Impact of Flood Control Systems on the Probability Distribution of Floods

. Detention dams are one of the most effective practices for flood mitigation. Therefore, the impact of these structures on the basin hydrological response is critical for flood management and the design of flood control structures. With the aim to provide a mathematical framework to interpret the effect of flow control systems on river basin dynamics, the functional relationship between inflows and outflows is investigated and derived in a closed-form. This allowed the definition of a 10 theoretically derived probability distribution of the peak outflows from in-line detention basins. The model has been derived assuming a rectangular hydrograph shape with a fixed duration, and a random flood peak. In the present study, the undisturbed flood distribution is assumed to be Gumbel distributed, but the proposed mathematical formulation can be extended to any other flood-peak probability distribution. A sensitivity analysis of parameters highlighted the influence of detention basin capacity and rainfall event duration on flood mitigation on the probability distribution of the peak outflows. The mathematical 15 framework has been tested using for comparison a Monte Carlo simulation where most of the simplified assumptions used to describe the dam behaviours are removed. This allowed to demonstrate that the proposed formulation is reliable for small river basins characterized by an impulsive response. The new approach for the quantification of flood peaks in river basins characterised by the presence of artificial detention basins can be used to improve existing flood mitigation practices, support the design of flood control systems and flood risk analyses. proposed be to any flood-peak probability distribution. of the are investigated under are tested with a numerical simulation of flood mitigation. This method can be applied to improve existing flood mitigation approaches, support the design of flood control systems, flood risk and damage analyses. explore the behaviour of the proposed formulation, we investigated the effects of different parametrization on the derived distribution starting from a single distribution of floods. In Figure 3, the influence of the storage capacity and hydrograph duration on the outflow of the reservoir is shown. We depicted the probability density functions (pdfs) associated with increasing storage capacity obtained by raising the crest level from two meters up to eight meters (moving top-down in 195 the figure) and considering two distinct equivalent event durations of 30 minutes and one hour. As expected, hydrographs with larger duration tend to saturate sooner the water storage capacity of the reservoir, reducing also the lamination effects. On the other hand, the increase in water storage capacity leads to a proportional growth of flood peak mitigation. These graphs describe the behaviour of a reservoir providing an output consistent with the dynamics of the process.

, which now requires renewed methods and techniques. Therefore, the need to find appropriate 30 measures for flood hazard prevention and mitigation is becoming more and more pressing.
Flood risk management is a complex challenge for hydrologists that need to identify dynamic solutions for flood mitigation. Plate (2002) underlined the need to update flood mitigation plans according to the changing conditions (e.g., climate, populations, land use). Hence, the design of flood mitigation systems requires rapid methodologies to evaluate scenarios and intervention measures and obtain effective flood risk management strategies (Franzi et al., 2016). 35 One of the most effective action for the attenuation of peak discharges is represented by a detention basin system or coordinated reservoir operations (e.g., UDFCD, 2016;Jacob et al., 2019;Seibert et al., 2014). The proper design of a detention basin is an extremely complex task given the large number of factors that are involved in the decision process: technical engineering features of the structure, site selection and environmental characteristics. For this reason, Bellu et al. (2016) proposed an innovative method to optimize the dimensioning and site selection of a flood mitigation system. The method follows three 40 steps that include a preliminary sizing, a site location and optimization according to environmental objectives. Nevertheless, the sizing of the detention volume is based, in most of the cases, on a given design flood event neglecting the random nature of the phenomena and the impact of the structure on a large spectrum of events.
The flood attenuation caused by the presence of artificial reservoirs is influenced by hydrologic and hydraulic factors, such as flood wave shape and duration, as well as by the storage capacity and geometric parameters of the detention basin. Therefore, 45 it is crucial to build a mathematical scheme able to interpret the functional relationships, even in a simplified form, among the mentioned variables.
With this aim, the present manuscript introduces a Theoretically Derived probability Distribution (TDD) of detention basins outflows, which is obtained assuming the incoming flood peaks randomly distributed and characterized by rectangular hydrographs of fixed duration (see i.e., Manfreda et al., 2018). In the present case, flood distribution is assumed to be Gumbel 50 for the sake of simplicity, but the proposed approach can be applied to any flood-peak probability distribution. Performances of the proposed method are investigated under different configurations and are tested with a numerical simulation of flood mitigation. This method can be applied to improve existing flood mitigation approaches, support the design of flood control systems, flood risk and damage analyses. This paper is organized as follow: in Section 2, the conceptual scheme of the detention basin adopted to derive the mathematical 55 formulation of outflows is described. Furthermore, the hydraulic concepts and assumptions to analytically compute the derived probability distribution of the peak outflows are introduced; in Section 3, the proposed methodology is tested under different reservoir configurations and compared with the results of the numerical simulations and a brief description of the numerical model is also provided; and in Section 4 main findings and results are discussed.

