Hydrological Modelling of Extreme Events in Ouergha Mediterranean Basin, Northern Morocco, Using a Deterministic Model and Gridded Precipitations

Abstract


Introduction
Morocco, as a North African country highly vulnerable to the effects of climate change, faces a significant threat from extreme precipitation events (Tramblay et al., 2012).Since the beginning of the twenty-first century, the country has experienced several floods in various regions, notably in 2002 (El Jadida, Tétouan, Settat), 2009 (Rabat, Casablanca, Kenitra, Essaouira, Agadir), and the recent floods of 2021 that impacted Tangier (Bouramtane et al., 2021), these floods were attributed to the occurrence of extreme precipitations and high flow rates.According to the World Bank Group, 2018, Morocco encountered 13 floods between 2000 and 2013, leading to the tragic loss of 263 lives.One of the most significant inundations, occurred in the floodplain of the Sebou river (Al-Gharb) in 2009-2010, causing several fatalities and extensive damage to both infrastructure and property; the recovery from this episode was pricey; according to the Organisation for Economic Co-operation and Development (OECD, 2016) it costs nearly 80 million dollars; which emphasizes the importance of developing accurate flood prediction models to reduce the risk posed by extreme weather events in the country.
Given that river flooding is a global natural hazard with high costs and risks (Blöschl et al., 2019), hydrological models have been developed to anticipate and forecast floods of extreme rainfall events, employing different approaches (Beven et al., 1984;Crawdord and Linsley, 1966;Todini, 1996;Zhao, 1992).In this context; the Hydrological Engineering Center-Hydrological Modelling System (HEC-HMS) is a widely utilized hydrological modelling system designed to simulate the behavior of diverse catchments.Developed by the United States Army Corps of Engineers (Feldman, 2000), it enables continuous and event simulations using empirical or conceptual modules (USACE, 2015).HEC-HMS is a straightforward model that relates rainfall to runoff processes, and it has broad applications for solving various hydrological problems related to floods across large geographic areas (Natarajan and Radhakrishnan, 2019).It has been successfully applied in many contrasting climates and has proven its suitability and efficiency (Aqnouy et al., 2018;Aziz et al., 2020;Chathuranika et al., 2022;Guta, 2021;Jin et al., 2015;Sampath et al., 2015).Moreover, many studies have carried out hydrological modelling in Morocco using HEC-HMS for different purposes: Bouabid and Elalaoui, (2010) performed monthly hydrological modelling on the Ouergha Basin using the IHACRES and HEC-HMS models; Daide et al. (2021) simulated the surface runoff of the Beht Basin using six extreme daily events; Khaddor et al. (2016) established hydrological model for Kalaya watersheds located southeast of Tangier city in order to simulate rainfall-runoff and to predict peak discharge for different return periods.Msaddek and Garouani, (2021) modelled the impacts of vegetation cover changes on the hydrology in the upper Oum Er-Rbia Basin, using an assembly of land use maps from 1984 to 2016 and four different events.While in Laassilia et al. (2021) paper, a continuous hydrological model was applied to the Bouregreg watershed to assess the inflow into the Sidi Mohamed Ben Abdellah (SMBA) dam near to Rabat city.
This paper aims to contribute to these efforts by modelling the hydrological response of the Ouergha basin, particularly during extreme events, using HEC-HMS and Geographic Information System (GIS).The Ouergha basin is home to several dams, one of which is Morocco`s largest (Al Wahda dam); however, during successive high rainfall occurrences, the storage capacity of the dam becomes limited leading to the necessity of releasing water downstream, which culminates in floods and widespread devastation along the Sebou river and particularly in the floodplain.In this study, enhanced accuracy and representativeness of rainfall data were achieved by utilizing the GageInterp tool to interpolate observed daily precipitation, the tool was applied to rainfall measurements obtained from 11 rain gauge stations within the Ouergha basin.while the recorded discharges from 4 hydrometric stations were used for calibration and optimization of significant daily events that occurred between 2003 and 2010.The results were evaluated using four commonly used statistical metrics: Nash-Sutcliffe Efficiency (NSE), PBIAS (percent bias), R2 (R-squared), and RSR (standard deviation of the root mean square error).

