Nature-based flood assessment for vulnerable educational infrastructure under tropical urban conditions

Worldwide, urban rainwater flooding is increasingly affecting small-scale public infrastructure in tropical cities. Consequently, drainage decisions are often made with limited hydrometric data. Therefore, this study aims to evaluate flood control alternatives using the Nature-based Solutions (NBS) approach through participatory assessment and modeling with the Stormwater Management Model (SWMM) at a vulnerable educational infrastructure (VEI) in Durán, Ecuador. The methodological design of this research involves case study analysis, rainfall data and the SWMM model. Then, a community workshop was developed with the VEI's stakeholders to identify areas of recurrent waterlogging, reported flood depths and drainage constraints. A reference condition and three Low-Impact Development (LID) scenarios were then simulated under 15 synthetic design storms, combining return periods of 5, 10 and 20 years with durations of 30 min, 1, 2, 3 and 4 hours. Finally, the model's performance was evaluated using variables such as runoff, outflow volume, peak flow, flood volume, maximum depth, the multi-criteria LID performance index, bootstrap confidence intervals and an exploratory robustness analysis. The findings of this study highlight the participation of 28 stakeholders in the workshop, including parents and teachers, which enabled the identification of flood-prone areas and the location of LID scenarios. This research demonstrated that individual LID scenarios reduced certain hydrological indicators but, in some cases, increased the flood volume or maximum depth. The combined LID scenario provided the most balanced response, reducing average runoff by 37.90%, flood volume by 50.90%, maximum water depth by 46.44%, and outflow volume by 17.37% compared with the baseline. Of particular note is the integration of hydrological modeling and the effectiveness of LIDs in reducing runoff and redistributing peak flows. The SUDS/LID alternative most widely accepted by the school community and demonstrating the best hydrological performance was the sustainable urban drainage system, which should be strategically located near the main runoff-generating surfaces or upstream of recurrent flooding areas. This study provides a flood risk management tool that can be adapted to other VEI environments in urban-coastal regions where hydrological data is scarce.

Urban and informal expansion poses a contemporary challenge for surface runoff management, as it increases soil sealing and creates conditions that heighten the risk of flooding (Davis and McCuen, 2005; Barbosa et al., 2012; Hirabayashi et al., 2013). In recent years, stormwater flooding has become a growing concern for urban water management in tropical cities Huang et al. (2020). Evidence from urban contexts with limited data shows that flood assessments must consider local drainage conditions and site-specific information. In Dar es Salaam, Tanzania, community-mapped data supported the development of detailed urban flood models where conventional drainage datasets were limited (Petersson et al., 2020). In Seattle, USA, high-resolution data on land use, impervious surfaces and stormwater infrastructure improved the representation of urban runoff processes (Halama et al., 2023). Although these cases differ in scale and context, they demonstrate the value of combining hydrological modeling with local evidence on drainage, particularly in contexts where monitoring networks are sparse.

Vulnerable Educational Infrastructures (VEIs) are susceptible to recurrent rainwater flooding because water accumulation can disrupt school activities, limit safe mobility, affect sanitation facilities, increase maintenance requirements and expose pupils and teachers to unhealthy environmental conditions (Munsaka and Mutasa, 2021). The risk of school flooding also has an equity dimension. Schools prone to flooding in the United States have been linked to geographical and socio-economic disparities (Miao et al., 2026), whilst multi-hazard assessments of school systems have shown that flooding can contribute to physical damage, reduced accessibility and disruption to educational provision (Vatteri et al., 2022). In tropical urban environments, these pressures may be exacerbated by high temperatures, inadequate sanitation, a lack of shaded areas and insufficient maintenance of the drainage system.

Nature-based Solutions (NBS) are increasingly seen as part of the response to urban water insecurity, particularly in cities in the Global South that are experiencing rapid urbanization (Abera et al., 2025). Low-impact development (LID) measures are local-scale applications of the approach that incorporate NBS. These measures are designed to retain, infiltrate, store or delay runoff before it enters the drainage network. Green roofs, bioretention basins, infiltration areas and distributed combinations of these measures are among the most common examples (Prince George's County Department of Environmental Resources, 1999; Woods Ballard et al., 2015; Ahiablame et al., 2012; Elliott and Trowsdale, 2007; Li et al., 2020; Weggemans et al., 2023; Corbett et al., 2024). However, their benefits cannot be taken for granted. The performance of stormwater control measures depends on how individual interventions are configured, connected and integrated with existing impervious surfaces and drainage systems (Bhaskar et al., 2025). A control measure that reduces runoff or peak flow at the sub-basin scale may still produce adverse local hydraulic effects if downstream conveyance capacity, storage through ponding or drainage connectivity remain limited (Li et al., 2020; Weggemans et al., 2023; Bhaskar et al., 2025).

The Storm Water Management Model (SWMM), developed by the U.S. Environmental Protection Agency (EPA), provides a well-established event-based framework for modeling urban runoff generation, flow within the drainage network, surface water pooling and LID controls under a constant rainfall forcing (Rossman and Simon, 2022; U.S. Environmental Protection Agency, 2026). Other hydrological and hydrodynamic models, such as HEC-HMS, HEC-RAS 2D, MIKE URBAN and InfoWorks ICM, can also support the assessment of urban runoff and flood dynamics, particularly where detailed data on land use and infrastructure are available (Halama et al., 2023; Lin et al., 2024; Kane et al., 2024; Apel et al., 2024; Jato-Espino et al., 2023).

