Assessing flash flood inundation from an extreme rainfall event: case study: Wadi Al Jizzi Oman

Flash floods are complex natural hazards that result from very intensive and localized rainfall storms among hydrometeorological, geomorphological, and anthropic complexities. While water assessments, management, and planning are important issues in urbanized communities, water scarcity is a major dilemma in arid regions and may damage their local economic growth. Extreme events such as drought and flash floods due to increasing or decreasing temperature trends lead to more heavy rainfall in a short time interval (El Kasri et al. 2021; Abushandi & Al Ajmi 2022). Therefore, there is a need to understand the hydrological processes that occur in arid regions although the lack of hydrological data and the difficulties of access to some areas make the task much more challenging. Reliable flash flood modelling requires sufficient, continuous, and high-intensity interval records not only for several years but even for a single storm event. Generally, governmental efforts have focused on strengthening water facilities through the establishment of dams for groundwater recharge and surface storage. Groundwater represents the main water resource that feeds various regions of the Sultanate of Oman. Networks and stations have also been established to monitor the water conditions in the Sultanate and to understand the characteristics of rainfall; these represent an essential input in hydrologic modelling. The behaviour of rainfall differs from storm to storm according to several factors, including:

• 1.

  Cloud diameter and thickness

• 2.

  Wind speed and direction

• 3.

  Rainfall intensity

• 4.

  Temperature gradients

Accurate rainfall estimates are crucial for accurate flood forecasting. A deep-learning-based Long Short-Term Memory (LSTM) model was introduced by Chen et al. (2022) to forecast monthly rainfall data. The model improved accuracy and efficiency in forecasting rainfall under different climatic conditions, making it a promising approach for rainfall forecasting in various global regions. Numerous studies have investigated the behaviour of single extreme storm events and their impact on the water sector. Using a three-dimensional computational fluid dynamics (CFD) model, Feng et al. (2022) studied flood behaviour during extreme storm events revealing the sensitivity of the flow distribution. Abdulla et al. (2002) modelled the hydrographic runoff for a single rainfall event in an arid basin using limited available meteorological and hydrological data. The results showed that the hydrographic simulation of the runoff was similar to the observed records. Abushandi & Merkel (2013) developed a new framework using two models – the HEC-HMS (hydrologic modelling system) and the IHACRES (Institute of Hydrology Center for Resource and Environmental Studies) – to assess a single runoff event for an arid catchment in Jordan. The results showed that both models could be used to calculate the flow volume. Primarily, the results of hydrological models are based on soil moisture antecedent conditions. Singh et al. (2022) proposed two hybrid Machine Learning-based Pedotransfer Functions (ML-PTFs), combining Genetic Algorithm with Multilayer Perceptron (MLP-GA) and Support Vector Machine (SVM-GA), to estimate saturated hydraulic conductivity (K_s) in soils. Tan et al. (2008) studied the performance of rainfall-runoff models from single storms for small catchments. The authors found that continuous-event calibration performed better in evaluating the overall shape of a hydrograph, the peak flow volume, and the delay time. Hossain et al. (2019) used an event-based stormwater management model (EPA-SWMM) performing continuous simulations for calculating individual runoff volumes. The calibration parameters were based on the physical characteristics of catchments such as Manning's roughness, saturated hydraulic conductivity, and unsaturated zone moisture contents. However, modelling flood events may overestimate reference data by up to 200% due to method uncertainties (Valeriya et al. 2019). A limited number of hydrologic models are currently capable of predicting short-duration runoff occurring during 30-minute-long storm events (Du et al. 2007; Abushandi & Merkel 2013). The simulated runoff in these models demonstrated close agreement with the observed events and a relative error of less than 10%.

The integration of remotely sensed data to complement the scarcity of ground hydrologic data has been a major modelling concern over the last 10 years. Chen et al. (2021) conducted a study using satellite estimates and sub-pixel changes in precipitation in a semi-arid region. The authors used the infrared (IR) method to detect clouds that produced rain through a multi-infrared channel and revealed that the spatial variations of precipitation depended on both the spatial scale and the type of precipitation. Massari et al. (2019) used a model conditional processor (MCP) to combine satellite precipitation products through a precipitation dataset. The method was developed for the forecasting of specific flooding events. Furthermore, Abushandi & Merkel (2013) developed a new framework using global satellite mapping of precipitation (GSMaP-MVK+) to feed two models, namely the HEC-HMS and IHACRES. The models were able to determine a single runoff volume for an arid catchment in Jordan. Croke et al. (2005) have also previously used the IHACRES to model rainfall-runoff for arid and semi-arid catchments. To reduce the parameter uncertainty, these authors developed a non-linear loss module which had stronger physical descriptors.