Study Area
Situated in central northern Morocco (Fig. 1 a,b), the Ouergha River is considered the primary tributary of the Sebou River (Combe, 1975).The Ouergha river drains the southern part of the Rif mountains (Fig. 1b), covering a distance of 300 km along the Prerif, with approximately twenty kilometers at the entrance of the Al-Gharb plain.The present study specifically focuses on the flood genesis zone, situated upstream of the Al Wahda dem (Fig. 1c).This area constitutes more than 80% of the total basin area, encompassing 6190 km 2 , and extends across a geographical range from 34.379° to 35.139° North and from 3.906° to 5.371° West; the lower regions of the basin lie at an elevation of 145m, while the highest points reach up to 2450m (Fig. 1c).The basin features challenging topography with rugged lands; around 86% of the area has a slope above 12%, while only 3% has a lower slope of 3%.The majority of the basin is made up of clay soils, Mesozoic shale, and marl formations, this considerable proportion of clay soils and the presence of disaggregated layers of marl and shale result in the high impermeability of the watershed, leading to minimal retention capacity and significant runoff.
According to Senoussi et al. (1999), the basin's climate is Mediterranean, and it's Morocco's most humid region; therefore, the implementation in this area of Al Wahda Dam was a crucial strategic decision, serving as a significant and valuable asset for the region; with its storage capacity of 3.7 billion m 3 , it's the largest reservoir in Morocco and the second-largest in Africa.
Otherwise, precipitations are concentrated in the Ouergha basin from October to April, and it has a rainy hydrological regime with very strong winter flows.Extremely high discharge intensities were recorded at Ouergha between 2008 and 2010, during this time, the runoff frequently exceeded the 3000 m 3 /s threshold; according to Msatef et al. (2018), the return period estimated by the Gumbel law for flow rates greater than 2500 m 3 is above (T > 1000).

Software and Model Description
HEC-HMS is a deterministic hydrological modelling system that enables events or continuous hydrological simulations through the integration of various modules, it can function as either a lumped (semi-distributed) or a distributed model (Feldman, 2000).
In this paper, to estimate losses during rainfall events the Soil Conservation Services Curve Number (SCS-CN) method was chosen it was basically developed by the Natural Resources Conservation Services (NRCS) (Heedan et al., 2017).This method, which is widely described by USDA-SCS (1985), is well-suited for event simulation and takes into account the initial abstraction conditions of the watershed.The precipitations excess is defined through the integration of the initial abstraction (Ia) which is rounded as 0.2 S; S is the parameter of retention in mm; and the curve number (CN) that represents the potential runoff (Al-Kubaisi and Al-Kubaisi, 2022).
For the base flow, which is the portion of the runoff that originates from the groundwater, the recession model was selected.According to Scharffenberg and Fleming (2010), this approach is appropriate for catchments where precipitation occurrences have a significant impact on the flood volume.and it depends on the important parameter (Rc) Recession constant, which is a coefficient that describes the exponential decay of discharge or flow rate over time.
The Clark unit hydrograph was used to transform precipitations excess to direct runoff; by representing two processes; translation and mitigation.The translation is based on a synthetic time-area histogram with a time of concentration Tc (hours).Attenuation is modeled by a linear reservoir, representing the impact of basin storage St (hours) (Feldman, 2000) One commonly used flood routing model is the Muskingum method, known for its simplicity (Chu, 2009).Its most widely used forms are the continuity and nonlinear storage equations, as seen in various studies (Gill, 1978;Tung, 1985;Yoon and Padmanabhan, 1993) The Geospatial Hydrologic Modelling Extension (HEC-GeoHMS), which integrates the HEC-HMS into a GIS environment as described by Fleming and Doan (2009) is utilized to determine the necessary previous parameters for initiation of the simulation.This hydrological extension enables the extraction of input data, including the physical characteristics of the basin and the drainage network.
Since the crucial step in modelling is ensuring the accuracy of input data representing the formations, GageInterp is used to create a sequence of grids that approximate the spatio-temporal variations of the recorded precipitations data.The interpolation used by GageInterp which is the inverse distance weight method was found to provide good results in a comparison of several hydrological interpolation methods conducted by Tramblay et al., (2016).

