South Jakarta is an economic hub of Indonesia’s capital, characterised by advanced infrastructure, extensive services and intensive horizontal and vertical development. Morphologically, it occupies alluvial plains composed mainly of tuff and clay. With clay occurring widely in the city, water rarely infiltrates (vertical) but instead flows on the ground (horizontal). The city has varying elevations from 5 m to 50 m above mean sea level (masl). However, it is mostly below 12.5 masl (covering 87.13% of the total area) or low-lying land with a slope gradient of 0% – 8% (flat to sloping terrain). These geographical and morphological profiles show the area’s high proneness to floods because runoff collects and accumulates quickly and prevents prompt infiltration into the soil.

The main drainage network of South Jakarta consists of the Ciliwung, Krukut, Pesanggrahan, Kali Grogol and Kali Baru Barat Rivers. On average, the city receives 2001 mm – 2500 mm of rainfall per year. Based on the pattern from 2013 to 2021, floods occur from November until March, peaking in January and February. A previous study revealed that the monthly average rainfall from December until March is always above 200 mm (Ginting 2015). The city lies in the upstream portion of Jakarta and is traversed by several rivers, which makes it highly prone to inundation during periods of extreme rainfall.

The primary and secondary data used in the research are summarised in Table 1. The primary data were collected through surveys and field observation. The secondary data were derived from satellite images, geospatial databases, statistical information from databases of responsible agencies, previous studies and regulations (Sugiyono 2015). The primary data are used to describe actual conditions in the field and validate current land-use types, and the secondary data were publicly accessible spatial data used as the basis for building the integration of flood-risk and land-use plans. All parameters were processed at a scale of 1:25 000 to match the spatial plan.

In this study, the GIS was used to compile, visualise and analyse numerous layers of geospatial data, including topography, land use, rainfall patterns, soil type, drainage network, elevation, etc. The GIS provides a platform for combining different forms of spatial data and performing MCA. In this context, the GIS was used to handle the spatial data inputs required by the FR and AHP models, as well as to generate risk of flood maps based on the models’ weighted outputs (Figure 1).

The FR model is a bivariate statistical analysis that can model the spatial distribution of flood events based on multiple conditioning factors. Frequency ratio model is a statistical method commonly used in flood-risk assessment to identify the link between previous flood occurrences and relevant elements (e.g. elevation, slope, rainfall). The FR model assumes that future floods will follow patterns similar to previous flood disasters (Rahmati et al. 2016; Ramesh & Iqbal 2022; Samanta et al. 2018; Tehrany et al. 2014). The FR model is useful for flood susceptibility modelling as it is simple and can handle spatial data. It assesses the possibility of flooding by comparing the presence of specific causal variables in areas with known flooding events to areas without flooding, resulting in a clear probability ratio.

For each factor, the historical flood locations were mapped, and the ratio between flood-prone and non-flood-prone areas for each factor category was calculated. A correlation analysis was performed to determine their relationship with floods occurrence. A greater bivariate probability value (more than 1) indicates a strong correlation between flood occurrences and their conditioning factors, whereas a low value (less than 1) suggests a weak correlation (Youssef et al. 2016). Based on historical data, the spatial distribution of flood events in South Jakarta is shown in Figure 2.

According to the flood inventory dataset, 30% (22 304 flood pixels) of the data was treated as the testing dataset, while the remaining 70% (74 347 flood pixels) of the dataset is treated as the training dataset. The spatial distribution of flood events was derived from the existing dataset, which records the number of neighbourhood units exposed to flooding. The dataset was randomly divided into training and testing subsets.

