Evaluation of X-Band Radar for Flash Flood Modeling in Guangrun River Basin

Abstract

Flash flood disasters occur frequently under the influence of climate change and human activities, with the characteristics of strong suddenness, a wide range of hazards, and difficult prediction. Obtaining high-spatial- and high-temporal-resolution and high-precision rainfall monitoring and forecasting data is of great significance for accurate early warnings for flash flood disasters. In order to evaluate the advantages of X-band radar inverted rainfall in flash flood simulations, two typical flood events (3 July 2024 and 13 July 2024) in the Guangrun River Basin were studied. A comparative study between X-band radar inversion-based rainfall and rainfall measured at rainfall stations in terms of the flooding process and inundation extent was carried out using the China Flash Flood Hydrological Model (CNFF) and the two-dimensional hydrodynamic model (FASFLOOD). The results indicated that the temporal and spatial distribution characteristics of rainfall inversion by X-band radar were highly consistent with the measured rainfall at rainfall stations; in terms of simulating flood processes, rainfall based on X-band radar inversion performed better in key indicators such as the relative error of runoff depth, relative error of peak flow, error in time of peak occurrence, and Nash–Sutcliffe efficiency coefficient (NSE). In terms of simulating flood inundation, the simulation results based on X-band radar inversion and the measured rainfall from rainfall stations were consistent in the trend of rising and falling water processes and inundation range changes, and X-band radar could more accurately capture the spatial heterogeneity of rainfall. This study can provide technical support for disaster prevention and reductions in mountain floods in small watersheds.

1. Introduction

Against the background of global warming, natural disasters triggered by extreme weather are becoming more and more frequent, seriously affecting social and economic development. Flash floods are a type of natural disaster triggered by heavy rainfall in mountainous areas, which are characterized by a high suddenness, high destructiveness, and high difficulty in providing early warnings. According to the World Meteorological Organization (WMO), more than 11,000 weather-, climate-, and water-related disasters have been reported globally over the past 50 years, resulting in more than 2 million deaths and economic losses of more than USD 3.64 trillion. This is equivalent to an average of 115 deaths and USD 202 million in economic losses per day [1]. Among these disasters, tropical cyclones cause the highest economic losses, followed by floods. China has become one of the countries most seriously affected by flash floods due to its diverse climatic conditions, complex topography, and frequent heavy rainfall, which makes it highly susceptible to flash floods. About two-thirds of China’s land area is exposed to different types and degrees of flooding. Statistics from the Ministry of Water Resources (MWR) and the National Disaster Reduction Center of China (NDRCC) show that, between 1991 and 2020, floods caused an average of 2020 deaths or disappearances per year, with the total number of deaths exceeding 60,000, and an average annual direct economic loss of CNY 164.4 billion (cumulatively about CNY 4.81 trillion) [2]. Integration with real-time monitoring data from ground rainfall stations will improve the monitoring accuracy of X-band radar for regional rainfall and flash flood disasters, ensure more accurate and timely rainfall forecasting and flash flood early warnings, and provide a scientific basis and technical support for regional disaster prevention and mitigation.

The triggering factors behind flash flood disasters are very complicated, among which rainfall is the most direct and threatening factor. An efficient and accurate hydrometeorological monitoring system is the basic prerequisite for the establishment of a high-precision flash flood disaster forecasting model. At present, China’s precipitation monitoring of flash floods mainly relies on rain gauges, and the operational meteorological radar network mainly consists of S-band and C-band systems deployed in plain areas. These traditional systems face limitations such as limited coverage, a low measurement accuracy, a low spatial resolution, and the inability to detect the fine-scale precipitation spatial structure for precipitation monitoring in mountainous areas. However, the traditional S-band and C-band weather radars have a long detection distance and high cost, and their effectiveness in detecting near-surface rainfall over long distances is difficult to meet the usage requirements [3]. In contrast, the X-band rainfall radar system is characterized by a high resolution, small scanning blind area, high measurement accuracy, etc. Meanwhile, the multipolarized antenna has unique advantages in high-precision precipitation measurement, precipitation structure detection, and precipitation particle-phase inversion [4,5,6]. With the more mature use of all-solid-state emission technology in the X-band, the United States, Japan, and some European countries have realized S/C-band radar and key areas of X-band radar networking precipitation monitoring [7,8,9]. Currently, X-band high-resolution radar rainfall measurement technology is gradually being used in urban hydrology, hydrological forecasting, and early flood warnings in initial applications. Liu et al. [10] used the X-band QX-60 weather radar equipment of MetaSensing from the Netherlands to study the application of full-polarization-frequency modulated continuous wave rainfall radar in the forecasting and warning of heavy rain and flash floods in small watersheds. Gao et al. [11] used precipitation data estimated by X-band all-solid-state rainfall radar in Xi’an to compare with precipitation data from six surrounding surface meteorological observation stations, analyzing the detection effect of rainfall radar in Xi’an and the accuracy of the data. Liu et al. [12] evaluated the reliability and accuracy of an X-band rainfall radar precipitation estimation product based on observation data from 34 ground rainfall stations around the lake basin area of Poyang Lake.