In Ecuador, recent studies on flood vulnerability in small coastal towns highlight the need to carry out data-driven and spatially explicit risk assessments under conditions of limited monitoring (Macas-Espinoza et al., 2026). This need is particularly relevant in the canton of Durán, an area characterized by terrain that is highly susceptible to flooding due to its valleys and alluvial terraces (Borbor-Cordova et al., 2020a; da Cruz e Sousa and Lopes Miranda, 2018) as it is situated in the lower reaches of the Guayas River catchment, the largest in the country. Other factors, such as urban expansion, surface sealing and drainage constraints, increase the pressure on stormwater infrastructure during heavy rainfall events (Matamoros et al., 2020)

In this study, SWMM is not treated as a fully calibrated predictive tool, but rather as a framework for comparing a reference condition with various intervention scenarios subject to the same rainfall conditions. This approach enables the analysis of local hydraulic constraints and the redistribution of water within the system using paired hydrological and hydraulic indicators (Elliott and Trowsdale, 2007; Ahiablame et al., 2012; Weggemans et al., 2023). Participatory input complemented the analysis by identifying critical flooding points, providing a spatial interpretation of the results and conducting qualitative plausibility checks, without replacing hydrometric calibration (Petersson et al., 2020; Mitze et al., 2025). La literatura reciente incluye estudios que combinan SWMM con procedimientos de optimización y evaluación multicriterio para comparar y clasificar configuraciones LID en sistemas urbanos de drenaje. Nazari et al. desarrollaron un marco SUSTAIN–SWMM–MCDM basado en criterios de reducción de la escorrentía, del caudal pico, de la capacidad hidráulica de la red y de los costos de implementaciónRecent literature includes studies that combine SWMM with optimisation and multi-criteria evaluation procedures to compare and rank LID configurations in urban drainage systems. Nazari et al. developed a SUSTAIN-SWMM-MCDM framework based on criteria for reducing runoff, peak flow, the hydraulic capacity of the network and implementation costs (Nazari et al., 2023); they used multi-objective optimisation and the TOPSIS method in multi-criteria analysis to prioritize LID alternatives according to their hydrological and economic performance (Javan et al., 2025).

The frameworks combining SWMM and multi-criteria assessment have primarily focused on the technical and economic optimisation of larger urban systems (Nazari et al., 2023; Javan et al., 2025), whilst participatory and sensitivity analysis approaches have been applied mainly to spatial vulnerability indices and variations in their weightings (de Brito et al., 2019; Moreira et al., 2023). Furthermore, the selection and weighting of indicators can alter the results of flood indices, making it necessary to examine the sensitivity and stability of classifications under different decision-making priorities (de Brito et al., 2019; Moreira et al., 2023).

The use of VEI-oriented frameworks is limited. Consequently, this study integrates socio-hydrological, technical and hydrological components. From a socio-hydrological perspective, a participatory assessment is carried out at the school level, considering key stakeholders such as the parents' committee and VEI teachers, through a multi-criteria evaluation of LID scenarios. The technical-hydrological component is incorporated through modeling using SWMM and the assessment of the stability of its classification in the face of various sources of uncertainty. This research provides a reproducible methodological framework that aims to contribute to governance processes in tropical urban environments where hydrological data is limited due to the absence of time series for flow or water depth. This is particularly the case in some areas of the Durán canton, which still lack modeling tools capable of comparing drainage interventions at school level. In this coastal city, 24 % of urban areas suffer the effects of chronic flooding (Borbor-Cordova et al., 2020b). Flood control at the school level is not merely a matter of hydraulic design, but also a question of infrastructure resilience and urban water equity (Field et al., 2012; Cornejo-Rodriguez et al., 2020; Miao et al., 2026; Vatteri et al., 2022). In this context, the following research questions are formulated: To what extent can NBS-based drainage strategies mitigate flooding in a VEI under tropical urban conditions and with limited data availability? And how stable does the ranking of these alternatives remain in the face of changes in decision-making preferences, the model's physical parameters and storm conditions?

This study aims to evaluate NBS alternatives for managing frequent flooding at an urban educational institution in Durán, Ecuador. The approach incorporates the following components: (i) a participatory assessment to identify flood-prone areas, reported water depths and local drainage constraints; (ii) the development of an SWMM-based framework to compare a baseline condition and three LID scenarios at a tropical urban school; (iii) a comparison of individual and combined LID alternatives under multiple design storms; and (iv) an assessment of performance using hydrological and hydraulic indicators, the multi-criteria indices LPI and FSI, sensitivity analysis of weightings, local and global sensitivity of the model's physical parameters, and paired bootstrapping to examine inter-storm variability and the robustness of the classification.

Methodological design of this study combined a qualitative approach, based on a workshop conducted with key stakeholders from the vulnerable educational infrastructure (VEI), with a quantitative approach involving SWMM modeling, hydrological characterization of the study site, and sensitivity analysis of the modeled scenarios (Figure 1).

Phase I: Case study, rainfall data, and SWMM model

Case study

The study site comprised the “Ciudad de Nazareth” Basic Education School, located in the urban parish of Divino Niño, Durán Canton, Guayas Province, Ecuador. According to the 2022 Population and Housing Census conducted by the National Institute of Statistics and Census (INEC), Durán has an estimated population of 303,910 inhabitants (Instituto Nacional de Estadistica y Censos, 2023). The canton covers an area of approximately 342 km 2 and contains 143 educational institutions, employing 2,494 teachers and serving 67,097 students (Ministerio de Educacion del Ecuador, 2024). These figures highlight the importance of assessing vulnerable educational infrastructure exposed to hydrometeorological hazards.