2.3.1.Rainfall, Run-off and paired data
The precipitations data used in this paper was collected from 11 stations throughout the Ouergha basin (Fig. 1c), 4 of which also provided streamflow observations.The data covers the period from January 2003 to December 2010, which is statistically significant for the purposes of this study.Moreover, the percentage of missing data is minimal, as only three stations exhibited less than 2% of data gaps (Table 1).These gaps were filled using the inverse distance weighting (IDW) method, which is commonly employed in hydrology to estimate missing, as per previous research (Eischeid et al., 2000;Lee and Kang, 2015).Additionally, the function of each dam was also integrated, these functions (storage-discharge, or elevation-discharge) control the release of a specific flow at a specific storage rate, which prevents the dam from overflowing at the time of high discharges.Finally, all the hydrological and meteorological data were provided by The Direction of Research and Water Planning (DRWP).

Field data
Fig. 2a exhibits the raw field data used in the hydrological modelling of the Ouergha basin.In this study we used the Advanced Spaceborne Thermal Emission and Reflection Radiometer -Global Digital Elevation Model (ASTER GDEM) (Aster, 2009) (Fig. 2a).It has an approximate horizontal resolution of 30m and a vertical accuracy of GDEM ± (Aster, 2009).It can be downloaded freely from https://search.earthdata.nasa.gov/.The study area's land use (Fig. 2b) was created and verified by the hydrological agency of Sebou basin based on Landsat ETM+ (Enhanced Thematic Mapper), with a resolution of 30 m, and then it was simplified to represent forest, agricultural, water, and residential.
The soil maps in this paper are provided by the African soil grid 250 m project, which is an update of the original 1 km grid map created by the World Soil Information in collaboration with other institutes  (Fleming and Neary, 2004) https://www.isric.org/.Fig. 2c,d, and e; are three maps at the depth of the bedrock representing different proportions of soil (clay, silt, and sand respectively) in the study area.
Each field data has been processed using different approaches to extract and prepare the input data required for the initialization of the hydrological modelling.
The combination of HEC-geoHMS and GIS allows the sequential application of the following functions on the DEM: fill, flow direction, flow accumulation, stream definition, stream segmentation, catchment grid delineation, catchment polygon, drainage line, adjoint catchment processing, and drainage point (Martin and Ovcharovichova, 2012).Fig. 3a,b exhibit the most significant map relevant to this procedure; Namely, flow direction, and the delineation of the subbasins respectively, this last, is based on the drainage network and the localization of the flow gauge stations.Besides, Fig. 3c illustrates the grid cell map (needed only in the case of gridded data) produced using the grid cell processing tool within HEC-geoHMS.Whereas Fig. 3d represents the model scheme of the study area, it includes subbasins, reservoirs, and channels; each entity enables the recording of their previously defined morphometric properties.Then, each entity is connected to the next to simulate the reality in a simplified manner which corresponds to the lumped aspect of the HEC-HMS model.
Moreover, the flow rate depends on the types of hydrologic soil group (HSG) (Al-Sulttani and Beg, 2020); to define the HSG the soil maps processing using GIS together with the texture triangle allowed the definition of the main soil textures of the Ouergha basin (Fig. 4a) (clay; clay loam; loam; and sandy clay loam).Also, the soil texture map was processed; thus, each texture zone is grouped into specific hydrological soil groups HSG (Fig. 4b); this classification is based on a minimum rate of infiltration obtained for bare soil after prolonged wetting (USDA, 1986).The groups are namely A, B, C, and D, whereas the A soil group is the most permeable ( sand, loamy sand, or sandy loam), and the D soil group is the most impermeable (clay, clay loam, etc.).More details about the proportions of sand and clay and the conductivity of each class are available on (USDA, 2009) In addition to the infiltration rate of each soil, the peaks of discharge depend also on the vegetation type and density, land cover, and land treatment.The specific data extracted from the land use map can be combined with the soil texture data in order to generate the curve number map that defines soil's capacity for runoff (Fleming and Doan, 2009).In this study, the land use map had previously been generated and processed, it required only a classification of the entities into specific categories: forests, agriculture, water, and residential using GIS (Fig. 4c).
The data derived from the union of hydrologic soil groups (HSG) and land use classes are called "hydrologic soil cover complex", and it's used to establish the curve number map.According to the study conducted by "Natural Resources Conservation Service" NRCS and detailed in (USDA- SCS, 1985), the soil conservation system (SCS) curve number method has been developed based on empirical equations, and each hydrological soil cover complex has been assigned to a runoff curve number defining the runoff potential of each soil.
The technical release of the USDA, ( 1986) study provides tables of curve number values for diverse hydrological soil groups versus land use combinations.
In the context of the Ouergha basin, Table 2 depicts the CN Look-Up table produced with the union of the attributes of the HSG and the land use maps.Hence, Fig. 4d represents the CN map of the Ouergha basin, generated by HEC-geoHMS using DEM, the union of the HSG map, the land use map, and finally the CN lookup table function.It is important to highlight that the curve number is dimensionless and varies from 0 to 100, where 0 represents insignificant runoff potential, and 100 represents maximum runoff potential, which is inversely proportional to the infiltration potential.Otherwise, using GIS and the grid cell map previously generated, the curve number map is transformed into the gridded Curve number map illustrated in Fig. 4e.which exhibits that the study area's potential runoff ranges from 58 to 97, increasing from the east to the west.

