies and evidences the benefits of using L2 and L1 regularization to dampen the effect of poisoning attacks, when hyperparameters are learned using a small trusted dataset. Additionally, our results show that the use of regularization plays an important robustness and stability role in complex models, such as deep neural networks (DNNs), where the attacker can have more flexibility to manipulate the decision boundary.

Proceedings of the 2023 Viral Clearance Symposium, Session 3: 2023 VCS New Modalities in Chromatography and Adsorptive Filters.

The session provided an update on the application and mechanistic understanding of intensified unit operations (e.g., mixed mode depth filters, mixed mode AEX) since the last conference in 2019. One of the key gaps identified in the 2019 Viral Clearance Symposium session on the topic was for more investigation required to achieve a clear understanding of the molecular mechanisms of virus removal and the relevance of different moleculés interactions including resin, virus, and product. Further investigation into worst-case conditions for these unit operations is also warranted. One of the key outcomes from that 2019 discussion was also that multimodal anion exchangers can have robust and effective virus removal, depending on process and impurities-an observation that was recapitulated with more specific case studies and evidenced by broader application of these chromatographic resins in late-stage regulatory filings.

Machine Learning Based Foreign Object Detection in Wireless Power Transfer Systems

As power electronics continue to drive advancement in electronic devices by managing the system's energy flow, it is advantageous to examine all areas of energy transfer. Wireless power transfer is one section seeing increased adoption in consumer, industrial, and healthcare applications. This consequently increases the concern for safety, reliability, and efficiency. Power transmission to any unintended load is a primary safety issue in high-power wireless power transfer applications, such as unmanned aircraft systems. Foreign object detection is utilized to selectively transmit power to only the intended target to mitigate this issue. This paper seeks to demonstrate a novel alternative method of using machine learning and artificial intelligence techniques for performing foreign object detection solely on the transmitter portion of the system by only monitoring the TX coil current. This method is first evaluated in SPICE simulation software. Then by transitioning to a closed-loop hardware-based testbed designed to be deployed in unmanned aircraft systems. During the test process for the novel FOD method, the aggregate detection accuracy achieved was 83%, including worst-case misalignment conditions with the RX coil. When the outlying worst-case conditions are excluded, the detection accuracy improves to 97%. To measure the FOD effectiveness in preventing temperature rise in a metallic object, the metallic object was placed at the peak power draw location on the transmitter coil. When the metallic object is left at this location for extended periods, the novel FOD method lowers the maximum temperature reached by 86.8°C when compared to no FOD being active. The primary advantage to this novel method is the simplistic hardware additions, which only require minimal alterations to a WPTS without any FOD implemented. Along with this, implementing this foreign object detection method in the transmitter would mitigate the need for a direct communication link between the receiver and transmitter. As a result, the entire system complexity would decrease, thus increasing the UAS power density while maintaining a safe, effective, and reliable wireless power transmission method.

Passive sampling of herbicides above sediments at sites with losses of submerged macrophytes in a mesotrophic lake

Declines of submerged macrophytes (SUM) were monitored in littoral zones of the deep, mesotrophic lake Suhrer See (Northern Germany) since 2017. Drastic losses coincided with intense agriculture in sandy sub-catchments and precipitation. All lines of evidence pointed to a causal connection with subsurface discharge indicating that herbicide application might have caused the effects. Passive sampling was applied in 2022 to elucidate, whether herbicides were really present at sites of losses and if so, in ecotoxicological relevant concentrations. Samplers were exposed on top of lake sediments in 2 m depth and under worst case conditions, i.e., at sites, known for losses of the whole functional group of SUM and at the beginning of the vegetation period. At this time, SUM diaspores were most vulnerable to repression of development and the subsurface discharge was high in the same instance. The potential ecotoxicological relevance of detected herbicide concentrations was assessed with a toxic units (TU) approach, with reference to acute effect concentrations (EC50 of green algae, 72 h, growth). The TU ranged from 0.001 to 0.03. Most concentrations exceeded the threshold of relevance set by an assessment factor of 1000, i.e., TU > 0.001. Locally applied herbicides acted by suppressing developmental stages, and the sum of TU exceeded 0.02 at all sites, mainly due to diflufenican. Not applied locally, terbuthylazine and its relevant metabolites, including terbutryn, acted by inhibiting photosynthesis, and the sum of TU reached 0.005. On this base, diflufenican was assessed to be likely a main stressor, all other detected herbicides to be potentially relevant. Uncertainties and knowledge gaps were specified. The result of the chemical risk assessment was counterchecked for consistence with biological monitoring data within a whole lake perspective. Concepts of empirical and advanced causal attribution methodology were applied to get a grip to the ecological causal field and to protection.