The spatial configuration of the study area and its representation within the SWMM model are presented in Figure 2. The site was selected according to the following inclusion criteria: (i) location within an urban environment; (ii) classification as vulnerable educational infrastructure; (iii) documented evidence of historical flooding; and (iv) the need to assess waterlogging conditions within the school grounds.

Rainfall data

Local rainfall was characterized using hourly records from the M1271 meteorological station, provided by Ecuador's National Institute of Meteorology and Hydrology (INAMHI), covering the period from July 2024 to August 2025 (Instituto Nacional de Meteorologia e Hidrologia, 2025). The highest rainfall totals were recorded between February and April 2025, with monthly values exceeding 450 mm and a maximum of 618.4 mm in February. Mean monthly rainfall during the study period was 130.3 mm, whereas rainfall was absent or very low between August and November 2024. The design rainfall used as input to the SWMM model was estimated using the regionalised Bernard relationship for the provinces of Los Ríos, Guayas and El Oro, proposed by Velasco Ramos et al. (2023):

where I represents rainfall intensity (mm/h), T the return period (years), D the storm duration (min), and K, m, and n the empirical parameters of the IDF relationship. For each combination of return period and duration, the resulting intensity was converted into cumulative rainfall depth as follows:

where P is the total rainfall depth (mm).

The calculated rainfall depths were temporally distributed using the Chicago method at 5-minute intervals. Fifteen synthetic storms were generated by combining three return periods of 5, 10, and 20 years with five storm durations of 30, 60, 120, 180, and 240 minutes. These storms were used as rainfall inputs to simulate the baseline and LID scenarios (Woods Ballard et al., 2015; Corbett et al., 2024). Under each storm condition, the same hyetograph was applied to all scenarios. This enabled paired comparisons and allowed differences in the model outputs to be attributed to the intervention configurations, while rainfall and drainage-network geometry remained constant. The design storms are summarized in Table 1.

SWMM model

Model construction and evaluation followed an integrated workflow involving UAV-derived information, geospatial processing, hydrological–hydraulic simulation, and statistical analysis. UAV imagery was processed using the DroneDeploy smartphone application (DroneDeploy, 2026) to generate the digital elevation model. Subsequently, the topographic information was prepared using the open-source QGIS software (version 4.0.3) (QGIS Development Team, 2026), which was used to delineate the subcatchments and represent the drainage network. Based on this spatial configuration, the hydrological and hydraulic response of the drainage system serving the vulnerable educational infrastructure (VEI) was simulated using EPA SWMM software (version 5.2.4) (Rossman and Simon, 2022). Finally, R was used to post-process the model outputs, conduct the statistical, multicriteria, and uncertainty analyses, and produce the graphical representations (R Core Team, 2026). The runoff and flow-routing continuity errors are presented in Table 2.

Following the integration of the topographic and drainage information, the study area was discretised into 23 subcatchments covering a total modeled area of 0.2721 ha, representing the roofs and other runoff-generating surfaces within the school grounds. The hydraulic network comprised 15 nodes, including 11 junctions and four outfalls, connected by 11 conduits. Among the four outfalls, node 13 was defined as the main system outlet and the hydraulic control point used to evaluate peak discharge. Seventeen subcatchments, covering 0.1979 ha and representing approximately 72.7% of the total modeled area, drained toward this node. In contrast, outfalls 12, 14, and 15 were incorporated as auxiliary boundary conditions to represent the direct discharge of runoff to the surrounding streets, primarily from roofs whose drainage systems were not connected to the internal school network. Hence, these auxiliary outfalls reproduced the observed drainage configuration, whereas node 13 provided the main control point for the comparative assessment of hydraulic performance. The adjustment of these components enabled the development of a model configuration consistent with the observed drainage dynamics. The spatial configuration of the model is presented in Figure 3.

To ensure comparability among the simulated interventions, the subcatchment geometry, network configuration, and outfall locations remained unchanged across the LID scenarios. Accordingly, only the components and parameters associated with the interventions were modified, ensuring that differences in the simulated responses could be attributed to the LID measures rather than to changes in the model structure.

Phase II: participatory workshop

Local knowledge and the social dimension were incorporated into the flood assessment through a participatory workshop held on 5 October 2025. The activity combined a participatory mapping exercise with a semi-structured group discussion, following recent applications of community-generated spatial information to complement technical assessments of flood risk and drainage conditions (Gnecco et al., 2024; Hummel et al., 2025).

Participant selection considered a target population of 236 adults from the school community, comprising 224 parents and 12 teachers (Escuelas Ecuador, 2026). From this population, an accessible population (N) of 36 representatives was defined, including 20 parents serving on school committees for grades one to four, 12 parents representing grades five to seven, and four teachers with the longest association with the vulnerable educational infrastructure (VEI). Participants were selected according to the following inclusion criteria: (i) direct knowledge of the school surroundings and recurrent flooding problems; (ii) familiarity with the affected access routes; and (iii) experience of the educational infrastructure during previous rainfall events. Based on this accessible population, the required sample size (n) was calculated using the finite-population equation (Alcaraz-Quiles et al., 2015):

where N represents the accessible population, z the value associated with the selected confidence level, p the expected probability of occurrence, q = 1− p, and ε the allowable error. The calculation used N = 36, z = 1.44, corresponding to an 85% confidence level, p = q = 0.50, and ε = 0.0649.

