Using a scenario-neutral approach to assess the impacts of climate change on flooding in the Ba River Basin, Viet Nam

Flood risk management in the Ba River Basin relies on a flood warning system based on the probability of flood peak occurrence. Level 1 flood warning (mildest) corresponds to a 50% flood frequency. Level 2 flood warning (medium) corresponds to a flood frequency between 55 and 25%. The highest flood warning level (level 3) corresponds to a flood frequency between 30 and more than 10% (MONRE 2021).

Climate change poses a deep uncertainty event and can have altering consequences on both the likelihood and consequences of flooding (Hallegatte et al. 2021). Therefore, the use of flood statistics in flood risk management is under question by the research community due to the hydrologic non-stationarity and uncertainty related to the probability assignment of flood peaks (Brown et al. 2020; DiFrancesco et al. 2020; Culley et al. 2021).

Thus, the research into the impacts of climate change on flooding should use a different approach. The literature on alternative approaches is vast with examples. Most of the approaches seek to evaluate the performance of a decision or strategy by determining the system's response under various conditions. This includes scenario-neutral approaches (Guo et al. 2017; Broderick et al. 2019), robust decision making (Lempert et al. 2013; McPhail et al. 2018), info-gap decision theory (Roach et al. 2015; Ben-Haim 2019), decision scaling (Brown et al. 2012; Poff et al. 2015; Tra et al. 2018), and robust optimization (Ray et al. 2013; Roach et al. 2015). Alternatively, other approaches seek to incorporate flexibility analysis within the planning processes. This includes dynamic adaptive planning (Kwadijk et al. 2010), adaptive policy pathways (Kwakkel et al. 2015), and engineering options analysis (Marchau et al. 2019).

One key component in the new approaches is the use of stress-tests. Stress-test involves determining the response of a system through identifying performance indicators and critical tipping points of these indicators (Broderick et al. 2019; DiFrancesco et al. 2020; Culley et al. 2021). For example, Culley et al. (2021) used a flood reliability criterion to assess the impacts of climate change in Lake Como, Italy. The flood reliability criterion is calculated as the fraction of days the reservoir height is below the flood threshold (1.24 m above minimum storage level). Likewise, Broderick et al. (2019) used a 20% flood peak increase allowance in project appraisal guidance issued by the UK Government's Department for Environment, Food and Rural Affairs. This guidance requires all flood management to include a sensitivity analysis of a 20% increase in flood peak. The 20% allowance is kept constant throughout England and Wales and makes no allowance for regional variation in climate change or catchment type. On the other hand, DiFrancesco et al. (2020) developed a bottom-up approach to assess the performance of flood management systems in the American River under climate uncertainty and non-stationarity. The expected annual damages ($/year) were plotted against different flood regimes to obtain a climate response space. A threshold value of expected annual damages of $38 million was obtained.

While these studies have identified useful performance indicators and threshold values for decision, these results could not easily be applied to the Ba River Basin. Firstly, there are 108 reservoirs, 29 hydropower dams, and 238 pumping stations in the Ba River Basin (MONRE 2022). The complex network of reservoirs and waterworks infrastructure in the Ba River Basin makes using a single reservoir performance indicator insufficient. Secondly, there is limited flood damage data from the river basin. This hinders the determination of a performance indicator based on damage value. Flood damage data in the country are often reported based on administrative units and do not necessarily overlap the natural river basin. In addition, flood damage includes non-monetary units (area of agriculture affected, volumetric landslide, extent of road damage, etc.) with limited insurance loss data to be used directly. Thirdly, using a ‘one size fits all’ measure of flood peak may invalidate the uncertainty analysis. This approach omits the uniqueness of the river basins' physical characteristics that can influence sensitivity to climate change.

This study, thus, aims to develop a performance indicator and a critical threshold and assesses the impacts of climate change on flooding in the Ba River Basin using a scenario-neutral approach. A performance indicator is a measurement to quantify system sensitivity to climate change. Using a performance indicator helps improve understanding of climate change's impact on flooding in the Ba River Basin through a quantifiable metric. The performance indicator allows the identification of a critical threshold that affects the vulnerability of the system. In this case, the critical threshold is the level at which the flood impact is beyond acceptable. Multiple Delphi questionnaire rounds and model simulations are used to obtain the performance indicator and the critical threshold. The scenario-neutral component in this study is derived from the departure from conventional scenario-led studies. In particular, instead of focusing on projected changes in climate in the Ba River Basin and using scenarios with historical flood frequency analysis, the study utilizes sensitivity analyses to a range of climate conditions. Hydrological and hydraulic models are thus used to determine the response of the system rather than tools to simulate projections.