Calibration and validation process
Calibration is a key step in successful hydrological modelling.It's the process of optimizing the most sensitive parameters to reproduce the hydrological hydrograph as accurately as possible, by reducing the gap between simulated and observed discharge in terms of intensity, flow rate, volume, and time to peak.
The sensitivity analysis is the evaluation of the sensitivity of the model's performance according to each different parameter (Hall et al., 2009;Sieber and Uhlenbrook, 2005;Tang et al., 2007).
There are two different types of calibration, the automatic proposed by HEC-HMS, and the manual calibration.The automatic algorithm is based on an optimization of defined sensitive parameters using a specific algorithm within HEC HMS, in order to reduce an objective function, such as percent error in peak, peak-weighted RMSE, the sum of absolute errors, or the sum of squared residuals.
In this study, both simple and complex events across the target period 2003-2010 were selected for modelling purposes; five of them were initially calibrated manually to identify and optimize sensitive parameters that impact the fitting between observed and simulated hydrographs; then, to improve the goodness-of-fit indices the automatic algorithm of calibration within HEC-HSM was carried out.It's noteworthy that the calibration performed in this study concerned several aspects of the simulation quality, including the surface runoff volume, the center of mass, as well as the peak discharge value and time.
Further, to assess the accuracy of the developed model in reproducing the hydrological behavior of the Ouergha basin during rainfall events, the optimal parameters were set and tested on an additional five events.

Evaluation metrics
The accuracy of a model to reproduce the hydrological behaviors of watersheds is judged according to statistical criteria dedicated to this kind of modelling; thus, to quantify the level of correspondence between the observed and simulated flow the following metrics were calculated.
The coefficient of determination R 2 , which is used to assess the degree of agreement between simulated and observed runoff (Moriasi et al., 2007).
Root Mean Square Error (RMSE) observations, standard deviation ratio (RSR), it's known that the lower RMSE is, the better the model performance, but to qualify what is considered to be a low RMSE, (Singh et al., 2004) have published a guideline, analyzing the RMSE performance on the basis of the observation's standard deviation.Thus, RSR is calculated as the ratio of the RMSE and standard deviation of measured data.Percent bias (PBIAS) is a statistical indicator commonly used in hydrology to assess the systematic tendency of simulated data to overestimate or underestimate the observed values (Gupta et al., 1999).