Based on these parameters, the estimated sample size was 28.03 and was rounded to 28 participants. Before the workshop, all participants were informed about the purpose of the activity, the voluntary nature of their involvement, and the use of the collected information for research purposes. Informed consent was obtained from all 28 participants, who were adults. The workshop activities, participatory mapping exercise, and group documentation are presented in Figure 4.

During the workshop, the guiding questions addressed areas affected by recurrent ponding, preferential surface-flow pathways, affected classrooms and circulation spaces, disruptions to access routes, perceived water depth, ponding duration, and interference with school activities. Accordingly, the collected information was used to interpret local impacts and spatially compare the critical locations identified by participants with the hydraulic behavior represented by the SWMM simulations. However, these participatory data were not used to calibrate the model.

Phase III: LID scenario design and comparative performance metrics

Establishing a consistent basis for the subsequent performance assessment required the extraction of five outputs from each simulation: total surface runoff, total outflow volume, peak discharge at outfall 13, flood volume at the nodes, and maximum water depth. Total outflow volume was retained as a complementary variable to support the interpretation of the hydraulic balance. In contrast, the remaining four outputs were used as performance indicators and subsequently incorporated into the LID Performance Index (LPI) and Flood Severity Index (FSI).

Because the four performance indicators were originally expressed in different units, they were converted into percentage changes relative to the baseline scenario. This transformation produced dimensionless and directly comparable values with a consistent direction of interpretation. It also provided the common input required to integrate the indicators into the LPI and FSI during the multicriteria evaluation conducted in Phase IV.

Comparisons were performed on a storm-by-storm paired basis, so that each LID result was compared with the corresponding baseline result under the same rainfall conditions. The resulting percentage changes were examined individually to evaluate performance across storm durations, intensities, and return periods, and were subsequently summarized using the mean and measures of variability. This aggregation assigned equal importance to each design storm and provided the basis for the overall classification of the LID alternatives.

Percentage change was calculated as follows:

where R i, s, j represents the percentage change in indicator i for LID scenario s and storm j; X i, ref, j is the value of the indicator under the baseline scenario; and X i, s, j is the value obtained for the same indicator under the LID scenario. Positive values represent a favorable reduction relative to the baseline scenario, zero indicates no change, and negative values represent an increase in the indicator.

Phase IV: multicriteria evaluation of performance and flood severity

Joint consideration of the four previously defined hydrological and hydraulic indicators enabled an integrated comparison of the LID alternatives. Two complementary indices were used: the LID Performance Index (LPI), which evaluated the overall performance of each alternative, and the Flood Severity Index (FSI), which focused specifically on the mitigation of local flooding. Uncertainty associated with decision priorities was examined using different weighting schemes. Within this framework, W 0 was adopted as the primary weighting scheme for the LPI, whereas the FSI retained fixed weights.

LPI calculation

Reductions in surface runoff, peak discharge, flood volume, and maximum water depth were integrated into the LPI through a weighted linear combination (Ni et al., 2025):

subject to:

where s represents the LID scenario, j the design storm, k the evaluated indicator, q the weighting scheme, R k, s, j the percentage change relative to the baseline scenario, and w k (q) the weight assigned to each indicator. Higher LPI values indicate better integrated performance. Four deterministic weighting schemes were considered:

Scheme W 0: assigned equal importance to the four indicators.

Scheme W 1: prioritized the reduction of local flooding in accordance with the concerns identified during the participatory workshop.

Scheme W 2: prioritized hydraulic safety.

Scheme W 3: prioritized control of the hydrological response.

The corresponding weights are presented in Table 3. Comparing these schemes allowed the dependence of the alternative rankings on specific decision priorities to be examined. Sensitivity of the LPI to uncertainty in the indicator weights was subsequently evaluated using scheme W 4. A total of 10 000 weighting vectors were generated from a symmetric Dirichlet distribution with parameters α = (1, 1, 1, 1), allowing the weighting space to be explored without favoring any of the four indicators. The LPI was recalculated for each vector, and the highest-ranked alternative was identified. The relative frequency with which each alternative occupied the first position was interpreted as a probability conditional on the adopted weight distribution rather than as a universal probability of superiority.

FSI calculation

Flood Severity Index (FSI) combined the reductions in flood volume and maximum water depth using equal weights (Sharma, 2021):

where R flood, s, j represents the percentage change in flood volume and R depth, s, j the percentage change in maximum water depth for alternative s and storm j. Higher FSI values indicate a greater reduction in local flood severity.

Both FSI components retained fixed weights of 0.50 and were not subjected to schemes W 0 – W 4, which were applied exclusively to the LPI. Nevertheless, FSI stability across storm events was evaluated using a paired bootstrap of the 15 design storms with 2,000 replicates. This procedure was used to estimate mean values, percentile-based 95% confidence intervals, the frequency of the ranking positions attained by each alternative, and the margin relative to the second-ranked alternative. Overall performance was obtained by averaging the FSI values across the 15 design storms.

Phase V: performance variability and robustness of the LID alternatives

Sensitivity and uncertainty analyses were used to determine how variations in the physical parameters and design storms affected the performance and ranking of the LID alternatives. Accordingly, local sensitivity analysis, global parameter-uncertainty assessment, and paired resampling were combined to evaluate ranking stability under the analyzed conditions. All stochastic components were executed using the fixed seed 20260713.