The conceptual scheme 60
In-line detention dams (also known as flow-through dams) are constructed solely with the purpose of flood control and mitigation of flood risks in downstream communities and ecosystems. Unlike reservoir dams, which are primarily built for water storage or power generation, the spillway (opening) is located at the same height as the riverbed level, allowing the river to continue its natural flow under normal conditions. When water levels rise above the spillway, the dam restricts the amount flowing through the opening, decreasing peak flow. Since detention dams minimally affect rivers natural flows, under normal 65 conditions negative environmental and socioeconomic impacts, such as sediment accumulation, restriction of water flow to downstream communities and ecosystems, and breaching during very extreme flood events, can be minimized or avoided altogether.
The schematization of the detention dam has been simplified with the aim to obtain a mathematical description of the outflows associated with a given hydrograph. In particular, we assumed that the dam body has two openings: a low-level opening at the 70 basement and a crest spillway. The former is assumed to let pass the flow below a given control value, while the latter starts functioning only when the volume of the dam is completely filled up to the crest level. Thereafter, the water starts to flow from both the openings, and the control is mainly exerted by the basin volume above the emergency spillway (crest level). This scheme can be described in closed form that may help the construction of a derived probability distribution of the outflow from a detention dam. 75

The hydraulic characteristics of the problem
Let's first introduce the key equations controlling the dynamics of a system like the one under study. The first equation to introduce is represented by the stage-storage capacity curve which is able to describe the morphology of the gorge closed by a specific dam. The function is generally represented by a power-law: where W(h) [L 3 ] is the water storage of detention basin, 1 [L 3-n ] is the parameter of the stage-storage capacity curve, h [L] is the water level in the reservoir and n [-] is the exponent influenced by the shape of the control volume. The exponent ranges between 1 and 4.5, where 1 is associated with a prismatic geometry with vertical surrounding walls and 4.5 is associated with a more complex morphology closed by more gentle lateral slopes.
The streamflow of the river system reaching the reservoir is altered by it based on the hydraulic characteristics of the dam. The 85 outflow will be controlled by the amount of water accumulated in the system according to the continuity equation where ( ) is the incoming flux and ( ) is the outflow from the reservoir.
The outflow can be computed using the traditional formulation of hydraulics based on the variation of the water level, h, in the reservoir, which can be derived from the stage-storage capacity curve (Equation (1)) and continuity equation (Equation (2)). 90 In particular, we can assume that the outflow for a simple scheme with two openings (the low-level opening and the emergency https://doi.org/10.5194/hess-2021-47 Preprint. Discussion started: 22 February 2021 c Author(s) 2021. CC BY 4.0 License. spillway) and the characteristics described in Figure 1 may vary according to four different discharge laws expressed as a function of the reservoir water level stage. The outflow can be described as follow: where, [-] is the coefficient of discharge of the low-level opening at the basement (suggested values may range between 0.6 95 (assuming a thin sharp-edged) and 0.8 (assuming a wall with thickness two times larger than the opening height), of the spillway crest, and g [LT -2 ] is the acceleration due to gravity.
The above expression includes the three main configurations that may occur in the proposed scheme with the increase of the water level stage in the reservoir. A detailed description of the scheme proposed along with graphical indication of the parameters meaning is given in Figure 1. In particular, the opening at the bottom does not exert a significant control on the incoming flow as long as the water level does not generate a submergence of the opening (i.e., as long as ℎ ≤ ℎ ). The water 105 flow starts to be limited when the water stage reaches the value hf, after which the opening is submerged. This allows to fill the storage volume of the reservoir up to the level of the crest spillway. After this stage, the reservoir tends to operate a mitigation that is influenced by the water storage capacity of the reservoir above the crest level and the hydraulic characteristics of the spillway. An example of flood mitigation obtained via numerical simulation is given in Figure 2 which provides a comparison between a synthetic hydrograph and the outflow from a detention dam. 110 For the scope of the present study, the outflow has been simplified through the following set of equations: where is the control value of discharge that is computed using the discharge equation of the submerged opening and assuming ℎ = ℎ .