The study uses evaporation, rainfall, river discharge, water level, river cross-section, topography, land use, and administration map data. Details on the sources and resolution are included in Table 1. Data are used to run hydrological and hydraulic simulations in the MIKE FLOOD model package. The time series of hydrological and meteorological data ranges from hourly to daily.

Data availability in the study region is limited. Firstly, there is a lack of up-to-date river cross-sections. Only data from 2016 are available, with 62 measured cross-sections. These sections are located downstream of the Ba River; therefore, only hydraulic modeling of these sections is possible. Additionally, the calibration and validation of the hydraulic model would thus be performed using cross-sections that may only partially reflect the current river channel characteristics. Secondly, there is a limited number of hydrological and meteorological monitoring stations. For example, only four flow monitoring stations exist for hydrological model calibration and validation. In addition, the sparse distribution of rainfall gauges affects the calibration and validation of hydrological parameters. Thirdly, only two water level gauges could be utilized for the study. The use of satellite data for improved river cross-section and rainfall data has been conducted with a certain degree of success (Lazri et al. 2020; Jiang et al. 2021; Lamine et al. 2021). However, this is beyond the scope of this study.

In the first step, experts are prescreened, and a list of potential experts is drawn up. Potential experts include people with experience in the field and the area, including local and national policymakers and decision-makers. At the national level, this includes experts working under the Ministry of Natural Resources and Environment, the Ministry of Agriculture and Rural Development, and the Viet Nam Academy of Science and Technology. At the local level, this includes experts working under the Department of Natural Resources and Environment, Department of Agriculture and Rural Development. Experts from research and education institutions were also considered. This includes experts from the University of Natural Resources and Environment, Thuyloi University, and the Mien Trung University of Engineering. Other retired and non-affiliated experts were also considered in the prescreening phase. Potential experts are contacted, and only those who could commit time to several rounds of questionnaires are selected.

In the second step, two questionnaires were sent out to the experts. For the first questionnaire, a list of all recorded historical flooding events was sent to the panel of experts. The experts were then given the option to select a representative flood event in the Ba River Basin by scoring these flood events on a scale of 1–5 (with 1 being the least likely and 5 being the most likely). After receiving the responses and scores of question items from the experts, an analysis was performed to determine whether the KAMET rules had been met. If the KAMET rules were met (Table 2), no further rounds of questionnaires would be required. However, if the KAMET rules are not satisfied, additional rounds of questions are involved. In the second questionnaire, the experts were given the option to select a threshold flood event, i.e., the flood event that is deemed most severe. The KAMET rules were then reapplied to determine whether additional rounds of questionnaires were required. It should be noted that the process requires a time commitment from the participants. Therefore, some stakeholders and experts who were initially involved in the questionnaire but could not commit time in later interview rounds were excluded from the analysis.

MIKE NAM is a lumped, conceptual rainfall-runoff model. The model simulates runoff from rainfall through vertically occurring processes. In particular, the model divides three layers: above ground, unsaturated zone, and saturated zone. MIKE NAM model includes 13 model parameters (Table 3).

The development of the MIKE NAM model for the study includes the calibration and validation of L_max, U_max, CQOF, CQIF, and CK1,2. MIKE NAM model calibration was performed using hourly data for October 1993, and validation was performed using hourly data for November 2009. Initial conditions in the model include initial water content in the surface and root zone storages, initial overland flow, interflow, and baseflow.

The MIKE 11 model development includes calibrating and validating Manning's n roughness coefficient at each river cross-section. The calibration process was performed using hourly data in October 1993. The validation process was performed using hourly data in November 2009. The upstream boundary conditions for the MIKE 11 model include river discharge at the Cung Son monitoring station and upstream using inputs from the MIKE NAM flow results. The downstream boundary conditions for the MIKE 11 model include water levels at the estuaries. The initial condition in the MIKE 11 is set at flow and water level on the onset of flooding. The simulation timestep used in the MIKE 11 model is 5 s.