One-at-a-time (OAT) analysis was used to identify the factors exerting the greatest influence on the model responses. Each factor was modified individually around its reference value while the remaining factors were held constant (Ahmed et al., 2026; Akdoğan and Güven, 2016; Del Giudice and Padulano, 2016). This procedure also supported the selection of the physical LID parameters that were subsequently varied simultaneously (Tansar et al., 2023).

Five groups of factors were analyzed: Horton infiltration, surface Manning roughness, depression storage, bioretention storage depth, and green-roof substrate depth. Only active parameters were modified, whereas values originally equal to zero remained unchanged. The configurations were generated automatically using PowerShell and executed in EPA SWMM. Each factor was modified according to:

where θ p, 0 represents the reference value of factor p, θ p, δ its modified value, and δ the applied relative change. The experimental design comprised 21 configurations: the original configuration and four perturbations for each factor. This resulted in 1 200 perturbed simulations which, together with the 60 original simulations, produced 1 260 records.

For each configuration, the hydrological and hydraulic responses, performance indices, and ranking positions of the alternatives were recalculated. Sensitivity was quantified using the normalized elasticity between the two extreme perturbation levels:

where E p, k represents the elasticity of response k to factor p, Y k, 0 the reference result, and Y k, p (- 0. 20) and Y k, p (+ 0. 20) the results corresponding to the extreme perturbations. The absolute elasticity value indicated the magnitude of sensitivity, whereas its sign indicated the direction of the response. Because each factor was modified separately, the OAT assessment represented a local analysis within the ±20% interval.

Simultaneous variation of the five factors was evaluated using Latin hypercube sampling (LHS), which enabled stratified exploration of the input intervals (Lambert et al., 2025). Parameter vectors and SWMM runs were generated using Python, while the input and output files associated with each sample were retained. For each sample m, the parameters were calculated as follows:

The 500 parameter vectors were applied to the baseline scenario and the three LID alternatives under the 15 design storms, producing 30 000 additional simulations. Scheme W 0 was held constant to separate the effects of physical-parameter uncertainty from those associated with the decision weights.

Statistical processing was conducted in R (version 4.5.3). The influence of each factor was examined using Spearman's rank correlation, partial rank correlation coefficients (PRCCs), and standardized regression coefficients (SRCs). Spearman's correlation identified monotonic associations, PRCCs estimated the association of each factor while controlling for the remaining factors, and SRCs enabled comparison of their relative influence. The p -values within each family of associations were adjusted using the Benjamini–Hochberg procedure.

Event-to-event variability was subsequently examined through a non-parametric bootstrap with 2,000 replicates, using the design storms as paired blocks. During each replicate, a sample equal in size to the original storm catalog was drawn with replacement, and the same storm identifiers were applied to the baseline scenario and all LID alternatives. Mean responses, percentage reductions, performance indices, and their percentile-based 95% confidence intervals were then calculated from the bootstrap distributions.

Paired t -tests were used to compare the baseline scenario with each alternative when the paired differences satisfied the normality assumption; otherwise, the Wilcoxon signed-rank test was applied. The magnitude of the differences was quantified using Hedges' g z or the rank-biserial correlation, depending on the statistical test employed. The p -values from each family of multiple comparisons were adjusted using the Holm procedure. These comparisons were used to determine whether the observed differences were sufficiently consistent to support the ranking of the alternatives. Finally, ranking robustness was assessed from the position attained by each alternative across the bootstrap replicates, using the frequency of each ranking position and the score margin between the highest-ranked alternative and its closest competitor as stability measures.

Estimation of implementation costs

Construction costs were estimated through a unit-price analysis based on wage references and input prices from the Ecuadorian market. The estimate was prepared at a pre-feasibility level and included the direct costs of materials, labor, machinery, and minor tools required to implement the bioretention cells and green roofs.

Labor costs were established using Ecuador's Unified Basic Salary and statutory sectoral minimum wages as references. For 2025, the Ministry of Labor set the Unified Basic Salary at 470 USD per month through Ministerial Agreement MDT-2024-300 and established the sectoral minimum wages through Agreement MDT-2024-301 (Ministerio del Trabajo, 2024; Registro Oficial del Ecuador, 2024). For 2026, the Unified Basic Salary was set at 482 USD (Ministerio del Trabajo, 2025). These values provided the basis for estimating the hourly labor costs of construction workers, assistants, and landscaping personnel.

Unit prices for materials, equipment, and machinery were obtained from unit-price analyses prepared for the Ecuadorian market (CYPE Ingenieros, S. A., 2026). The principal components considered were topsoil, filter gravel, geotextiles, waterproofing membranes, drainage systems, substrates, and vegetation cover. The temporal consistency of these prices was assessed using the Construction Price Index (IPCO), an official indicator that measures monthly changes in the prices of materials, equipment, and machinery used by the Ecuadorian construction sector (Instituto Nacional de Estadistica y Censos, 2026).

Finally, total implementation costs were calculated by multiplying the corresponding unit cost by the treated area. The cost of LID001 was estimated for 569 m 2 of bioretention cells, whereas LID002 was estimated for 710 m 2 of green roofs. The cost of LID003 was obtained by combining the costs of both interventions.

Participatory diagnosis and its implications for the evaluation

The 28 participants identified flooding as the school community's main infrastructure problem. The perceived causes included a lack of internal drainage, debris accumulation in drains, deficiencies in the existing infrastructure, and local weather conditions. Exposure to heat was also recognized as a significant issue. The effects noted included health impacts, thermal discomfort, and reduced participation in school activities. The activities carried out during the workshop are shown in the Figure 4.