In assessing the timeliness of rainfall radar for flash flood warnings, researchers have predominantly utilized C-band and S-band radar rainfall-driven hydrological models to conduct studies on flash flood alerts, whereas the application of X-band radar remains relatively uncommon. For instance, Lopper et al. [13] employed a distributed hydrologic model incorporating physical mechanisms driven by operational rain-gauge-adjusted radar rainfall, as opposed to rainfall solely monitored by rain gauges, to assess the accuracy of forecasts for a significant flood event that occurred on 7–8 September 2010 in Austin, Texas. The results indicated that the time error in the operational rain-gauge-adjusted radar-based rainfall predictions of flash flooding was reduced by 1.9 h. Kong et al. [14] used the Modular Distributed hydrological model (MDHM) to simulate flash floods in Liulin Town in Hubei Province on 12 August 2021 based on data from a ground rain measurement station and rain measurement radar. The results indicated that the effectiveness of early warnings using rain-radar-forecasted rainfall data was significantly enhanced. Hu et al. [15] assessed the hydrological application of radar-estimated rainfall by utilizing S-band radar-estimated rainfall data to drive the WRF-Hydro model during a typical heavy rainfall flooding event in the Erhe River Basin, located in central Chongqing. The findings indicated that the radar-estimated rainfall-driven WRF-Hydro model was capable of effectively simulating the flooding process, the flooding flow rate, and the timing of the peak occurrence. Paz et al. [16] demonstrated that X-band-radar-modeled flash flood results exhibited a greater accuracy compared to those derived from measured sites, particularly in their ability to capture small-scale rainfall spatial variability and the responses of complex terrain.

Jianshi County, located in Enshi Prefecture of Hubei Province, is situated in the northern part of the mountainous region in southwestern Hubei. This region is characterized by a steep terrain and complex geological conditions, which form the natural topographic basis for the occurrence of flash floods. Additionally, the region experiences frequent heavy rainfall, making it susceptible to flash floods and landslides, with the risks of such events being particularly pronounced. The peak of heavy rainfall typically occurs at regular intervals of approximately three years. During these peak periods, flash floods are likely to occur, resulting in significant damage to arable land, transportation infrastructure, water conservancy projects, and residential buildings [17]. Jianshi County established an X-band radar system in October 2023. This radar possesses the capabilities of high precision, high frequency, and extensive coverage monitoring, effectively addressing the challenges of monitoring localized heavy rainfall in mountainous regions. It also offers significant technical support for refined early warnings for mountain torrent disasters. An accurate assessment of flash flood warnings, derived from the X-band radar inversion of rainfall, can offer a scientific foundation for disaster prevention and mitigation decisions in Jianshi County. Accurately assessing the impact of flash flood simulations derived from X-band radar rainfall inversion can offer a scientific foundation for decision making regarding disaster prevention and mitigation in Jianshi County. Therefore, this study selected two typical flash floods events in the Guangrun River Basin on 3 July 2024 and 13 July 2024. The X-band radar data of Jianshi County and rainfall data measured by rain gauges were used to drive the distributed hydrological model and the two-dimensional hydrodynamic model, respectively, and the effects of the two types of rainfall data on refined early warnings for flash floods in the Guangrun River small watershed were compared and analyzed. The main contents are as follows: first, based on the dynamic calibration of the Z–R relationship of X-band radar reflectivity, a quantitative precipitation estimation product with a 1 km/10 min resolution is generated, and multi-scale calibration is completed with the joint rainfall station data; second, the impacts of measured rainfall and X-band-radar-inverted rainfall on runoff depth, peak flow, and peak appearance time are evaluated based on the China Mountain Flood Hydrological Model (CNFF); and third, a two-dimensional hydrodynamic model (FASFLOOD) is used to reveal the differences between the two types of rainfall data in indicators such as flood inundation range and maximum water depth.

2. Study Area

The Guangrun River basin is located between 109°32′–110°12′ E longitude and 30°06′–30°54′ N latitude, with the terrain declining in steps from northwest to southeast (Figure 1). The basin crosses the townships of Yezhou Town and Changliang Town in Jianshi County, Enshi Prefecture, Hubei Province, with a total watershed area of 164.07 km². The main channel is 38.7 km, and the average slope of the riverbed is 15.8‰. The climate of the Guangrun River Basin is a subtropical monsoon mountain humid climate with an average annual precipitation of 1707 mm. The annual distribution of precipitation is extremely uneven, with April to October accounting for 84.7% of the total annual precipitation and the monthly precipitation in July reaching over 300 mm. Extreme rainstorms frequently occur in the basin, with the maximum 24 h rainfall recorded at 286 mm (Yezhou Town Station, 19 July 2016) and the maximum hourly rainfall intensity reaching 72 mm (27 June 2020) [18].