The flood inventory data were then evaluated against seven conditioning factors. Seven conditioning factors representing topography, hydrology and human activities were used for the FR-based flood susceptibility mapping: topography (elevation, slope gradient, slope aspect and TWI), hydrology (rainfall and distance from the river) and human activities (land-use type and zoning based on spatial plan documents), as depicted in Figure 1 (Dottori et al. 2015; Nasiri et al. 2019; Nkwunonwo et al. 2016; Papaioannou et al. 2016). The first step was to calculate the FR value for each factor on the training dataset using the tabulate area and raster calculator tools in ArcGIS. Equation 1 was used in the calculation (Rahmati et al. 2016; Siahkamari et al. 2018):

Then, FR was mathematically normalised into relative frequency (RF) with a value in the range of 0 to 1 using Equation 2 (Youssef et al. 2016). Relative frequency was used to determine the correlation between each class of a conditioning factor and flood using a predictor rate (PR). Predictor rate acted as the weight value of each factor obtained with Equation 3. Both RF and PR were calculated in Microsoft Excel:

Flood Susceptibility Index (FSI) was assessed by summing a series of multiplication results of the weight (PRi) and the RF value of each factor (RFi) for the total number of the considered factors (n = 7), as formulated in Equation 4. Afterward, FSI values were categorized into three classes of susceptibility: low, medium, and high, and then mapped in ArcGIS 10.3:

The ROC–AUC method is an effective tool for assessing the effectiveness of predictive models, especially in flood susceptibility assessments (Lin, Wu & Liang 2019). It is a frequently accepted technique for evaluating the validity of binary classification models, such as flood-risk predictions, by comparing predicted flood-prone areas to actual flood occurrences (Fawcett 2006). In this work, the ROC–AUC method was used to test the accuracy of flood susceptibility models created using the FR methodologies. The ROC–AUC approach was chosen because it provides a convenient and quantitative way to assess model performance, providing that both models are evaluated not only for technical execution but also for practical applicability in flood susceptibility mapping.

The modelling results were validated by compiling a no-flooding dataset equal in size to the testing dataset.

No-flooding data were placed randomly within non-flooded areas using Create Random Points tool in ArcGIS.

Two datasets were created using the flood history data: 70% (training data, 74 347 flood pixels) were used to build the susceptibility mapping model, and 30% (testing data, 22 304 flood pixels) were used to assess the model’s prediction accuracy. Afterward, the no-flooding and testing data were combined and classified into ten classes to facilitate the mathematical operation of the validation. The values of the classified dataset at the randomly created points were extracted using the ArcGIS tool extract values to points and then calculated in Microsoft Excel to process the model validation.

The flood susceptibility map was evaluated using the ROC–AUC method, comprising the receiver operating characteristic (ROC) curve and the area under the ROC curve. The area under the curve was used to assess the prediction accuracy and model development quality (Hong et al. 2017), and the ROC curve determined the model’s level of success in predicting flood-prone areas based on previous flood events (Siahkamari et al. 2018). The ROC curve was generated by plotting true positive rates (TPR/sensitivity), that is, location points of flood events, against false positive rates (FPR/1-specificity), that is, location points of predicted floods (Ramesh & Iqbal 2022). The qualitative relationship between AUC and prediction accuracy can be classified into five categories: poor or unsatisfactory (0.5–0.6), fair or satisfactory (0.6–0.7), good (0.7–0.8), very good (0.8–0.9) and excellent (0.9–1) (Rahmati et al. 2016).

Urban flood vulnerability encompasses physical, social, economic and environmental vulnerabilities. The four parameters explain elements at risk of damage, loss and destruction in the event of urban flooding. The parameters and variables used in the assessment are summarised in Table 2. The spatial distribution of social vulnerability was investigated with the district (administrative boundaries) as the unit of analysis. The AHP approach in the vulnerability research module was performed in the ArcGIS environment in three stages. The first stage was to classify variables for every vulnerability parameter and rank their classes. Secondly, the weight value of each variable was assessed with the AHP approach and calculation. Thirdly, the four variables were calculated as total vulnerability using the overlay analysis in ArcGIS and then classified into low, medium or high.

This integrated approach combines historical data (through the FR method) and expert judgement (via the AHP), enabling a more comprehensive analysis of urban flood risk. It is particularly suitable in contexts with limited data availability or complex planning requirements. The urban flood-risk levels were determined and mapped by combining and integrating susceptibility and vulnerability assessments (Dandapat & Panda 2017), as presented in Equation 5 (Apel et al. 2009).