The MIKE 21 model development includes calibration and validation of Manning's M roughness coefficient (inverse of Manning's n) at each grid cell. The calibration process was performed using hourly data in October 1993, while calibration was performed using hourly data in November 2009. The simulation of the MIKE 21 model used a closed boundary condition. The initial condition used in the MIKE 21 model is the initial water levels and initial flow velocity in each grid cell, which are set at 0. The simulation timestep used in the MIKE 21 model is 5 s with a Courant number of 0.8.

Three different accuracy measurements for model calibration and validation processes are used, namely the Nash–Sutcliffe efficiency (NSE), the water balance error (%WBL), and the mean absolute error (MAE). The NSE is used for MIKE NAM and MIKE 11, %WBL is used for MIKE NAM, and MAE is used for MIKE 11 and MIKE 21. The NSE measures the variance of the model errors as a fraction of the variance of the observations (Nash & Sutcliffe 1970; Broekhuizen et al. 2019). An NSE value of 1 indicates perfect performance, i.e., simulated results fit perfectly to observed data. On the other hand, the water balance error (%WBL) measures the difference in percentage of the average observed data versus the simulated results (Teshome et al. 2020; Pareta 2023). The MAE measures the mean absolute difference between the simulated and observed flood peaks (Kara et al. 2016; Abdul Kadir 2019). For the model to be valid, an NSE value of greater than 0.8, %WBL less than 10%, and MAE less than 0.2 m is required (Bessar et al. 2020; Pareta 2023).

To simulate climate change conditions in the Ba River Basin, two potential approaches could be used, namely stochastic weather generators and parametrically varying climate conditions. Weather generators allow a larger climate space to be evaluated and have been used successfully in previous studies (Spence & Brown 2016; Guo et al. 2017). On the other hand, parametrically varying climate conditions is a more straightforward method and could be used without weather generators.

This study opted for parametrically varying climate conditions, given that the climate projections for the study area are within a limited range of climate conditions. This implies that using more computing intense weather generators would not be necessary. The range of climate projections is illustrated utilizing the climate future concept (CSIRO and Bureau of Meteorology 2021), as shown in Tables 4–6. The numbers in the cells represent the number of climate projections within a range of climate conditions. These projections are analyzed using the CMIP5 and CMIP6 with 105 climate projection outputs. The inclusion of both the CMIP5 and CMIP6 is to explore a larger climate future given that both CMIPs have been well compared and although CMIP6 outputs showed improved skills, CMIP5 outputs are generally considered acceptable (Chen et al. 2021; Kamruzzaman et al. 2021; Carvalho et al. 2022). The increase in rainfall and temperature is relative to the Intergovernmental Panel on Climate Change's (IPCC) baseline period of 1986–2005.

Table 4

Climate future for the 2021–2040 period

.

.

View Large

Table 5

Climate future for the 2041–2060 period

.

.

View Large

Table 6

Climate future for the 2081–2100 period

.

.

View Large

Climate conditions are parametrically varied for hydrological and hydraulics simulations in the MIKE model package. This includes simulating changes in temperature and rainfall. A change from 0.5 to 2.5 °C in 0.5 °C increase increments is used for temperature. A change from −20 to +20% rainfall in 10% increments is used for rainfall. The selection of the temperature and rainfall increments is to accommodate the 1, 1.5, and 2 °C warming, and the 10–20% heavier rain bursts benchmarks set by the IPCC and the increments used by previous studies (CSIRO and Bureau of Meteorology 2021; IPCC 2022, 2023). This is illustrated in Table 7. KB_n denotes the combination of change in rainfall and change in temperature. The combination of scenarios is then used to drive the simulation in the MIKE model package in the form of input rainfalls and input temperatures. Other climate variables are not considered in the study.

Table 7

Combination of climate conditions used during simulation

.

.

View Large

Two outputs are obtained using the Delphi questionnaire, namely, a representative flood event and a threshold flood event. The flood event in November 2009 was selected as a representative flood event, while the flood event in December 1993 was selected as the critical threshold for flood impacts.