Floods in the Guangrun River Basin have the typical characteristics of rain flash floods in mountainous areas, with a high intensity of rainstorms, fierce floods, and great losses caused by flood disasters. The flood process rises and falls sharply, with a main single peak. The peak time is generally 5–10 h. Affected by the terrain, the center of the rainstorm tends to be upstream. According to survey statistics, since the 1990s, flooding has occurred every year in the Guangrun River basin, especially in 2008, 2016, 2017, and 2020, with very serious disasters occurring. In July 2016, a heavy rainstorm occurred in Jianxi County, with a rainfall of 355 mm in 24 h. The flood caused extensive agricultural land on both sides of Malan Creek to be destroyed, inundating an area of 250 acres of farmland, interrupting road traffic, and flooding more than 20 residential houses. The direct economic loss amounted to over CNY 800,000. Starting from 23:00 on 25 July 2020, heavy rain fell in Jianshi County. From 3:00 to 7:00 a.m. on July 26, the water level at the Jianshi hydrological station rose sharply by 5.86 m, and the peak water level reached 560.70 m, exceeding the warning water level by 3.70 m. The corresponding maximum peak flow was 1190 m³/s, a flood that occurs once in a century. The disaster affected more than 160,000 people, damaged more than 300 houses, and affected 3291 hectares of crops. The direct economic loss was CNY 248 million.

3. Materials and Methods

3.1. Data and Sources

The rainfall radar used in this study was selected from the X-band radar in Jianshi County, Enshi Prefecture, Hubei Province. The data were obtained from the Hubei Provincial Department of Water Resources and passed the data quality control. As shown in Figure 2, the radar has a maximum detection radius of 75 km and an effective rainfall detection range of 45 km, which completely covers the Guangrun River basin and ensures data quality. There is a hydrometeorological observation network consisting of 7 rainfall stations, 5 water level stations, and 1 hydrological station in the basin (Figure 3). Long-term rainfall runoff data provided by the Hubei Hydrology and Water Resources Center were used for parameter calibration and validation of the distributed hydrological model. The X-band radar rainfall data for the two rainfall events on 3 July and 13 July 2024 and data from the rainfall monitoring stations in the region were collected for the refined early warning simulation study of flash flood warnings in small watersheds. Among them, the radar inversion rainfall data were obtained from the gridded rainfall products of the X-band radar in Jianshi County, with a time interval of 10 min and a spatial resolution of 75 m, including the instantaneous rainfall intensity and a cumulative rainfall of 5–120 min. In addition, the rainfall data are from the surface rainfall station in the Guangrun River sub-basin, with the same time interval of 10 min.

The basic datasets collected for the study included basic geographic information and event-specific rainfall flood records. The geographic data infrastructure consists of a 1:50,000 digital elevation model (DEM), 1:1,000,000 hydrologic network data, 1:250,000 land use classifications, and 1:500,000 soil texture maps, all from the National Geographic Information Center of China (http://www.ngcc.cn/).

3.2. Attenuation Correction

Attenuation correction is an important quality control algorithm for X-band rain measurement radar. We used the H-B attenuation correction method, which does not require polarization and directly calculates the attenuation coefficient using the reflectivity factor for correction.

The attenuation correction of high-frequency dual-polarization radar is to find the two-way $PI{A}_H$ and add it to the measured reflectivity factor to obtain the corrected reflectivity factor. The correction formula is as follows:

$$Z_m(r)=Z(r)\exp [-0.46{\displaystyle {\int }_0^{r}A(l)dl}]$$

(1)

Among them, $Z_m$ and $Z$ are the reflectance factors before and after correction (unit: mm⁶/m³).

Using the attention and reflection relationship in the H-B method, we obtain the following:

$$A=a{Z}^b$$

(2)

Among them, a and b are coefficients that depend on the raindrop spectrum.

$$Z_m(r)=Z(r)g^{1/b}(r)$$

(3)

$$g(r)=\exp [-0.46b{\displaystyle {\int }_0^{r}A(l)dl}]$$

(4)

By differentiating the above equation, we can obtain the following:

$${\displaystyle \frac{dg(r)}{dr}}=-0.46bg(r)A(r)=-0.46bg(r)a{Z}^b(r)=-0.46ba{Z}_m^b(r)$$

(5)

Then,

$$dg(r)=-0.46ba{Z}_m^b(r)dr$$

(6)

Integrating the results in the following:

$$g(r)=1-0.46ba{\displaystyle {\int }_0^r{Z}_m^b(l)dl}$$

(7)

The final reflectance factor after attenuation correction is as follows:

$$Z(r)={\displaystyle \frac{Z_m(r)}{g^{1/b}(r)}}={\displaystyle \frac{Z_m(r)}{{[1-0.46ba{\displaystyle {\int }_0^r{Z}_m^b(l)dl}]}^{1/b}}}$$

(8)

The above equation is the H-B method. For the X-band, the coefficients in this article are $a=1.26378\times {10}^{-4}$, $b=0.75419$.

3.3. Spatio-Temporal Dynamic Z–R Radar Rainfall Inversion

The spatio-temporal dynamic Z–R radar precipitation retrieval algorithm is an advanced method to improve the accuracy of precipitation estimation, especially in heavy rainfall areas [19]. The algorithm utilizes single-radar volume scan data to automatically classify precipitation types and applies different Z–R relationships accordingly to improve the quantitative precipitation estimation (QPE) accuracy. The technical implementation involves the following four key computational processes:

(1)

Convective kernel identification. Based on the single-radar body-scan data, the identification of flow cores is realized in polar coordinates. The convective nuclei in the area of heavy precipitation are accurately identified by setting different criteria (e.g., reflectivity factor, vertical layer water content, etc.).