The susceptibility and vulnerability maps transformed into raster data were spatially overlaid and used as the input for risk index calculation with the raster calculator tool (Abdrabo et al. 2020). Consequently, the risk assessment results contain maps of urban flood susceptibility factors and physical, social, economic and environmental vulnerability factors. Risks were classified into three levels: low, medium and high. Fuzzy membership was used to normalise the spatial data in flood susceptibility, vulnerability and risk assessments, resulting in a homogenous index range of 0–1.

This article followed all ethical standards for research without direct contact with human or animal subjects.

The quality and resolution of spatial data are vital factors in flood-risk assessment, especially when using GIS-based models such as the FR and AHP. Those variables have a substantial impact on the accuracy, reliability and application of the results (Chen et al. 2005). The quality of spatial data is determined by its accuracy, consistency and relevance to the studied area. High-quality spatial data should accurately reflect real-world conditions, with minimal errors, as flaws may influence flood hazard predictions and mislead decision-makers (Bilskie et al. 2014; Sanyal & Lu 2005). This research uses the type, resolution and data source of each variable in Table 1 to produce a modelling scale used in regional planning at a scale of 1:25 000. This scale was chosen because it is a mapping need according to the applicable regulations in Indonesia.

Frequency ratio was calculated for each class of each conditioning factor (Table 2), in which FR > 1 indicates a strong correlation between flood and a particular class, and PRs were calculated to assign weight values to these factors, as depicted in Figure. The slope gradient of 8–16° was found to have the highest RF of 1.092, indicating a higher flood frequency at the foot of slopes compared to other terrains (Tehrany & Kumar 2017). Similarly, the correlation between elevation and flood frequency showed FR > 1, indicating a higher probability of occurrence in low-lying areas (Khosravi et al. 2016).

The topographic parameters or terrain attributes are strongly related to flood events, as evident from their PRs. However, the correlation between slope aspect and floods did not significantly differ across the classes. The slightly varied FRs resulted from a homogenous slope aspect that is distributed relatively homogenously with similar area percentages throughout South Jakarta (see Table 2 and Figure 3). The slope aspect also had a low PR compared to other conditioning factors; hence, it can be concluded that it is a terrain attribute with less significant control over flood occurrences in the study area (Figure 4). The same case applies to the secondary topographic parameter, topographic wetness index (TWI). Topographic wetness index showed no substantial difference between the FR values of its classes and had the lowest PR of 1.186 among the conditioning factors because South Jakarta has a homogenous relief (Figure 4). These results indicated a direct correlation between soil moisture and flooding (Samanta et al. 2018).

The amount of rainfall (FR > 1) is shown as a major determinant of flood formation (Sarkar & Mondal 2020).

Rainfall had a PR of 12.566, making it the most significant conditioning factor of flooding (Figure 4). Therefore, extreme rainfall in Ciliwung, Pesanggrahan and Krukut watershed systems is always an indication of flooding in the city. Similarly, distance from river networks had a PR higher than 1 for all its classes, except the 100 m distance. Areas located closer to the river had a higher FR. The probability of flooding is higher with a shorter distance to the river (Kabenge et al. 2017). In addition, this factor had a high PR, or weight value, indicating a strong determinant.

Land use plays an essential role in identifying areas prone to flooding (Samanta et al. 2018). Land use identification helps assess the extent of impervious surfaces, which contribute to increased surface runoff in urban areas (Al-Zahrani, Al-Areeq & Sharif 2017). An extensive impenetrable surface accelerates runoff velocity and thus has the highest susceptibility to flooding (Ullah & Zhang 2020). Land use had the highest PR in areas associated with bodies of water, that is, aquaculture, and green belts on river banks (Figure 4). Moreover, built-up lands had high FRs, especially government offices and residential zones (FR > 1). In terms of area, residential zones are the type of land use most widely associated with floods because up to 63.84% of the total housing area had a high probability of flooding. Land use, as a conditioning factor of susceptibility, showed the second-highest PR, indicating a strong relationship between urban flooding and land use patterns (Figure 4). Built-up areas near river networks tend to be the most susceptible to flooding, mainly because they generally have a large population size and are centres of economy, services and infrastructure (Khosravi et al. 2016; Sarkar & Mondal 2020).