However, modelling rainfall-runoff in arid catchments is associated with difficulties in defining a threshold because the information commonly contains a large number of zero data points. Ali et al. (2019) conducted a study to examine the capability of the analytic hierarchy process (AHP) for flood vulnerability mapping incorporating geographic information systems (GIS). Seven factors were used for this purpose, namely land elevation, slope angle, topographic wetness index, rainfall deviation, land use land cover, clay contents in soil, and distance from rivers. The most important factor determining the occurrence of floods in the AHP model according to Souissi et al. (2020) was the elevation (22.5%); around 75% of the observed flood areas were located in moderate to very high elevation districts. Similarly, a flood susceptibility map generated using an AHP based on the elevation factor has previously shown that flood potential (moderate to very high classes) is generally concentrated in low-altitude areas (Hammami et al. 2019). The simplicity, flexibility, and ability to mix multi-criteria in the final decision framework make the AHP method much more popular (Pohekar & Ramachandran 2004).

Isolation of specific hydrologic events allows for a more detailed investigation (Bermúdez et al. 2017), such as storm duration, intensity, and distribution which can significantly influence the flood magnitudes and dynamics. In addition, analysing single storm events enables the collection of high-resolution data. These detailed datasets explain how the catchment responds to the rapid changes in hydrological conditions during the storm. However, analysing individual storm events allows us to identify the critical factors that trigger flash floods in arid catchments and determine the thresholds of rainfall and soil antecedent conditions that lead to flash flooding (Abushandi & Merkel 2013). Modelling individual storm events provides an opportunity to validate hydrological models and refine the sensitive parameters, making the models more reliable and accurate tools for predicting and managing flash floods in arid regions.

A combination of the Google Earth Engine (GEE) platform, the AHP, and GIS has been used to generate susceptibility maps (Swain Singha & Nayak 2020) showing that major floods may accumulate preferentially in zones with lower elevations. Lallam et al. (2018) applied the AHP to identify surface runoff coefficients based on three different criteria, namely soil type, slope, and vegetation cover. The Wadi Al Jizzi arid catchment has been experiencing flash floods almost every year, causing significant damage to infrastructure, human and animal losses, and erosion of fertile soils. This paper aims to assess flood magnitudes for a single storm event that occurred on 4 May 2021, using the AHP and IHACRES models.

A diver data logger for water level measurements was used to collect high-resolution temporal data at 10 min intervals with a memory capacity of 0.5 million records. These instruments are generally used for groundwater table monitoring rather than for the monitoring of surface water levels. The device required calibration and proper installation in the Wadi. The device was used to measure the water pressure column and therefore, the cross-sectional area of the Wadi was also measured to determine the water volume per unit time (flow).

Five parameters used in the AHP model affected the runoff volume, namely the slope, soil group, rainfall intensity, temperature, and land cover type. Accordingly, the comparison matrix was created based on several important evaluations defined as numbers. This step consisted of assigning weights, checks, and parameter balances to ensure that ratios were consistent when comparing the relative importance of the criteria. A matrix was created for all the factors affecting the flow as shown in Table 2. Each entry in the column was then divided by the column sum to yield its normalized score. The indices of each column were calculated, and the results are listed in Tables 3 and 4.

The values of Pearson correlation always range between −1 and 1, and if the relation between the observed and simulated records is weak, the correlation approaches zero.

The results obtained from the two models showed their ability to estimate a single flood event in arid areas, and the accuracy of the results ranged from satisfactory to very good. The IHACRES method has been successfully used for arid catchments, i.e. Jordan, Saudi Arabia, Oman, and Australia where the major advantage of using this model is the few data inputs it requires. However, the IHACRES model was not able to simulate flooding when rainfall was less than 1 mm due to a high infiltration rate. Therefore, as an initial stage to run the model, the rainfall records were divided into two categories: low (5 mm or less) and high (more than 5 mm). The calibrated parameters of the IHACRES model are summarized in Table 5.