Numerical generation of omnistrain failure envelopes

Traditional failure criteria for composites are usually formulated in material coordinates and depend on all three inplane stresses, hence failure evaluation depends on the ply angle. The omnistrain failure envelope describes the most critical failure envelope in strain space irrespective of ply orientation. This independence of ply orientation leads to an isotropic failure criterion that depends only on the principal strains. Omnistrain envelopes greatly simplify the task of design and optimisation of composite laminates. This paper proposes a numerical technique to generate omnistrain failure envelopes for different composite failure criteria. The failure index, describing how far a point in strain space is from the failure boundary, is used to describe the failure surface. Assuming convexity of the failure surface, a set of points is generated on the surface, and the convex hull algorithm is used to generate a polygonal approximation of the failure surface. Representing strains in terms of principal strains and the angle between the principal and material coordinates, allows us to eliminate the angle analytically by considering the worst case condition. The omnistrain envelope is thus directly generated from the approximate three-dimensional failure surface. The proposed algorithm does not require analytic expressions of the failure surface. An adaptive algorithm is proposed to generate the omnistrain envelope with relatively small number of points. As demonstration of the proposed algorithm, the omnistrain envelopes for various composite materials are generated for a number of composite failure criteria. The omnistrain envelopes generated for the Tsai-Wu criteria accurately match to existing analytic expressions.

Aseptic microwave sterilization and validation of food containing particles

Microwave and ohmic heating are becoming a more popular choice for heating in aseptic processing due to rapid heating for bacteriological destruction and better product quality. Due to regulatory constraints and the complexities of validation requirements, the food industry has been struggling to launch aseptic low-acid products containing particles into the market with confidence. This study aims to find an alternative approach to simulate and mimic processing conditions by modeling and validating the model process by using surrogate particles and microorganisms. A mathematical model was used to simulate the heat transfer in the carrier fluid and solid particles. For the validation of the process, a simulated particle was designed to create the worst-case conditions, being both the fastest moving and slowest heating compared to the real food particles. These particles with conservative thermal and physical properties were inoculated with Geobacillus stearothermophilus spores as a surrogate microorganism for the microbiological validation of the thermal process. Magnetic implants were embedded in simulated spherical particles (12.77 mm and 15.88 mm) to measure the residence time distribution. Total accumulated lethality using the surrogate system was estimated at F0 > 7.7 min. The microwave aseptic system was microbiologically validated to produce shelf-stable particulate food products.

Sources of bias and limitations of thrombinography: inner filter effect and substrate depletion at the edge of failure algorithm

BackgroundFluorogenic thrombin generation (TG) is a global hemostasis assay that provides an overall representation of hemostasis potential. However, the accurate detection of thrombin activity in plasma may be affected by artifacts inherent to the assay-associated fluorogenic substrate. The significance of the fluorogenic artifacts or their corrections has not been studied in hemophilia treatment applications.MethodsWe sought to investigate TG in hemophilia plasma samples under typical and worst-case fluorogenic artifact conditions and assess the performance of artifact correction algorithms. Severe hemophilic plasma with or without added Factor VIII (FVIII) was evaluated using commercially available and in-house TG reagents, instruments, and software packages. The inner filter effect (IFE) was induced by spiking elevated amounts of fluorophore 7-amino-4-methylcoumarin (AMC) into plasma prior to the TG experiment. Substrate consumption was modeled by adding decreasing amounts of Z-Gly-Gly-Arg-AMC (ZGGR-AMC) to plasma or performing TG in antithrombin deficient plasma.ResultsAll algorithms corrected the AMC-induced IFE and antithrombin-deficiency induced substrate consumption up to a certain level of either artifact (edge of failure) upon which TG results were not returned or overestimated. TG values in FVIII deficient (FVIII-DP) or supplemented plasma were affected similarly. Normalization of FVIII-DP resulted in a more accurate correction of substrate artifacts than algorithmic methods.ConclusionsCorrection algorithms may be effective in situations of moderate fluorogenic substrate artifacts inherent to highly procoagulant samples, but correction may not be required under typical conditions for hemophilia treatment studies if TG parameters can be normalized to a reference plasma sample.