The proposed solutions included drainage works, land grading, canals, shade structures, tree planting, and green spaces. Among these options, the participants prioritized an intervention based on SuDS/LID principles, a summary of which is presented in the Figure 5. This prioritization linked the community assessment with the quantitative evaluation, which aimed to determine the extent to which the alternatives could reduce flooding in educational infrastructure.

Hydrological, hydraulic, and multi-criteria performance of the LID alternatives

The hydraulic response of the conduits differed among the interventions. The maximum duration of limited-capacity conditions increased from 0.42 h in the baseline scenario to 0.45 h with LID001, but decreased to 0.35 h with LID002 and to 0.29 h with LID003 (Table 4). Therefore, bioretention alone did not alleviate these constraints, whereas green roofs and especially the combined approach reduced the duration during which the system operated under limiting hydraulic conditions.

This pattern was consistent with the reductions observed in the performance indicators (Table 5). LID001 produced the greatest decrease in peak discharge. However, it increased the flood volume, indicating that the attenuation of discharge at the main outlet did not guarantee reduced flooding in the modeled area. LID002 primarily improved the flood volume and depth. LID003, on the other hand, recorded the greatest reductions in three of the four indicators: surface runoff, flood volume, and maximum depth. Overall, the results show that the combined alternative favored internal hydraulic control, although the reduction in peak discharge was smaller.

The integration of these responses using the equal-weighting scheme W 0 placed LID003 first, followed by LID002 and LID001 (Table 6). LID003 obtained the highest mean values for the LPI and the FSI. At the same time, LID001 showed differing results between the two indices: its reductions in runoff and peak discharge resulted in a positive LPI, but its confidence interval for the FSI included zero. This confirms that the hydrological improvement achieved by LID001 did not translate into a consistent reduction in the severity of local flooding. Although the index values varied across events, the ranking of the alternatives remained constant (Figure 6). Furthermore, LID003 retained first place during bootstrap resampling, and its margins relative to LID002 remained above zero (Figure 7).

Sensitivity of the classification and parametric robustness

The ranking showed low sensitivity to the priorities assigned to the indicators. The LID003 model remained in first place among the four deterministic models in the Table 3 and in 99.45% of the 10 000 weighted random vectors. The few investments favored LID001 and were concentrated in configurations that assigned a predominant weight to maximum flow, the indicator in which this alternative demonstrated its main advantage. In contrast, LID003 remained the preferred option under more balanced weight distributions (Figure 8), which demonstrates that its ranking did not depend on a single weighting scheme.

Local sensitivity showed similar stability in response to individual parameter perturbations. LID003 ranked first in 98.10% of the event-by-event evaluations and maintained that position in all aggregated rankings. Specific inversions favored LID002 and were primarily related to increases in storage in depressions, which had the greatest influence on LID003's LPI, with a mean absolute elasticity of 0.408. The substrate depth of the green roofs had a lesser influence, with an elasticity of 0.066, and the other factors produced minor changes (Figure 9). Consequently, surface storage emerged as the primary parameter to be examined in the overall assessment.

The overall analysis confirmed this pattern. LID003 ranked first in 7 029 of the 7 500 cases and maintained that position in all 500 aggregated rankings, with a minimum margin of 6.16 percentage points of LPI over the second-ranked alternative. Storage in depressions showed the strongest association with its performance, with a PRCC = −0.988, in accordance with local sensitivities (Figure 10). Although the parametric variations resulted in reversals in specific cases, these were not sufficient to change the classification within the evaluated ranges.

Performance variability and preliminary feasibility of LID alternatives

The distribution of reductions differed among indicators and alternatives (Figure 11). Surface runoff exhibited the most stable behavior, with consistently greater reductions in LID003, followed by LID002 and LID001. In contrast, peak discharge showed a more variable response: LID001 achieved the highest and most consistent reductions, while LID002 and LID003 exhibited greater variation, with events showing reduced or near-zero benefits. Local flooding indicators showed a different pattern. LID001 exhibited increases in flood volume and minimal changes in maximum depth; LID002 showed improvements in both indicators, albeit with high variability; and LID003 recorded the greatest reductions. Taken together, these results demonstrate a trade-off between peak flow control and the reduction of runoff and local flooding, which supported the selection of LID003 in the integrated assessment. The preliminary economic estimate for Phase VI complemented this assessment and examined the feasibility of its implementation.

Synthesis of trade-offs and alternative selection

The distribution of the reductions made it possible to identify the trade-offs that led to the final ranking (Figure 11). Surface runoff showed a comparatively stable response, while flood volume and depth exhibited greater variation, particularly with LID002 and LID003. This indicates that the benefits regarding local flooding were more sensitive to the characteristics of the events. However, the observed variability did not alter the overall result: LID003 maintained the most balanced performance because it combined runoff control with the greatest improvements in flood indicators.

A higher economic cost accompanied this technical advantage. LID003 had an estimated construction cost of 9 619 USD, corresponding to the sum of the individual interventions, compared to 7 854 USD for LID002 and 1 765 USD for LID001 (Table 7). Since costs were not part of the LPI or the FSI, LID003's top ranking should be interpreted as a hydrological and hydraulic preference, not as a demonstration of economic superiority. Consequently, the combined alternative constitutes the most robust technical option under the evaluated conditions. At the same time, its final selection requires consideration of budget availability and confirmation of the model's performance through field observations.

