Spatial Planning and Settlement Resilience for Disaster Mitigation in Coastal Banda Aceh: A Perception-Based SEM-PLS Assessment

Banda Aceh City is a coastal urban area with high exposure to tsunami, coastal flooding, abrasion, and earthquake-related hazards. Settlement expansion following the 2004 tsunami has increased spatial pressure in northern coastal districts, creating a need for more risk-sensitive spatial planning and development control.

Previous studies on coastal disaster mitigation in Banda Aceh and comparable tsunami-prone settlements have generally treated hazard mapping, structural coastal protection, spatial planning regulation, and community preparedness as separate analytical domains. Hazard-based studies [1-3] identify tsunami inundation zones and physical exposure but stop short of linking exposure to settlement-level resilience outcomes. Regulatory and land-use studies [4-6] evaluate zoning compliance and urbanization pressure without incorporating participation or education as explanatory mechanisms, and comparable regulatory analysis has also been applied to spatial planning in tsunami-prone zones [7]. Structural and ecosystem-oriented studies [8-10] emphasize infrastructure and coastal buffers, and related work on integrated, community-based coastal management [11, 12] similarly emphasizes these measures, but such studies rarely test these dimensions against psychosocial and governance factors using a validated causal model. Spatial studies of coastal flooding also demonstrate the value of hazard mapping for evaluating settlement planning [13]. Existing models thus remain fragmented, limiting their use for prioritizing planning and budgetary resources such as the Regional Spatial Plan (RSP) and Detailed Spatial Plan (DSP). This study instead combines disaster potential (X1), settlement arrangement (X2), settlement resilience (X3), community participation (X4), climate change (X5), ecosystem-based coastal protection (X6), and public education (N1) in one structural framework so their relative relationships with disaster mitigation effectiveness (Y) can be examined together.

Beyond structural and regulatory measures, coastal ecosystems play a direct role in reducing disaster exposure. In particular, mangrove systems provide ecosystem-related benefits for coastal settlements by buffering wave energy, stabilizing shorelines against abrasion, and supporting fisheries-based livelihoods. These functions make mangrove restoration a nature-based solution that can complement engineered protection such as seawalls [8]. In this context, the role of coastal ecosystems extends beyond hazard reduction to supporting the ecological and livelihood conditions of coastal communities. In Banda Aceh, this ecological buffering capacity is unevenly distributed: protective structures are concentrated in areas such as Ulee Lheue, while mangrove cover in Alue Naga remains limited, leaving parts of the coastline more exposed to abrasion and tidal flooding [14, 15].

Effective disaster mitigation requires integrating spatial planning, resilient infrastructure, and hazard-sensitive development controls [16], an approach echoed in other urbanizing coastal contexts where modelling projects future growth and its environmental consequences [17]. Coastal settlement vulnerability is further shaped by population density, poverty, and lack of protective vegetation, as well as flooding compounded by alluvial soils and flat topography [15]. This study addresses this gap by integrating theoretical spatial planning principles with empirical evidence on Y. The analysis combines a theoretical model, Structural Equation Modeling–Partial Least Squares (SEM-PLS) validation using respondent data, and an empirical model developed from the validated relationships. The empirical model is based on data from the doctoral dissertation underlying this study.

This structural model draws on several complementary theoretical perspectives. The Pressure and Release (PAR) framework explains disaster risk as the interaction of root causes, dynamic pressures, unsafe conditions, and triggering hazards, which is reflected in X1. Community and settlement resilience theory, which emphasizes the capacity of social and physical systems to absorb disturbance, self-organize, and adapt, underlies X3, X4, and N1 as capacity-related constructs. X2 and X5 are grounded in spatial planning theory, particularly its treatment of land-use control and climate-related risks as determinants of exposure. X6 extends the framework by representing household perceptions of the effectiveness and benefits of nature-based coastal measures. These strands are combined to justify modeling all eight constructs together as a single testable structure (X1–X6, N1, Y).

Within this framework, N1 was originally hypothesized to operate as a mediator rather than a parallel predictor, on the reasoning that household knowledge and awareness form the mechanism through which X1, X2, X4, and X5 are translated into concrete mitigation behavior. The mediation test reported in Section 3.2 indicates that this translation mechanism is not supported in the present sample, since none of the indirect paths through N1 reach significance. The revised model therefore treats N1 as a direct explanatory construct operating in parallel with X3 rather than as a mediator. N1 and X3 represent two distinct capacity dimensions: X3 represents the physical and institutional capacity of a settlement to withstand and recover from hazards, whereas N1 represents the cognitive and behavioral capacity of households. X6 is included as a separate construct using the dedicated ecological indicators in the questionnaire. This distinction prevents X6 from being treated merely as a general perception of environmental quality and allows its relationship with Y to be tested explicitly in the revised SEM-PLS model.

