A Bayesian Hierarchical Gating Model for Culturally Grounded Spatial Decision-Making: Evidence from Kampung Adat Kuta, Indonesia

Sustainable spatial planning increasingly requires decision-support systems that integrate ecological processes, socio-cultural values, and governance institutions within a coherent analytical framework [1, 2]. This requirement highlights the inherent politics of sustainability, where transformations are viewed not as manageable technical shifts but as processes drawing on 'unruly politics' and the integration of diverse forms of knowledge [3]. In traditional settlements, spatial organization is shaped not only by environmental suitability but also by ritual obligations, customary authority, sacred boundaries, and collective memory [4, 5]. These conditions demand planning models that can account simultaneously for measurable spatial variables and culturally embedded rules.

Recent advances in geographic information system (GIS), GeoAI, and Bayesian spatial modelling have improved the capacity of planning systems to combine heterogeneous datasets and represent uncertainty [6, 7]. These developments in geoAI facilitate the processing of increasingly diverse and complex data sources, yet they often rely on 'black box' machine learning methods that lack transparency regarding the specific purposes and inherent qualitative properties of the geographic space they analyze [8].

However, many computational tools remain technocentric because they prioritize optimisation and standardized data while giving limited attention to subjective experience, customary knowledge, and social legitimacy [9, 10]. This is particularly problematic in traditional settlements where culturally prohibited spaces cannot be treated as ordinary suitability variables.

Planning theory has long addressed the limitations of technical rationality. Rational-comprehensive planning offers analytical rigor, communicative planning emphasizes deliberation and consensus, and phenomenological perspectives foreground lived experience, meaning, and place attachment [11-13]. Yet these paradigms remain weakly connected at the operational level: participation does not automatically become decision authority, and cultural meaning is often acknowledged conceptually without being translated into reproducible decision rules [14, 15].

The unresolved gap is therefore not whether cultural values matter, but how they can be transformed into an explicit, traceable, and testable spatial decision mechanism. Participatory GIS and participatory Bayesian networks show that local knowledge can be spatialized and linked to probabilistic reasoning, but a clear architecture for representing non-compensatory customary restrictions remains absent [16, 17]. This study addresses that gap by developing and validating a culturally grounded Bayesian decision-support model.

Kampung Adat Kuta, a customary village (kampung adat) in Ciamis Regency, West Java, provides a critical case because its spatial governance combines customary authority, sacred zoning, social deliberation, and environmental constraints. In this context, cultural permissibility precedes social acceptance and physical evaluation. A spatial unit that violates pamali or overlaps with a sacred zone is rejected regardless of its technical suitability, making the case suitable for testing a hierarchical gating model.

This study develops the Transcendental Integrative System (TIS) as an operational extension of Transcendental Integrative Planning (TIP). TIP provides the conceptual ontology, TIS translates it into an operational architecture, and the Bayesian Hierarchical Gating Model functions as the computational engine. Naïve Bayes is used only as a benchmark because its conditional-independence assumption cannot represent the sequential dependency among value, communication, and physical variables [18, 19].

The central research question is: how can cultural legitimacy, social consensus, and physical feasibility be operationalised into a Bayesian Hierarchical Gating Model for spatial decision-making in Kampung Adat Kuta? The study aims to identify the value, communication, and physical dimensions of spatial decision-making; operationalise them into computable variables; and evaluate the model by comparing predicted decisions with observed spatial decisions across 115 spatial units.

The novelty of the study lies in three related contributions. First, cultural legitimacy is formalized as a deterministic Value Gate rather than a compensatory weight. Second, social consensus is represented as a conditional posterior probability. Third, physical feasibility is evaluated conditionally only after cultural and social legitimacy are satisfied. This sequential structure makes the model fully auditable and aligns computational planning with culturally grounded decision authority [17, 20].

The research design adopts a phenomenological-computational approach to bridge qualitative spatial knowledge and structured decision systems. This orientation is methodologically supported by mixed-method research design, participatory GIS, and participatory Bayesian modelling because each enables local knowledge, spatial evidence, and probabilistic reasoning to be integrated in a traceable workflow [16, 17, 25].