Trade-offs between hydrological control and local flooding

The main finding was that reducing runoff or peak flow did not, on its own, guarantee a reduction in flooding at educational facilities. LID001 illustrates this difference: although it reduced surface runoff and produced the greatest attenuation of peak flow, it increased flood volume and slightly prolonged periods of limited capacity. This result indicates that the flow recorded at the main outlet and the accumulation of water within the network represent distinct aspects of hydraulic performance. An intervention may alter the magnitude or timing of the discharge without resolving the conveyance and storage constraints that govern local flooding. A comparable behavior was documented by Luan et al. (2017), who observed that LID measures could control runoff while simultaneously proving less effective at eliminating ponding when limitations in the drainage infrastructure persisted. The school case study expands this discussion by showing that such a divergence can also occur in a small network, where the consequences are concentrated in functional areas and not necessarily at the discharge point.

LID003 demonstrated the most balanced hydrological and hydraulic performance by combining green roofs and bioretention. This configuration affected both roof and ground surfaces, allowing runoff control to be distributed across different areas and drainage pathways. The literature has shown that location, treated area, and the spatial combination of measures can be just as important as the individual type of LID (Sadiqi et al., 2025; Gaurkhede and Adane, 2025; Mai et al., 2025). However, the results alone do not demonstrate a synergistic interaction between the two measures. LID003 covered 1 279 m 2,compared to 569 m 2 with LID001 an 710 m 2 for LID002. Therefore, part of its advantage may also be due to the larger treated area. To distinguish between complementarity, an additive effect, and the effect of area, it would be necessary to compare configurations with equivalent surface areas or to use a factorial design that separates the contributions of each intervention.

The multi-criteria ranking made it possible to represent these trade-offs without obscuring the performance of the individual indicators. The contrast between the positive FSI and the inconclusive LID001 shows that a generally favorable assessment can coexist with an insufficient response to local flooding. In contrast, LID003 ranked first in both indices, supporting its selection when the objective includes both hydrological control and floodwater reduction. This preference, however, depends on decision-making priorities: the observed investments, with weightings concentrated almost exclusively on peak flow, favored LID001. Therefore, LID003 should be interpreted as the most balanced alternative among multiple priorities, not as a universally superior option for any management objective.

Robustness of the classification and physical sensitivity

The stability of LID003 in the face of event resampling, deterministic schemes, and 99.45% of the random weighting vectors demonstrates that its top ranking did not depend on any particular storm or on a specific weighting assignment. Unlike a comparison based solely on mean values, this result shows that the preference held despite variations in hydrological evidence and in the priorities represented. The limited investments do not contradict this stability but rather define the decision space within which another alternative might be preferable: LID001 gained an advantage when peak flow was prioritized almost exclusively, while LID003 dominated under more balanced weighting structures. The ranking is therefore robust within the formulated multi-criteria problem, but it continues to depend on the objectives considered legitimate for school infrastructure.

The parametric sensitivity analysis identified storage in depressions as the main factor associated with the relative performance of LID003. Its high local elasticity and the negative PRCC obtained in the global analysis indicate that variations in this parameter substantially altered the LPI within the explored range. One possible explanation is that greater surface storage in the reference scenario reduces the relative advantage attributable to the interventions; however, this interpretation must be verified through field observations. The lower sensitivity observed for the substrate depth of green roofs does not imply that this parameter is irrelevant, but rather that it had a lesser influence in the intervals, storms, and configurations considered. Studies using experimental data have shown that the simulated response of green roofs can change significantly depending on calibration, initial moisture conditions, and the representation of the drainage layer (Jeffers et al., 2022; Weggemans et al., 2023).

Numerical stability, bootstrap analysis, and sensitivity analyses provide evidence of internal robustness, but they do not replace empirical validation. The results show that the classification remained consistent across the storms, weightings, and parametric ranges evaluated; they do not demonstrate that the simulated magnitudes will necessarily be replicated in the field under different temporal rainfall distributions or following the deterioration of the infrastructure. This distinction is particularly important because available evidence shows that the retention capacity of green infrastructure varies with precipitation magnitude, design, and baseline conditions (Dobkowitz et al., 2025; Bhaskar et al., 2025). Therefore, the robustness obtained strengthens the relative comparison among alternatives, but its transfer to real-world conditions remains contingent on calibration and monitoring.

Contributions and limitations of participatory diagnosis

Participatory diagnosis added a social dimension that hydraulic modeling alone could not address. The community identified flooding as the main infrastructure problem and pointed out the areas where water disrupted access, movement, and school activities. This evidence made it possible to focus the assessment on the volume and depth of the flooding, rather than limiting it to total runoff or outfall flow. In contexts with limited data, community mapping can help identify drainage networks, obstructions, and recurring flood zones, as well as improve the relevance of modeling decisions (Petersson et al., 2020; Mitze et al., 2025). In this study, the main contribution was defining the problem and linking technical indicators to socially relevant locations.

Participants' perceptions do not constitute calibration or hydraulic validation data. Identifying a courtyard or access point as a problem area confirms its social significance but does not, on its own, provide a set of observed depths, durations, or volumes. Similarly, respiratory illnesses, allergies, and other health issues mentioned should be interpreted as perceived impacts, not as epidemiological effects causally attributable to flooding. The literature demonstrates that exposure to flooding can interact with issues of accessibility, sanitation, health, and educational continuity (Miao et al., 2026; Vatteri et al., 2022; Coral-Almeida et al., 2026). However, verifying these relationships at the evaluated institution would require specific health, educational, and environmental records. Maintaining this distinction prevents overinterpretation of participatory evidence while preserving its value for guiding management decisions.

