Dam break simulation with HEC-RAS: Chepete proposed dam

Dam failures are catastrophic events in which a dam overflows out of control causing massive fatalities and economic losses. 
represa hidroelectrica Chepete inundacion

Although nobody wants a dam failure, dam failure and dam breaks occur. Cannata & Marzochi note that between the years 2000 and 2009 more than 200 notable dam failures happened worldwide. Moreover, in the last years we had several dam failure events in different countries such as:

Thus, every dam should consider at least a dam failure analysis. The cost of a dam failure analysis is negligible compared with the cost of dams.

Note: Dam failure does not necessarily refer to a structural collapse (This collapse is dam break or dam collapse). Dam failure refers to any event in which water overflows out of control.

Simulating a dam break

When we simulate a dam break scenario, we have to consider three issues:
  • What happens upstream the dam. This can be simulated by assuming a level pool routing (0D) or by an unsteady simulation solving the Saint Venant equations (either 1D or 2D). Although the unsteady simulation provides the most accurate results, it also requires more data and more pre-processing work. 
  • What happens in the dam itself. This refers to the dam break process. It describes the trigger mechanism, the moment when the dam begins to breach, the geometry progression of the dam break and the flow. Although there are different guidelines for estimating the final geometry and the time to break, every case is unique and requires a detailed analysis.
  • What happens downstream the dam. This is the so-called flood routing and it can be simulated by solving the Saint Venant equations (either 1D or 2D). In some cases, a kinematic wave and a dynamic wave assumption may also provide good results.
There are different combinations to simulate a dam break based on the approach used to simulate each issue. The most common approaches are:
  • 0D – 1D (0D upstream the dam and 1D downstream the dam)
  • 1D – 1D (1D upstream the dam and 1D downstream the dam)
  • 0D – 2D (0D upstream the dam and 2D downstream the dam)
  • 1D – 2D (1D upstream the dam and 2D downstream the dam)
  • 2D – 2D (2D upstream the dam and 2D downstream the dam)

Early versions of HEC-RAS were able to simulate the 0D and 1D combinations. A few years ago, HEC-RAS v5 was released including 2D capabilities. Hence, HEC-RAS allows simulating a dam break considering all the 5 combinations.

Simulating a hypothetical Chepete dam break

In the present case we analyze a hypothetical dam break scenario in the proposed mega dam Chepete (Bolivia). Chepete is a 200 m (height) proposed hydropower dam in the Chepete narrow creek (Image 1). 
Chepete El Bala megarepresas hidroelectricas
Image 1. Proposed Chepete dam location

This dam is still a proposed one with limited information. Thus, some assumptions were necessary.
  • Shape of the dam. Due to the geographical conditions and the height of the dam, it is reasonable to assume that it would be a concrete arch dam (200 m dam (h) in a 150 m width creek). 
  • Dam break. Guidelines suggest that this type of dams usually experience a dam break in less than 0.1 h, and at least 80% of the dam fails. Dams with large volumes of water may experience breaches bigger than the mentioned by most guidelines. Although in the present case a full collapse may be reasonable, we assume an 80% width breach with 50% height breach. 

The most common dam progression models are the linear and the sinusoidal (image 2). In the present case, we assume the linear model with rectangular shape. (NOTE. A better breach geometry evolution requires a more detailed analysis)
dam break geometry
Image 2. Linear and sinusoidal breach models
  • Reservoir volume. The reservoir would cover at least 680 km2, flooding jungles, creeks and small towns.  The reservoir area-volume relationship (Image 3) was obtained from Molina et al (2018).
hydropower reservoir volume
Image 3. Elevation - volume ralationship (Source: Molina et al., 2018)
  • Climatic conditions. Dam break studies should consider dam break scenarios under different climatic initial conditions such as: clear sunny day in dry season (optimistic scenario), rainy day during wet season (worst case scenario), and other intermediate scenarios. In the present case we assume an optimistic scenario during dry season and neglecting flow from tributaries.
  • Upstream simulation. Due to data limitations, the reservoir was simulated assuming level pool routing (0D). Although the best option would be an unsteady routing, that option requires additional data not available such as: detailed topography and flows from different tributaries. An important detail of this assumption, is that its connection with the dam requires a dummy 1D river with 2 dummy cross section: 
  1. One cross section to connect the storage area with the reach
  2. One cross section to connect the reach with the dam
HEC RAS storage area inline structure
  • Downstream simulation. Two simulations were performed: one 1D and one 2D. In both cases the downstream topography was based on the SRTM 30 DEM. Although some may consider that this DEM is quite coarse for a dam break, it is important to remember that dam break simulations require a minimum spacing between 50 ft to 100 ft. Otherwise, the derivatives with respect to distance may be overestimated, especially on the rising side of the flood wave. This may cause the leading edge of the flood wave to over steepen, to the point at which the model may become unstable.
An important advantage of 1D simulation is that it is much faster than the 2D simulation. This 1D simulation is important not only because it provides a first idea of the flood wave and its magnitude, but it also allows testing different boundary conditions. Hence, a sensitivity to the downstream boundary condition was performed using the tool to automate HEC-RAS AHYDRA.