Data collection is stratified across three domains: value, communication, and physical feasibility. Participatory mapping is central to this design because it records geographic and non-geographic knowledge that conventional spatial datasets often miss, while customary governance studies show that ritual norms and local authority can operate as binding spatial rules rather than informal preferences [16, 21, 26].

The Bayesian-spatial modelling approach is justified by its capacity to integrate heterogeneous evidence, represent uncertainty, and preserve dependencies among spatial, social, and cultural variables [6, 7, 17]. The hierarchical structure of the model ensures that spatial outputs are conditioned by cultural and social filters before environmental feasibility is assessed, reflecting the decision logic of customary spatial governance [21, 26, 27].

3.1 Research design and epistemological foundation

A convergent mixed-method design is appropriate because the study combines qualitative community knowledge, participatory mapping, and quantitative spatial analysis [16, 17]. This design captures the cultural depth of customary legitimacy while providing a reproducible structure for modelling value, communication, and physical feasibility.

This integration aligns with the framework of TIP, which conceptualizes planning as a relational process shaped by value, communication, and physical conditions. To operationalise this conceptual framework, the study develops the TIS, enabling the translation of these dimensions into a probabilistic decision-support model.

The model is designed as a decision-support system rather than an autonomous replacement for customary authority. Its purpose is to make culturally grounded spatial logic explicit and auditable while ensuring that final legitimacy remains accountable to community validation [16, 17].

3.2 Case selection and contextual justification

The empirical case of Kampung Adat Kuta, West Java, Indonesia, is selected as a critical case representing a traditional village system with strong customary governance and transcendental spatial logic. Spatial decision-making in this context is regulated through ritual validation, sacred zoning, and collective deliberation, making it highly suitable for examining the interaction between value, communication, and physical dimensions.

The selection follows a theoretical sampling logic, where the case represents an extreme or information-rich example of culturally embedded planning. This allows the study to test the applicability of the TIP–TIS framework in a context where the “missing mechanism” between values and operational decisions is most visible.

3.3 Data collection and informant profile

To strengthen reproducibility, the qualitative component was organized through a structured informant profile. Data were collected from 25 key informants selected purposively according to their authority, knowledge, and direct involvement in Kampung Adat Kuta spatial practices. The informant composition was designed to capture customary authority, collective memory, social legitimacy, household experience, and formal administrative perspectives.

Data collection was organized into three domains corresponding to the TIS layers: value data were obtained from ritual practices, sacred norms, and customary restrictions; communication data from deliberation, participatory mapping, and stakeholder interaction; and physical data from spatial, environmental, and geophysical conditions. This structure ensured that symbolic and transcendental knowledge could be transformed into structured variables without detaching them from their local context [16].

3.4 Operationalization of the Value Gate and consciousness framework

To operationalize the Value Gate as a deterministic constraint, sacred zones and restricted spatial units were identified through direct consultation with customary authorities. The Kuncen and Sesepuh Adat delineated Leuweung Larangan (sacred forest), sacred water-source areas, and pamali (prohibited) zones on high-resolution base maps during participatory mapping sessions. This narrative and ritual knowledge was subsequently converted into spatial data and linked to the 115 candidate spatial units. The operational dimensions and proxy indicators for this process are presented in Table 2, while the specific qualitative coding scheme for these variables is detailed in Table 3.

Table 2. Operationalization of Transcendental Integrative System (TIS) variables

Table 3. Qualitative coding scheme for Transcendental Integrative System (TIS) variables

The Value Gate is formulated not as a probabilistic score, but as a binary exclusion rule; consequently, units coded as V = 0 are excluded from all subsequent computation regardless of their communication or physical scores. Spatial units assigned a code of V = 0 represent restricted, sacred, or prohibited areas that are not ritually approved; these are automatically classified as rejected and do not proceed to subsequent layers of the model. Conversely, spatial units coded as V = 1 are considered permissible, neutral, or ritually approved, allowing them to proceed to the Communication and Physical layers for further evaluation.