Two scope limitations are stated at the outset. First, X6 is measured through household perceptions of the effectiveness and the ecological and socioeconomic benefits of nature-based coastal measures. It does not measure mangrove extent, mangrove condition, or ecosystem service delivery in the field, and this study therefore does not empirically test coastal ecosystem resilience as an ecological state. Second, the Geographic Information System (GIS) material presented below provides spatial context for describing the study area and interpreting the survey results; no spatial variable is entered as a numerical indicator in the SEM-PLS estimation and no quantitative spatial analysis is performed. The study is accordingly described throughout as a spatial-planning-oriented, perception-based mitigation assessment rather than as a spatially integrated model.

3.1 Measurement model validation

The theoretical model assumes that Y in coastal settlements depends on interacting physical, social, spatial, climatic, educational, and ecosystem-based dimensions. The revised model therefore examines X1–X6 as explanatory constructs, with N1 representing household knowledge and preparedness and Y representing disaster mitigation effectiveness. The SEM-PLS analysis tests the validity and reliability of these constructs and the significance of their hypothesized relationships. The GIS-based spatial analysis is used alongside the SEM-PLS results to provide geographic context for interpreting the implications of the statistically tested relationships.

The revised measurement model reports the outer loadings of all retained indicators together with convergent and discriminant validity. The three indicators of X6 have outer loadings of 0.842, 0.872, and 0.855, respectively, and an AVE of 0.733. These results support the convergent validity of the X6 construct. The HTMT assessment also shows that the X6 ratios with the other constructs are below 0.90, with the highest value observed for X6 and X5 (0.843), supporting discriminant validity.

The SEM-PLS analysis was performed using SmartPLS 4.0. The measurement model was evaluated through convergent validity, discriminant validity, and reliability. Tables 3–10 present the measurement and structural results supporting the transition from the theoretical model to the empirical model.

To verify that the reported results originate from the intended model and dataset, the final structural model was cross-checked directly against the SmartPLS 4.0 output. The imported dataset comprised exactly 400 observations, consistent with the sample size reported in Section 2.2. Figure 3 presents the complete structural model exported from SmartPLS, showing all eight constructs, their indicators, and the corresponding outer loadings and path coefficients. Because the SmartPLS diagram uses single-letter construct labels (A–F, N, Y) for display purposes, Table 3 provides the correspondence between these labels and the construct codes (X1–X6, N1, Y) used throughout this manuscript.

Figure 3 confirms that the exported SmartPLS model reflects the full set of eight constructs and their indicator structure used throughout the measurement and structural analysis reported below.

Table 3. Correspondence between SmartPLS diagram labels and manuscript construct codes

The path coefficients shown in Figure 3 match those reported in Table 8: A → Y = −0.020 (X1 → Y); B → Y = 0.620 (X3 → Y); C → Y = 0.061 (X4 → Y); D → Y = 0.043 (X5 → Y); E → Y = 0.103 (X2 → Y); F → Y = −0.111 (X6 → Y); N → Y = −0.003 (N1 → Y). This confirms that the diagram originates from the same model and dataset reported throughout this manuscript. Table 4 reports the outer loadings of all retained indicators for each construct.

Table 4. Outer loadings of the retained indicators

All outer loadings exceed the 0.60 threshold, supporting the indicator reliability of the retained measurement items. Table 5 reports the average variance extracted (AVE) for each construct to assess convergent validity.

Table 5 shows that all constructs meet the AVE criterion of 0.50. The AVE values are 0.727 for Y, 0.624 for X4, 0.636 for X2, 0.634 for X5, 0.563 for X1, 0.556 for X3, and 0.733 for X6. The X6 value is therefore above the recommended threshold and supports convergent validity for the X6 construct. Table 6 reports the reliability coefficients, including Cronbach’s alpha, rho_A, and rho_C, for each construct.

Table 5. Average variance extracted (AVE) for each construct

Table 6 shows that the constructs generally meet the reliability criteria, with rho_C (ρC) values above 0.70. However, Cronbach's alpha for X5 (0.465) is below the commonly accepted threshold. Because X5 is measured with only two indicators, Cronbach's alpha may underestimate its internal consistency. Although its rho_A (ρA) is 0.609, its rho_C (ρC) remains acceptable at 0.771. X6 shows Cronbach's alpha = 0.818, rho_A (ρA) = 0.818, and rho_C (ρC) = 0.891, indicating acceptable internal consistency and reliability.

