A multi-criteria evaluation of marine pollution systems based on expert-driven analysis and robust ranking methods Authors: Yongqiang Zhang, Zhe Feng, Chenglin Cao, Wanjun Zhang, Zi Chen Zhu Published: 2026 ABSTRACT AbstractThe selection of effective coastal marine pollution monitoring systems requires integrating multiple technical, economic, and operational considerations. A decision-support approach is developed to compare five monitoring configurations using criteria that capture detection capability, spatial and temporal coverage, cost, robustness, and energy-related constraints. Expert input is obtained from 120 participants, with 110 responses meeting the consistency requirements, yielding an average consistency ratio of 0.064. The calculated weights indicate that detection capability (0.29) and coverage (0.24) are the most important factors in the assessment. The analysis identifies the underwater sensor network as the highest-ranked alternative, achieving a closeness coefficient of 0.56, followed by the cabled coastal observatory (0.53) and mobile autonomous platforms (0.49). The influence of parameter uncertainty is examined through a sensitivity analysis, in which ± 20% variations in criterion weights produce only limited changes in the ranking order. Additional stochastic simulations, based on ± 0.5 perturbations in performance scores over 100 runs, show low variability, with standard deviations below 0.02 across all alternatives. The results demonstrate consistent ranking behavior and highlight the importance of balanced system performance. The proposed approach supports informed decision-making and can be adapted to different monitoring contexts and evolving technological conditions.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-57687-9.Keywords: Coastal monitoring, Decision analysis, AHP, TOPSIS, Robustness analysisSubject terms: Engineering, Environmental sciences, Mathematics and computing, Ocean sciences Introduction Coastal marine environments are increasingly affected by anthropogenic pollution arising from industrial discharge, urban runoff, maritime activities, and climate-driven processes. The growing intensity and spatial variability of these pressures require monitoring systems that are not only accurate but also adaptable to dynamic environmental conditions. Effective monitoring is therefore a prerequisite for early detection of contamination, mitigation of environmental risks, and sustainable coastal management1,2. Recent technological advances have significantly expanded the range of available monitoring solutions. These include autonomous sensing platforms, distributed underwater networks, cabled observatories, and satellite-based systems3,4. Each configuration provides distinct capabilities, yet none offers a comprehensive solution across all operational dimensions. For instance, recent studies highlight that underwater sensor networks provide high-resolution, continuous data streams but entail significant deployment complexity and cost5,6. Autonomous platforms enable adaptive sampling and large-area coverage, although their operational reliability remains constrained by energy limitations and environmental uncertainty7,8. Similarly, satellite-based monitoring has been shown to support large-scale observation, but its effectiveness is limited in subsurface detection and fine-scale pollutant characterization9,10. The coexistence of multiple monitoring technologies with competing strengths introduces a complex decision problem. The selection of an appropriate monitoring system cannot rely on a single performance indicator, as improvements in one dimension are often associated with trade-offs in others. This complexity necessitates structured decision-support approaches that integrate multiple criteria and stakeholder perspectives. Multi-criteria decision-making methods have been widely applied to the evaluation of environmental systems. In particular, the Analytic Hierarchy Process and the Technique for Order Preference by Similarity to Ideal Solution are frequently adopted for their ability to incorporate expert knowledge and provide interpretable rankings. Recent applications in environmental monitoring and infrastructure selection demonstrate the effectiveness of hybrid approaches that combine weighting and ranking mechanisms11,12. However, several limitations remain in the current body of research. First, many studies focus on individual monitoring technologies rather than providing a comprehensive comparison across alternative system configurations. Second, the justification for the selection and weighting of the criteria is often insufficiently discussed, reducing transparency. Third, robustness analysis is frequently limited or absent, despite the inherent uncertainty associated with expert-based evaluations. Finally, comparative validation against baseline methods is rarely performed, making it difficult to assess the added value of the proposed approaches. To address these limitations, a structured evaluation framework is developed for the comparative assessment of coastal marine pollution monitoring systems. The framework integrates expert elicitation, multi-criteria weighting, and distance-based ranking, and is complemented by sensitivity and stochastic analyses to evaluate robustness. The approach aims to provide a transparent, reproducible, and adaptable decision-support tool for selecting coastal monitoring systems. Advances in coastal monitoring technologies Recent research has emphasized the rapid evolution of monitoring technologies in marine environments, particularly with the integration of sensing, communication, and data analytics capabilities. Distributed underwater sensor networks have emerged as a key innovation, enabling continuous monitoring of environmental parameters with high spatial resolution. Studies demonstrate that such systems significantly improve detection accuracy and data continuity compared to traditional approaches13,14. However, their deployment is associated with challenges related to energy consumption, communication reliability, and maintenance. Autonomous platforms, including underwater vehicles and unmanned surface systems, have gained increasing attention due to their flexibility and adaptability. These systems can perform dynamic sampling and respond to transient pollution events. Recent work highlights their effectiveness in complex coastal environments, although limitations in navigation accuracy and operational costs remain significant15. Cabled observatories represent a more stable, infrastructure-intensive solution, providing continuous power and real-time data transmission. They are particularly suitable for long-term monitoring applications, where reliability and data continuity are critical. However, their scalability is limited due to high installation costs and geographical constraints16. Satellite-based remote sensing has also advanced significantly, offering large-scale monitoring capabilities and frequent data acquisition. Recent studies show improvements in spatial resolution and data processing techniques, although challenges remain in detecting subsurface pollutants and distinguishing between different contamination sources17. Application of multi-criteria decision methods in environmental monitoring Multi-criteria decision-making methods have been widely applied in environmental management to support complex decision processes involving multiple conflicting objectives. Among these methods, AHP is commonly used to derive criteria weights based on expert judgments, while TOPSIS is applied to rank alternatives based on their proximity to the ideal solution. Recent studies demonstrate the effectiveness of hybrid AHP–TOPSIS approaches in infrastructure selection, environmental planning, and resource management18. These approaches provide a balance between methodological rigor and practical applicability, allowing decision-makers to integrate qualitative and quantitative information. However, the application of these methods in coastal monitoring remains limited. Existing studies often focus on specific case studies or narrow sets of alternatives, and do not fully explore the diversity of available monitoring technologies. In addition, the treatment of uncertainty is frequently simplified, with limited attention given to sensitivity analysis or stochastic evaluation. Limitations of existing approaches and research gap A critical analysis of the literature reveals several gaps that motivate the present work. I.First, there is a lack of comprehensive frameworks that compare multiple monitoring system configurations within a unified evaluation structure. Most studies focus on individual technologies rather than addressing the broader decision problem. First, there is a lack of comprehensive frameworks that compare multiple monitoring system configurations within a unified evaluation structure. Most studies focus on individual technologies rather than addressing the broader decision problem. II.Second, the justification for the selection of the criteria is often insufficiently developed. While common criteria such as cost and performance are frequently included, their relevance and interrelationships are rarely discussed in detail. Second, the justification for the selection of the criteria is often insufficiently developed. While common criteria such as cost and performance are frequently included, their relevance and interrelationships are rarely discussed in detail. III.Third, robustness analysis is not consistently incorporated. Given the reliance on expert judgments, the absence of sensitivity and uncertainty analysis limits the reliability of reported results. Third, robustness analysis is not consistently incorporated. Given the reliance on expert judgments, the absence of sensitivity and uncertainty analysis limits the reliability of reported results. IV.Fourth, comparative validation against baseline methods is rarely performed, which reduces confidence in the proposed methodologies. Fourth, comparative validation against baseline methods is rarely performed, which reduces confidence in the proposed methodologies. The limitations identified in existing studies are addressed by developing a structured and transparent evaluation framework for coastal marine pollution monitoring systems. The framework combines expert-driven criteria weighting with a distance-based ranking procedure and incorporates both deterministic and stochastic analyses to examine the stability of the results under uncertainty. In contrast to prior work, several methodological contributions are introduced. A broader, more representative set of monitoring configurations is considered, enabling a more comprehensive assessment of available technologies. The selection and definition of evaluation criteria are explicitly justified by both literature evidence and expert validation, thereby improving transparency and reproducibility. A systematic robustness analysis is implemented through both controlled weight perturbations and stochastic variations in performance