The community's prioritization of SuDS/LID also enhances the legitimacy of a potential implementation, but does not guarantee its long-term acceptance or maintenance. Participatory processes can link alternatives to local needs and improve ownership, although subsequent sustainability depends on institutional responsibilities, resources, and monitoring mechanisms. Therefore, participation should continue throughout the final selection, construction design, and monitoring phases, rather than being limited to the initial assessment.

Economic feasibility and implications for implementation

The highest estimated cost accompanied LID003's best integrated performance, as this alternative incorporated bioretention and green roofs. Since costs were not included in the LPI, the FSI or the weighting analyses, its top ranking reflects a hydrological and hydraulic preference, not a demonstration of economic superiority. An alternative with higher absolute performance does not necessarily provide the greatest benefit per unit of currency. Studies that integrate hydrological modeling and economic analysis show that LID selection must also consider life-cycle costs, maintenance, and avoided damages (Ferreira and da Rocha, 2026; Kumar et al., 2022). Therefore, the economic feasibility of LID003 requires an assessment of the cost per unit of reduced flooding, the present value of the investment, and the benefit-cost ratio.

Differences in performance and cost make a phased implementation a viable option. LID002 could serve as an intermediate option when the priority is to reduce local flooding, with a lower initial investment than the combined alternative. LID001 would require a review of its location and connection to critical nodes before being implemented in isolation. In contrast, LID003 could be implemented in phases, starting in the most exposed areas and expanding after verifying its performance. This strategy would allow for budgetary constraints to be accommodated without assuming that a reduction in the treated area will proportionally maintain the simulated performance.

Implementation also depends on conditions not addressed in the classification. Green roofs require verification of structural capacity, waterproofing, drainage, and access for maintenance; bioretention requires evaluation of inflows, overflow, silting, and sediment accumulation. These conditions are particularly relevant in an educational institution with limited technical and budgetary resources. Future effectiveness will therefore depend on the hydraulic design, the availability of maintenance, and the allocation of institutional responsibilities (Velásquez Fernández, 2025; Gaurkhede and Adane, 2025; Cerda et al., 2025).

Scope and areas for future research

The results support LID003 as the most technically balanced alternative under the simulated conditions, but they do not yet constitute a definitive construction recommendation. The lack of local measurements prevents verification of the runoff, flow, and depth values reproduced by the model. At the same time, the use of design storms does not represent the sequence of wet and dry periods that affects the recovery of storage in green roofs and bioretention basins. Deterioration, clogging, water quality, and changes in effectiveness associated with maintenance were also not evaluated.

The next steps should prioritize monitoring precipitation, flow, and depth at critical points identified by the community; calibrating the most influential parameters; comparing alternatives under equivalent treated surface areas; and conducting an economic evaluation that incorporates construction, operation, and maintenance. An instrumented pilot study would allow for verifying the simulated reductions, reviewing the location of measurement points, and determining whether the advantage of the combined configuration holds up under real-world conditions. These actions would transform the current classification–based on comparative robustness–into a more solid foundation for design and investment.

The study identified the LID alternative with the most balanced performance in reducing runoff and local flooding at the educational facilities evaluated. Under the simulated conditions, LID003 which combined green roofs and bioretention cells ranked first in the multicriteria evaluation by achieving the greatest reductions in surface runoff, flood volume, and maximum depth, as well as the shortest duration of limited-capacity conditions. The comparison also showed that the alternatives affected the system differently: LID001 achieved the greatest reduction in peak flow but increased flood volume, while LID002 primarily improved local flooding indicators. These results confirm that flow attenuation at the main outlet alone did not ensure a reduction in flooding at the site and that selecting an alternative required a combined consideration of the hydrological response and local hydraulic performance. Results are based on a single case study and require field validation across additional sites before broader generalization to tropical urban environments such as Durán canton. Therefore, this research represents a baseline model for future studies, in which field checkpoints could be selected based on social validation conducted in collaboration with key school stakeholders (VEI).

The ranking of LID003 remained consistent despite resampling of events, changes in weightings, and variations in physical parameters, indicating that its classification did not depend exclusively on a single storm, a decision scheme, or a particular parametric combination. However, this stability represents a measure of internal robustness and not a validation of its field performance. LID003 also covered a larger area and had the highest construction cost, so its advantage cannot be attributed solely to a synergistic interaction between green roofs and bioretention. Because costs were not incorporated into the multi-criteria indices, this alternative's top ranking should be interpreted as a hydrological and hydraulic preference, not as a demonstration of economic superiority. Consequently, the results allow for the identification of a technically favorable candidate but do not yet support a definitive construction recommendation or its generalization to other educational institutions.

The participatory assessment links the technical evaluation to the problems and areas considered priorities by the school community. However, it did not replace calibration or validation through hydrometric measurements. Its contribution consisted of contextualizing the flooding, guiding the selection of relevant indicators, and supporting the interpretation of scenarios based on local needs. The next step should integrate this social evidence with observations of precipitation, flow rate, and water depth, model calibration, structural testing of roofs, maintenance assessment, and a life-cycle economic analysis. Furthermore, comparing alternatives with equivalent treated areas and developing an instrumented pilot project would make it possible to determine whether the simulated advantage of LID003 holds up under real-world conditions. Therefore, the study provides a comparative basis for continuing the design and evaluation at the institution analyzed, with favorable results within the model. However, these are still subject to technical, economic, and empirical verification.