To maintain ethical alignment with indigenous sovereignty, a Community Validation Protocol was implemented. If the model predicts a location as accepted but customary authority deems it restricted during the validation phase, the customary decision is treated as the final cultural constraint. In such instances, the discrepancy is documented in a log, and the Value Gate coding is recalibrated to preserve the role of TIS as a culturally grounded decision-support tool rather than an autonomous decision-maker.

Figure 3 illustrates the flow of data collection, transformation, and integration across value, communication, and physical layers, leading to probabilistic decision outputs.

Figure 3. Spatial decision-making process in the Transcendental Integrative System

3.5 Qualitative coding scheme and inter-coder reliability

The transformation of narrative interview data, focus group discussion (FGD) notes, and participatory mapping records into computational variables followed a deductive-inductive coding procedure, as structured within the multi-stage participatory GIS workflow presented in Table 4. Deductive codes were derived directly from the TIS framework, while inductive codes were employed to capture field-specific expressions of customary restriction, boundary disagreements, and localized environmental concerns.

Table 4. Participatory geographic information system (PGIS) workflow and outputs

To ensure the reliability of this qualitative-to-quantitative transformation, two independent coders processed the interview transcripts and mapping records. Inter-coder reliability was measured using Cohen’s Kappa, producing a score of 0.82, which indicates a high level of consistency and algorithmic readiness before the variables were integrated into the Bayesian model. Any identified disagreements between coders were resolved through deliberative discussion and meticulous cross-checking with original field notes and participatory mapping outputs to ensure the model accurately reflected indigenous spatial logic.

3.6 Participatory geographic information system workflow and probability estimation

The spatial data for the 115 units were generated through a multi-stage participatory GIS workflow. The process was designed to connect qualitative indigenous knowledge with quantitative spatial modelling and to make each transformation step traceable.

The Bayesian Hierarchical Gating model estimates the final spatial decision through the probability chain P(D) = P(V) × P(C|V) × P(P|C,V), where P(V) represents cultural permissibility as a binary gate, P(C|V) represents social legitimacy for units that pass the Value Gate, and P(P|C,V) represents physical feasibility conditioned on cultural and social approval. To ensure full reproducibility, Table 5 presents three concrete illustrative cases showing how each probability is estimated for individual spatial units.

Table 5. Probability estimation procedure

The operational logic of the Bayesian Hierarchical Gating Model is demonstrated through three representative cases of probability estimation presented in Table 5. These examples illustrate the sequential probability chain P(D) = P(V)×P(C∣V)×P(P∣C,V). For Unit SU-01, the unit successfully passed the Value Gate (V = 1). Its Likert consensus score of 5/5 was converted to a communication posterior of 0.95 using frequency-based calibration derived from FGD records and historical deliberation data.

Furthermore, SU-01 was assigned a physical class of high feasibility based on a slope gradient of less than 8%, a transitional zone land cover, low hydrological risk, and flood hazard category class I. This physical profile was converted to a physical posterior of 0.78 through a conditional lookup table derived from GIS suitability classification, resulting in a combined probability of 0.74 and a final Accepted status. In contrast, Unit SU-02 illustrates the deterministic nature of the cultural filter within TIS. Because the unit was identified as being in a sacred or restricted zone (V = 0), the computational process was terminated immediately. Consequently, consensus and physical scores were not computed, yielding a combined probability of 0.00 and an automatic Rejected status at the Value Gate.

Finally, Unit SU-03 represents a culturally permissible unit (V = 1) that failed to reach the acceptance threshold due to social and physical constraints. A consensus score of 3/5 produced a lower communication posterior of 0.65, reflecting recorded boundary disagreements among stakeholders. Combined with a low physical feasibility posterior of 0.42, the final probability P(D) was calculated at 0.27, falling below the classification threshold and resulting in a Rejected decision. This hierarchical sequence ensures that final spatial recommendations are grounded in both indigenous legitimacy and biophysical feasibility.

Figure 4 presents the probabilistic interaction between value, communication, and physical layers, demonstrating how decision outcomes are derived through conditional dependencies.