Table 6. Reliability values for each construct

X5 requires a specific justification. It is measured by two indicators, D1 and D2, covering perceived sea-level rise, tidal flooding, abrasion, and general climate vulnerability as defined in Table 2. Their outer loadings are 0.784 and 0.831 (Table 4), the construct AVE is 0.634 (Table 5), and its reliability coefficients are Cronbach's alpha = 0.465, rho_A (ρA) = 0.609, and rho_C (ρC) = 0.771 (Table 6). Convergent validity is therefore satisfied, but internal consistency is weak. Cronbach's alpha is downward-biased for a two-item construct because it is a function of item count, yet rho_A (ρA) also remains below 0.70 here, so the weakness cannot be attributed to item count alone. X5 is retained for two reasons: removing it would eliminate the climate dimension from a coastal-hazard model in which climate pressure is a stated planning concern, and its structural path to Y is in any case non-significant, so retention does not inflate any supported conclusion. Given the measurement quality, X5 is treated as an exploratory construct throughout: its coefficient is not interpreted substantively, no comparison of effect sizes involving X5 is made, and no climate-related recommendation in Section 3.4 rests on it. Strengthening this construct with additional items is identified as a requirement for future work.

3.2 Structural model validation

The coefficient of determination (R²) was evaluated to assess the explanatory power of the revised structural model. With X6 included as an explanatory construct, the revised SmartPLS estimation gives an R² of 0.536 and an adjusted R² of 0.528 for Y.

Table 7 reports the R² and adjusted R² values for the dependent construct.

Table 7. Coefficient of determination (R²) for disaster mitigation effectiveness (Y)

The R² value of 0.536 means that 53.6% of the variance in Y can be explained by the predictors in the revised model. The remaining 46.4% may be associated with factors outside the model, including economic capacity, institutional and government support, emergency response capability, objective ecosystem condition, and changes in coastal land use. These factors were not included as explanatory constructs in the present SEM-PLS model and therefore represent potential directions for further model development. The revised model thus provides a moderate level of explanatory power for the perception-based evidence from coastal households.

Table 8 presents the bootstrapping results of the revised structural model. The results include path coefficients, t-statistics, and p-values used to assess the significance of the relationships between the explanatory variables and Y.

Table 8. Bootstrapping results of the structural model

The bootstrapping results in Table 8 show that X3 has the strongest statistically significant relationship with Y (β = 0.620; t = 11.147; p < 0.001). The relationships for X1, X2, X4, X5, X6, and N1 are not statistically significant at the 5% level. X6 has a negative coefficient of −0.111 with p = 0.051, which is slightly above the 0.05 threshold. Therefore, X6 is retained as an empirically tested construct, but its relationship with Y should be interpreted as non-significant at the 5% level rather than as a confirmed determinant.

To evaluate whether N1 functions as a mediator, indirect effects from X1, X2, X4, X5, and X6 to Y through N1 were assessed using bootstrapping in SmartPLS 4.0 with 5000 resamples. The results are presented in Table 9.

All five indirect paths are not significant, since every p-value is above 0.05. This indicates that N1 does not mediate the relationships between X1, X2, X4, X5, or X6 and Y. The direct path from N1 to Y is also not significant (β = −0.003; t = 0.062; p = 0.950). Therefore, N1 is retained in the structural framework as a direct explanatory construct, but the present sample does not provide statistical evidence of either a mediating or direct effect.

Table 9. Bootstrapping results of indirect effects through public education (N1)

Table 10 reports the goodness-of-fit statistics for the revised model. The Standardized Root Mean Square Residual (SRMR) value is 0.065 for both the saturated and estimated models. The Normed Fit Index (NFI) is 0.752, while d_ULS, d_G, and chi-square are 2.817, 0.867, and 2041.676, respectively. These values represent the model after the inclusion of X6 and should be interpreted together with the measurement and structural results.

Table 10. Model fit results

The non-significance of X1, X2, X4, X5, and X6 means that statistically significant relationships with Y were not established for these constructs in the present sample at the 5% level. This does not imply that these dimensions are unimportant in disaster mitigation practice. Their interpretation should instead consider the way they were operationalized through household perceptions and the spatial context of Banda Aceh. X3, by contrast, is the only construct that shows a statistically significant relationship with Y and therefore provides the strongest empirical basis for the current planning recommendations.

The non-significance of X1 may partly reflect measurement design: it was captured through perception-based survey items rather than objective hazard-exposure indicators, and perceived risk can vary independently of a household's actual spatial exposure across the five coastal districts. This may introduce measurement noise that attenuates the statistical relationship. Similar caution applies to X2, X4, and X5 because the present analysis measures perceptions rather than direct observations of land-use compliance, participation intensity, or physical climate impacts.