CN Lookup
Nash-Sutcliff (NSE), Hydrologists commonly use this parameter as an objective function, according to Servat and Dezetter (1991), NSE is the optimal objective function for providing a comprehensive assessment of the overall goodness-of-fit of a hydrograph.It assesses the predictive accuracy of not only the calibrated set of parameters but also of the entire hydrological model built for the studied basin.A hydrographic model is deemed powerful only when the estimated flow matches the observed flow (Kouassi et al., 2013), and thus the Nash criterion approaches 1.
Each metric has its own performance intervals, which vary from one metric to another.Table 3 displays the performance rating of each statistical metric.
Table 3. Performance measurements for stream flow simulation (Moriasi et al., 2015) To summarize and schematize the general methodology adopted for the purpose of this study, all the processing and the application of raw data are illustrated in Fig. 5.

Model Calibration
Initially, the highest impact on the model's performance parameters were identified and then adjusted.The curve number CN, the time of concentration Tc, and the recession constant Rc were found to have the greatest influence on the results.CN influences the amount of surface run-off volume (Silva et al., 2015); while Tc, and Rc, increase the concordance between the simulated and observed hydrographs.between the simulated and observed hydrograph.For the curve number, its minimum value has increased by 20 percent Table 4 while the maximum value has increased by 7.23 percent.

Table 4. Ratings of the recommended parameters for stream flow simulation
Compared to the main parameters, the curve number has undergone a relatively minor change, but it has had a significant impact, and this is due to the hypersensitivity of the CN in this study.This finding is consistent with the lithological aspect of the Ouergha watershed; the very high CN refers to the marly lands that dominate the basin.On the other hand, Tc has known the most important adjusting: its minimal and maximal initial values have decreased by 80.91% and 30%, respectively; this is mainly due to the fact that estimating Tc is based on empirical methods and the difficulty of adopting the right calculation equation; it's also due to the rugged topography of the basin and its significant slope.The calibration has also impacted the Rc initial values, with changes of 50% and 10% for the min and max initial values, respectively.
Further, the daily rainfall events listed in Table 5 were calibrated.They are distinguished by their peak discharge values that range from moderate (691 m 3 /s) to extremely high (2529.6m 3 /s); events 1, 2, 3, and 4 are characterized by a single flow wave, while event 5 is two merged rainfall occurrences leading to nested flow waves where the first is pronounced while the second is attenuated.Fig. 6. presents the graphical results of the calibration process, which demonstrate the model's ability to accurately reproduce the basin's hydrological behavior for most calibration events.
The results from the scatter plots align with the graphical observations, and the general pattern of the observed hydrograph is well simulated, although there may be slight differences in timing and peak discharge values.
The statistical details of the calibration results are presented in Table 6, providing further insight into the accuracy of the model.
Event 1: is characterized by a high level of agreement between the simulated and observed discharge, with R 2 exceeding 0.73, and a low flow residual variation RSR <0.6, however, the high bias of 22% has impacted the quality of the simulation.Therefore, an NSE of 0.60 is deemed satisfactory.Event 2: the agreement in this simulation is excellent with R 2 0.98, the residual variation is rather low with RSR<0.6, and NSE is deemed good with a value of 0.69, meanwhile the bias shows a discharge overestimation that exceeds 25 %.
Event 3: This simulation has an unsignificant discharges gap with Pbias of 0.63; however, the correlation, residuals, and Nash indices are considered satisfactory at 0.66, 0.61, and 0.60, respectively; we assume that the one-day lag, shown graphically in Fig. 6, is the main reason for the simulation's performance.
Event 4: With an excellent agreement between simulated and observed flow R 2 of 0.91, a negligible residuals variation RSR of 0.36, and an unsignificant discharge bias, Pbias = 3.9, this simulation is deemed to be the most accurate, resulting in an NSE of 0.85.
Event 5: With its two nested waves and notably the underestimation of the first one, the fifth event is the most complex simulation.The implementation of multiple calibration processes resulted in a high degree of concordance between the simulated and observed flow, R 2 of 0.82, a low residual variation RSR, of 0.52; as well as a good NSE of 0.69, although its graphical results remain less relevant than those of the previous events.To assess the overall performance of this calibration, an average of all indices was calculated; so on the basis of this finding: the NSE indicating a good overall performance (0.69), the excellent correlation between the simulated and calibrated hydrographs with an R 2 of 0.82, as well as the residual variation that remains below its lowest threshold, the performance of this calibration is rated as good to excellent.Fig. 7 displays the graphical illustration of the performance of each calibrated simulation: it indicates that the best performance belongs to event 4, while the overall performance is rated as good to excellent since 50% of the metrics show very good results, another 25% show good results, and only 25% show satisfactory; this last refers to the bias index that highlights either an overestimation or an underestimate impacting the overall performance.In fact, Boyle et al. (2000) specify that during model calibration, an optimization based on the reduction of RMSE (decrease of the residual variance) might result in a minimum error variance at the detriment of a significant bias.

