Modeling compound flood risk and risk reduction using a globally-applicable framework: A case study in the Sofala region
Abstract. In low-lying coastal areas floods occur from (combinations of) fluvial, pluvial, and coastal drivers. If these flood drivers are statistically dependent, their joint likelihood might be misrepresented if dependence is not accounted for. However, few studies have examined flood risk and risk reduction measures while accounting for so-called compound flooding. We present a globally-applicable framework for compound flood risk assessments using combined hydrodynamic, impact and statistical modeling and apply it to a case study in the Sofala province of Mozambique. The framework broadly consists of three steps. First, a large stochastic event set is derived from reanalysis data, taking into account co-occurrence and dependence between all flood drivers based on a vine copula structure. Then, both flood hazard and impact are simulated for different combinations of drivers at non-flood and flood conditions. Finally, the impact of each stochastic event is interpolated from the simulated events to derive a complete flood risk profile. Our case study results show that from all drivers, coastal flooding causes the largest risk in the region despite a more widespread fluvial and pluvial flood hazard. Events with return periods larger than 25 year are more damaging when considering the observed statistical dependence compared to independence, e.g.: 12 % for the 100-year return period. However, the total compound flood risk in terms of expected annual damage is only 0.55 % larger. This is explained by the fact that for frequent events, which contribute most to the risk, limited physical interaction between flood drivers is simulated. We also assess the effectiveness of three measures in terms of risk reduction. For our case, zoning based on the 2-year return period flood plain is as effective as levees with a 10-year return period protection level, while dry proofing up to 1 m does not reach the same effectiveness. As the framework is based on global datasets and is largely automated, it can easily be repeated for many other regions for first order assessments of compound flood risk.
Dirk Eilander et al.
Status: final response (author comments only)
RC1: 'Comment on nhess-2022-248', Anonymous Referee #1, 07 Nov 2022
- AC1: 'Reply on RC1', Dirk Eilander, 16 Feb 2023
RC2: 'Comment on nhess-2022-248', Anonymous Referee #2, 05 Dec 2022
- AC2: 'Reply on RC2', Dirk Eilander, 16 Feb 2023
Dirk Eilander et al.
Dirk Eilander et al.
Viewed (geographical distribution)
The manuscript presents a global framework for assessing flood risk and risk reduction strategies for compound flooding. The framework was applied to Sofala province in Mozambique. The Authors showed that coastal flooding causes the greatest impact regardless of the other drivers.
The manuscript is well-written and with a clear structure. However, there are a few criticalities that need to be addressed.
The novelty of this work compared to the current literature and to previous works from the same Authors is difficult to grasp. In its current form, the manuscript reads as a case study which is not enough. The advise is to highlight how this work addresses the limitations of previous studies and goes beyond what has been already done.
The global applicability claimed by the Authors is not really proven since they validated and applied it to the same location (line 82). How does this model apply and perform in other locations?
The description of the probabilistic model needs improvements. First, the Authors should clarify the data used in the vine-copula models, whether they are annual maxima (annual maxima obtained independently in each time series) or whether such data are shifted relative to each other (see table 1). It seems like the annual maxima of 5 different variables occur within a window of +-10 days, which seems a bit unlikely. Moreover, the introduction of a rate of occurrence of annual maxima should be better explained in the relation to the copula model. Why is this necessary? The vine-copula is built to generate sets of dependent variables, including sets in which all of the variables are extremes. This point also relates to the distinction the Authors made between compound events and non-compound events. What does make an event compound? Is this related to the impacts? How can it be defined a priori then?
Line 36: Do the Authors refer to the joint likelihood or the joint probability? The two concepts are different.
Line 49: Specify what are the “four drivers”.
Lines 68-74: this paragraph is difficult to read. For example, what is an event set? How is a “model event set from univariate distribution” different from a “stochastic event set from a multivariate probabilistic model”? (see also general comment)
Lines 82: Please further elaborate on the reason why global models are useful in data-scarce regions
Line 89: Add some information on the boundary conditions. For example, are the boundary conditions generated independently? How are the normal and extreme boundaries selected and combined? Are all normal or are all extremes?
Section Discharge and Total water level. Clarify the link between annual maxima analysis and hydrograph generation. What are the raw data used for annual maxima analysis and how this annual maximum relates to the hydrograph?
Line 150: why did the Authors use annual maxima if the temporal resolution of the data is hourly?
Line 155: Relative timing between drivers: it seems like the annual maxima of each driver occur around the same time, is it the case? Or the correlation in table 1 is the correlation obtained between the annual max Buzi discharge and the corresponding driver around that period (even if not extreme)? Discuss whether the timing in Table 1 makes sense. Also, why the Authors selected river discharge at Buzi and not a precipitation event? Precipitation might drive high water in the river unless other processes are of relevance.
Line 171: Add some discussion on the rate at which different combinations of drivers co-occur. Should not this come from the vine-copula? Please, clarify this second step. (see also general comment)
Lines 179: Do you try all the possible vine copulas? It would be good to show somewhere what the vine-copula selected looks like and the associated bi-variate copulas.
Line 189: How is an event defined?
Line 240: The Authors made a distinction between exposure and vulnerability. However, their definitions are missing. How is exposure defined? How is vulnerability defined? How do they contribute to the impacts? It is not clear which variables have been used to quantify these two concepts.
Line 243: Why is a bias correction needed?
Line 266: Is this return period associated with the univariate case? Is 5-years realistic for the case study? It would be good to justify the choices made.
From lines 287 – flood drivers. It is a bit unclear how an event is defined and the time series used in the vine-copula models. Why would a compound event be the event in which one variable is extreme? When generating a set of dependent variables, any results in terms of water depths, in this specific case, can be classified as compound (see also general comment)
Flood Hazard: Please, specify how the 100-year fully dependent event is identified, i.e., how the value of each variable is quantified. Also, are the drivers 4 or 5?
Figure 9: The definition of the percentage of base risk is not fully clear. What is the component of the total risk? Is the total risk different per strategy?