Prior to mapping, the FSI was normalised using an ArcGIS geoprocessing tool called fuzzy overlay to create a scale of 0–1 to match the risk and vulnerability index range. Therefore, the derived index values were transformed from 193.399–903.571 to 0.0534–0.8085. The FSI was later classified into low, medium and high susceptibility using the equal method, according to Head of National Disaster Management Agency Regulation No. 2 of 2012. Figure 3(h) indicates that South Jakarta is exposed to three classes of susceptibility: low (22.451%, 28.845 km²), medium (73.171%, 94.101 km²) and high (54.377%, 5.624 km²). Areas highly susceptible to flooding were characterised by high rainfall, settlements or other types of buildings close to rivers and flat to gently sloping terrain. The map also shows that these areas were allocated for residential purposes, government offices and economic facilities and services and situated along the river channels (see Figure 3). Meanwhile, low susceptibility was primarily attributed to elevation, that is, more than 40 m.

The developed FR model was tested for its quality and accuracy of predictions using the ROC curve (Samanta et al. 2018; Sarkar & Mondal 2020; Ullah & Zhang 2020). The ROC curve plots (Figure 4) false positive rates on the x-axis against true positive rates on the y-axis according to the FR analysis. The prediction’s accuracy was assessed using the area under the ROC curve (AUC). The model cannot be used to predict and map flood susceptibility if the AUC is less than or equal to 0.5, but the predictions are accepted if the AUC is higher than 0.75 (Saha et al. 2021; Sarkar & Mondal 2020), as shown in Figure 4. The flood inventory (historical data) was divided into: 70% (training data, 74 347 flood pixels) to build the model for susceptibility mapping and 30% (testing data, 22 304 flood pixels) to test for the accuracy of the model’s prediction. The accuracy test produced an ROC curve with a success rate of 0.798 (Figure 4), indicating that the FR-based flood susceptibility map for South Jakarta has good predictive performance and can be applied to estimate flood trends as well as to support further analyses such as flood-risk assessment.

The total vulnerability index was calculated from four weighted parameters: social, physical, economic and environmental vulnerability. Each parameter consisted of variables that were ranked based on expert-defined criteria and analytical considerations. Social and environmental vulnerability variables were ranked according to the Head of National Disaster Management Agency Regulation No. 2 of 2012, which explains the social conditions of people exposed to disaster impacts and the environmental characteristics of an area prone to being severely affected. For the physical vulnerability, the variables were ranked by considering the type and level of physical infrastructure in urban areas that would suffer losses in the event of a flood. As for the variables of economic vulnerability, the ranks were determined by assessing financial losses indirectly from potentially affected urban economic activities (i.e. land-use type) based on gross domestic product (GDP) in current prices in South Jakarta (Table 3).

Analytic hierarchy process assigned a relative weight value to each parameter and variable. The social vulnerability was given the largest weight, 40%, because it represents the most variable exposed to flooding, population characteristics (Bigi et al. 2021). Meanwhile, the physical and economic vulnerability was 25% each, and the environmental vulnerability was 10%. Land-use zoning was used to capture the significance of urban economic activities and measure potential financial losses.

The variable overlay analysis with fuzzy membership resulted in a total vulnerability index ranging from 0.182 to 0.670. The distribution of total vulnerability levels is mapped in Figure 4. Overall, most of the South Jakarta area had medium (47% of the study area) and low vulnerability (53%). The total vulnerability showed a similar spatial pattern to the physical and social vulnerability. The medium vulnerability was observed in areas where the distributions of the number of houses and population densities overlapped, particularly in the districts of Pesanggrahan, Kebayoran Lama, Cilandak, Pasar Minggu and Jagakarsa. These results correspond to existing land-use conditions in South Jakarta, where the majority (61%) of the area is used for residential zones.