Model Validation
Five occurrences listed in Table 7 were carefully selected to validate the model, the first one is a common event in this study area, and the remaining ones are extremely exceptional events.Event 6 is defined by a singular wave with a peak discharge that does not exceed 1100 m 3 /s.In contrast, events 7, 8, 9, and 10 are characterized by a significantly higher peak discharge ranging from 2908.7 m 3 /s to 3361.7 m 3 /s, which represents a millenary event that resulted in inundations in the Sebou floodplain.These events are characterized by their extensive or prolonged rainfall episodes with multiple nested waves of different amplitudes.
The graphical results obtained from the validation process are exhibited in Fig. 8; While the statistical details of the validation results are presented in Table 8 which provides further insight into the accuracy of the model.
Event 6 represents an ideal simulation case where the two curves (observed and simulated) are almost in full agreement, with the same peak rate and a marginal variation in the discharge amount.The statistical findings support the graphical evaluation mentioned above.With an excellent agreement between the observed and simulated discharge of 0.92, a negligible residual variation and bias of less than 0.26 and 3.65, respectively, and an NSE of 0.92, this event is deemed to be the ideal case for this study Event 7 is the second-best graphical validation result; it's characterized by a pronounced wave with a substantial flow rate exceeding 2900 m 3 /s.There are only a few negligible graphical differences between the two peak flow rates, only the computed hydrograph was one day behind the observed one.The statistical analysis of the simulation results shows similarities with the calibration events, as there is a difference between the correlation, residual variation, and NSE, which have very good rates of 0.92, 0.41, and 0.81, respectively, and the Pbias, which shows a significant overestimation of the simulated runoff up to 33.56% revealed by the scatterplot.Event 8 which consists of multiple nested waves and a high peak discharge, was the most complex simulation in this study.The rapid rise of the observed hydrograph to its highest peak discharge (2943 m 3 /s) is attributed to heavy daily precipitations over a short duration.Despite the model's incorporation of physical soil properties and its ability to infiltrate water, the first wave was poorly reproduced, with a significant difference in both the time to peak and discharge rate between the observed and simulated results.This can be attributed to the soil's poor permeability, which reduces its ability to absorb water when heavy rainfall occurs over a brief period.However, the model improved its performance for the December 15th wave as it became aware that the soil was saturated from the previous episode; thus the simulated and observed flow peaks occurred on the same day; and the gaps between the two volumes decreased remarkably.The statistical results of this complex event are deemed acceptable.The correlation between the simulated and observed discharge of 0.62 is considered satisfactory.The residual variation and bias of 0.67 and 23% respectively, can be considered acceptable.The NSE with 0.51 is rated as average.
In event 9 an exceptional discharge of over 3000 m 3 /s was observed; it starts with a high value of 900 m 3 /s, while the simulated discharge shows ordinary base flow values.Despite a one-day lag and an underestimation of the surface runoff volume, the shape of the observed hydrograph is well captured.But although the pattern of the event was captured by the model, the performance was mediocre due to the one-day lag and overestimation of discharge values.Despite this, the correlation between the observed and simulated discharges, as indicated by the R 2 of 0.75, was good.However, the bias of 30 % and residual variation of 0.67 were noticeable, and the Nash index of 0.54 was deemed acceptable.Overall, this simulation can still be considered acceptable, and the result could be explained by either a substantial initial abstraction from a previous event or an unplanned release of a dam due to the unusual rainfall depth in 2010.The final simulation event in the study is identified with two phases: an accurate rising limb and an overestimated recession limb.The simulated discharge pattern is deemed to be more logical, as the recorded rainfall data does not indicate an increase in intensity that would explain the nested peak in the observed discharge on December 2nd, 2010.It is possible that the nested peak in the observed discharge data was due to an anomaly in the data recorded on December 1st.However, the agreement was found to be very good, as indicated by an R 2 of 0.89.The residual variation was also low, with an RSR slightly exceeding the threshold of 0.5.Nevertheless, the overestimation in the recession limb was expressed by a Pbias metric exceeding the 25% threshold, and the Nash index was deemed satisfactory with a value of 0.63.Taking into account the overall performance of NSE and the strong correlation, this simulation can be judged as satisfactory.
The overall performance of the validation process is referred to the average of all the previous results, the strong correlation in the calibration phase, as reflected by an R 2 value of 0.81, has been

