Numerical simulation and validation of heavy rainfall flood inundation in small watersheds in undocumented mountainous areas

With global warming, the north-western region of China is experiencing a process of change from warm-dry to warm-wet, and this transition is most pronounced in areas such as the Qilian Mountains in the northern part of the Tibetan Plateau (Shi et al. 2003). The potential threat of this change to the safety of human life and property has become increasingly concerning (Peters 1990; Daniel et al. 2007; Manzanedo & Manning 2020). The Qilian Mountains are an important ecological barrier and watershed in the north-western region of China, and play an important role in safeguarding the ecological security of China. In the last 57 years, temperature and precipitation in the Qilian Mountains have increased significantly with a trend of 0.35 °C/decade and 14.7 mm/decade, respectively (Jing et al. 2022). Several phenomena such as glacier melting, permafrost thawing, and increased runoff can be attributed to the increase in extreme rainfall events. The Qilian Shan region contains a large number of mountainous sub-watersheds, which greatly increase the chances of flash floods caused by extreme rainfall in the region (Ye et al. 2022).

Flash flooding is one of the most catastrophic natural disasters. It may originate in a small watershed in mountainous regions due to rainfall and sudden fluctuations in surface runoff. It is characterised by fierce, fast, and destructive runoff, thus making prediction and prevention difficult (Follum et al. 2017; Song et al. 2023). This causes huge negative socio-economic impacts and loss of tens of thousands of lives each year globally (Grimaldi et al. 2019). Flash floods and mudslides were triggered by heavy rainfall in the Shadai River basin on 18 August 2022, which caused numerous casualties. Considering its surrounding environment, disaster distribution characteristics, and the availability of data for research, this area can serve as a typical representative watershed for flash flood analyses in the Qilian Mountains.

Research on flash flood disasters has mainly focused on their spatial and temporal distribution characteristics, cause analysis, inundation simulation, forecasting and early warning, risk management, assessment of losses, emergency response, and prevention and control measures (Beillouin et al. 2021; Peng et al. 2023). With global warming causing the melting of snow and ice and thereby increasing weather extremes, damage-costs from future flood events in high-altitude mountainous regions is expected to increase exponentially. It is critical to accurately predict the hydrological lines of flooding and the areas of inundation in order to fundamentally mitigate damage caused by such floods and increase flood resilience. Areas prone to flash floods are mostly remote mountainous and upland areas with harsh weather, complex landscapes, and economic underdevelopment, where it is very difficult to build and maintain hydrological stations or detect runoff, and it is impossible to obtain hydrological flow data (Huo et al. 2021). These areas are broadly referred to as undocumented areas, especially in areas such as the Shadai River basin. This poses a significant challenge for flood modelling and the subsequent development of mitigation measures. The current proliferation of remotely sensed observations and open-access high-resolution hydrological datasets can circumvent limitations in data availability. Therefore, current flood modelling based on geographic information system (GIS) is the mainstream direction of flood modelling. Contrarily, hydrological–hydrodynamic modelling shows great promise in the field of water resources, where complex hydrological phenomena and processes can be transformed into simple scientific models, greatly improving the efficiency and accuracy of hydrological calculations; therefore, this is widely used in watersheds where hydrological data is scarce. Currently, studies on the dynamic simulation of flash flood inundations often use a single hydrological or hydrodynamic model. Distributed hydrological models such as the VIC (Nanda et al. 2019), SWAT (Maru et al. 2023), TOPMODEL (Vincendon et al. 2010), HEC-HMS (Wang et al. 2018), and Hebei rainwater models (Li & Zhang 2017) were widely used to simulate flash floods in small watersheds. Some studies (Sun et al. 2018) have used several synthetic satellite precipitation products from the Huai River basin as input data for the VIC model, providing a strong support for water management in undocumented areas. With the continuous development of hydrological simulation to mitigate the impacts of floods, researchers have developed hydrodynamic models to simulate the flood time with continued improvement of performance (Teng et al. 2018; Bates 2022). A large number of hydrodynamic models were gradually used in the simulation and analysis and research of flash flood processes, such as FloodMap-HydroInundation2D (Gao et al. 2023), HEC-RAS (Iuliia et al. 2019), LISFLOOD (Komi et al. 2017; Li et al. 2021), and other models, which are mainly used by solving complex differential equations on high-resolution numerical grids to simulate flooding (Razavi et al. 2012). A single hydrological or hydrodynamic model has certain limitations, whether it is in regard to the calculation of the flood flow or the change in water level; it is difficult to reproduce the entire process of flood occurrence and development in a refined way; only by coupling the two models with each other can we compensate for the shortcomings of the two models, which is also the development trend of storm flood simulation (Hao et al. 2023). Most case studies of urban flood modelling in China focused only on one-dimensional validation by comparing the results of simulations with those of the corresponding stage records (Yin et al. 2015), predicting that two-dimensional inundation information is not always validated mainly due to the lack of available validation data. Nandi generated high-resolution flow simulations as well as accurate flood inundation maps by coupling the VIC hydrological model with the LISFLOOD 2D hydrodynamic model in the upper Krishna basin (Nandi & Reddy 2022), and used it to validate the simulated flood extents with Sentinel-1 SAR data, which were in good agreement. Some people (Li et al. 2019) used TOPMODEL coupled with the MIKE FLOOD model to simulate the maximum water depth and duration of a flood event, and validated the results of the simulation model by comparing the calculated and field-measured maximum floodplain inundation, and obtained detailed flash flood risk maps in undocumented areas.