The AHP is a decision-making process that prioritises factors based on expert assessment. In flood-risk assessment, AHP assists in determining the relative relevance of various flood-related factors. Analytic hierarchy process helps the methodical integration of expert knowledge, which is especially useful when measurable data are limited or several aspects of different importance must be evaluated. The AHP is applied to minimise subjectivity to decide the weighting factors of each variable employed in the analysis. It is commonly employed in multi-criteria decision analysis (MCDA) for spatial planning, particularly when subjective judgements are required to assign relative weights to factors.

The risk modelling used two components: susceptibility based on the FR model and vulnerability assessed with AHP. Both quantitative methods included adjustments to the risk index assessment requirements described in Head of National Disaster Management Agency Regulation No. 2 of 2012, including an index scale of 0 to 1 (Van Westen et al. 2011). Results showed that the risk index values ranged from 0.144 to 0.690 and were classified into low, medium and high. The South Jakarta area mainly had medium flood risk, covering 108.64 km² or 74.55% of the total area, while the rest was at low (37.08 km² or 25.45%) and high risk (0.004 km² or 0.003%).

In this study, flood-risk modelling was carried out at a scale of 1:25 000 to align with the South Jakarta Spatial Plan (Figure 5). By comparison, the Provincial Government of Jakarta currently produces flood-risk maps at a coarser scale of 1:100 000. Therefore, the modelling developed in the research can be used to evaluate spatial planning and revise regional spatial plans to incorporate disaster aspects in the future. The current challenge, then, is to incorporate this detailed information into the spatial plan. As the previous study suggests to incorporate risk information into spatial plan, there are four procedures, such as strategic integration, substantive integration, technical integration and procedure integration (Setiawan 2008).

According to previous studies, risk information at a 1:25 000 scale allows for substantive integration into a detailed spatial plan. The strategic integration of spatial planning with risk information is mandated by Indonesia’s National Spatial Planning Law No. 26/2007, which requires local governments (urban or city level) to produce risk information at scales of 1:25 000, 1:10 000 and 1:5000. The national law also mentioned that the risk area should be categorised as the spatial pattern of conservation area for risk. Two key components of a detailed spatial plan are the spatial structure and spatial pattern. Incorporating susceptibility, vulnerability and risk information into these components can help prevent further damage and losses in areas exposed to flooding. Within the spatial pattern, such areas can be designated as conservation zones for risk mitigation. However, if nearly 70% of the area is classified as medium risk, an important question arises: how should development proceed under these conditions?

This study shows that high-resolution spatial data can provide useful information to urban planners and decision-makers. Incorporating flood-risk assessments into urban development initiatives will improve disaster risk reduction and strengthen communities in South Jakarta. The spatial plan is a guideline to develop the area. Development permits submitted to the local government are granted only if the application complies with the spatial plan guidelines. Therefore, any proposed development in flood-prone areas must adhere to mitigation regulations. These findings highlight the need for policymakers to prioritise flood risk in urban planning to minimise the socioeconomic impacts of flooding. To effectively reduce flood risk in South Jakarta, practical strategies based on risk mapping must be implemented. Strategies such as improving drainage systems, supporting green infrastructure (e.g. parks, green roofs), flood insurance system (municipality, household, individual, etc.) and encouraging community participation in flood preparedness activities can all help to reduce flood damage. Local governments and stakeholders play an important role in implementing these initiatives, ensuring they are integrated into current urban planning frameworks.

However, it is critical to recognise the limitations of the approach used in this research. While using high-resolution spatial data improves the accuracy of flood-risk assessments, factors such as potential biases in data sources, assumptions made in the FR and AHP models, and spatial data resolution may limit the applicability of these results to other areas of Jakarta or to different contexts within Indonesia. This caution is particularly important in urban environments, where local conditions can significantly influence flood behaviour and risk. Future research should address these limitations by incorporating ground-truth data and examining the heterogeneity of flood hazards across different urban contexts. Additionally, frequent updates to the data – both spatial and temporal – as well as the addition of capacity data could enhance the thoroughness of risk assessments. Building on this study, the FR–AHP framework could be further refined by integrating hydrodynamic modelling and higher-resolution spatial data to improve predictive accuracy and support more adaptive urban planning.