Figure 4. Bayesian integration structure of Transcendental Integrative System (TIS)

3.7 Model and community validation protocols

Model validation was conducted through retrospective classification validation across all 115 spatial units, comparing model-predicted decisions with observed spatial decision outcomes. The primary metric is Accuracy, defined as the proportion of correct matches between model predictions and field observations. Precision, Recall, and F1-score were also calculated. Model performance was compared with two baselines: a physical-suitability-only model and a standard Naïve Bayes benchmark using the same V, C, and P variables but treating them as conditionally independent predictors [20, 21].

A structured community validation protocol was implemented through a dual-track verification process, as detailed in Table 6. First, a technical-social check was performed using participatory map-based verification involving customary leaders, village officials, community figures, and representative residents. This stage allowed for the cross-checking of spatial boundaries, social consensus scores, and physical suitability assessments. Second, a transcendental legitimacy check was conducted through the Nepus ritual forum with the Kuncen and Sesepuh Adat to confirm whether any model outputs violated sacred zones or pamali (customary prohibitions).

Table 6. Community validation protocol

To maintain ethical alignment with indigenous sovereignty, the protocol included a specific procedure for discrepancy handling. If a model prediction conflicted with customary authority during validation, the customary decision was treated as the final cultural constraint. In such instances, the conflict was documented in a discrepancy log, and the Value Gate coding was recalibrated accordingly to ensure the TIS remains a culturally grounded decision-support tool rather than an autonomous decision-maker.

3.8 Ethical and epistemological considerations

The study maintains ethical and epistemological integrity by ensuring that cultural knowledge is not reduced to purely technical variables. All data transformations were conducted through participatory validation, and the model is positioned as a decision-support system rather than a replacement for customary authority. This ensures that TIS remains aligned with the principles of TIP and preserves the relational and cultural foundations of spatial decision-making.

5.1 Cultural legitimacy as a non-compensatory constraint

The results as a whole demonstrate that the principal contribution of TIS is not simply higher classification accuracy, but the formalisation of culturally embedded decision authority into an auditable computational sequence in which each rejection is traceable to a specific cultural, social, or physical constraint. The Value Gate results demonstrate that cultural legitimacy operates as a non-compensatory planning constraint rather than as an additional preference in a weighted suitability model. The rejection of 30 spatial units through customary rules confirms that sacred boundaries, pamali, and ritual validation function as binding decision thresholds, consistent with indigenous planning and sacred ecology scholarship [4, 5, 21]. This finding strengthens the central novelty of TIS: cultural legitimacy is encoded as a gate that determines whether a spatial unit may enter further analysis.

The nine units that would have been considered physically suitable under conventional analysis but were rejected by the Value Gate show the limitation of technocratic suitability models. Unlike multi-criteria decision analysis (MCDA)-based approaches that allow trade-offs among variables, the TIS architecture prevents culturally restricted space from being compensated by high physical scores [17, 28, 29]. This non-compensatory structure is the main distinction between TIS and conventional spatial suitability modelling.

5.2 Probabilistic representation of social legitimacy

The Communication Posterior results indicate that social legitimacy is strongly shaped by prior cultural permissibility. The high proportion of units with P(C|V) >= 0.95 should therefore be interpreted as evidence that customary legitimacy structures the range of socially acceptable options, rather than as a simple statistical artifact [12, 14, 30]. This supports the argument that participation becomes meaningful only when embedded within recognized institutional and cultural authority, a concept central to reaching meaningful public participation in spatial planning [16]. This is further reinforced by recent research suggesting that integrating multi-stakeholder perspectives is essential for quantifying cultural indicators and identifying the trade-offs required for culturally grounded landscape management [31].

By translating consensus and disagreement into posterior probabilities, TIS preserves social uncertainty rather than erasing it. This is important for explainable and accountable planning systems because contested units can be traced to specific boundary disputes, stakeholder disagreement, or institutional ambiguity [20, 32]. The Communication layer therefore extends communicative planning by turning deliberative legitimacy into an auditable computational output.

5.3 Conditional physical feasibility and limits of technocratic suitability models

The variation in Physical Posterior values across eligible units demonstrates that environmental feasibility remains decisive but conditional. The lowest posterior values were observed in environmentally constrained target cases, particularly where slope, hydrological exposure, and flood risk coincided, confirming that physical data must be interpreted within culturally and socially validated decision sequences [6, 7, 17].