Environmental noise predictions using a statistical approach

Noise pollution from industrial premises has the potential to have an impact on neighbouring properties. In New South Wales the Environment Protection Authority has developed the Noise Policy for Industry (NPfI). The NPfI requires noise assessments to be conducted using a typical assessment procedure that results in setting project noise trigger levels (PNTLs) which are the benchmark levels against which potential noise impacts from industrial developments are judged. The PNTLs are then used in Environmental Protection Licences to regulate the premises. The NPfI requires noise assessments to consider worst-case (i.e., noise-enhancing) meteorological conditions. In addition, regulators and consent authorities often require noise assessments to include a range of other conservative assumptions such as worst-case source levels and locations. This process appears to be very conservative, particularly considering that the PNTL in each assessment period (day, evening and night) are based on lowest 10th percentile background noise levels. This paper, through a statistical approach of using random probability distributions, will identify the likelihood that modelled noise levels and associated impacts are likely to occur and comment on the appropriateness of the conservative assessment process.

Robust actions for improving supply chain resilience and viability

It is vital for supply chains (SCs) to survive the dramatic and long-term impacts from severe disruptive events, such as COVID-19 pandemic. SC viability, an extension of SC resilience, is increasingly attracting attention from both academics and practitioners. To improve SC viability, the government can perform a series of costly interventions on SCs. Due to data scarcity on unpredictable disruptive events, especially under the pandemic, the information related to SC partners may not be accurately obtained. In this paper, we investigate a novel SC resilience and viability improving problem under severe disruptive events, in which only the probability intervals of SC partners’ states are known. The problem consists of the selection of appropriate intervention actions, respecting a limited capital budget. The objective is to minimize the worst-case disruption risk of the manufacturer. Specifically, Causal Bayesian Network (CBN) is applied to quantify the SC ripple effects; Do-calculus technique is used to measure the benefits of government intervention actions; and robust optimization is employed to minimize the disruption risk under the worst-case condition. For the problem, a new robust optimization model that combines the CBN and the Do-calculus is constructed. Based on analyses of problem features, an efficient problem-specific branch-and-bound (PS-BAB) algorithm is proposed to solve the problem exactly. Experimental results show the efficiency of our methodology and managerial insights are drawn.

Experimental and numerical investigation of leakage and diffusion accidents based on nuclear energy hydrogen production system

The high-temperature gas-cooled reactor, as a fourth-generation nuclear power system, possesses a distinct advantage of high core outlet temperature, making it a viable option for large-scale, low-carbon hydrogen production. However, before this technology can be widely adopted, it is crucial to conduct a safety assessment of the nuclear hydrogen production system, considering the unique physical and chemical properties of hydrogen. This study aimed to address this requirement by implementing two experimental setups, utilizing helium as a substitute for hydrogen. Concentration distribution measurements were performed in both open spaces and inside a wind tunnel. The concentration decay rates were found to be 0.177 under windless conditions and 0.313 in windy conditions. Subsequently, numerical models were employed to validate the feasibility of leakage and diffusion calculations for subsonic leakage jets and high-pressure underexpanded leakage jets, respectively. Moreover, multiple simulation calculations were conducted to design leakage accident scenarios based on the nuclear hydrogen production system. The corresponding concentration decay patterns were summarized, and evaluation relational expressions were proposed to determine diffusion distance and explosion distance for the nuclear energy hydrogen production system, accounting for the influence of buoyancy effects and incorporating the average concentration of combustible hydrogen clouds. Lastly, considering the worst-case accident conditions of the current system, a fan reverse blowing scheme was suggested to mitigate safety risks. The findings indicated that this scheme could potentially reduce the separation distance from 479 m to 183 m. Overall, this research contributes valuable insights into the safety assessment and design of nuclear hydrogen production systems.