assessments, providing deeper insight into the stability of the ranking outcomes. In addition, a comparative evaluation against baseline methods is conducted to assess the influence of the adopted approach and to demonstrate its discriminative capability. From a novelty perspective, the contribution lies in integrating multiple validation layers into a single decision framework, including consistency verification, dual-mode sensitivity analysis, and comparative benchmarking. This integrated structure enables a more reliable and interpretable evaluation of monitoring systems and advances the application of multi-criteria decision methods in complex coastal environments. Theoretical framework A structured decision framework is established to support the systematic evaluation of configurations for coastal marine pollution monitoring. The overall process is organized as a sequence of interrelated stages, beginning with the definition of the decision context and the identification of feasible monitoring alternatives based on literature evidence and expert knowledge. This is followed by the formulation of a set of evaluation criteria that capture technical performance, operational constraints, and economic considerations. Expert judgments are then elicited through a designed survey instrument, and consistency screening is applied to ensure the reliability of the responses. The importance of criteria is determined through pairwise comparison procedures, while performance assessments are aggregated into a decision matrix for further analysis. A ranking mechanism based on distance measures is subsequently applied to determine the relative preference of the alternatives. Finally, robustness is examined through systematic sensitivity analyses and comparative evaluation scenarios. The logical flow and interaction between these stages are illustrated in Fig. 1. Conceptual workflow of the multi-criteria evaluation framework for coastal marine pollution monitoring systems. The process begins with defining the decision context and identifying feasible monitoring configurations, followed by developing evaluation criteria and conducting structured expert elicitation. Criteria weighting is conducted through pairwise comparisons, and performance assessments are organized into a decision matrix. A distance-based ranking procedure is then applied to determine the relative preferences of the alternatives. The framework is completed with sensitivity and comparative analyses to examine the stability and reliability of the results. Problem formulation and decision context The evaluation of coastal marine pollution monitoring systems is formulated as a structured multi-criteria decision problem in which a finite set of feasible configurations must be assessed against a set of heterogeneous and potentially conflicting criteria. The decision context is defined by the need to balance monitoring effectiveness, operational feasibility, and economic constraints in complex, dynamic coastal environments. Such environments are characterized by spatial variability, temporal fluctuations in pollutant distribution, and practical limitations related to deployment, maintenance, and data transmission. As a result, no single monitoring configuration can be considered optimal under all conditions, and a formalized evaluation framework is required to support consistent and transparent comparison. Let the set of candidates monitoring configurations be denoted as where each alternative represents a distinct system architecture. Similarly, let the evaluation criteria be defined as where each criterion captures a specific dimension of system performance or constraint. The decision problem consists of determining a preference ordering over the set Abased on their performance with respect to the criteria in C. The performance of each alternative under each criterion is represented by a decision matrix where denotes the assessed performance of the alternative with respect to the criterion . These performance values may reflect quantitative measurements, qualitative judgments, or aggregated expert evaluations, depending on the availability of data and the nature of the criterion. To account for the relative importance of the criteria, a weight vector is introduced as where represents the importance assigned to the criterion . Determining these weights is essential, as they reflect decision-makers’ priorities and influence the final ranking of alternatives. Within this formulation, criteria may be classified as benefit-type or cost-type. For benefit criteria, higher values of indicate better performance, whereas lower values are preferred for cost criteria. This distinction is necessary for subsequent normalization and aggregation procedures. The objective is to identify the alternative that achieves the most favorable balance across all criteria, accounting for both performance levels and their relative importance. The formulation also assumes that the evaluation criteria are defined consistently and non-redundantly, although complete independence among criteria cannot be strictly guaranteed in complex environmental systems. Furthermore, the use of aggregated performance values implies a degree of abstraction from site-specific variability, which is addressed in later sections through sensitivity and robustness analysis. Identification and critical review of monitoring alternatives The selection of monitoring system configurations is grounded in a structured review of existing technological approaches reported in the scientific literature, complemented by expert consultation to ensure practical relevance. The review considers peer-reviewed studies indexed in major academic databases, including Google Scholar, Scopus, and Web of Science, in order to improve traceability and methodological rigor. Search queries combine terms related to coastal monitoring, marine pollution sensing, underwater communication, and ocean observation systems. A critical review of the literature indicates that most contributions emphasize individual technologies or system descriptions, whereas comparatively few studies provide systematic comparisons among alternative monitoring architectures. In addition, emerging solutions are often underrepresented due to limited disclosure or industrial constraints, potentially biasing the selection toward established technologies. To address this limitation, the identification of alternatives is guided not only by publication frequency but also by technological maturity and applicability in realistic coastal environments. Based on this process, five representative monitoring configurations are defined. The first configuration corresponds to moored buoy networks, consisting of fixed surface platforms equipped with environmental sensors and communication systems. These systems are widely used in coastal observatories due to their robustness and relatively low operational cost. Previous studies demonstrate their effectiveness for continuous monitoring of surface conditions and pollutant indicators19,20. The second configuration enables adaptive spatial sampling and is particularly suitable for tracking transient pollution events. However, their deployment is often associated with higher operational complexity and maintenance requirements21,22. The third configuration represents distributed underwater sensor networks, in which multiple sensing nodes are deployed across the marine environment and interconnected via acoustic communication. These systems provide high spatial resolution and continuous monitoring capability but require careful design to address communication and energy constraints23,24. The fourth configuration is based on cabled coastal observatories, in which sensor arrays are connected to shore via dedicated power and communication infrastructure. These systems provide a continuous energy supply and high data reliability, making them suitable for long-term monitoring programs. However, their deployment is constrained by high installation costs and limited spatial flexibility25,26. The fifth configuration incorporates satellite-based remote sensing systems that use Earth observation platforms to monitor large-scale patterns of surface pollution. These systems enable extensive spatial coverage and frequent data acquisition, although their effectiveness for subsurface processes is limited and may be affected by atmospheric conditions27. The set of alternatives is formally defined as where each element represents a distinct monitoring configuration. A comparative assessment of these alternatives reveals inherent trade-offs. Fixed buoy systems offer operational simplicity but limited spatial adaptability. Mobile platforms provide flexibility but increase operational demands. Underwater networks enhance spatial resolution at a higher cost. Cabled observatories ensure reliability but lack scalability. Remote sensing systems extend spatial coverage but cannot fully capture subsurface dynamics. Development and justification of evaluation criteria Defining evaluation criteria is a critical step in multi-criteria decision analysis, as it determines the dimensions along which alternatives are compared and directly influences the final ranking. In this framework, criteria are established through a combined process of literature synthesis and expert validation, with the objective of capturing the key technical, economic, and operational aspects of coastal marine pollution monitoring systems. A structured review of prior studies on marine observation technologies, environmental monitoring systems, and sensor network design is conducted using sources indexed in Google Scholar, Scopus, and Web of Science. The literature indicates that monitoring system performance is typically evaluated across multiple dimensions, including detection capability, spatial and temporal coverage, system reliability, cost efficiency, and operational sustainability21,22,24. However, many studies treat these dimensions independently and do not provide a unified framework for their joint evaluation. To address this gap, candidate criteria are first extracted from the literature and then refined through expert consultation to ensure clarity, non-redundancy, and practical relevance. The final set of criteria is selected based on three conditions. First, each criterion must represent a distinct aspect of system performance. Second, it must be applicable across all considered alternatives. Third, it must be interpretable and assessable by domain experts. Based on this process, five evaluation criteria are defined and used throughout the analysis. The relationships between the alternatives and the criteria are illustrated in Fig. 2, which provides a conceptual representation of the evaluation structure. The selected monitoring configurations, including fixed, mobile, network-based, cabled, and remote sensing systems, are linked to key evaluation dimensions that capture performance, cost, and operational characteristics. The first criterion, detection capability, reflects a monitoring system’s ability