Table 5

IHACRES input parameter values for the low and high cateogries

Low .		High .
Parameters .	Value .	Parameters .	Value .
	0.00680000		0.01000000
	2.500		1.000
	0.060		0.020
	0.800		0.030
	0.340		0.030
	0.570		0.700
	0.301000		0.291000
Area	870.6	Area	870.6
Delay	1	Delay	1

Low .		High .
Parameters .	Value .	Parameters .	Value .
	0.00680000		0.01000000
	2.500		1.000
	0.060		0.020
	0.800		0.030
	0.340		0.030
	0.570		0.700
	0.301000		0.291000
Area	870.6	Area	870.6
Delay	1	Delay	1

View Large

Based on further analyses, the IHACRES maximum flow rate was 2.09 m³/s while that obtained with the AHP model was 2.10 m³/s. Both models were able to capture the peak flow with a longer response delay for the AHP model.

Although the AHP model used more input data, the IHACRES performed better for this event. The main advantage of applying the AHP model is that it allows the consideration of numerous criteria based on pairwise comparisons for estimating the runoff volume. The AHP method was moreover highly efficient for modelling floods in arid areas based on multi-criteria decision-making (MCDM) tools. It was noted that the rainfall intensity and the type of soil had a clear effect on the surface runoff in the study area. In general, the results of the three efficiency methods (E_f, IoA, and KGE) depicted a better performance of high category rainfall events (Table 6).

A reliable assessment of short-term flood risk using the multi-criteria analysis model with the AHP-entropy method by Wu et al. (2022) showed similar storm spatial characteristics as the long flood risk with a correlation coefficient of more than 0.9.

According to the results obtained by Lallam et al. (2018) who investigated the effects of the AHP operative criteria, the vegetation cover had the greatest impact on runoff volume simulations. Generally, the AHP model is subjected to expert judgement of ranking importance while consistency measures are the reference to compare the selected criteria. Accordingly, the error propagation due to the expert judgement may have caused uncertainties in the model outputs (Fernández & Lutz 2010). Similarly, nature-based solutions (NBSs) are increasingly being utilized to mitigate flood risks in diverse areas, but the lack of quantitative assessment approaches limits their acceptance; however, NBS performance assessment provides valuable guidance for flood mitigation (Pugliese et al. 2021).

Peleg et al. (2022) studied high-resolution individual rainfall storms coupled with the future climate scenario. The spatial and temporal patterns of extreme rainfall storm will be increased by 20–30%.

Here, the application of hydrologic models using numerical quantification provided a better understanding of the behaviour of flood events. The results of both IHACRES and AHP models allowed us to predict a single storm event albeit with some limitations. Extreme rainfall magnitudes in the upper parts of the catchment were not recorded due to their topographic complexity. Mountains cover around 68% of the catchment area and, therefore, spatial rainfall distributions cannot be monitored by meteorological stations. However, the satellite images from TRMM used here indicated the presence of a stormy area extending from the western part of the catchment southward of the Wadi Al Jizzi catchment. The two models applied to the study area were able to reproduce the important hydrologic and geo-physical features associated with flood events in arid areas. The average flows determined using the IHACRES and AHP models were 0.47 and 0.45 m³/s, while the average observed flow was 0.55 m³/s using the AHP model. Efficiency measures provide a good tool for comparing model performances. In this paper, the three efficiency proxies (E_f, IoA, and KGE) concluded that the IHACRES model performed better in evaluating the single storm event. The proposed research framework is evolutionary and can be adapted to different climatic areas.

The study relies on limited datasets collected from three weather stations, which may not fully represent the region's climatic diversity. Expanding the data collection to include more meteorological stations and extending the study period could enhance the accuracy and representativeness of the results. Additionally, the study primarily focuses on temperature and rainfall trends, leaving room for further exploration of other climatic parameters and factors influencing extreme events, such as land use changes and socioeconomic drivers.

The authors are thankful to the Directorate General of Regional Municipalities and Water Resources in Al Batina North/Oman, particularly to the Surface Water Unit in Sohar for providing hydrologic data.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.