The selection of 2009 and 1993 as representative and critical thresholds is reasonable. In particular, the 1993 flood corresponds to a 3% flood frequency and is the most severe flood event ever recorded in the Ba River Basin, reflecting the uppermost threshold for flood impacts. The 2009 flood event corresponds to a 10% flood frequency and is considered a severe flood event. The selection of 2009 is also interesting, given the reported damage of the flood event (Tuan et al. 2014). Furthermore, 2009 is a milestone in water resources systems in the Ba River Basin. The flood event in 2009 was one of the large floods that occurred before the development of the current reservoir system. After 2009, major river basin reservoirs were completed and operationalized. This includes large reservoirs such as Ka Nak, Krong H'Nang, and Song Ba Ha (see Table 8). However, one of the critical components of these reservoirs is the lack of flood control, which has minimal impact on flooding downstream (compared to the condition where no reservoir exists) (Tuan et al. 2014).

The calibration and validation of the MIKE NAM rainfall-runoff model and the MIKE FLOOD model package were deemed satisfactory (Table 9). For the MIKE NAM model, the NSE for the calibration process is 0.89, and the validation process is 0.92. In addition, the %WBL for the calibration process is 4.9%, and for the validation process is 4.1%. The NSE for the MIKE 11 model is 0.93 and 0.97 for the calibration and validation process, respectively. The MAE in the MIKE 11 model for the flood peak is 0.07 m for the calibration process and 0.03 m for the validation process. For the MIKE 21 model, the MAE is 0.17 m for the calibration process and 0.025 m for the validation process.

The spatial distribution of flooding downstream of the Ba River Basin for KB1, KB6, KB11, and KB16 representing changes in rainfall of −20, −10, +10, and +20% under a 0.5 °C warmer conditions are shown in Figure 5. As rainfall increases, so does both the flood depth and flood area. Under KB1 and KB6, the flooded area is limited to areas close to the mainstream of the Ba River. However, as rainfall increased from 10 to 20%, the flooded area extended northwards and southwards of the river.

Figure 6 depicts the flood curve under various scenarios compared to the representative flood and flood warning levels 2 and 3. Under all scenarios simulated, the water level exceeds flood warning level 3. However, the peak flood levels simulated under KB16, KB17, KB18, KB19, and KB20 (i.e., 20% rainfall increase condition) are both the highest and earliest. The results also showed that all simulations of flood peaks under the same change in rainfall scenario tend to follow a similar pattern. In the 20% increased rainfall scenario, all four simulations obtained the same flood peak curve. This implies that rainfall is the primary factor affecting floods.

Similarly, Figure 7 depicts the duration of the flood peak above flood warning level 2 and 3. As can be seen, there is an increase in flood peak duration in KB11, KB12, KB13, KB14, KB15, KB16, KB17, KB18, KB19, and KB20. These scenarios simulated increased rainfall conditions in the Ba River Basin. As rainfall increases up to 10%, there is a significant increase from 48 to 64 h of flood duration above warning level 2 and from 39 to 43 h above warning level 3. There is little contrast between the increase in flood peak maintenance duration in the 10% rainfall increase and 20% rainfall increase scenarios.

Figure 8 shows the flooding threshold and flooded area under various scenarios. Flooded areas in all scenarios remained under the threshold except for KB16, KB17, KB18, KB19, and KB20. This corresponds to scenarios where rainfall increases by 20% compared to the baseline period. On the other hand, an increase of rainfall up to 10% increases the flooded area just below the flooded threshold (KB11, KB12, KB13, KB14, KB15). The increase in temperature has little impact on the change in flooded areas. The flooded areas are equally the same throughout the scenarios that simulate the same change in rainfall. This is expected since flooding is heavily influenced by rainfall.

From the results, a few implications could be formed. Firstly, rainfall change is the primary factor affecting flood impact in the Ba River Basin. This is expected since temperature has little impact during a comparatively short flood event. However, as a long-term process, the increase in temperature creates more severe and extreme rainfall events. This in turn increases the impact of flooding. Secondly, the flood peak duration in the Ba River Basin is highly sensitive to an increase in rainfall by up to 10%. As rainfall increases up to 10%, flood peak duration increases significantly and then plateaus off with minimal change as further rainfall increases. Thirdly, the flooded area in the Ba River Basin remains beyond the flooded threshold as rainfall increases up to 10%. However, as rainfall increases up to 20%, the flooded threshold area is exceeded. Fourthly, the flood extent and depth increase as rainfall increases. While a decrease in rainfall will only likely see flooding in areas close to the mainstream of Ba River, as rainfall increases, the flooded area is further expanded northwards and southwards.

