Flood Modelling of June 2023 Flooding of Léogâne City by the Overflow of the Rouyonne River (Haiti)

1. Introduction

Urban flooding is a subject of major concern to both the scientific community and the public. To improve urban planning and manage these risks, it is important to have digital tools capable of simulating flood events for better decision-making. Hydraulic modelling tools can predict the spatial and temporal evolution of river behavior during a flood event, and the resulting flood extent [1,2,3], provided that prior calibration of the model can be achieved. There are various approaches for modelling river flooding events, including simpler one-dimensional models [4,5], requiring less data, and more detailed two-dimensional [6,7,8] models. The choice of approach must be consistent with the scale under study and with the risks involved [9].

This research study concerns a stretch of the Rouyonne River along the town of Léogâne in Haiti (Figure 1a). Fed by a mountainous upper watershed of 46.45 km² (Figure 1b), the river flows to the Caribbean Sea via an 8.2 km-long channelized reach that begins at the upper watershed outlet (X= 754280 m; Y= 2044471 m, coordinate system WGS 84/UTM zone 18N) and ends at the sea (X= 747693 m; Y= 2049408 m). While the river is rather sinuous in the mountainous upper watershed, the downstream part of the Rouyonne River (the channel) is nearly straight with an average slope of 0.62%. The lower course of the river was indeed artificially channelized and confined by embankments, precisely to prevent inundations. The upper watershed is very hilly, with 75% of its surface area situated between 250 and 950 meters above sea level (Figure 1c). The city of Léogâne lies in the flood plain zone, close to the channelized part of the river. Unfortunately, the embankments are frequently overflown, which explains the recurrent flooding of the city.

The topographical and bathymetric data available for the site were the official data from CNIGS (Centre National des Informations Géospatiales – National Center for Geospatial Information) covering the entire country, dating back to 2016 and issued from a measurement campaign that was started in January 2014 by IGN France International (IGN FI), with funding from the World Bank. However, since the data acquisition, many strong morphological changes were observed in the main rivers of Haiti, caused by extreme climatic events.

The June 3, 2023, event resulted in the death of almost twenty people in the city of Léogâne, a lot of injuries and considerable damage (Figure 2a-b). In just 32 hours and 30 minutes, the average rainfall recorded in the watershed amounted to 142.1 mm. This value exceeds the monthly average rainfall of around 100 mm for the month of June in the municipality of Léogâne. Consequently, significant overflows have been observed during this event, inundating the flood plains and finally a part of the city. Immediately after the event, high-water marks have been measured in the city and along the river (Figure 2c).

The aim of this research study is to develop hydraulic modelling of the study site combined with hydrologic tools to reconstruct missing data in order to reproduce the June 2023 event in Léogâne and from there, initiate a methodology to characterize the flood risk in the region. The paper is organized as follows: first the data collection is presented, including drone-based measurements for the bathymetry and hydrological modelling to evaluate the discharges; then a hydraulic model is constructed, and simulations of the June 2023 event are conducted. The results are compared to field measurements and finally conclusions are drawn on the long-term perspective regarding possible flood forecasting systems for the city of Léogâne, to predict what might happen when heavy rainfall is announced, and to establish risk maps.

2. Materials and Methods

The simulation of June 2-3, 2023, event is performed by adopting the following approach: (i) topographic data acquisition; (ii) determination of the discharge as upstream hydraulic boundary condition and (iii) construction of the hydraulic model used for the simulation. All these points are presented in the following sub-sections.

2.1. Topographic Data

The 1.5 m resolution digital terrain model (DTM) from the CNIGS was used for hydrological modelling of the upper watershed. However, due to important morphological changes in the last years in the lower reach, this DTM has become too inaccurate for hydraulic modelling of this reach. Therefore, a drone-based survey of the river was conducted during the dry season, when the river is empty (Figure 3a), and photogrammetry was applied using the image processing software “Agisoft Metashape pro”. Aerial images were taken from a drone (DJI marvic Air 2) equipped with a GoPro Hero 4 camera (Figure 3b), which are all low-cost and easy to handle devices, well suited for such field conditions.

Each drone flight had a duration of about 10 minutes, depending on battery autonomy, with a speed of 5 m/s and a capture frequency of one image every two seconds. Image overlap between successive flights is 80% at the front and 60% at the side [10]. These overlaps provide sufficient information for the spatial reconstruction of the scenes studied, with a total of 300 images on average for a flight plan. The accurate georeferencing of the imagery was performed by the GNSS (Global Navigation Satellite System) positioning of at least 15 GCPs (Ground Control Points) for each flight plan, used for image georeferencing during photogrammetric processing. For more information on the processing procedure and assessment of this method at sites in Haiti, see [11,12,13,14,15,16].