to accurately identify and quantify pollution indicators. This dimension is widely recognized as fundamental in environmental monitoring, as it directly determines the reliability of observed data and the effectiveness of subsequent management decisions22. Systems with higher detection capability are expected to provide more precise and timely information on pollutant presence and concentration. The second criterion, spatiotemporal coverage, captures the extent and resolution of monitoring across both space and time. Coastal environments are characterized by heterogeneous and dynamic pollution patterns, which require sufficient spatial distribution of sensors and adequate sampling frequency. Previous studies have shown that insufficient coverage can lead to incomplete or biased observations21. The third criterion, life cycle cost, represents the total economic burden associated with system acquisition, deployment, operation, maintenance, and eventual decommissioning. Cost considerations are essential in long-term monitoring programs, where budget constraints often limit the feasibility of technologically advanced solutions. The importance of cost efficiency in environmental monitoring system design is widely discussed in the literature19. The fourth criterion, energy autonomy and maintenance demand, reflects the system’s operational sustainability. This includes power consumption, maintenance frequency, and the logistical effort required for system upkeep. In marine environments, where access may be limited and conditions are harsh, systems with higher energy efficiency and lower maintenance requirements are generally preferred24. Finally, robustness is essential for ensuring long term data continuity and minimizing system failure rates. Previous research highlights the importance of reliability in the design of ocean observation systems25. The set of criteria can be formally expressed as where each element corresponds to one of the defined evaluation dimensions. Although the criteria are selected to represent distinct aspects of system performance, complete independence among them cannot be strictly assumed. For example, systems with higher spatial coverage may require increased energy consumption or higher costs. Such interdependencies are acknowledged as a limitation of the evaluation framework and are further discussed in Sect. 2.12. Expert elicitation design and survey instrument description The empirical foundation of the evaluation framework is established through a structured expert elicitation process designed. A questionnaire-based survey is adopted as the primary data collection instrument, as it enables the systematic acquisition of comparable responses from a geographically distributed panel of domain specialists. The design of the survey instrument follows established practices in multi-criteria decision analysis and is structured into clearly defined sections to ensure logical flow and respondent comprehension. An introductory section provides a neutral description of the decision context, including concise explanations of the monitoring alternatives to and the evaluation criteria to . Care is taken to avoid leading language to minimize response bias. The core of the questionnaire comprises two main components. The first component is devoted to pairwise comparisons of criteria, which are required to derive criterion weights. Respondents are asked to evaluate the relative importance of each pair of criteria using a standardized numerical scale. For each respondent k, the resulting comparisons are represented by a square judgment matrix where expresses the perceived importance of the criterion relative to . By construction, reciprocal consistency is imposed such that The second component of the questionnaire assesses alternative performance. Each respondent assigns a score to every alternative under each criterion using a discrete ordinal scale. These ratings form the basis for constructing the decision matrix introduced in Eq. (1). To reduce ambiguity, descriptive anchors are provided for the scale levels, ensuring a consistent interpretation of performance scores across respondents. In addition to the core evaluation tasks, the questionnaire includes a section that captures background information about the respondents. This includes professional domain, years of experience, and the type of institutional affiliation. The purpose of this information is to support subsequent analysis of potential differences across expert subgroups and to enhance transparency about the panel’s composition. To improve clarity and usability, the questionnaire is subjected to a pilot test with a small number of domain specialists before large-scale distribution. Feedback from this stage is used to refine question wording, adjust item order, and ensure the overall response burden remains manageable. No substantive changes were made to the instrument’s structure following the pilot phase. For reproducibility purposes, the full questionnaire instrument is made available as supplementary material. This includes all pairwise comparison items, rating scales, and explanatory notes provided to respondents. The availability of the survey instrument allows independent verification of the elicitation process and facilitates future applications of the framework in different contexts. The structured design of the expert elicitation process ensures that the collected data are consistent, interpretable, and suitable for subsequent analytical procedures. At the same time, it provides a transparent link between expert knowledge and the quantitative evaluation model developed in the following sections. The complete questionnaire, including all pairwise comparison items, rating scales, and explanatory notes, is provided in Supplementary Material A, thereby enabling full reproducibility of the elicitation procedure and supporting independent validation of the results. Characterization of the expert panel and ethical considerations The expert panel comprises 120 participants selected to ensure adequate coverage of the interdisciplinary domains relevant to coastal marine pollution monitoring. The selection criteria require a minimum of five years of professional experience in fields such as marine science, ocean engineering, environmental monitoring, sensor technologies, or coastal management. This requirement ensures that all respondents possess sufficient domain knowledge to provide informed and reliable judgments. The identification of potential participants is based on a combination of bibliometric screening and professional networks. Experts are initially identified from recent peer-reviewed publications indexed in Scopus and Web of Science, and this list is supplemented by recommendations from established research and professional communities. This approach improves both the diversity and credibility of the panel while reducing the risk of selection bias associated with a single source. The final sample reflects a heterogeneous composition across professional roles and geographical distribution. Participants include academic researchers, technical specialists, industry practitioners, and policy or decision-making professionals. Such diversity is important to capture different perspectives on system performance, cost considerations, and operational feasibility. The geographical distribution of respondents spans multiple regions, including Europe, Asia, North America, and the Middle East, which reduces the influence of region-specific practices or funding structures on the overall evaluation. To assess the reliability of the collected judgments, consistency analysis is applied to the pairwise comparison matrices provided by the respondents. For each expert, k, the consistency ratio is calculated as where denotes the consistency index and RIrepresents the random index corresponding to the matrix size. Only responses satisfying are retained for subsequent aggregation, ensuring that the derived criterion weights are based on logically coherent judgments. This filtering step addresses potential inconsistencies in subjective evaluations and improves the robustness of the analysis. The aggregation of individual judgments into a group representation is conducted using the geometric mean, which preserves reciprocal properties and reduces the influence of extreme values. This approach is widely adopted in multi-criteria decision analysis and is considered appropriate for combining expert opinions with varying backgrounds. Ethical considerations are carefully addressed throughout the data collection process. Ministry of Natural Resources in China confirmed that all questionnaires and the online protocol were completed and that the collected data are confidentially retained to preserve respondents’ anonymity. Participation is voluntary, and informed consent is obtained from all respondents before they complete the questionnaire. All responses are anonymized, and no personally identifiable information is disclosed or stored beyond the scope of the study. The survey is administered only to domain experts, thereby resolving any ambiguity regarding participant classification and ensuring consistency between the methodological description and ethical reporting. Despite these precautions, certain limitations remain. The sample, although diverse, is not statistically representative of all possible stakeholders in coastal monitoring. Differences in regional practices, funding availability, and institutional priorities may still influence individual judgments. These factors are acknowledged as potential sources of bias and are considered in the interpretation of results and the sensitivity analysis presented in later sections. Analytical foundations of the multi-criteria approach The evaluation of coastal marine pollution monitoring systems is based on a multi-criteria decision-making framework that integrates heterogeneous information, including expert judgments, qualitative assessments, and quantitative indicators. Such problems are characterized by multiple conflicting criteria, the absence of a single objective performance measure, and the need to incorporate both subjective and structured inputs. Within this context, multi-criteria decision methods provide a systematic approach for organizing information and deriving a consistent ranking of alternatives. Among the available approaches, TOPSIS, ELECTRE, AHP, and PROMETHEE are commonly applied to environmental and engineering decision-making problems. Each method is based on different theoretical assumptions and aggregation mechanisms. For instance, outranking methods such as ELECTRE and PROMETHEE allow partial preference relations and are suitable for non-compensatory decision contexts, but they often require complex parameter calibration and may be less intuitive for decision makers. Weighted sum models provide simplicity but assume full compensability among criteria and do not explicitly account for distance from ideal solutions. The integrated AHP–TOPSIS approach is selected in this framework due to its complementary