Assuming a rectangular hydrograph of the incoming flow, it is possible to derive the peak flow associated with an incoming 115 flood peak. Following the simplifying assumption given in Equation (4), the outflow is not affected by the presence of the dam for lower streamflow values, while it is modified when the inflow exceeds the control discharge. In particular, the outflow remains almost constant as long as the reservoir is filled and, thereafter, it is controlled by the crest spillway. In this last https://doi.org/10.5194/hess-2021-47 Preprint. Discussion started: 22 February 2021 c Author(s) 2021. CC BY 4.0 License. configuration, it is possible to use the linear reservoir concept for the water volume accumulated above the elevation of the crest spillway. 120 In order to estimate the peak flow associated with a specific rectangular hydrograph of constant discharge equal to , we should recall the expression of the peak flow generated by a simple linear reservoir, which can be described as follow: where [T] is the event duration, and [T] is the delay constant of the conceptual linear reservoir.
The above equation should be modified, in the present case, considering that the flood event should fill the dam water storage 125 capacity (or detention basin) before reaching the crest spillway. Therefore, the crest spillway will be activated only after a time where is the volume of water accumulated in the dam at the crest level ℎ .
When the volume below the spillway crest is totally filled, the crest spillway starts functioning for discharge values above the control discharge that is released at the bottom. Therefore, the peak outflow, , , assumes the following form: 130 where [T] is the equivalent delay constant of the conceptual reservoir associated with the outflow. This parameter can be derived exploiting the characteristics of the spillway and the stage-storage capacity curve. In particular, according to the linearity concept, the two functions should have the same exponent. Under such a hypothesis, the parameter can be estimated as: 135 where 2 is the coefficient of the rescaled stage-storage capacity curve above the crest level. Such parameter should be computed in order to get the best approximation of the function describing the volumes above the mentioned level, ℎ , imposing a coefficient n equal to 1.5 for the rescaled stage-storage capacity curve (referred to the stage-storage capacity curve above the crest level). With this aim, the parameter can be computed comparing the two functions and setting that they are 140 equal in a point ℎ which is representative of the range of variability of the water level above the crest level. [8] Within the present manuscript, we assumed the parameter ℎ was set equal to ℎ .
These assumptions allowed to derive a functional relationship between the reservoir inflow and outflow, which can be used to invert the function respect to the incoming flow and associate a probability to each flow discharged value by exploiting the 145 theory of derived distributions (Benjamin and Cornell, 2014).

Estimation of the event duration
The assumption of a rectangular hydrograph may produce a significant overestimation of the flood volume. Therefore, the parameter should be defined accounting for the real volume associated with a realistic flood hydrograph. In this contest, we can rely on the Flow Duration Frequency Reduction curve (FDF) proposed by the NERC (Natural Environment Research 150 Council, 1975), which describes the maximum average discharge ( ) as a function of the event duration D: where ω represents the characterising basin time response that is frequently associated with the lag-time of the river basin.
Adopting the above formulation, Fiorentino (1985) suggested a simple form of synthetic hydrograph redistributing the volume symmetrically respect to the time of the peak. This leads to the following form of hydrograph 155 [10] Based on the above formulation, it is possible to impose that the duration of the equivalent rectangular event has the same volume of the synthetic hydrograph of Equation (10) in the temporal window of around the peak flow. This led to the following equivalence: ≅ 0.632 .
[11] 160 Given the above assumption, the term will be named equivalent event duration from now on.