Two drone-based surveys were conducted in 2020 and 2022, respectively. The drone-based technique was found to be very accurate, with a RMSE of 0.18 m on 80 verification points. A comparison between a manual survey of cross section 54-54 and photogrammetric data is illustrated in Figure 4a-b. The strong morphological evolution of the river at this cross-section is illustrated in Figure 4c, where the official DTM (2014-2016) is compared to the 2020 and 2022 surveys. In particular, it can be observed that the left bank, that was destroyed during hurricane Matthew in 2016 and then reconstructed, was again completely eroded between 2020 and 2022.

One day after the June 2023 event, 21 high-water marks were recorded at locations illustrated in Figure 5. These will be used to validate the hydraulic model.

2.2. Discharge Estimation

To run the hydraulic model, we need the flood hydrograph for the June 2-3, 2023, event at the system inlet. Ideally, the discharge should be obtained from a hydrometric station, which is unfortunately not the case at Léogâne. Only data from a limnimeter (OTT PLS probe: pressure sensor) installed at the measurement section (at the channel inlet) since April 2022 are available. At this location, an OTT “MF Pro” magnetic induction current meter was used to take a range of flow measurements at shallow water depths, below 40 cm. Beyond this depth, measurements become dangerous, and therefore the number of stage-discharge points available to construct a rating curve is not sufficient. It is thus necessary to extrapolate the few points of the rating curve for greater water depths. The construction of the rating curve is based on the assumption of uniform flow in the measurement section and was complemented with the BaRatin (Bayesian Rating curve) method [17,18] developed by INRAE since 2010, allowing to assess the uncertainty associated with this approach.

However, it is also important to be able to predict the flow from rainfall measurements when the water level station is dysfunctional: this was indeed the case during the June 2-3, 2023, event, as the probe was destroyed. To achieve this, the rainfall data recorded at three rain gauges on the Rouyonne river catchment were used to establish a rainfall-discharge relationship for this watershed using the AtHyS (Atelier Hydrologique Spatialisé) software developed at IRD Montpelier [19]. The key parameters of this hydrological model are: V₀ (m/s) that can be estimated as the velocity of flood wave propagation; S (mm) the total ground reservoir capacity which depends on soil characteristics; ds (day^-1) the proportionality coefficient between reservoir level and emptying intensity; ω (dimensionless) the fraction of the underground reservoir emptying that participates in hypodermic flow; and K₀ (dimensionless) a proportionality constant between diffusion time and transfer time. These parameters are calibrated and validated using two other events before the June 2023 event. The performance of the hydrological model is assessed using the classic criteria most widely used in hydrology studies [20,21,22,23], i.e. NSE (Nash-Sutcliffe Efficiency) proposed by [24] and KGE (Kling-Gupta Efficiency) [25], used in other hydrological studies [26,27,28]. The KGE criterion was used to select the best parameter set during calibration and assess the hydrological model performance. Once validated, the rainfall-discharge relationship can be used to predict flows resulting from given rainfall events.

2.3. Hydraulic MODEL AND SIMULATION TOOL

Based on the DTM, a hydraulic model is built using a triangular mesh (Figure 6). Different sizes are defined for the computational cells: (1) the riverbed with a 1 m resolution), (2) the floodplain with a 40 m resolution and (3) the transition zone between the Rouyonne riverbed and the floodplain, where the dikes can be identified, with an intermediate 3 m resolution. This mesh is then used for numerical simulations within the WATLAB environment developed at UCLouvain/iMMC (https://sites.uclouvain.be/hydraulics-group/watlab/index.html), solving the two-dimensional shallow-water equations using a finite-volume scheme for unstructured meshes.

The hydraulic model roughness was characterized through two different values of Mannings’ coefficient: n_c = 0.048 sm^-1/3 for the main channel and n_f = 0.05 sm^-1/3 for the floodplain. As regards the upstream boundary conditions, the flood hydrograph generated using AtHyS was used because no complete hydrograph could be established from the limnimetric data recording as this had stopped at the time when the probe cable broke.

2.4. Summary of Data Used In This Study

All data used in this study are summarized in Table 1, that describes the origin of the data and the time or space resolution, as applied in the simulations.

3. Results

3.1. Rating Curve for the Rouyonne River

The cross-section of the Rouyonne River at the location of the limnimetric station is illustrated in Figure 7. It has an almost regular trapezoidal shape, and based on the uniform flow assumption, a stage-discharge relation could be constructed using Manning’s equations, after calibrating the roughness using the 8 manually measured stage-discharge values for water depths of up to 0.40 m. As the maximum discharge that was manually measured was much below the probable discharge that occurred during the June 2023 event, it was necessary to assess the quality of the rating curve established using the uniform flow assumption. Therefore, the BaRatin method [17,18] combining all sources of uncertainties related to rating curves, was used.