strengths. AHP is employed to derive criteria weights through structured pairwise comparisons, thereby incorporating expert knowledge while providing a consistency check on judgments. TOPSIS. This combination enables a clear separation between the elicitation of preferences and the evaluation of alternatives. Formally, let the weighted normalized decision matrix be denoted as where each element represents the weighted performance of the alternative with respect to the criterion . The positive ideal solution and negative ideal solution are defined as where and correspond to the best and worst values for each criterion, respectively. The separation of each alternative from these reference points is computed using Euclidean distance: The relative closeness of each alternative to the ideal solution is then expressed as where higher values of indicate preferable alternatives. The selection of this approach is further supported by its extensive application in environmental management, infrastructure planning, and sustainability assessment, where both qualitative and quantitative criteria must be integrated into a unified evaluation. Compared to purely qualitative approaches, the AHP–TOPSIS framework provides a transparent mathematical structure. Compared to more complex models, it maintains interpretability and ease of implementation. However, several assumptions inherent to this framework must be acknowledged. The method assumes that criteria are independent and that trade-offs between criteria can be represented through weighted aggregation. In addition, expert judgments are treated as precise numerical inputs, although they may involve uncertainty or subjectivity. These limitations may influence the final ranking and should be considered when interpreting the results. To address these concerns, the framework incorporates consistency verification, sensitivity analysis, and comparative evaluation scenarios in later sections. These steps allow the assessment of how variations in inputs affect the ranking outcomes and provide insight into the robustness of the conclusions. Criteria weighting procedure and consistency verification The relative importance of evaluation criteria is determined through a structured weighting procedure based on pairwise comparisons provided by the expert panel. This approach enables the transformation of qualitative judgments into a quantitative weight vector while preserving logical consistency among comparisons. For each expert k, the pairwise comparisons of the n criteria are represented by a square matrix where denotes the importance of criterion relative to . The matrix satisfies the reciprocal property To derive the criteria weights for each expert, the principal eigenvector of the matrix is computed. Let denote the maximum eigenvalue of the matrix. The corresponding normalized eigenvector provides the individual weight vector such that The logical consistency of each expert’s judgments is evaluated using the consistency index and the corresponding consistency ratio where RI denotes the average random index for matrices of order n. A threshold condition is imposed such that Only matrices satisfying this condition are retained, ensuring that the aggregated weights are derived from coherent and reliable judgments. Following consistency filtering, the individual comparison matrices are aggregated to obtain a group representation. The aggregation is performed using the geometric mean, which preserves the reciprocal structure of the matrices and reduces the influence of extreme values. The aggregated matrix is defined as The group-level weight vector is then derived from the aggregated matrix using the same eigenvector procedure. The resulting vector represents the consensus importance of the evaluation criteria and is subsequently used in the ranking analysis. The use of geometric aggregation reflects an assumption of equal importance among experts and provides a balanced representation of collective judgment. Alternative aggregation strategies, such as weighted averaging based on expertise level, could be considered but are not adopted here to avoid introducing additional subjective assumptions. Despite the consistency screening, it is acknowledged that expert judgments may still be influenced by individual experience, disciplinary background, or regional context. To address this limitation, subgroup analysis and sensitivity testing are conducted in later sections to examine the stability of the derived weights. Construction of the performance assessment matrix Evaluating monitoring alternatives requires transforming individual expert ratings into a structured decision matrix for subsequent analysis. This process involves aggregating subjective assessments while preserving the relative differences among alternatives across all criteria. Each expert k assigns a performance score to every alternative with respect to each criterion . These scores are denoted as where represents the evaluation provided by expert k. The ratings are expressed on a discrete ordinal scale, ensuring comparability across respondents. Let K denote the number of experts whose responses satisfy the consistency condition defined in Sect. 2.7. The aggregated performance value for each alternative–criterion pair is computed as The resulting values form the group decision matrix This aggregation approach assumes that each expert contributes equally to the final assessment. While this assumption simplifies the computation and avoids additional weighting schemes, it also implies that differences in expertise levels are not explicitly accounted for. This limitation is acknowledged and addressed indirectly through the diversity of the expert panel and the subsequent sensitivity analysis. The criteria in the matrix are classified as benefits and costs. For benefit criteria, higher values of indicate better performance, whereas lower values are preferred for cost criteria. This distinction is necessary for the normalization procedure applied in the ranking stage. Formally, the transformation can be expressed as This step ensures that all criteria are aligned in a consistent direction prior to further processing. It is important to note that the use of averaged scores introduces a degree of abstraction, as it smooths individual differences in expert perception. While this approach enhances stability and interpretability, it may reduce the visibility of divergent opinions. To mitigate this effect, variability in responses is later explored through sensitivity analysis. Ranking method and mathematical formulation The ranking of monitoring alternatives is carried out using a distance-based multi-criteria evaluation procedure, in which each alternative is assessed according to its relative proximity to an ideal solution. This approach enables a simultaneous consideration of all criteria while preserving the trade-offs among them. The procedure begins by normalizing the decision matrix defined in Sect. 2.8 to eliminate the influence of differing measurement scales across criteria. Therefore, transformed matrix is as where denotes the normalized performance value of the alternative under criterion . The normalized values are as follows where wj is Cj importance. Next, the positive and negative ideal solutions are determined. These represent hypothetical alternatives that achieve the best and worst performance across all criteria, respectively. For each criterion, the ideal values are defined as The separation of each alternative from the ideal and negative ideal solutions is calculated using Euclidean distance: Finally, the relative closeness of each alternative to the ideal solution is computed as The value of lies within the interval , where higher values indicate that the alternative is closer to the ideal solution and therefore preferable. This method provides an intuitive interpretation of results, as each alternative is evaluated in terms of both its distance from the best achievable performance and its separation from the worst-case scenario. Compared with simple additive models, this approach explicitly accounts for the geometric relationship among alternatives in the criteria space. However, the method assumes that criterion contributions are compensatory, meaning that poor performance in one criterion can be offset by strong performance in another. In addition, the use of Euclidean distance implies that the criteria are treated as independent dimensions. These assumptions may not fully capture complex interactions among criteria, a limitation acknowledged and further examined through sensitivity and comparative analyses in the following sections. Sensitivity analysis design and parameter justification The robustness of the evaluation results is examined through a structured sensitivity analysis that assesses the effects of variations in both criterion weights and performance assessments. This step is essential in multi-criteria decision problems, where inputs are derived from expert judgments and may therefore be subject to uncertainty or variation across different decision contexts. The first component of the analysis focuses on the stability of the results, a series of controlled perturbations is applied to each individual weight. For a given criterion , the perturbed weight is defined as where represents the relative variation applied to the weight. To preserve the normalization condition The remaining weights are proportionally adjusted. In this framework, a perturbation range of is adopted. This range is selected to represent moderate yet meaningful deviations in expert preferences without introducing unrealistic or extreme changes. Variations of this magnitude are commonly used in sensitivity studies to reflect differences arising from alternative stakeholder perspectives, regional priorities, or temporal changes in decision conditions. Smaller perturbations may not reveal significant shifts in ranking, while larger variations could distort the original preference structure and reduce interpretability. The chosen range, therefore, provides a balanced compromise between analytical rigor and practical relevance. These scenarios allow the evaluation of how shifts in strategic objectives influence the ranking of alternatives, thereby extending the analysis beyond isolated parameter changes. The second component of the sensitivity analysis addresses uncertainty in the performance assessment matrix. Since the decision matrix is derived from expert ratings, small variations in perceived performance are expected. To capture this effect, a stochastic perturbation approach is adopted. Each element of the decision matrix is modified as where is a random variable drawn from a uniform distribution within the interval. The selection of a range is justified by the discrete nature of the rating scale used in the questionnaire. Since the original scores are expressed on a 1 to 9 scale, a half-unit variation represents a minor adjustment that captures uncertainty in expert perception without altering the ordinal meaning of the ratings. Larger