(2)

Convective area identification. After identifying the convective nuclei, the entire convective zone is identified using the method of area growth. The accurate identification of the convective zone is ensured by comprehensively judging multiple physical quantities (e.g., combined reflectivity, maximum reflectivity height, vertical gradient of reflectivity, etc.).

(3)

Identification of laminar cloud bright zones. After identifying the entire convective cloud area, bright band identification is performed for the stratiform cloud precipitation area. The regional growth method is used to accurately identify the regions affected by the bright bands and avoid overestimation of the stratiform cloud precipitation area.

(4)

Inversion formula selection. According to the identification of strong precipitation areas and non-strong precipitation areas, different Z–R relations are used to invert the rainfall, which can effectively improve the accuracy of rainfall inversion. Among them, the stratiform cloud formula is Z = 200I^1.6 and the convective cloud formula is Z = 300I^1.4, where, Z is the radar reflection factor (unit: mm³/m⁶) and I is the rainfall intensity (mm/h).

3.4. Chinese Flash Flood Hydrologic Modeling Approach

3.4.1. Introduction to China’s Flash Flood Hydrological Modeling

The China Flash Flood Hydrological Model (CNFF) is a distributed hydrological model independently developed by the China Institute of Water Resources and Hydropower Research (IWHR), which has been widely used for flash flood analysis and flash flood forecasting in small- and medium-sized watersheds across China [20,21,22,23]. The model conceptualizes the studied watershed through the following seven hydrological elements: sub-basin, reach, node, watershed, water source, depression, and reservoir. The model uses sub-watersheds as the basic computational unit (meteorological conditions and underlying surface characteristics remain relatively consistent) and uses modular modeling techniques to construct a distributed hydrologic framework. The framework systematically performs rainfall runoff generation, flow concentration, channel routing, and reservoir regulation calculations.

Compared with other distributed hydrological models, the CNFF model has the following significant advantages: (1) the model adopts multiple runoff generation mechanisms, including saturated runoff generation, infiltration runoff generation, and mixed runoff generation, in order to adapt to the different regional and climatic conditions of the sub-basin; (2) the model utilizes high-resolution topographic data and, at the same time, takes into account the rainfall intensity effect and adopts the distributed unit hydrograph technique; and (3) the model can continuously simulate and forecast multiple consecutive flash flood events. Based on the CNFF framework, we developed a customized distributed hydrological model for the Guangrun River basin, taking into account the specific characteristics of the basin and the available data.

3.4.2. Flash Flood Hydrological Modeling

The saturation excess runoff module of the Xin’anjiang hydrological model was used to calculate the surface flow and soil moisture, which is suitable for the hydrological characteristics of the humid climate zone. A hybrid model combining the distributed unit hydrograph method and the linear reservoir method was developed for calculating the discharge. The discharge of surface runoff was calculated using the comprehensive unit hydrograph of small watersheds in the study area, as shown in Figure 4. The flood evolution of the river channel was nonlinearly simulated by the dynamic Muskingum method, which can effectively reflect the spatial and temporal change characteristics of the hydraulic features of the river channel.

The formulas for slope flow velocity and time of convergence are given as follows:

$$\left\{\begin{array}{c}V=K_s{S}^{0.5}{i}^{0.4}\\ T_j={\displaystyle \sum _{m=1}^{M_j}}{\displaystyle \frac{c{L}_m}{V_m}}\end{array}\right.$$

(9)

where $V$ is the velocity of water flow, m/s; $Ks$ is the flow coefficient, m/s, it is an empirical parameter mainly reflecting the influence of land use characteristics on the friction of flow; $S$ is the slope along the direction of water flow of a grid in the watershed; $i$ is the dimensionless rainfall intensity; $Tj$ is the time of convergence of the $j\mathrm{t}\mathrm{h}$ grid, s; $Lm$ is the length of the flow path of the $m\mathrm{t}\mathrm{h}$ grid, m; $c$ is the coefficient $c=1$ or $c=\sqrt{2}$; and $Mj$ is the convergence path of the $j\mathrm{t}\mathrm{h}$ grid number of grids on it.

3.5. Approaches to Hydrodynamic Modeling

FASFLOOD is a domestic high-performance flood simulation software with independent intellectual property rights developed by China Institute of Water Resources and Hydropower Research [24]. The software integrates rainfall runoff calculations, one-dimensional pipeline and river network hydrodynamic modeling, two-dimensional surface hydrodynamic simulation, one-dimensional–two-dimensional coupling calculations, sediment scouring calculations at the breakout point, and other comprehensive calculation modules. Its flexible multi-module coupling capability can accurately and rapidly simulate various complex flooding processes, such as dam failure, flash flooding, and urban flooding [25]. By integrating GPU-accelerated parallel computing technology into the numerical simulation of hydrodynamics, FASFLOOD realizes the minute-level computation of models with millions of computing units. Currently, the software is listed in the “2023 Key Promotion Catalog of Advanced Water Resources Technology” of the Ministry of Water Resources, and has been successfully implemented and promoted in many national projects, such as urban flood control and drainage planning, flood risk mapping, and digital twin-system development.