This finding sharpens the critique of physical-suitability-only models. Environmental optimisation is necessary for sustainable planning, but it is insufficient when spatial decisions are also governed by ritual authority, social legitimacy, and customary restrictions [1, 19]. The TIS model therefore integrates environmental feasibility without allowing technocratic optimisation to override cultural and social constraints.

5.4 Bayesian Hierarchical Gating as a planning architecture

The Bayesian Hierarchical Gating Model advances spatial decision-support methodology by replacing independent-variable classification with a sequential decision architecture. Standard Naïve Bayes is useful as a benchmark because of its computational simplicity, but its conditional-independence assumption cannot represent the relational dependency between customary legitimacy, consensus, and physical feasibility in traditional settlements [7, 18, 19].

The model performance comparison supports this architectural choice. The TIS model achieved higher accuracy than both the Naïve Bayes benchmark and the physical-only baseline, indicating that the deterministic Value Gate, the Communication Posterior, and the conditional Physical layer jointly improve classification performance [17, 20]. This performance gain is not merely statistical; it demonstrates that the model structure better reflects the actual hierarchy of decision authority in Kampung Adat Kuta.

The model also contributes to explainable spatial AI because each decision can be traced through the sequence V -> C -> P. Unlike black-box prediction, the TIS output can be audited by identifying whether a unit is rejected because of cultural restriction, social disagreement, or physical constraint [20, 32, 33].

Figure 9 synthesizes the TIS as a layered spatial computational architecture in which cultural, communicative, and environmental knowledge flows through the sequential decision chain V→C→P. The diagram shows the deterministic Value Gate, the Bayesian Communication Posterior, and the conditional Physical Feasibility layer, with the probability chain P(D) = P(V) × P(C|V) × P(P|C,V) annotated at each transition. All labels and legend elements are reproduced at 300 dpi for print clarity.

Figure 9. Transcendental Integrative System architecture

5.5 Theoretical contributions to planning theory

The study contributes to planning theory by operationalizing three planning paradigms within a single computational architecture. Rational-comprehensive planning is represented through the Physical layer, communicative planning through the Communication Posterior, and phenomenological planning through the Value Gate [11-13]. The contribution is not an additive combination of paradigms, but a hierarchical translation of different planning rationalities into different computational functions.

The core novelty is therefore that cultural values are not merely included in the model; they are assigned a distinct computational role. Transcendental value becomes a deterministic rule, social legitimacy becomes a posterior probability, and physical feasibility becomes a conditional environmental probability [17, 20, 22]. This shows that phenomenological and communicative insights can be operationalised without reducing them to arbitrary numerical weights.

5.6 Policy and planning implications

For planning policy and practice, TIS offers a decision-support framework for traditional settlement planning, heritage-sensitive zoning, village spatial planning, and customary land governance [34, 35]. Its main policy value lies in making the basis of each spatial decision explicit: planners can see whether a unit is rejected because of customary restriction, social disagreement, or physical constraint, while communities can use validation forums to review and recalibrate model outputs [16, 17, 20, 36].

The model should be understood as transferable after local recalibration rather than universally generalisable. Each settlement possesses distinct customary authorities, sacred geographies, social deliberation structures, heritage morphologies, and environmental risks; therefore, TIS should be understood as a replicable protocol and process innovation framework rather than as a universal formula [4, 27, 37, 38]. This position preserves the sovereignty of local knowledge while allowing formal planning institutions to engage with it systematically.

5.7 Limitations and future research

Several limitations bound the scope of the study. The model is calibrated and validated within a single case study, the 115 spatial units reflect the spatial and governance structure of Kampung Adat Kuta, and the probability thresholds remain context-specific. Application in other settlements requires recalibration of priors, thresholds, and Value Gate rules through local participatory validation [6, 19, 28].

Future research should test the TIS architecture across multiple traditional settlements with different governance structures, develop dynamic Bayesian updating for temporal change, and examine integration with formal Indonesian spatial planning instruments at district and provincial levels. Additional work should explore how explainable AI interfaces can help communities inspect and contest model outputs without surrendering decision authority to algorithmic systems [17, 22].