MIKE (DHI MIKE) is a suite of software product developed by the Danish Institute for Water Resources and the Water Environment (DHI) (Patro et al. 2009) as a specialised engineering package that directly integrates high-resolution digital terrain elevation data with a range of numerical models, including 1D MIKE11, 2D MIKE21, and other models (Taylan & Damçayırı 2023). The 1D MIKE11 model can be used to predict the capacity of water environments in lakes, rivers, and estuaries (Petroselli et al. 2020; Wu et al. 2020), as well as for inundation risk assessment (Luan et al. 2016). The hydrodynamics module in the 2D MIKE21 model is the most fundamental module of the model, which can be used to simulate flood currents, sediment, and river water quality (Sun et al. 2024) in oceans, estuaries, lakes (Jiang & Peng 2023), and mountainous areas (Wang et al. 2019). Additionally, 2D MIKE21 was used to simulate downstream river flows and extreme flows to account for the internal relationships between simulated flows, measured water levels, and tidal ranges (Song et al. 2024), which effectively modelled the extent and depth of flash flood inundation (Li et al. 2019). The 2D MIKE21 model can simulate changes in water level and flow due to various forces, and can also simulate any 2D free surface where stratification is ignored (Wang et al. 2020). The coupling of the 1D and 2D models provides a great advantage for real-time simulation of flooding events.

Therefore, the objectives of this study were as follows: (1) To generate river network files in the river channel as well as the one-dimensional flow through the 2D MIKE21 model, coupling the hydrodynamic model to reproduce the dynamic process of flash flooding, and reveal the spatial and temporal characteristics of inundation. (2) The measured flood scar data were used to project the flood inundation area in the affected regions, and validate the results of the model simulation by combining post-flood high-resolution remote sensing images.

The Shadai River basin is located in Datong county, Xining city, and north-eastern Hung River valley of Qinghai province. Qilian Mountains encompass the northern and southern parts of the Tibetan Plateau, and also the Loess Plateau transition zones. The basin terrain comprises the northwest high and southeast low; the river is about 35 km long, the basin area is approximately 124 km², the ditch elevation ranges from 2,550 to 3,970 m. The river's average specific fall of 44.1% is from the Heilin River, which is a first-class tributary of the Beichuan River (Yellow River, second level) Tributary Grade II, originating from Bahahu, Shadai village, Qingshan township, and Datong county. The region has an inland plateau continental climate, and according to the 1981–2010 meteorological element statistics in the Thirty-Year Compilation of Surface Climatic Data of Qinghai Province (1981–2010), the annual average temperature of the region is 3 °C, and the extreme minimum and maximum temperatures are −33 and 28.8 °C, respectively. The extreme maximum temperature is 28.8 °C, and the annual average relative humidity is 65%. The annual sunshine hours are 2,555.4 h, the average evaporation for many years is 1,158.1 mm, the average annual wind speed is 1.6 m/s, and the annual precipitation is 525.2 mm. The precipitation is mainly concentrated in June–August. The soil types were mainly alpine meadow, mountain brown-brown, black calcium, and chestnut calcium soils. The vegetation distributed in the region comprises mainly trees and shrubs, and the herbaceous plants include pearl tooth polygonum, tussock grass, and oriental strawberries.