The results and the implications obtained are useful for flood management in the Ba River Basin. Development in the river basin could benefit from better planning with due consideration to the flooding threshold and the spatial distribution of floods under climate change. A 20% increase in rainfall will likely severely impact flooding in the river basin, and planning of significant development should consider mitigation measures of this threshold. Additionally, flooding under climate change in the Ba River Basin under extreme cases can extend beyond areas close to the mainstream. Therefore, development planning and response activities must consider these vulnerable areas.

The study developed a MIKE FLOOD model package for the downstream of the Ba River Basin. This includes the development of a MIKE NAM rainfall-runoff model and the MIKE 11 and MIKE 21 hydraulics models. Calibration and validation of the model package were deemed satisfactory using the NSE, %WBL, and MAE measures. Simulations were carried out to simulate varying conditions at the Ba River Basin, including a representative flood event, a threshold event, and 20 combinations of changes in rainfall and temperature to simulate climate change conditions.

A Delphi questionnaire method with a set of KAMET rules was used to obtain a representative flood event and a threshold flood event. Several rounds of interviews were carried out, and the panel of experts selected the representative flood event to be 2009, the event before the completion of water infrastructure in the river basin. The threshold flood event of 1993 was selected as this was the most severe flood historically recorded in the river basin.

The results obtained from the study have several implications. Firstly, rainfall change is the primary factor affecting flood impact in the Ba River Basin. Secondly, the flood peak duration in the Ba River Basin is highly sensitive to an increase of rainfall up to 10%. As rainfall increases up to 10%, flood peak duration increases significantly. Thirdly, the flooded area in the Ba River Basin remains beyond the flooded threshold as rainfall increases up to 10%. However, the flooded threshold area is reached as rainfall increases up to 20%. Fourthly, the flood extent and depth increase as rainfall increases.

The study, however, also entails limitations. Firstly, the limited data availability affects the results of the study. Due to the lack of cross-sections in the river basin, only the downstream reach of the basin is simulated. Limited rainfall gauge data and river geometry data affect the development of hydrological and hydraulic models. Without additional data to compare, little inference could be made between the data used in the study and additional data sources such as satellite data. Secondly, the downstream reach of the river basin simulated using the MIKE FLOOD package might not be optimal. Using a MIKE URBAN to simulate floods in urban areas would provide better results. Additional urban flood simulation models would also be able to offer more insight as well as a more exact flooded area in the river basin. With the lack of infrastructure and elevation data, the MIKE URBAN model could not be developed. Thirdly, flood is simulated with limited inclusion of the reservoir operational rule. The reservoir operating rule is incorporated into the river discharge and water level in the gauging stations. Fourthly, only a limited number of simulations were conducted in the study. This includes simulating the representative flood event, the threshold, and 20 additional conditions of varying rainfall and temperature. This explores only a limited climate space that could impact the Ba River Basin. Nonetheless, the climate space explored provided useful information on the plausible climate change future in the Ba River Basin.

Future research to address the limitations of the study could potentially explore improving the input data, using additional hydraulic modeling tools, and the use of stochastic weather generators. Satellite rainfall data and satellite digital elevation data could potentially be used to improve rainfall input data and river cross-sections. Improved input data could then be used to develop a more sophisticated urban flooding area using the MIKE URBAN model package. This would allow flood simulation in urban areas that are currently not explored. Using a more comprehensive range of plausible climate futures through stochastic weather generators would allow a greater climate space to be evaluated, hence, more information on flood impacts.

This research is funded by the Viet Nam Ministry of Science and Technology under the independent research project: ‘Establishing a methodology based on the combined top-down and bottom-up approach to assess water resources risks as a result of changes in the hydrological regime under global and regional changes’, code number DTDL.CN-60/21. The authors wish to acknowledge the financial assistance received. We further acknowledge the World Climate Research Program's Working Group on Coupled Modelling, which is responsible for CMIP5 and CMIP6 data, and we thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP and ESGF. Finally, we are deeply grateful to the anonymous reviewers for their contribution to improving the manuscript to its current form.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.