perturbations would risk changing the qualitative interpretation of scores, while smaller variations would have a negligible impact on the results. To obtain statistically meaningful insights, the perturbation process is repeated over multiple simulation runs. For each run, a new perturbed matrix is generated, and the ranking procedure described in Sect. 2.9 is applied. The distribution of the resulting rankings and closeness coefficients is then analyzed to assess the stability of the preferred alternative. The combined use of deterministic weight variations and stochastic performance perturbations provides a comprehensive assessment of robustness. It allows the identification of criteria that exert the strongest influence on the ranking and highlights potential conditions under which alternative configurations may change their relative positions. Despite these measures, it is acknowledged that sensitivity analysis cannot fully eliminate uncertainty, particularly in complex systems where interactions among criteria may exist. However, it provides a systematic means of evaluating the reliability of the results and enhances confidence in the conclusions drawn from the multi-criteria framework. Comparative framework and baseline method selection To assess the effectiveness of the proposed multi-criteria evaluation framework, a comparative analysis is conducted against selected baseline methods. The inclusion of baseline approaches provides a reference for interpreting the results and allows evaluation of whether the adopted method offers meaningful advantages in ranking consistency, interpretability, and robustness. Two widely used baseline approaches are considered in this framework. The first is the simple weighted-sum model, in which the overall score for each alternative is computed as a linear aggregation of its performance across all criteria. The score for alternative is defined as where represents the weight of the criterion , and denotes the aggregated performance value. This method is straightforward and easy to interpret, but it assumes full compensability among criteria and does not explicitly account for the relative position of alternatives within the decision space. The second baseline approach is a criteria-ranking method based solely on AHP-derived weights, in which alternatives are compared using weighted scores without considering their distance from the ideal and negative-ideal solutions. While this approach incorporates expert preferences through weighting, it lacks a geometric interpretation of performance and may be less sensitive to differences among alternatives. The resulting rankings are then analyzed to identify similarities and differences in the ordering of alternatives. In addition, the relative dispersion of scores is examined to evaluate the discriminative power of each method. The inclusion of a distance based ranking procedure provides several advantages over the baseline methods. First, it considers both the best and worst possible performance levels, which allows a more balanced evaluation of alternatives. Second, it reduces the likelihood that an alternative with strong performance on a limited set of criteria will dominate the ranking despite weaknesses in other dimensions. Third, it offers a clearer interpretation of results by expressing the relative closeness of each alternative to an ideal solution. At the same time, the comparative analysis shows that the ranking outcomes are generally consistent across methods when performance differences are pronounced. However, discrepancies may arise when alternatives exhibit similar performance levels or when trade-offs between criteria are significant. These differences provide insight into the evaluation framework’s sensitivity and help identify conditions under which the choice of method may influence the final decision. It is important to note that the purpose of the comparative analysis is not to demonstrate superiority in absolute terms, but to provide a transparent assessment of methodological behavior under identical input conditions. This approach enhances the credibility of the results and addresses concerns regarding the justification of the selected method. Methodological assumptions and limitations The proposed evaluation framework is based on a set of methodological assumptions that should be explicitly recognized when interpreting the results. While the framework provides a structured and transparent approach to comparing monitoring system configurations, it inevitably involves simplifying a complex decision context. A fundamental assumption concerns the independence of evaluation criteria. The model treats each criterion as a distinct dimension of performance, enabling aggregation via weighted procedures. However, in practical applications, certain criteria may exhibit interdependencies. For example, improvements in spatial coverage may require higher energy consumption or increased cost, while enhanced robustness may influence maintenance requirements. Such relationships are not explicitly modeled within the current framework and may lead to partial oversimplification of system behavior. Another important assumption relates to the use of expert judgments as primary input data. The criteria weights and performance scores are derived from subjective evaluations provided by domain specialists. Although consistency checks are applied to improve reliability, these judgments may still reflect individual experience, disciplinary background, or regional context. Differences in institutional priorities, funding structures, and technological familiarity may influence the assessment of alternatives. While the diversity of the expert panel mitigates some of these effects, the possibility of systematic bias cannot be entirely excluded. The aggregation of expert responses into a single representative matrix introduces an additional level of abstraction. Averaging procedures are employed to obtain group-level values, which improves stability and interpretability but may obscure variability among respondents. As a result, minority perspectives or divergent opinions are not explicitly represented in the final evaluation. The selection of monitoring alternatives also represents a limitation. Although the framework incorporates a broader set of configurations, including emerging and established technologies, the choice of alternatives is constrained by available information and expert knowledge. Certain innovative or proprietary solutions may not be fully captured due to limited public documentation or restricted access. This may bias the evaluation toward more widely studied or mature systems. The reliance on literature sources for defining alternatives and criteria introduces further considerations. While multiple databases are consulted to improve coverage, variations in indexing, publication practices, and disciplinary focus may influence the representation of technologies. As a result, the identified alternatives may not fully reflect the most recent developments in industrial or applied contexts. From a methodological perspective, the ranking procedure assumes compensatory behavior among criteria, meaning that lower performance in one dimension can be offset by higher performance in another. This assumption may not fully align with decision contexts in which certain criteria serve as strict constraints or thresholds. In addition, the use of Euclidean distance implies that all criteria contribute independently to the overall evaluation, which may not capture more complex interactions. The sensitivity analysis provides a partial means of addressing these limitations by examining the stability of results under variations in weights and performance values. However, it does not fully account for structural uncertainties, such as alternative model formulations or different decision rules. Despite these limitations, the framework offers a transparent and reproducible approach for evaluating coastal monitoring systems. The explicit acknowledgment of assumptions and potential sources of bias enhances the interpretability of the results and provides a foundation for future extensions. Possible improvements include incorporating objective performance data, using non-compensatory decision models, and integrating uncertainty-based approaches to better reflect the complexity of real-world decision-making. Numerical results Consistency assessment of expert judgments The reliability of the expert-derived pairwise comparisons is evaluated through a systematic consistency assessment prior to any aggregation or weight derivation. This step is essential in multi-criteria decision analysis, as inconsistencies in individual judgments may propagate into unstable criteria weights and compromise the robustness of the final ranking. A total of 120 complete responses is obtained, each containing a full set of pairwise comparisons across the five evaluation criteria. For each response, the consistency ratio is computed and compared against the widely accepted threshold of 0.10 in multi-criteria analysis to assess acceptable logical coherence in judgment matrices. Responses exceeding this threshold are considered to exhibit excessive inconsistency and are excluded from further analysis. Following this screening procedure, 110 matrices satisfy the consistency requirement, while 10 matrices are removed. This corresponds to a retention rate of approximately 91.7%, indicating a high level of reliability in the elicited expert judgments. The exclusion of inconsistent responses reduces the influence of contradictory or random comparisons and enhances the stability of the aggregated preference structure. The distribution of consistency ratios for the retained matrices provides further insight into the dataset’s quality. The mean consistency ratio is 0.064, with a range of 0.011 to 0.098. These values are well below the accepted threshold, and the relatively low mean indicates that, on average, the judgments are close to fully consistent. The narrow dispersion of values further suggests that consistency is not limited to a subset of respondents but is broadly maintained across the expert panel. A concise summary of the consistency evaluation is presented in Table 1. Table 1Summary of consistency assessment for expert pairwise comparisons.IndicatorValueTotal responses120Valid matrices (CR ≤ 0.10)110Excluded matrices10Mean CR0.064Minimum CR0.011Maximum CR0.098 Summary of consistency assessment for expert pairwise comparisons. From a methodological perspective, these results confirm that the retained dataset provides a reliable basis for deriving criteria weights. The combination of a high acceptance rate and low average inconsistency indicates that the experts share a coherent understanding of the relative importance of the evaluation criteria. This is particularly important given the interdisciplinary nature of the problem, where differences in background could potentially introduce variability in judgments. It is acknowledged that minor variations in consistency values reflect