The basic data include digital elevation models (DEMs), land cover data, soil texture, meteorological data, high-resolution remote sensing imagery, and unmanned aerial vehicle imagery of flood-affected areas. This study used DEM data generated from GF-7 imagery with a resolution of 2 m. The land cover data were obtained from GF-7 imagery with a resolution of 2 m. Soil texture data were sourced from the National Science and Technology Resource Sharing Service Platform (http://soil.geodata.cn), which was divided into two parts. Data from 11 meteorological stations around the Shadai River basin were collected. These included 3-hourly precipitation, maximum, and minimum temperatures; wind speed was acquired from the National Oceanic and Atmospheric Administration (https://www.ncei.noaa.gov/) website, and hourly rainfall data as monitored by the rainfall station in Shadai village was obtained. Validation data from remote sensing imagery observed by GF-7 after the 11 September 2022 disaster, with a resolution of 0.8 m, and from drone imagery of the flood-affected areas, with a resolution of 0.046 m, were obtained and are listed in Table 1.

As flash floods occur in mountainous areas with complex subsurface conditions, hydrological processes, such as precipitation, evaporation, and infiltration, as well as hydrodynamic processes, such as surface runoff, must be considered simultaneously (Gai et al. 2021). Therefore, in this study, the MIKE21 model was selected to simulate the inundation extent of floods; the hydrodynamic module MIKE21 FM in the MIKE21 model is the core basic module of the model, and the Green-Ampt model was coupled with this to calculate the net rainfall. The estimation of evaporation was based on the empirical sinusoidal equation (Calder et al. 1983). The surface flooding process was modelled using the Saint-Venant system of equations (Zhu et al. 2018), taking into consideration the influence of water-blocking buildings and hydraulic engineering on flood evolution. This model has a simple structure, relatively few parameters, high computational efficiency, and strong applicability and was successfully applied to floodplain simulations in many areas (Wei et al. 2013; Banasiak 2019; Ren et al. 2019). Additionally, this model can utilise high-precision DEM data, land use, soil infiltration, and other subsurface information to achieve a refined simulation of flooding in undocumented mountainous areas.

The MIKE21 model was selected in this study to simulate the extent of flood inundation in the Shadai River basin in Datong county using the MIKE ZERO software platform. The initial input conditions of the model included the grid division of the study area and elevation values corresponding to each grid, simulation time step, initial water depth, roughness, precipitation, and evapotranspiration. The study area was gridded, and to ensure the accuracy of the calculations and to reduce the time required for the calculations, the grid was encrypted in the river channel and in the areas where the flood water may overflow. The rest of the area was gridded with larger grids (Zuo et al. 2020), with the total number of grids being 152,228. The entire flash flood simulation was set to start at 22:30 PM on 17 August and end at 05:30 AM on 18 August when the flood inundation process reached a steady state. The measurement was divided into 140 steps, with a b-step length of 3 min; Manning's coefficient was adopted to reflect the roughness ‘n’ of different underlying surfaces (Zaifoglu et al. 2019), and the roughness ‘n’ of the whole region was determined based on the fact that different features corresponded to different values of roughness, and empirical values were adopted in combination with the land use in the study area (Zhang 2012), i.e., river n = 0.03 (Zhang & Li 2012), construction land n = 0.04, agricultural land n = 0.025, grassland n = 0.03, and forest n = 0.06 (Guo et al. 2013).

In addition, this paper uses the relative error (Re) to assess the gap between model-simulated and calculated flows, and fitted statistical values (F) to reflect the consistency between the model-simulated inundation area and the actual observed inundation extent.

The actual inundation of houses, roads, and farmland was thus determined. According to mapping statistics, 45 houses, 73 roads, and 0.426 km² of farmland were completely flooded. The inundation of each element in Shadai village is shown in Figure 6(a).

The model results in this study were validated using the flow values at the flood mark points and the observed inundation extent.

The model results were validated by observing the inundation areas. Considering the inaccuracy of image-observed flooding in densely populated areas, the inundation area inferred from flood marks was chosen for validation in disaster-prone villages. After the field visit, flood inundation was shown to be overflowing from the river channel, except for the area where the flood scar points were measured, and the extent of the flood inundation could be clearly observed from the images.