Functional relationship between the incoming discharge and the outflow
The possibility to identify the analytical relationship between two processes where one represents the stochastic forcing allows to determine the derived probability distribution of the variable. This approach has been used several times for flood maxima (Eagleson, 1972;De Michele and Salvadori, 2002;Gioia et al., 2008); soil moisture (Rodriguez-Iturbe and Porporato, 2007;165 Manfreda and Fiorentino, 2008); and scour process (Manfreda et al., 2018). In the present case, the methodology has been applied to the laminated flood peak.
Given the above approximations, it is possible to derive the inverse function of the peak discharge function of the peak outflow.
This equation can be obtained mathematically inverting Equation (6)  . [12] This expression can be used to analytically compute the derived probability distribution of the peak outflow from a detention dam characterized by a storage capacity , an equivalent delay constant , invested by flood hydrograph of equivalent event duration . With this aim, any probability distribution of the flood peaks can be used given the monotonic nature of the 175 above expression. The expression of the TDD will be (see Benjamin and Cornell, 2014): where the derivative of −1 ( , ) assumes the following form: . [14] In order to describe the probability distribution of the outflows, we should divide it according to the three potential 180 configurations of the detention dams: 1) undisturbed flow; 2) accumulation of water in the reservoir; 3) activation of the crest spillway. Based on these assumptions, the probability distribution of the outflow is subdivided into three components and modelled by the following set of equations: . Assuming that the lower opening is able to control the outflow around after submergence, there is a mass probability in depending on the storage volume of the reservoir ( , = ). After these two phases, the outflow is affected by the lamination due to the water volume accumulation above the crest level ( , > ).

Examples of application of the TDD of the detention basin outflows under different configurations
In order to explore the behaviour of the proposed formulation, we investigated the effects of different parametrization on the derived distribution starting from a single distribution of floods. In Figure 3, the influence of the storage capacity and hydrograph duration on the outflow of the reservoir is shown. We depicted the probability density functions (pdfs) associated with increasing storage capacity obtained by raising the crest level from two meters up to eight meters (moving top-down in 195 the figure) and considering two distinct equivalent event durations of 30 minutes and one hour. As expected, hydrographs with larger duration tend to saturate sooner the water storage capacity of the reservoir, reducing also the lamination effects. On the other hand, the increase in water storage capacity leads to a proportional growth of flood peak mitigation. These graphs describe the behaviour of a reservoir providing an output consistent with the dynamics of the process. In Figure 4, we modified the coefficient of the stage-storage capacity curve, 1 , using the values of 5000 and 10000 and 200 explored equivalent event durations ranging from half an hour to two hours. Graphs display how the pdfs of outflows are altered by the presence of a dam with these characteristics. In the present example, the impact of the lower opening can be better appreciated with a mass probability around that is equal to 52m 3 /s. It must be clarified that the cross-section of the opening has been increased in this second example on purpose to emphasize its impact on the proposed mathematical scheme.