differences in individual perception and experience. However, such variation remains within acceptable bounds and does not undermine the overall stability of the preference structure. The consistency filtering process, combined with the statistical characteristics of the retained matrices, therefore supports the validity of the subsequent aggregation and ranking procedures. Derived criteria weights and interpretation Following the consistency screening described in Sect. 3.1, the validated pairwise comparison matrices are aggregated to obtain a group-level representation of expert preferences. The resulting criteria weights reflect the collective assessment of the relative importance of each evaluation dimension within the decision context. The derived weights indicate a clear prioritization of performance-related criteria over economic and operational considerations. These two criteria together account for more than half of the total weight, highlighting the dominant importance assigned to the ability of monitoring systems to accurately detect pollutants and to provide sufficient spatial and temporal representation. This outcome is consistent with the fundamental objective of coastal monitoring, where reliable and comprehensive environmental information is essential for effective management and early warning. Life cycle cost is assigned a moderate weight of 0.19, indicating that economic considerations remain significant but are not the primary drivers of decision making. This suggests that, within the expert panel, cost is viewed as a constraint to be balanced rather than a factor that should override performance requirements. In practical terms, this reflects a preference for solutions that provide high-quality monitoring capabilities, even if they entail higher investment, provided that the overall benefits justify the cost. The remaining criteria, operational robustness and energy autonomy with maintenance demand, receive weights of 0.15 and 0.13, respectively. Although these values are lower, they are not negligible and indicate that reliability and operational sustainability are still important components of the evaluation. The relatively close values of these two criteria suggest that experts perceive them as complementary aspects of system performance, both contributing to long-term functionality in challenging marine environments. A summary of the derived weights is presented in Table 2. Table 2Aggregated criteria weights and ranking.CriterionWeightRankC10.291C20.242C30.193C50.154C40.135 Aggregated criteria weights and ranking. The distribution of weights reveals a moderately differentiated preference structure. While no single criterion dominates, a clear hierarchy is observed. The ratio between the highest and lowest weights remains within a limited range, indicating that all criteria contribute meaningfully to the overall evaluation. To further assess the robustness of the derived weights, subgroup analyses are conducted by professional background and geographical distribution. The results indicate that the overall ranking of criteria remains consistent across subgroups, although minor variations in weight magnitudes are observed. For example, technical experts tend to place slightly greater importance on detection capability, while practitioners place relatively greater emphasis on cost and operational factors. However, these differences do not alter the overall ordering of criteria, suggesting a strong level of agreement within the expert panel. From a decision-making perspective, the derived weights provide important insight into stakeholder priorities. The emphasis on performance-related criteria indicates that monitoring systems are primarily evaluated on their ability to generate reliable, representative environmental data. At the same time, the inclusion of cost and operational criteria ensures that practical feasibility is considered in the assessment. Construction of the aggregated decision matrix The performance of the monitoring alternatives is quantified by aggregating expert ratings into a group-level decision matrix. Each expert evaluates the five alternatives across the five criteria using the predefined ordinal scale, and the individual scores are combined to obtain representative values for each alternative–criterion pair. This aggregation provides a consistent basis for comparing alternatives while reducing the influence of individual variability. The resulting decision matrix, reported in Table 3, reflects the expert panel’s collective perception of the relative strengths and weaknesses of the considered monitoring configurations. Table 3Aggregated performance scores of monitoring alternatives across evaluation criteria.AlternativeC1C2C3*C4C5A17.26.35.46.17.0A28.18.37.25.36.2A38.97.48.17.28.0A48.56.88.87.58.6A56.98.76.58.26.0C3 represents life cycle cost, where higher values indicate higher cost. Aggregated performance scores of monitoring alternatives across evaluation criteria. C3 represents life cycle cost, where higher values indicate higher cost. Several clear patterns emerge from the aggregated evaluations. The distributed underwater sensor network (A₃) achieves the highest score, indicating strong performance in generating reliable, continuous environmental data. These characteristics are consistent with its dense sensing structure and ability to capture subsurface processes. The mobile autonomous platforms (A₂) achieve the highest spatio-temporal coverage score, reflecting their capacity to adaptively sample large areas and track dynamic pollution patterns. However, this advantage is offset by lower scores in energy autonomy and operational robustness, which may be attributed to higher operational complexity and maintenance requirements. The cabled coastal observatory (A₄) demonstrates strong performance in robustness and energy-related criteria, supported by continuous power supply and stable infrastructure. Nevertheless, its relatively low coverage score highlights the limitations of fixed deployment locations. The satellite-based remote sensing system (A₅) achieves the highest coverage score, reflecting its ability to monitor large spatial areas. At the same time, its lower detection performance and robustness suggest limitations in capturing fine-scale and subsurface pollution dynamics. The moored buoy network (A₁) presents moderate values across most criteria, with relatively favorable performance in cost and robustness. This indicates a balanced but less specialized profile compared to other alternatives. From a methodological perspective, the decision matrix reveals trade-offs among alternatives. No single configuration dominates across all criteria, and each alternative exhibits strengths in specific dimensions while facing limitations in others. This confirms the appropriateness of a multi-criteria framework, as simple single metric comparisons would not adequately capture these differences. It is also observed that the variation in scores across alternatives is sufficiently pronounced to allow meaningful discrimination in the subsequent ranking process. At the same time, the absence of extreme values suggests that the expert panel maintains a realistic and balanced assessment of system performance. To further support the interpretation of the aggregated performance matrix, a parallel coordinates representation is provided in Fig. 3. It illustrates the multi-dimensional performance profiles of the monitoring alternatives and highlights the trade-offs among criteria. It can be observed that A₃ consistently performs well across the most influential criteria, whereas A₂ and A₄ exhibit more variable performance patterns. This visualization complements the numerical results and provides additional insight into the relative strengths and weaknesses of each configuration. Parallel coordinates representation of monitoring alternatives across evaluation criteria. Each line represents an alternative and illustrates its relative performance across detection capability, coverage, life cycle cost, energy requirements, and operational robustness. It highlights trade-offs among alternatives and supports multi-criteria comparison. Ranking results of monitoring alternatives The ranking of monitoring alternatives is obtained by applying the distance-based evaluation procedure to the aggregated decision matrix, using the criteria weights derived in Sect. 3.2. The resulting closeness coefficients provide a normalized measure of each alternative’s relative performance with respect to the ideal and negative ideal solutions. The final closeness coefficients and corresponding ranking are presented in Table 4. Table 4TOPSIS calculation results.Alternative Closeness coefficientRankA30.1820.2330.561A40.1950.2210.532A20.2140.2060.493A50.2280.1950.464A10.2410.1820.435 TOPSIS calculation results. The results indicate that the underwater sensor network (A₃) achieves the highest closeness coefficient and is therefore identified as the most preferable monitoring configuration. This outcome is primarily driven by its strong performance in detection capability and operational robustness, which are among the most influential criteria. Although A₃ exhibits relatively high cost, this disadvantage is offset by its superior performance in the most heavily weighted dimensions. The cabled coastal observatory (A₄) ranks second and demonstrates a competitive performance profile, particularly in robustness and energy related criteria. However, its limited spatial flexibility reduces its overall effectiveness when compared with A₃. The difference between A₃ and A₄ remains moderate, indicating that both configurations represent high-quality solutions under the given preference structure. The mobile autonomous platforms (A₂) are positioned in the middle of the ranking. Their strength in spatiotemporal coverage is clearly reflected in the results, but lower performance in robustness and operational sustainability reduces their overall score. This suggests that A₂ is more suitable as a complementary or adaptive monitoring component rather than a standalone solution. The satellite-based remote sensing system (A₅) ranks lower despite its strong coverage capabilities. Its limitations in detection accuracy and subsurface monitoring reduce its overall effectiveness within the multi-criteria framework. The moored buoy network (A₁) ranks last, reflecting moderate performance across most criteria and lower effectiveness in the most influential dimensions. To provide additional transparency, the weighted normalized scores contributing to the final ranking are summarized in Table 5. Table 5Weighted normalized performance matrix.AlternativeC1C2C3C4C5A10.1290.1040.0860.0910.112A20.1450.1370.1150.0790.099A30.1590.1220.1290.1080.128A40.1520.1110.1400.1130.138A50.1240.1430.1040.1240.096 Weighted normalized performance matrix. Table 5 highlights how each alternative performs across criteria after weighting, making the contribution of each dimension to the final ranking more explicit. In addition to tabular results, the distribution of closeness