From the results of the simulation, the area with the highest flood level is located between Shadai and Qingshan villages, where the ditch ratio drop is 35.3%, and the average slope is 36°. The area is surrounded by mountains on three sides, which makes drainage difficult (Mingrong et al. 2007). This, when accompanied by super-intense rainfall, quickly raises the water level in this area and produces a very strong destructive impact force on the houses and roads. Additionally, for the same rainfall duration, the water levels in the forested land area increased, those in the cultivated land area decreased, and both flood peaks and volumes showed a decreasing trend (Li et al. 2014). Contrastingly, in the Shadai River basin, the area of cultivated land is 53.13 km² (Table 2), occupying about half of the area of the basin, and the area of forested land is very small (5.62 km²), which is weak against floods (Zhai et al. 2022). This is one of the reasons for the huge losses caused by the flood event of 18th August.

However, some differences were observed between the results of the model simulation and the actual results; the simulation results show that on 18 August at 01:00 AM. the Shadai River downstream of the Lijiamo village section of the flood reached its peak, ‘Datong county 18th August-flash flood disaster review report’ shows that at approximately 01:40 AM, the Lijiamo village section of the flood reached its peak. The reason for the time difference may be that in the actual flood process, the flood is accompanied by a large amount of sediment and a reduction in rainfall; the resistance becomes larger and the power becomes smaller, resulting in slower water flow. Furthermore, the eddy viscosity coefficient in the model had a fixed value of 0.28, which does not take into account the resistance change, and hence, a time difference between the actual and simulated results is generated. Additionally, it is difficult to obtain rainfall station data, and the spatial distribution is not uniform. In the process of interpolation, errors may occur in the rainfall data between different regions, resulting in certain differences between the final and actual simulation results.

In this study, the MIKE21 model was used to simulate the inundation of heavy rainfall floods in small watersheds in undocumented mountainous areas, providing important methodological support and decision-making basis for the numerical simulation of flash floods in small watersheds in undocumented mountainous areas and for the forecasting of disaster situations. The results of the study show that the model can better simulate the whole process of dynamic inundation of small watersheds in mountainous areas with no information about heavy rainfall flash floods, and the inundation time of flash floods in the ‘18th August-Flash Flood Disaster Review Report of Datong County’ is basically in line with the results of the model simulation. Spatially, the accuracy of the MIKE21 model validation can be verified by the inundation extent inferred and observed from the measured flood scar data and high-resolution remote sensing imagery, respectively, and good results were obtained, with an F-statistic value of 0.82 and a relative error of 8%.

This study also has certain shortcomings, and future research can be improved and perfected with respect to the following aspects: First, the precipitation data are obtained by the inverse distance weight interpolation method through the meteorological stations near the Shadai River basin and the rainfall station data in Shadai village, and a certain degree of variability will be present in the amount of precipitation depending on the region. Furthermore, the meteorological stations are far away from the Shadai River basin, and the meteorological stations have a resolution of 3 h, which is a low resolution for flood simulation, resulting in less than optimal inversion of spatial and temporal distributions of precipitation. In the future, more accurate radar rainfall data or precipitation data from multiple local rainfall stations can be used to input the model; Second, due to the specificity of the location of the study area, complete high-resolution imagery data were not collected, and the images used in the study were a mosaic of two views, dated 11 September 2022 and 22 December 2022, with a much longer gap between the secondary images, which may have resulted in inaccuracies in the inundation extent of the actual flood observations and Manning's coefficients, and time close to the occurrence of flash floods would be chosen to reduce the inaccuracies in the process of subsequent validation of the flash flood simulation. Finally, with the impact of climate change and human activities, the losses caused by flash floods will show an intensifying trend, and the early warning and forecasting of flash floods will face greater challenges in the future. Simulation studies on the scope of flash floods, the establishment of flood prevention measures for flash flood-prone areas, and studies on the pre-establishment of emergency rescue and relief programmes will provide an important scientific basis for the risk management of flood disasters in uninformed mountainous areas.

The authors thank all the members of the Second Tibetan Plateau Scientific Expedition and Research Program for their hard work. This work was supported by grants from the Second Tibetan Plateau Scientific Expedition and Research Program (2019QZKK0903) and the Special Funds of the National Natural Science Foundation of China (Grant No. 42041004).

All relevant data are available from an online repository or repositories at The National Oceanic and Atmospheric Administration (https://www.ncei.noaa.gov/) website and The National Science and Technology Resource Sharing Service Platform (http://soil.geodata.cn).

The authors declare there is no conflict.