Non-Contact Monitoring of Operating Conditions for Solar Cells in a Photovoltaic Module Using a Surface Potential Meter for Detecting the Risk of Fire

Fires in photovoltaic modules are caused by hot spots, which are typically monitored by thermal images. This method helps visualize the hot spot, but it is affected by the environment (solar irradiance, wind, ambient temperature) and is not reproducible. Assessing the heat dissipation of the hot cell can be used for alternative assessment of the fire risk. This method was validated by comparing the value measured by the surface potential meter and the module potential measured directly by adding a bypass measurement circuit. The substantial reverse-bias voltage caused by mismatching or partial shading (depending on the operating conditions) leads to local heat consumption of the partially shaded solar cells and potentially causes fire. The fire risk can be assessed in the worst-case conditions (ex. 1380 W/m2 solar irradiance) by non-contact measurement of the reverse-bias voltage and calculating the heat dissipation and temperature rise. This work suggested that −13 V is the criterion and was close to the known value of reverse voltage for Si cells. The current technology inspects solar cells before assembly to the module, and there is no way of inspecting in the product test or detecting after degradation that can be covered by the proposed method in this work.

Analysis of lubricating oil degradation and its influence on brake specific fuel consumption of a light-duty compression-ignition engine running a durability cycle on a test stand

The Euro 6 emission standard required compliance with legal exhaust emissions limits for newly registered vehicles and obligates light duty vehicle manufacturers to respect the 160,000 km durability requirements for in-service conformity. Although there is no legal limit set for fuel consumption, manufacturers are obligated to decrease the carbon footprint of vehicle fleets in order to obtain carbon neutral mobility beyond 2035. The aim of this paper is to analyse the impact of various oils’ and viscosity grades’ degradation on the change in break specific fuel consumption (BSFC) measured over a standardized durability test cycle. Each oil candidate underwent 300 h of durability test running performed on a test bed, without any oil changes. The purpose of the laboratory test was to reproduce the worst-case operating conditions and degradation process of the long-life engine oil type that can be experienced during extreme real life driving of a vehicle. In order to define the influence of the engine oil deterioration on the BSFC profile, the engine operation parameters were continually monitored throughout the test run. Additionally, chemical analysis of the oil was performed and the solid deposits formed on the turbocharger’s compressor side were evaluated. The test results revealed differences up to 3.5% in the BSFC values between the oil candidates tested over the durability cycle. The observed BSFC increase was directly related to the decrease in engine efficiency and can cause higher fuel consumption of the engine, which in turn has an adverse effect on environmental protection goals.

Fish Monitoring from Low-Contrast Underwater Images

A toolset supporting fish detection, orientation, tracking and especially morphological feature estimation with high speed and accuracy, is presented in this paper. It can be exploited in fish farms to automate everyday procedures including size measurement and optimal harvest time estimation, fish health assessment, quantification of feeding needs, etc. It can also be used in an open sea environment to monitor fish size, behavior and the population of various species. An efficient deep learning technique for fish detection is employed and adapted, while methods for fish tracking are also proposed. The fish orientation is classified in order to apply a shape alignment technique that is based on the Ensemble of Regression Trees machine learning method. Shape alignment allows the estimation of fish dimensions (length, height) and the localization of fish body parts of particular interest such as the eyes and gills. The proposed method can estimate the position of 18 landmarks with an accuracy of about 95% from low-contrast underwater images where the fish can be hardly distinguished from its background. Hardware and software acceleration techniques have been applied at the shape alignment process reducing the frame processing latency to less than 0.5 us on a general purpose computer and less than 16 ms on an embedded platform. As a case study, the developed system has been trained and tested with several Mediterranean fish species in the category of seabream. A large public dataset with low-resolution underwater videos and images has also been developed to test the proposed system under worst case conditions.

ESD Events to Wearable Medical Devices in Healthcare Environments—Part 1: Current Measurements

Wearable medical devices are widely used for monitoring and treatment of patients. Electrostatic discharge can render these devices unreliable and cause a temporary or permanent disturbance in their operation. In a healthcare environment, severe electrostatic discharge (ESD) can occur while a patient, lying down or sitting on a hospital bed with a wearable device, discharges the device via a grounded bedframe. To protect the devices from ESD damage, the worst-case discharge conditions in the usage environment need to be identified. Previous studies by authors revealed that such events could be more severe than the conventional human metal model (HMM). However, the impact of various body postures and device location on the body and the severity of the discharge current compared with HMM have not been investigated for healthcare environments. This study is an attempt to address the gap in the literature by investigating severe discharges in such environments and characterizing their current waveforms for three postures (standing on the floor, sitting, and lying down on a hospital bed), two device locations (hand and waist), and four body voltages (2, 4, 6, and 8 kV). This study highlights that the IEC 61000-4-2 standard may not be sufficient for testing wearable medical devices.