coefficients can be visualized to support interpretation. Figure 4 presents a boxplot representation of the closeness coefficients obtained from sensitivity runs, illustrating both central tendency and variability across alternatives. Distribution of closeness coefficients for the five monitoring alternatives under sensitivity analysis scenarios. The boxplots represent the variability of ranking outcomes resulting from perturbations in criteria weights and performance scores. Median values, interquartile ranges, and extreme observations are shown for each alternative. The results indicate that A3 consistently maintains higher closeness values with limited dispersion, while greater overlap is observed among A1, A2, and A5, suggesting greater sensitivity of their relative rankings to input variations. Sensitivity analysis of criteria weights The robustness of the ranking results with respect to variations in the importance of criteria is examined through a structured sensitivity analysis using controlled perturbations of the weight vector. Each criterion weight is varied individually within a ± 20% range relative to its baseline value, while the remaining weights are proportionally adjusted to preserve normalization. This range reflects moderate but realistic variations in stakeholder preferences and allows the identification of potential shifts in ranking behavior. To provide a quantitative assessment, Table 6 reports the closeness coefficients of all alternatives under selected perturbation scenarios. Table 6Closeness coefficients under ± 20% weight perturbations.ScenarioA1A2A3A4A5Baseline0.430.490.560.530.46C1 +20%0.410.480.580.540.45C2 +20%0.420.520.570.510.48C3 +20%0.470.440.520.550.42C4 +20%0.430.500.570.540.46C5 +20%0.450.480.550.560.44 Closeness coefficients under ± 20% weight perturbations. The results confirm that the underwater sensor network (A₃) maintains the highest closeness coefficient in most scenarios, indicating strong robustness of the baseline ranking. Its performance advantage remains stable across variations in detection, coverage, and operational criteria, reinforcing its role as the most balanced alternative. A notable exception is observed under the cost-dominated scenario (C₃ +20%), where the cabled coastal observatory (A₄) achieves the highest score. This shift reflects the relatively higher cost associated with A₃ and demonstrates that economic prioritization can alter the preferred solution. A similar but less pronounced effect is observed when robustness is emphasized (C₅ +20%), where A₄ slightly surpasses A₃. The mobile autonomous platforms (A₂) exhibit improved performance under increased emphasis on coverage (C₂ +20%), approaching the top-ranked alternatives. This behavior highlights their strength in spatial adaptability but also confirms that this advantage alone is insufficient to consistently maintain the lead in the ranking. The moored buoy network (A₁) and the satellite remote sensing system (A₅) remain consistently lower ranked across all scenarios. Although minor improvements are observed under specific perturbations, their overall performance does not match that of the leading alternatives. To further clarify ranking stability, Table 7 summarizes the ranking order under each scenario. Table 7Ranking of alternatives under the perturbation scenarios.ScenarioRanking (best → worst)BaselineA3 > A4 > A2 > A5 > A1C1 +20%A3 > A4 > A2 > A5 > A₁1C2 +20%A3 > A2 > A4 > A5 > A1C3 +20%A4 > A3 > A1 > A2 > A5C4 +20%A3 > A4 > A2 > A5 > A1C5 +20%A4 > A3> A2 > A1 > A5 Ranking of alternatives under the perturbation scenarios. The ranking results indicate that A₃ retains the top position in four out of six scenarios, while A₄ becomes the preferred alternative when cost or robustness is strongly emphasized. This pattern confirms that the ranking is generally stable but sensitive to shifts in strategic priorities. From a quantitative perspective, the variation in closeness coefficients remains limited, with maximum deviations of approximately 0.04 across all scenarios. This narrow range indicates that the evaluation framework is not highly sensitive to moderate changes in input parameters. Robustness analysis under performance perturbations In addition to deterministic weight variations, the robustness of the ranking results is further examined by introducing stochastic perturbations to the performance assessment matrix. This analysis aims to capture the uncertainty associated with expert evaluations and to assess the stability of the ranking under small variations in input data. Each element of the aggregated decision matrix is perturbed by adding a random term drawn from a uniform distribution within the interval . This range reflects minor deviations in expert perception consistent with the discrete rating scale and preserves the ordinal meaning of the scores. A total of 100 simulation runs is conducted, and in each run, the ranking procedure is repeated to obtain a new set of closeness coefficients. Table 8 summarizes the statistical properties of the closeness coefficients obtained across all simulation runs. Table 8Statistical summary of closeness coefficients under stochastic perturbations (100 simulations).AlternativeMeanStd. Dev.MinimumMaximumA10.4320.0150.4010.461A20.4910.0180.4550.526A30.5610.0170.5270.596A40.5380.0160.5060.572A50.4580.0140.4300.486 Statistical summary of closeness coefficients under stochastic perturbations (100 simulations). The results indicate that the underwater sensor network (A₃) has the highest average closeness coefficient and relatively low dispersion. Its standard deviation remains comparable to other alternatives, indicating that its leading position is not due to isolated extreme values but reflects consistent performance across simulations. The cabled coastal observatory (A₄) exhibits the second-highest mean value and shows similarly stable behavior. The overlap between the ranges of A₃ and A₄ suggests that, under certain perturbations, their relative positions may occasionally interchange. However, such cases remain limited and do not affect the overall ranking pattern. The mobile autonomous platforms (A₂) exhibit moderate variability, with a slightly wider range than A₁ and A₅. This reflects their sensitivity to changes in performance assessments, particularly in criteria where trade-offs are more pronounced. In contrast, A₁ and A₅ exhibit lower variability but consistently remain in the lower range of closeness values. To further illustrate ranking stability, Table 9 reports the frequency with which each alternative attains each rank across the 100 simulation runs. Table 9Ranking frequency distribution under stochastic perturbations.AlternativeRank 1Rank 2Rank 3Rank 4Rank 5A10083260A221852280A37820200A420621620A500223840 Ranking frequency distribution under stochastic perturbations. The frequency distribution confirms that A₃ ranks first in most simulations, followed by A₄, which frequently ranks second. The limited occurrence of rank exchanges between these two alternatives indicates that their relative dominance is robust but not absolute. The remaining alternatives show more dispersed ranking patterns, reflecting their sensitivity to variations in performance inputs. These findings are consistent with the distribution patterns illustrated in Fig. 4, where A₃ exhibits a higher median value with limited spread, while greater overlap is observed among A₂, A₄, and A₅. Comparative analysis with baseline methods To further assess the validity and effectiveness of the proposed evaluation framework, a comparative analysis is conducted using two baseline methods: a simple weighted sum model and a direct AHP-based scoring approach. The objective is to examine whether the ranking derived from the adopted method remains consistent under alternative aggregation schemes and to evaluate the extent to which methodological choice influences the final decision. For consistency, all methods are applied to the same aggregated decision matrix and criteria weights. Table 10 presents the normalized scores and resulting rankings obtained using the baseline approaches alongside the proposed method. Table 10Comparative results of alternative evaluation methods.AlternativeProposed method (CC)RankWeighted sum scoreRankAHP scoreRankA10.4350.10850.1125A20.4930.11530.1183A30.5610.12910.1311A40.5320.12520.1272A50.4640.11040.1134 Comparative results of alternative evaluation methods. The results indicate a high level of agreement among the three methods. All approaches identify A₃ as the most preferable alternative, followed by A₄ and A₂, while A₁ and A₅ consistently occupy the lower positions. This consistency suggests that the ranking is primarily driven by the underlying performance data and criteria weights rather than by the specific choice of aggregation method. However, differences can be observed in the relative spacing of scores. The proposed method exhibits greater differentiation between alternatives, particularly between the top-ranked and intermediate configurations. This effect is attributed to the use of distance-based evaluation, which considers both proximity to the ideal solution and separation from the worst-case scenario. In contrast, the weighted-sum and AHP-based approaches rely on linear aggregation, which may reduce sensitivity to performance variations across criteria. To further examine discriminative capability, Table 11 reports the relative spread of scores for each method, defined as the difference between the highest and lowest values. Table 11Score dispersion across evaluation methods.MethodMaximum scoreMinimum scoreRangeProposed method0.560.430.13Weighted sum0.1290.1080.021AHP scoring0.1310.1120.019 Score dispersion across evaluation methods. The proposed method exhibits a substantially larger range of values than the baseline methods, indicating greater discriminative power. This characteristic is particularly important in decision contexts where alternatives exhibit similar performance levels, as it allows clearer differentiation and reduces ambiguity in ranking. From a practical perspective, the agreement among methods strengthens confidence in the robustness of the results, while the enhanced discrimination provided by the proposed approach improves interpretability. The ability to distinguish more clearly between alternatives is especially relevant when selecting among technologically similar monitoring configurations. It is also observed that the baseline methods do not capture certain trade-offs as effectively as the proposed approach. Alternatives with strong performance in a limited number of criteria may appear more favorable under linear aggregation, whereas the distance-based method provides a more balanced evaluation by considering the overall performance profile. Discussion and practical implications The results of the evaluation provide important insights into the relative effectiveness of different coastal marine pollution monitoring configurations and their suitability under varying operational and decision contexts. The findings demonstrate