Event
Flow peak m sustained during the validation phase.This indicates that the model effectively captures the pattern of the observed discharge.Additionally, the residual variation showed a minor increase, with an RSR value of 0.51 compared to 0.50 in the calibration phase.However, the bias index experienced a significant increase from 19.34 in the calibration phase to 27.27 in the validation phase.This can primarily be attributed to the exceptional nature of the runoff events during the validation phase, which were rated as uncommon in the study area.It may also be due to unexpected releases from the dam during exceptional events 8 and 9, which did not align with the intended function of the dam.As a result, the model's performance with regard to the bias index did not meet our expectations.Finally, the Nash-Sutcliffe efficiency index, which provides a comprehensive view of the overall performance, remained consistent with the calibration phase.With an NSE of 0.69, the overall performance of the validation phase can be classified as "good".This is supported by Fig. 9, which indicates that 50% of the metrics validate the performance of the model and classify it as "good", and 20% classify it as "excellent".The same Figure indicates that the best performance belongs to event 6.Given the context of climate change, the occurrence of floods has emerged as a pressing issue owing to the mounting frequency and magnitude of extreme precipitation events.In Morocco, several studies indicate a trend in the frequency and intensity of extreme rainfalls.According to Benabdelouahab et al., (2020), there is an annual trend of accumulating daily precipitation that is exceptionally concentrated in time and space, which leads to floods and soil degradation.Also, Bouizrou et al., (2022) found that highlands with a Mediterranean climate (Which is the case of the Ouergha basin) are prone to extreme precipitations and thus flooding events.Indeed, after dry conditions, the Ouergha basin experienced a series of extreme precipitation episodes during the hydrological year of 2009-2010, with a cumulative count of rainy days exceeding 80 (Alanord et al., 2017), this resulted in a significant runoff and severe flooding.Moreover, the predominant composition of clay loam soils in the Ouergha basin played a significant role in the hydrological dynamics observed during this period.These soils have limited infiltration capacity, leading to rapid saturation and a high volume of runoff.
The hydrological modelling conducted using the HEC-HMS model has provided valuable insights into the hydrological processes which permit the anticipation of water flow dynamics.This modelling approach allows the simulation of flow patterns and the detection of anomalies in water discharge.However, underestimations in the simulated flow were observed during specific episodes, coinciding with controlled releases from the reservoirs.These underestimations may be attributed to the complexity of accurately representing the interactions between reservoir operations and the natural flow regime.
In fact, reservoirs in the Ouergha basin were already nearing their maximum storage capacity due to the cumulative effect of the rainy season.Thus, to prevent overtopping and mitigate the risk of dam failure, unscheduled releases through the spillway became necessary.While these releases aimed to maintain the reservoir levels within a safe operating range, balancing the inflow from the intense precipitation events, they also resulted in floods in Al-Gharb plain, causing fatalities and economic damages.Similar situations have been reported in other studies around the world, such as in Malaysia (Ishak and Hashim, 2018), India (Mishra et al., 2018), and Japan (Nohara et al., 2022), often due to reduced flood control capability of the reservoir.
The complexity of reservoir flood control originates from the need to consider the safety of upstream catchments, the reservoir itself, and downstream catchments.Consequently, determining the operation rule for flood control in reservoirs becomes a multi-objective decision-making process (Hsu et al., 2015), involving the balancing of dam and surrounding area safety, flood control management, and effective utilization of water resources (Ding et al., 2015;Lei et al., 2018).Several studies have proposed optimization methods to address this challenge, including pre-release models to enhance flood control storage capacity at the early stage of floods, multi-objective optimization models to optimize flood control operating rules considering resilience and risk objectives, and fuzzy optimum selecting models to determine operating rules that satisfy forecast uncertainty and decision-makers preferences (Ding et al., 2023;Wei et al., 2022).