Figure 8 shows the rating curve obtained with the uniform-flow assumption calibrated on the few measured points and the rating curve calculated with the Baratin method, as well as the associated uncertainties. The parametric uncertainty expresses the uncertainty related only to the estimation of the curve parameters, while the total uncertainty is the combination of parametric uncertainty and the uncertainty induced by the remnant error.

Unfortunately, the water level gauge failed during the June 2023 event when the water level reached 2.43 m, making it impossible to use the rating curve to calculate the corresponding hydrograph. Therefore, a rainfall-discharge relation is required.

3.2. Rainfall-Discharge Relationship

The AtHyS hydrological model was run on one event that occurred in August 2022 to calibrate the model parameters, and then applied to an event in September 2022 for validation. The comparisons between the estimated flows using the rating curve of Figure 8 and the hydrological model results are shown in Figure 9. The values of the calibrated parameters are given in Table 2, as well as the KGE and NSE scores.

Using this validated model, the complete hydrograph of the June 2-3, 2023, event was built and the results are illustrated in Figure 10. It can already be observed that the agreement between the observations and the model is very good in the period when the limnimetric station was working properly. The simulated maximum flow (Q=168 m³/s) for this event, at the measuring section, is two times larger than the maximum flow (Q=84 m³/s) that the river can convey without overflowing as this point.

3.3. Hydraulic Modelling of the June 3, 2023 Flood Hydrograph

The simulation results are illustrated in Figure 11: it can be observed that the Rouyonne River indeed overflows the dikes and induces severe flooding of the surrounding areas, including the city of Léogâne. The extent of the inundated area appears to be well reproduced by the model, as the points where high-water marks were recorded are indeed located in the inundated area (yellow points in Figure 11).

The comparison of simulated and measured water depths at the 21 high-water marks (Figure 12) shows that the model provides a good estimate of the water depth in the study area. With KGE and NSE values of 0.890 and 0.882 respectively, based on these 21 comparison points, the model is considered as satisfactory.

4. Discussion

The rating curve was established using a few measured points and considering uniform flow assumption, which is a questionable assumption as the flow is highly unsteady, especially during rapid floods. Still the validity of the curve was confirmed by the use of Baratin method, that allowed to estimate the related uncertainty, especially considering the limited number of gauging points. A similar situation is observed in [29], where the authors verified the beginning of their rating curve with water depths of no more than 0.50 m for a total section height of around 1.20 m.

This rating curve was then successfully applied to determine the parameters of the AtHyS hydrological model. The optimal parameters (Table 2) allow for a very good reproduction of both the validation case and the beginning of the June 2023 hydrograph. Still, the very high value of the velocity V₀, equal to 5.43 m/s, is questionable. It could be explained by the very steep slopes in the upper catchment, as well as by significant deforestation that prevents slowing down runoff water by vegetation. The calibrated value of S = 82.44 mm corresponds to a CN close to 75, which is in agreement with classical values for the type of soil and vegetation cover in the catchment, where the hydrological conditions of the cultivated surfaces were medium.

The simulation identified three major overflow points on the Rouyonne river, that are highlighted in Figure 13: two points on the right bank (Figure 13a), responsible for flooding the town center, and one point on the left bank (Figure 13b), close upstream of the gauging station. This last overflow point explains why the probe was torn off, interrupting water depth recording at the measuring section.

From the good agreement between simulated and measured water depth (Figure 12), the hydraulic model can be considered as satisfactory. However, the accuracy of the results could be further improved by using a more refined mesh, for example reducing the size of the cells in the floodplain, however at the cost of a great increase in calculation time. Also, the buildings in the city of Léogâne are not properly accounted for. As shown in the work of [6,7], various local aspects can influence the modelled water height, such as spatialized roughness in the floodplain and the effect of buildings in the city. Consequently, our results may underestimate the water depths in some places. This is a prospect for further studies in Haiti.

5. Conclusions

Combining different sources of data and low-cost field measurement techniques, it was possible to develop a hydraulic model for the Rouyonne catchment that reproduced the inundation of the city following the June 2023 event. This shows that the model could now be used to draw up flood zone maps for the city of Léogâne. Using rainfall events, or synthetic rainfalls, with given return periods, it is possible to derive the corresponding hydrographs and to characterize the possible flooded area for different return periods. Such rainfall data with known return periods and not available for the moment but could be used in the future.