In this study, the numerical simulation of flood evolution in the Guangrun River basin is carried out based on the two-dimensional water surface hydrodynamic computation module of FASFLOOD, which takes the physically meaningful two-dimensional shallow water equation set as the controlling equations in the form shown as follows:

$${\displaystyle \frac{\partial }{\partial t}}h+{\displaystyle \frac{\partial }{\partial x}}hu+{\displaystyle \frac{\partial }{\partial y}}hv=0$$

(10)

$${\displaystyle \frac{\partial }{\partial t}}uh+{\displaystyle \frac{\partial }{\partial x}}h{u}^2+{\displaystyle \frac{\partial }{\partial y}}huv+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\partial }{\partial x}}g{h}^2+gh{\displaystyle \frac{\partial }{\partial x}}B\left(x,y\right)+{\tau }_{bx}=0$$

(11)

$${\displaystyle \frac{\partial }{\partial t}}vh+{\displaystyle \frac{\partial }{\partial y}}h{v}^2+{\displaystyle \frac{\partial }{\partial x}}huv+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\partial }{\partial y}}g{h}^2+gh{\displaystyle \frac{\partial }{\partial y}}B\left(x,y\right)+{\tau }_{by}=0$$

(12)

where $h$ is the water depth, m; $u$ and $v$ are the flow velocities in the $x$ and $y$ directions, m/s, respectively; $t$ is the time, s; $B(x,y)$ is the elevation of the bottom slope, m; $\tau$ is the friction term, and $bx$ and $by$ are the friction components of the friction term in the $x$ and $y$ directions, respectively.

The shallow water equation is a system of nonlinear hyperbolic partial differential equations, which can effectively describe complex free surface flow phenomena, but it is still challenging to obtain its analytical solution. The 2D flood calculation module of FASFLOOD employs a Godunov-type finite volume method with shock-capturing capabilities to discretely solve the two-dimensional shallow water equations on unstructured grids. In this context, the Riemann problem is addressed using an approximate Riemann solver that integrates the HLLC (Harten–Lax–Leer–Contact) formulation [26]; the bottom slope source term is discretized implicitly and hydrostatic reconfiguration is used to correct the negative water depth problem at the dry and wet boundaries in order to improve the stability of the model. MUSCL (Monotonic Upstream-centered Scheme for Conservation Laws) spatial reconfiguration and a prediction correction method are used to ensure the model has a second-order accuracy in time and space [27,28,29].

3.6. Assessment Methodology

3.6.1. X-Band Radar Rainfall Inversion Accuracy Assessment

The relative error (RE) and root mean square error (RMSE) of 24 h cumulative rainfall were selected as indicators in the methodology for assessing radar rainfall results [30].

(1)

Relative error (RE)

The relative error (RE) serves as an indicator of the extent of deviation between the predicted value and the measured value, and it is a dimensionless quantity. A positive RE value signifies that the forecasted value is excessively high, while a negative RE value indicates that the forecasted value is insufficiently low. Typically, the acceptable error range is defined as −20% < RE < 20%. The formula for calculating the relative error (RE) is as follows:

$$\mathrm{R}\mathrm{E}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^M}\left({P’}_i-P_i\right)}{{\displaystyle {\sum }_{i=1}^M}{P}_i}}\times 100\%$$

(13)

where $P_i$ and ${P\mathrm{’}}_i$ represent the ith measured surface rainfall value and the predicted surface rainfall value in the sequence, respectively, mm, and M represents the number of measured values in the series.

(2)

The root mean square error (RMSE) is calculated as follows:

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}={\displaystyle \frac{\sqrt{\frac{1}{M}{\displaystyle {\sum }_{j=1}^M}{(P_j^{\prime }-P_j)}^2}}{\frac{1}{M}{\displaystyle {\sum }_{j=1}^M}{P}_j}}$$

(14)

where ${P^{\prime }}_j$ and $P_j$ are the radar precipitation inversion value and the rainfall value observed by the ground rainfall station, respectively. When evaluating on a spatial scale, $M$ is the number of rainfall stations; ${P^{\prime }}_j$ and $P_j$ are the radar inversion value of the accumulated rainfall and the observed value of the ground rainfall station for the whole observation period at a particular spatial location $j$, respectively, and when evaluating on a temporal scale, $M$ is the length of the rainfall period; and ${P^{\prime }}_j$ and $P_j$ are the radar inversion value of the surface-averaged rainfall for the study area and the observed value of the ground rainfall station for the observation moment $j$, respectively. In order to remove the effect of different field rainfall, the final RMSE value is the average value of the ground rainfall station observations divided by the corresponding dimension, respectively.