Testing the reliability of the proposed method 205
In the current work, a numerical simulation of flood mitigation through a detention basin with the characteristics reported in Figure 1 was carried out using the same forcing adopted for the proposed TDD. Therefore, we adopted the Gumbel distribution as reference distribution to generate random values of discharge and numerically simulate the dynamics of the detention dam and its peak outflows. It must be clarified that any probability distribution of floods (e.g., generalised extreme value, threeparameter log-normal, generalised logistic and Gumbel distributions) can be applied. An example of the numerical simulation 210 is given in Figure 2 with the consequent attenuation of the hydrograph due to the detention dam.
The numerical simulation was carried out with the main scope of testing the theoretically derived probability distribution of laminated peak flows and also quantify the impact of the approximations adopted to obtain a closed-form of the solution.
Therefore, the comparison of the theoretically derived distribution and numerical outflows helps understanding the reliability of the proposed methodology. Results are given in the following graphs. 215 Figure 5 provides a comparison of different pdfs obtained modifying the maximum water storage capacity of the dam and its height. It can be noted that the values of the probability distribution replicate fairly well those obtained with the numerical simulations. The approximation of a fixed discharge from the submerged opening induces a small dispersion of values around the control value of discharge, , that the theoretical probability distribution is not able to capture. Moreover, the adopted approximations slightly overestimate the outflows which can be interpreted as a safety approximation for flood mitigation 220 planning.
Comparing the different pdfs, it should be clarified that the parameters have been changed looking for combinations of dam heights, ℎ , and coefficient 1 leading to similar water storage capacity on each row. This allows to demonstrate that it is much more effective to increase the area flooded by the reservoir (the parameter 1 represent the rate of increase of the water storage with the water level) rather than increase the height of the dam. 225 Last analysis performed is given in Figure 6 where we tested the performances of the TDD assuming the lower opening closed.
Such a condition may be representative of an ordinary dam with an assigned flood retention volume or used for water supply purposes, where the volume above the crest spillway leads to a lamination of floods. This configuration has been compared with the scheme described in the previous sections that includes the presence of the lower opening. The difference between the two configurations is given in Figure 6.A (closed opening) and Figure 6.B (lower opening active). The two graphs show 230 https://doi.org/10.5194/hess-2021-47 Preprint. Discussion started: 22 February 2021 c Author(s) 2021. CC BY 4.0 License. the ability of the mathematical formulation to properly interpret also the present configuration, offering a wide spectrum of potential applications in hydraulic design.
Finally, in order to test the impact due to the approximation adopted by the rectangular hydrographs, we also compared the results of a numerical simulation where the hydrograph is assumed to be a symmetric exponential one according to the expression given in Equation (11). The comparison is given in Figure 7 for three different equivalent event durations. It can 235 be appreciated how the use of an equivalent event duration allowed to reproduce correctly the flood mitigation that looks very similar to those obtained with a symmetric exponential hydrograph. It must also be underlined that with the increase of the duration of the event such an approximation tends to deteriorate the result of the proposed model.

Conclusion
The present manuscript introduces a new formulation useful to quantify the impact of detention dams on the probability 240 distribution of floods. We must acknowledge that the formulation was obtained with several simplifying assumptions that include the shape of the incoming hydrograph; the approximation used to interpret the flow through the lower opening; the approximation of a linear reservoir for the flow above the crest level. Summing all these, it is really satisfying to see that the obtained formulation can fairly well interpret the dynamics of such hydraulic infrastructures providing an analytical description of the impact of artificial reservoirs on flood dynamics. This may be extremely useful in properly addressing the effects of 245 water infrastructures on floods. The TDD can be used for detention dams, but the formalism can also be applied to ordinary dams just setting the control discharge to zero and assigning a given value of the water level in the dam. Therefore, the formalism is versatile and can be applied in different contexts. The strongest assumption is represented by the rectangular hydrograph which can be realistic for small river basins characterised by relatively small concentration time. With the aim to minimize the impact of such a choice, we adopted an equivalent event duration in the formulation that allowed to account for 250 the flood variability during a specific event. However, this assumption may become limitative in large river basins where hydrograph evolves over large areas and its shape is also not simple to be predicted. In fact, large river basins may display complex hydrographs with multiple peaks that require a specific approach.
The proposed method may be used in some contexts such as the projects and design of small lamination dams and detention dams realized in small river basins. For instance, there are several river basins along the coastline, which drain a high amount 255 of water in short durations, affecting cities and towns developed along the waterfront. These areas are typically exposed to frequent flood events that may impose the need to properly identify potential solutions for flood mitigation.
The scheme can be used to carry out preliminary dimensioning of these structures and eventually could be coupled with other tools to identify optimal configurations for flood mitigations. Moreover, the scheme can be applied to any probability distribution of floods including the case of floods that are already subject to laminations allowing the description of a scheme 260 of nested dams. This topic is still under investigation and its study will be applied to identify optimal solutions in flood control systems quantifying the impact of structure on the full spectra of floods.

Notations α [-]
Scale parameter of Gumbel distribution 265 Location parameter of Gumbel distribution