Conclusions
The results of the hydrological modelling of the Ouergha basin using HEC-HMS and gridded precipitations have been found to be satisfactory.However, the simulation of the basin's hydrological response to different events reveals two distinct behaviors.The model accurately reproduces events with a maximum discharge rate of less than 1500 m 3 /s and gives good to excellent results.But, in the case of heavy precipitations events with prolonged episodes and multiple hydrographs, although the observed pattern is well captured, there is an underestimation of the initial hydrograph, causing a difference between the measured and calculated runoff; which leads to an increase in the bias index; however, the subsequent hydrographs are modeled more accurately even for high discharge rates.This underestimation could be due to the unscheduled releases of the upstream dams, heavy precipitations over prolonged episodes, or an imprecise calculation of the basin's initial abstraction.
To improve the model's performance for such events, it is recommended to simulate and calibrate hypothetical heavy rainfall and the associated release from upstream dams.Overall, these findings suggest that the model has the potential to predict the hydrological behavior of the Ouergha basin using forecasted and validated precipitation data, which can help in identifying potential floods.Furthermore, considering the impact of climate change, which is marked by increased frequency and intensity of precipitations; often occurring in irregular patterns over time; It is essential to improve the management of water volumes received by dams, which involves optimizing storage capacity and implementing proactive measures to address potential flood occurrences.

Fig. 5 .
Fig. 5.A summary of the methodology

Fig. 6 .
Fig. 6.Graphical results of daily rainfall-runoff calibration events.Correlation scatterplots of each calibrated event

Fig. 7 .
Fig. 7. Graphical representation of statistical performance evaluation of each calibrated simulation

Fig. 8 .
Fig. 8. Graphical results of daily rainfall-runoff validation events.Correlation scatterplots of each validation event

Fig. 9 .
Fig. 9. Graphical representation of statistical performance evaluation for each validation event

Table 1 .
Summary, and characteristics of Ouergha's stations

Table 2 .
CN Lookup table function

Table 5 .
Description of the calibration events

Table 6 .
Statistical assessment and performance of the calibration results

Table 7 .
Description of the validation events

Table 8 .
Statistical assessment and performance of the validation results