3.6.2. Flood Forecast Accuracy Assessment

In this study, the following four key performance metrics [31] were used to assess the accuracy of flash flood simulations: (1) relative error in runoff depth ($R_{\mathrm{r}\mathrm{e}}$); (2) relative error in peak discharge ($R_{\mathrm{q}\mathrm{e}}$); (3) peak time lag ($T_{\mathrm{e}}$); and (4) Nash–Sutcliffe efficiency coefficient (NSE). The optimal values of these metrics are 0 for $R_{\mathrm{r}\mathrm{e}}$, $R_{\mathrm{q}\mathrm{e}}$, and $T_{\mathrm{e}}$ and 1 for NSE. The simulation results are considered acceptable when the absolute values of $R_{\mathrm{r}\mathrm{e}}$ and $R_{\mathrm{q}\mathrm{e}}$ are less than 20%, $T_{\mathrm{e}}$ is less than 2.5 h, and NSE exceeds 0.6. The formulas for each index are as follows:

$$R_{\mathrm{r}\mathrm{e}}={\displaystyle \frac{\mid R_{\mathrm{s}}-R_0\mid }{R_0}}$$

(15)

$$R_{\mathrm{q}\mathrm{e}}={\displaystyle \frac{\mid Q_{\mathrm{s}}-Q_0\mid }{Q_0}}$$

(16)

$$T_{\mathrm{e}}=T_{\mathrm{s}}-T_{\mathrm{o}}$$

(17)

$$NSE=1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n}\left[Q_{\mathrm{s}}\right(i)-Q_0(i){]}^2}{{\displaystyle {\sum }_{i=1}^n}[Q_0-{\bar{Q}}_0{]}^2}}$$

(18)

where $R_s$ is the simulated runoff depth, mm; $R_0$ is the measured runoff depth, mm; $Q_s$ is the simulated peak flow, m³/s; $Q_0$ is the measured peak flow, m³/s; $T_s$ is the simulated peak present time, h; $T_0$ is the measured peak present time, h; $Q_s\left(i\right)$ is the simulated flow at the i moment, m³/s; $Q_0\left(i\right)$ is the measured flow at the i moment, m³/s; ${\overline{Q}}_0$ is the measured flow rate average, m³/s; and $n$ is the length of the flood sequence.

4. Results

4.1. Flash Flood Events

This study investigated two major heavy rainfall flood events that occurred in the Guangrun River basin after the deployment of the X-band radar monitoring system. On 3 July 2024, a heavy precipitation event occurred in the basin, with a cumulative rainfall of 114.3 mm at the Jianshi station and 188.9 mm at the Longping station. Subsequently, on 13 July 2024, an extreme rainfall event occurred in the Guangrun River basin in Enshi Prefecture, with a clear bimodal precipitation pattern in Gaoping Township, Jianshi County, with Gaoping Station recording 169.4 mm of rainfall in 24 h, and the center of the storm showed a clear pattern of migration from north to south. This extreme weather triggered a chain reaction characteristic of flash floods, leading to a rapid rise in river levels, with 13 river sections exceeding the alert level for a once-in-twenty-years flood and 54 sections exceeding the alert level for a once-in-five-years flood, with a clear risk of slope destabilization throughout the affected area.

4.2. Rainfall Inversion Based on X-Band Radar

Figure 5 and Figure 6 show the 24 h X-band radar inversion cumulative rainfall versus the rainfall-station-monitored cumulative rainfall for 3 July and 13 July 2024, respectively. For the 3 July rainfall in the Guangrun River basin (Figure 5), the cumulative rainfall from the X-band radar inversion (Figure 5a) and the cumulative surface rainfall monitored by the rainfall stations (Figure 5b) are consistent in terms of spatial distribution and rainfall class. Heavy rainfall levels (50–100 mm/24 h) are primarily concentrated in the middle and lower reaches and parts of the northwest and northeast regions of the basin. The analysis of surface-accumulated rainfall recorded by the rainfall stations in the Guangrun River Basin on 13 July (Figure 6a) indicates that the heavy rain level (100–150 mm/24 h) is primarily concentrated in the western part of the Guangrun River Basin, whereas other areas experience rainfall at the heavy rain level (50–100 mm/24 h). In the accumulations of the X-band radar inversion (Figure 6a), the spatial distribution of heavy rainfall and torrential rainfall is basically consistent with the surface-accumulated rainfall monitored by the rainfall stations, except for the downstream area of the basin; in the downstream area of the basin, the accumulated rainfall from the X-band radar inversion is one level higher than that of the surface accumulated rainfall monitored by the rainfall stations, and reaches the torrential rainfall level.

In addition, the rainfall station observation is taken as the true value, the accuracy index is used for quantitative assessment, and the results are shown in

Table 1. Results of rainfall assessment based on X-band radar inversion.

Rainfall Event	RE (%)	Spatial Scale RMSE	Temporal Scale RMSE
3 July 2024	3.58	0.26	0.31
13 July 2024	4.23	0.28	0.37

. From this table, it can be seen that the relative errors of cumulative rainfall for the two rainfall events 20240703 and 20240713 are 3.58% and 4.23%, respectively; the relative errors in the spatial scales are 0.26 and 0.28, respectively; and the relative errors in the temporal scales are 0.31 and 0.37, respectively. Generally speaking, rainfall inversion utilizing X-band radar can provide a more accurate representation of rainfall levels and their spatial distribution.