3.3 Empirical model and comparative positioning

Based on the revised SEM-PLS results, the empirical model identifies X3 as the strongest statistically supported determinant of Y. The path coefficients represent standardized structural relationships rather than a predictive regression equation. The revised model can be summarized as: Y = −0.020X1 + 0.103X2 + 0.620X3 + 0.061X4 + 0.043X5 − 0.111X6 − 0.003N1 + ζ. X6 is included explicitly because X6 was measured through dedicated questionnaire indicators and tested as an explanatory construct in the revised model

The empirical model translates the SEM-PLS results into priorities for disaster mitigation in coastal settlements. X1 remains a contextual hazard variable, but merely recognizing it is insufficient. Y depends more on whether settlements are physically and socially resilient-road access to evacuation routes, vertical evacuation buildings, disaster-resistant housing, and integration of mitigation infrastructure into statutory spatial plans.

Table 11. Comparative positioning of this study relative to prior coastal mitigation research

Table 11 positions the present SEM-PLS findings against the four strands of prior Banda Aceh and comparable coastal-mitigation research already discussed above (hazard mapping, regulation, structural/coastal protection, and community or ecosystem-based management). The revised comparison also reflects the explicit inclusion of X6 as a construct.

Read across the comparison, prior studies emphasize hazard exposure, regulation, and structural protection [10, 19]. Related studies further address community preparedness and risk-sensitive planning [7, 11, 20], while other work emphasizes integrated coastal management [12, 13] and highlights the importance of considering social and psychological consequences in spatial planning [21]. Unlike these predominantly hazard-planning or case-study-centered approaches, this study uses a household-based SEM-PLS framework to test several of these dimensions together through a perception-based model, with X6 tested explicitly rather than discussed only descriptively. X3 is statistically significant, whereas X1, X2, X4, X5, and X6 are not statistically significant at the 5% level. The X6 result should therefore be interpreted as empirical evidence from household perceptions of X6, but not as a statistically confirmed determinant of Y.

3.4 Implications for spatial planning

It should be emphasized that, based on the revised structural results, X3 is the only statistically significant determinant of Y at the 5% level. X6 has a negative but non-significant relationship with Y (β = −0.111; t = 1.950; p = 0.051). Recommendations concerning spatial planning enforcement, N1, X4, climate adaptation, and ecosystem-based protection are therefore presented as planning and policy implications rather than as claims of statistically significant effects.

The empirical model therefore recommends strengthening X3 as the main empirically supported intervention area, while continuing to consider spatial planning enforcement, N1, X4, climate adaptation, and X6 in coastal mitigation planning. These dimensions are presented as complementary planning considerations unless supported by statistically significant relationships in the present model. For X6 specifically, the non-significant result at the 5% level does not remove its relevance as a planning consideration; rather, it indicates that the household-perception data do not establish a statistically significant direct relationship with Y. Research on spatial impacts in dense hazard-prone settlements further supports attention to social and psychological dimensions when planning recovery and risk reduction [21].

N1 remains part of the conceptual framework as a direct explanatory construct representing household knowledge and awareness. The direct path from N1 to Y is not statistically significant (β = −0.003; t = 0.062; p = 0.950), and its indirect effects are also not significant. Therefore, N1 is not presented as a confirmed determinant or mediator in the current empirical evidence. Instead, N1 is retained as a practical planning consideration that may support preparedness, participation, and appropriate responses when implemented together with physical and institutional measures [12].

The role of N1 is specified consistently across the model, the tables, and the structural equation. N1 enters the final model as a direct explanatory construct in parallel with X1-X6; its direct path to Y is reported in Table 8 (β = −0.003; t = 0.062; p = 0.950), its indirect paths are reported in Table 9, and it appears in the structural equation as the −0.003N1 term. Both the direct and the indirect effects of N1 are non-significant in this sample. N1 is therefore retained in the conceptual framework and in the reported model for completeness, but it is not claimed as a determinant or as a mediator of Y.

These findings align with prior research emphasizing hazard-sensitive zoning, structural measures, integrated coastal management, and differentiated spatial controls [7, 12, 13]. The empirical model therefore yields five planning priorities: (1) stricter regulation of disaster-prone coastal zones near the shoreline and within historical tsunami inundation areas; (2) integrated evacuation infrastructure, including improved roads, signage, vertical evacuation buildings, and gathering points; (3) strengthened X3 through better building quality, disaster-resistant construction, and public facility preparedness; (4) institutionalized X4 and education through regular training and local disaster forums; and (5) consideration of X6, including mangrove restoration and other nature-based measures, within the RSP and DSP. These recommendations are framed as planning implications, while the SEM-PLS result for X6 is reported as non-significant at the 5% level.

The authors declare that there are no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.