Operational measures to maintaining physical distancing at railway stations

The corona virus disease 2019 (COVID-19) pandemic has increased awareness towards maintaining physical distancing during transportation-related activities. This study presents a microsimulation model to explore operational measures to maintain physical distance among railway station passengers. The secondary and primary data obtained from field surveys are utilized to construct and calibrate the model. The peak hour data is employed to investigate the worst-case conditions. The calibrated model is then utilized to evaluate several operational measures, i.e., changing the headway, increasing the train capacity, increasing the train door duration, and changing the train door rules. From the simulation, it is found that changing the train door rules was ineffective if it was individually implemented. It is concluded that a combination of operational measures provides additional benefits for maintaining physical distancing among passengers.

Mega-Hertz High-Voltage EMAT Pulser Based on LCC Resonant Inverter Using SiC MOSFETs

Pulse power supply is an important field of high-frequency power electronics. In advanced applications, such as ultrasonic sensors, Lidar waves and laser generators, the voltage supply with high amplitude and high frequency is required to achieve sufficient technical performance. Meanwhile, the high power quality is pursued in some sensor applications to obtain better performance. This paper develops a ±1000 V/1.5 MHz LCC resonant pulse inverter for Electro-Magnetic Acoustic Transducer. First, the LCC filter is optimized to achieve the goals of low THD, small gain variation and ZVS soft-switching in a wide load range. The design guideline of LCC filter for non-resistive load conditions is proposed, and the design considerations for burst mode converters are summarized. Second, the start-up and ending transients occur repetitively, which is a common problem for burst mode converters. The proposed pre-charging and damping strategy solves this transient problem. Moreover, by analyzing the circuit state equations, this strategy provides a general method for eliminating transients of resonant converters. Next, the signal receiving circuit uses depletion mode MOSFETs to attenuate the high transmitting load voltage while amplifying microvolt receiving signals. Finally, the experimental results under four worst-case conditions validate the LCC filter design. A control experiment with or without the proposed transient mitigation method has been conducted to verify the effectiveness of the proposed strategy.

Fast Determination of Transient Short-Circuit Current of PM Synchronous Machines via Magnetostatic Flux Maps

The three-phase symmetric short-circuit is a reference fault condition for the qualification of a newly designed permanent magnet (PM) synchronous machine against the risk of irreversible demagnetization. To date, the accurate determination of the peak transient short-circuit current condition requires coupled circuital and transient Finite Element Analysis (FEA) to properly account for magnetic saturation, and several simulations to determine the worst-case pre-fault conditions. This work presents a method for the fast evaluation of the transient short-circuit current of a PM synchronous machine by manipulation of extended flux linkage maps, obtained with magnetostatic FEA or experimental measures. Besides providing a fast and accessible computational tool, the flux-map based method gives insights into the effect of pre-fault conditions, showing that the higher the pre-fault torque and thus flux amplitude, the higher the transient peak current after a fault. Moreover, the paper shows that braking is a more severe pre-fault condition. Finally, the hyper-worst-case short circuit current is also defined as the magnetic property of the machine under test and computed by means of quick FEA iterations without the need for pre-determined flux-linkage maps. The proposed methods are validated against transient FEA using a commercial software and verified experimentally on a commercial motor for traction applications.