4.3. Comparison of Flash Flood Process Simulation Results

4.3.1. Model Construction and Calibration

A distributed hydrological model of the Guangrun River basin was developed using the China Flash Flood Hydrological Model (CNFF) (shown in Figure 7). The watershed was divided into 192 sub-watersheds ranging from 0.27 km² to 27.63 km². In total, 15 historical flood events with a flood peak modulus greater than 0.3 were selected to calibrate and verify the CNFF model. A total of 10 events were used for parameter calibration and the remaining 5 events were used for model validation. Flood events with a flood peak modulus greater than 0.3 were selected to calibrate the parameters of the CNFF model. Based on the collected historical rainfall and flood data of the small watershed, a combination of parameter automatic optimization (SCE-UA) and manual optimization was used to optimize the parameters of the distributed hydrological model of the small watershed and determine the optimal model parameters, and the calibration results of the relevant parameters are shown in

Table 2. Calibration results of relevant parameters in the CNFF model.

Parameters	Meaning	Parameter Value	Parameter Value Lower Limit	Parameter Value Upper Limit
B	Storage Capacity Distribution Curve Exponent	0.2	0	1
IMP	Impervious Area Proportion	0.01	0	1
WUM	Upper Layer Soil Water Storage Capacity	20	1	20
WLM	Lower Layer Soil Water Storage Capacity	60	6	90
WDM	Deep Layer Soil Water Storage Capacity	40	20	100
EX	Free Water Storage Capacity Curve Exponent	1.2	1	2
SM	Free Water Reservoir Capacity	25	0	50
KS	Interflow Daily Outflow Coefficient	0.5	0.1	1
KG	Groundwater Daily Outflow Coefficient	0.2	0	10
KKS	Interflow Daily Recession Coefficient	0.1	0	1
KKG	Groundwater Daily Recession Coefficient	0.1	0	1

.

Figure 8 and Figure 9 show the rainfall and flooding processes in selected fields during the rate and validation periods. The relative error of the mean runoff depth in the study area was 2.57%, the relative error of the mean peak flow was −3.09%, the relative error of the mean peak time was 0.33 h, and the mean NS efficiency coefficient was 0.92. The mean runoff depth in the study area was 2.57%, the relative error of the mean peak flow was −3.09%, the relative error of the mean peak time was 0.33 h, and the mean NS efficiency coefficient was 0.92. The mean values of the indicators in the validation period were 7.12%, −4.2%, and −4.2%, respectively. The NS efficiency coefficient was 0.92. The mean values of each index in the validation period were 7.12%, −4.2%, 0.54 h, and 0.93, respectively. Overall, the Chinese flash flood hydrological model (CNFF) could accurately simulate the field rainfall and flood response process, and it had a good applicability in the study area.

4.3.2. Simulation of Flooding Processes Based on X-Band Radar Inversion and Measured Rainfall at Rainfall Stations

Based on the X-band radar inversion rainfall data and rainfall-station-measured rainfall data, respectively, the distributed hydrological model of the Guangrun River basin is driven to simulate floods, and the simulated flood processes are shown in Figure 10 and Figure 11. The results of the two flood simulations are shown in

Table 3. Flood simulation effect of X-band radar inversion rainfall and rainfall station measured rainfall.

Rainfall Event	Relative Error of Runoff Depth (%)		Relative Error of Peak Flow (%)		Time Error of Peak Occurrence (h)		NSE
	Rainfall Stations	X-Band Radar	Rainfall Stations	X-Band Radar	Rainfall Stations	X-Band Radar	Rainfall Stations	X-Band Radar
3 July 2024	11.9	4.3	5.0	1.2	2	1	0.90	0.96
13 July 2024	7.0	13.8	1.2	0.9	2	1	0.54	0.86

, from which it can be seen that for the rainfall event on 3 July 2024, the relative error of flood runoff depth based on X-band radar rainfall inversion simulation is 4.3%, the relative error of peak flow is 1.2%, the error in time of peak occurrence is 1 h, and the NSE is 0.96. The relative error of runoff depth based on the rainfall measured at the rainfall station is 11.9%, the relative error of peak flow is 5.0%, the error in time of peak occurrence is 2 h, and the NSE is 0.90. For the rainfall event on 13 July 2024, the relative error of runoff depth based on X-band radar inversion rainfall simulation is 13.8%, the relative error of peak flow is 0.9%, the error in time of peak occurrence is 1 h, and the NSE is 0.86. The relative error of runoff depth based on rainfall station real measurement rainfall simulation is 7.0%, the relative error of the peak flow is 1.2%, the error in time of peak occurrence is 2 h, and the NSE is 0.54. It can be seen that the flood simulated based on the X-band radar inversion rainfall is better than that simulated based on the rainfall measured by rainfall stations in terms of the relative error of runoff depth, peak flow, the time of peak occurrence, and the NSE.

4.4. Comparison of Flood Inundation Results

In this study, a two-dimensional hydrodynamic model was built using FASFLOOD (http://cdr.iwhr.com/fhkhjzzx/shzhzyfxrjxz/webinfo/2022/11/1668157350789555.htm, accessed on 11 November 2022), a high-performance flood analysis software developed by IWHR, to simulate the 3 July 2024 and 13 July 2024 flood events using X-band radar data and rain gauge measurements. In order to balance the model computational accuracy and efficiency, a 10 m triangular unstructured grid was used in this study to discretely dissect the study area. The simulation results are shown in Figure 12, Figure 13, Figure 14 and Figure 15, which indicate that the upwelling and receding processes simulated based on the two types of rainfall data were highly consistent, and the trend of the inundation extent was also relatively consistent. Meanwhile, the inundation ranges simulated based on the X-band radar inversion rainfall were slightly larger than those simulated based on the measured rainfall from the rainfall station at the same moment. This indicates that radar data are more advantageous in capturing the spatial distribution of rainfall.