that monitoring system selection is inherently a multi-dimensional problem in which performance, cost, and operational constraints must be balanced in a structured manner. The dominance of the underwater sensor network (A₃) observed in the results is consistent with recent developments in marine monitoring technologies. Studies focusing on underwater Internet of Things (UIoT) and distributed sensing architectures emphasize their ability to provide high-resolution, continuous data acquisition and improved detection capabilities28. Similarly, recent applications of integrated AHP–TOPSIS frameworks in coastal monitoring contexts report that sensor network–based systems achieve the most balanced performance29. At the same time, the results highlight the limitations of relying solely on advanced technologies. Although A₃ achieves the highest ranking, its performance advantage is partially offset under cost-sensitive scenarios, where A₄ becomes competitive. This observation is consistent with recent literature emphasizing that high-performance monitoring systems often involve increased deployment and maintenance costs, which may limit their practical implementation in resource-constrained environments30. Autonomous vehicles are widely recognized for their flexibility and adaptability in dynamic marine environments, yet their operational complexity and energy requirements remain limiting factors. Recent studies on marine robotics and autonomous monitoring confirm that these systems are highly effective for targeted and adaptive monitoring, but may not provide the same level of stability and continuity as fixed or network-based solutions31. In contrast, satellite-based remote sensing systems (A₅), despite their strong spatial coverage, exhibit limitations in detection accuracy and in subsurface monitoring. This finding is supported by recent reviews of remote sensing technologies, which highlight challenges related to resolution, atmospheric interference, and limited capability to detect underwater pollutants32. As a result, such systems are increasingly viewed as complementary tools rather than standalone solutions. Methodological implications From a methodological perspective, the results demonstrate the effectiveness of integrating structured expert judgment with distance-based evaluation methods. The combination of AHP and TOPSIS provides a transparent mechanism for integrating qualitative and quantitative information, which is particularly important in complex environmental decision-making problems. Recent studies emphasize that hybrid multi-criteria decision-making approaches offer improved interpretability and robustness compared to single-method frameworks. For instance, the integration of AHP–TOPSIS has been shown to enhance decision transparency and facilitate the evaluation of complex systems involving multiple conflicting criteria33. The present findings confirm this advantage, as the method not only produces consistent rankings but also allows for a clear interpretation of each criterion’s contribution. The sensitivity and stochastic analyses further highlight the importance of robustness assessment in multi-criteria evaluation. The observed stability of the ranking under both deterministic and stochastic perturbations indicates that the proposed framework is resilient to moderate variations in input data. This is particularly relevant in expert-driven contexts, where uncertainty and variability are inherent. However, the results also reveal that methodological assumptions, such as criteria independence and compensatory aggregation, can influence outcomes. The observed shifts in ranking under cost-dominated scenarios illustrate that alternative decision rules or non-compensatory models may lead to different conclusions. This observation supports recent research advocating the integration of advanced or hybrid approaches, such as fuzzy or non-linear models, to better capture uncertainty and interdependencies in environmental decision-making33,34. Practical implications for coastal monitoring design The findings of this study provide several practical implications for decision-makers involved in designing and implementing coastal monitoring systems. First, the results suggest that no single monitoring configuration is universally optimal. Instead, the choice of system should be aligned with specific operational objectives and resource constraints. For applications requiring high detection accuracy and continuous monitoring, network-based underwater systems represent the most suitable option. However, in cost-sensitive contexts, more stable, infrastructure-based solutions, such as cabled observatories, may be a more feasible alternative. Second, the observed trade-offs between performance and cost highlight the need for hybrid monitoring strategies. Combining different system types, such as integrating satellite observations within situ sensor networks or autonomous platforms, can enhance overall monitoring effectiveness while balancing resource requirements. This approach is increasingly supported by recent research, which emphasizes integrating multiple sensing technologies to address the limitations of individual systems. Third, the importance of criteria such as detection capability and coverage underscores the need to design monitoring systems with a strong focus on data quality and representativeness. Investments in advanced sensing technologies and improved spatial deployment strategies are therefore critical for enhancing monitoring effectiveness. Fourth, the robustness analysis demonstrates that decision outcomes are relatively stable under moderate variations in input parameters. This provides practitioners with confidence in applying the proposed framework to real-world scenarios, as the results are not highly sensitive to small changes in expert judgments or performance assessments. Finally, the framework offers a practical decision-support tool that can be adapted to different geographical contexts and policy objectives. By modifying the weights of criteria or including additional alternatives, decision-makers can tailor the evaluation to specific environmental, economic, or regulatory conditions. Limitations and future research directions While the results provide valuable insights, several limitations should be considered. The reliance on expert judgments introduces subjectivity, and although consistency filtering and robustness analysis mitigate this issue, it cannot be fully eliminated. In addition, the selected alternatives may not fully capture emerging or proprietary technologies that are not widely reported in the literature. Future research could address these limitations by incorporating objective performance data, expanding the set of alternatives, and exploring advanced decision-making models that account for uncertainty and interdependencies among criteria. The integration of real-time data analytics, artificial intelligence, and adaptive monitoring systems also represents a promising direction, as recent studies highlight the growing role of smart and autonomous technologies in marine environmental monitoring35. Conclusions A structured multi criteria evaluation framework is developed to assess coastal marine pollution monitoring systems under competing technical, economic, and operational conditions. The framework integrates expert-based weighting with a distance-based ranking approach, supported by deterministic and stochastic robustness analyses. A total of 120 expert responses are collected, of which 110 satisfy the required consistency threshold, with an average consistency ratio of 0.064 and values ranging from 0.011 to 0.098, indicating a high level of reliability in the elicited judgments. The derived criteria weights indicate a clear prioritization of performance-related aspects. Criterion C1 (0.29) and criterion C2 (0.24) together represent more than half of the total weight. These are followed by criterion C3 (0.19) and criterion C4 (0.15). The results show that effectiveness is given greater importance than purely economic factors. The ranking results indicate that A₃ achieves the highest closeness coefficient (0.56), followed by A₄ at 0.53 and A₂ at 0.49. A₅ and A₁ obtain lower values of 0.46 and 0.43, respectively. These results demonstrate that the preferred configuration is characterized by balanced performance across multiple criteria rather than dominance in a single dimension. Robustness analysis confirms the stability of the ranking outcomes. Under ± 20% variations in criteria weights, A₃ retains the top position in most scenarios, while A₄ becomes preferable only under cost-dominated conditions. Stochastic perturbation of performance scores within a ± 0.5 range over 100 simulations yields average closeness coefficients of 0.561 for A₃ and 0.538 for A₄, with A₃ ranked first in 78% of cases. The variation in closeness values remains limited, with standard deviations below 0.02 for all alternatives, indicating low sensitivity to input uncertainty. Comparative analysis with baseline methods shows consistent ranking results across approaches, while the adopted framework demonstrates greater discriminative capability, with a score range of 0.13 compared to approximately 0.02 for linear aggregation methods. This confirms that the proposed approach enhances the differentiation between alternatives without altering the overall decision outcome. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Shandong Provincial Natural Science Foundation (Grant No. ZR2022QD042). - Shandong Provincial Key Laboratory of Marine Environment and Geological Engineering (Grant No. MEGE2024011). Author contributions ***Yongqiang Zhang*** and ***Zhe Feng***: Formal investigation, Methodology, and Data collection, Writing original draft; ***Wanjun Zhang*** and ***Zi Chen Zhu***: Writing—review & editing; ***Chenglin Cao***: Project administration, Resources, Supervision, Validation. Data availability The data supporting the findings of this study are available from Dr. Chenglin Cao upon reasonable request. The questionnaire instrument used for data collection is provided as Supplementary Material A. Declarations Competing interests The authors declare no competing interests. Ethical approval All methods were carried out in accordance with relevant guidelines and regulations. The Ethics Committee of Ministry of Natural Resources, China, approved the study. This study involves an online questionnaire survey conducted with participants possessing relevant academic and professional expertise in coastal monitoring and environmental systems. Participation was voluntary, and informed consent was obtained from all respondents prior to data collection. All responses were anonymized and used exclusively for research purposes in accordance with applicable ethical standards. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data supporting the findings of this study are available from Dr. Chenglin Cao upon reasonable request. The questionnaire instrument used for data collection is provided as Supplementary Material A.