Influence of living shoreline elements on wave run up elevations

Nature-based coastal protection, also known as engineering with nature or living shorelines, is becoming increasingly popular due to its dual benefits of reducing coastal flooding and providing ecological and recreational opportunities. In many coastal areas experiencing chronic erosion, changes in sediment supply, composition, and grain size are significant contributing factors to shoreline recession. One living shoreline strategy to consider includes the application of cobbles over more traditional sand nourishments. On sandy beaches that experience high-energy wave conditions, the introduction (or reintroduction) of cobbles can mitigate backshore erosion. Cobble-backed beaches have been found to mitigate the effect of coastal erosion and flooding in laboratory settings and field observations, and they have recently been piloted in locations such as Cape Lookout State Park in Tillamook County, Oregon, and Surfers Point in the City of Ventura, California. However, there are no formal engineering guidelines stipulating the calculation of wave run-up on cobble-backed beaches. This study applies three different wave run-up equations on a living shoreline design (i.e. mixed sand and cobble berm-backed beach) in Malibu, California, and compares the predicted run-up levels with existing condition flood levels for typical and eroded conditions. The different wave run-up equations were designed for cobbles only, revetments, and composite beaches, respectively, where the composite beach equation was most applicable to project design. For typical beach conditions (higher levels of sediment accretion resulting in shallower beach face and berm slopes), all three equations showed a reduction in wave run-up values. When applied to worst-case conditions (i.e. scoured by a creek channel and steeper fronting beach slopes), the equation most applicable to the design showed the highest reduction of total water levels. A sensitivity analysis found that the cobble-backed beach equation predicted the most consistent values of run-up (run-up values changed the least), even when input parameters (slope and water depth) changed. This study shows that cobblebacked beaches hold promise to mitigate coastal flooding in appropriate areas, in addition to being a natural solution for areas experiencing erosion. This study also points to the need for more studies and field observations to validate the run-up levels determined here.

84 Effectiveness of Protective Measures—Diesel Exhaust in Garages of Fire Engines Vehicles and Rescue Services

Abstract Diesel exhaust is a complex mixture of soot particles and many gaseous components. Meanwhile, occupational exposure limits (OEL) have been derived for many of these constituents in Europe. Normally, vehicles of fire brigades and rescue units are parked in garages. To ensure readiness for action, they are connected to the power supply system. If they vehicles move out or return to the garage employees may be exposed to diesel engine emission. We have performed measurements in 7 emergency rescue services and 5 fire-fighting stations to assess the exposure to hazardous substances and compare the effectiveness of protective measures.During the measuremnts the different exhaus emission standards and different type of technical protection measure have been compared. The results of the measurements show, that exposure of the employees should be considered as short-term exposure. The most critical substances are diesel soot particles and nitrogen oxides. Especially, for nitrogen monoxide exceedings of the short-term exposure limit value (STELV) have been measured. It was found, that exposure to diesel soot particles decreased from exhaust emission standard EURO 1 to EURO 6. On the other hand, exposure to nitrogen oxides increased for the newer emission standards. The most effective technical protective measure is a trailing extraction system, which should be applied when driving into tho garage, during parking up to moving out, when it should be removed automatically. Alternatively, a vertical transverse system may be effective, but should be assessed for each garage under worst case conditions.

Statement of the Scientific Panel on Plant Protection Products and their Residues (PPR Panel) on the design and conduct of groundwater monitoring studies supporting groundwater exposure assessments of pesticides.

Groundwater monitoring is the highest tier in the leaching assessment of plant protection products in the EU. The European Commission requested EFSA for a review by the PPR Panel of the scientific paper of Gimsing etal. (2019) on the design and conduct of groundwater monitoring studies. The Panel concludes that this paper provides many recommendations; however, specific guidance on how to design, conduct and evaluate groundwater monitoring studies for regulatory purposes is missing. The Panel notes that there is no agreed specific protection goal (SPG) at EU level. Also, the SPG has not yet been operationalised in an agreed exposure assessment goal (ExAG). The ExAG describes which groundwater needs to be protected, where and when. Because the design and interpretation of monitoring studies depends on the ExAG, development of harmonised guidance is not yet possible. The development of an agreed ExAG must therefore be given priority. A central question in the design and interpretation of groundwater monitoring studies is that of groundwater vulnerability. Applicants must demonstrate that the selected monitoring sites represent realistic worst-case conditions as specified in the ExAG. Guidance and models are needed to support this step. A prerequisite for the regulatory use of monitoring data is the availability of complete data on the use history of the products containing the respective active substances. Applicants must further demonstrate that monitoring wells are hydrologically connected to the fields where the active substance has been applied. Modelling in combination with (pseudo)tracer experiments would be the preferred option. The Panel concludes that well-conducted monitoring studies provide more realistic exposure assessments and can therefore overrule results from lower tier studies. Groundwater monitoring studies involve a high workload for both regulators and applicants. Standardised procedures and monitoring networks could help to reduce this workload.