5. Discussion

The monitoring of surface rainfall in flash floods in China is mostly performed using traditional rain gauge equipment. Although the rainfall measured by rain gauges is more accurate, due to the uneven distribution and low density of rain gauge stations, the current rainfall data struggle to accurately reflect the spatial distribution of rainfall, which leads to a low precision of surface rainfall in the study area [32,33]. Compared with single-site rainfall monitoring, rain radar, as a means of active remote sensing, can obtain accurate, wide-ranging, high-spatial- and high-temporal-resolution real-time rainfall information, which can effectively prevent the risk of leakage of localized heavy rainfall measurement [34]. Weather radar can be divided into S-band, C-band, and X-band radar according to the band. X-band weather radar can provide a higher spatial and temporal resolution than traditional S-band or C-band radar because of its shorter wavelength. It has significant advantages in capturing strong local convective weather processes, thereby improving prediction accuracy of extreme weather events [35]. In this study, two typical rainfall events of 3 July 2024 and 13 July 2024 were selected to evaluate the effect of rainfall retrieved by X-band radar. Compared with the rainfall monitored by the rainfall station, the relative error of cumulative rainfall based on X-band radar inversion was smaller (3.58% and 4.23%), and it could better reflect the rainfall level and its spatial distribution at the time scale and spatial scale.

The flash floods have the characteristics of suddenness and uncertainty, and traditional early warning methods are mainly based on real-time rainfall monitored by rainfall stations for early warning, which is prone to false alarms, missed alarms, and poor timeliness of early warning. The application of precipitation radar technology offers high-precision rainfall information with a high spatio-temporal resolution, thereby enhancing the early warning and forecasting capabilities for flash floods. Regarding rainfall radar in flash flood early warning assessment, scholars mostly base this on S-band and C-band radar inversion rainfall driven hydrological model simulation. The results show that flash flood simulations based on rainfall radar inversion can not only can better simulate the flood process, flood flow, and peak time [15,36], but also reduce the time error of flash flood prediction and significantly improve prediction accuracy [13,14]. In addition, X-band radar technology shows a superior accuracy performance in flash flood simulation. It can not only analyze the spatial distribution characteristics of local rainfall with a high resolution, but also accurately portray the hydrological response process under complex terrain conditions [16]. In this study, the flooding process and inundation of two rainfall events, 3 July 2024 and 13 July 2024, were simulated based on X-band radar inversion rainfall and the rainfall measured at rainfall stations. In terms of flood process simulation, the flood simulated based on X-band radar inversion rainfall was better than that simulated based on rainfall-station-measured rainfall in terms of the relative error of runoff depth, error of peak flow, error in time of peak occurrence, and NSE. The results of this study are in line with the results of the study by Paz et al. [16]. In terms of flash flood inundation simulation, the inundation range of the simulation based on the X-band radar inversion rainfall was consistent with that based on the simulation of the rainfall measured from rainfall stations in terms of the process of rising and receding and the range of inundation, and the range of inundation was slightly larger than that of the latter.

There are still some shortcomings in this study, mainly in the following two aspects. First, the X-band radar station was built in a short period of time, resulting in fewer typical floods applied to this study, and the results of this study will be further demonstrated when there are more typical floods at a later stage. Secondly, due to the lack of actual inundation data for the two floods, this paper did not compare the flood inundation range based on the X-band radar inversion rainfall and measured rainfall simulation with the actual inundation, which is to be further discussed later.

6. Conclusions

In this study, the X-band rainfall radar inversion rainfall and ground-station-measured rainfall data of two flash floods (20240703 and 20240713) in the Guangrun River Basin were compared to evaluate the simulation effects of the two types of rainfall data in the spatial distribution, flooding process simulation, and flooding inundation process of the two flash flood events. The main conclusions are as follows:

(1)

The cumulative rainfall from the X-band radar inversion and the surface cumulative rainfall measured at the rainfall station show a high match in rainfall level and spatial distribution, indicating that the X-band rainfall radar has a better ability to monitor the rainfall process and is able to show the spatial and temporal distribution of rainfall in the basin.

(2)

Comparing the simulation results of the distributed hydrological model based on X-band radar inversion rainfall and measured rainfall, the results of the former are better than the latter in terms of the relative error of runoff depth, error of peak flow, error in time of peak occurrence, and NSE. This indicates that the X-band radar inversion of rainfall data has a higher accuracy and reliability in flood simulation.

(3)

A two-dimensional hydrodynamic model based on FASFLOOD is used to simulate the flood inundation of 3 July 2024 and 13 July 2024. The research shows that the simulation results based on X-band radar inversion and the rainfall measured by rainfall stations are consistent regarding the process of rising and falling water and the trend of submergence range, and radar data has more advantages in capturing the spatial distribution of rainfall, which provides more reliable technical support for flood warning.