This 1D simulation is also important to define the 2D simulation period. As 2D models are time consuming, we try to avoid running unnecessary time steps.
  • Sediment in the reservoir. The presence of sediment in the reservoir may have some influence. If the sediment volume is large, it may result in a short concentrated flow, hence modifying the Newtonian conditions of water and increasing the internal friction. In the present case sediments were neglected (clear water assumption).


Dam break hydrograph

A dam break acts as a variable size rectangular reservoir. The break shows a high peak, higher than 180 000 m3/s (Image 4). Another important detail of this hydrograph is the long duration of the flood. Because of the reservoir's volume, the flood would last more than 160 h (about 7 days). 
peak flow flood volume
Image 4. Breach hydrograph

The 1D simulation video also shows an initial idea of the magnitude of the catastrophe with flood depths over 20 m. 

HEC RAS simulation water profile
Image 5. 1D flood profile

The 2D simulation

The 2D simulation video provides a detailed view of the flood extent and the flood depth. It is important to remember that this case was an optimistic scenario assuming dry season and neglecting flow from tributaries. In a real case, the flood would also create backwater flows in all the tributaries that flow in the flooded area, while upstream tributaries would provide additional water, hence worsening the flood. The aerial view identifies at least 4 important rivers that would get be affected. 
inundacion Chepete Rurrenabaque
Image 6. 2D flood extent
The flood would cover more than 500 km2. Within the first 14 hours after the breach the flood would cover about 450 km2 (Image 7). A limitation of this model is that we only considered the effects until Rurrenabaque and San Buenaventura (about 60 km downstream). However, results show that considerable flood may propagate downstream, maybe even until Reyes or at least close to Reyes (about 24 km from Rurrenabaque).
Inundacion rurrenabaque bolivia
Image 7. Flood extent in the first 14 h

The flood would be disastrous not only for the environment, but also for cities like Rurrenabaque and San Buenaventura. In some urban areas the flood may reach 15 m depth and would last for several days (Image 8).
Inundacion Beni Bolivia
Image 8. Flood depth in Rurrenabaque

This short post shows that even considering optimistic conditions the sole presence of the proposed Chepete dam increases of the exposure and the vulnerability of the areas and cities located downstream. 

Not only the flood peak is important, but also the flood volume. Hydrologic studies must consider a multivariate statistical analysis considering at least a peak - volume joint probability.

This initial simulation provides an initial idea of the flood exposure and vulnerability of the area and potential consequences of a dam failure event.

Note that this is an initial simulation with several assumptions. Additional simulations should be performed with more data and considering different scenarios. As already mentioned, dam failure also considers any water overflow out of control (or beyond the expected volume).


Gera said...

Hi and thanks for the useful article.
Can you please elaborate on the statement:
"..it is important to remember that dam break simulations require a minimum spacing between 50 ft to 100 ft. Otherwise, the derivatives with respect to distance may be overestimated, especially on the rising side of the flood wave. This may cause the leading edge of the flood wave to over steepen, to the point at which the model may become unstable"
any references I can read on that?
I have 1m DEM. Should I make the grid downstream of the dam coarser to avoid this instabilities?
any references on this subject?

Vadyman said...

hi. You can use the 1m grid, but pay attention to the time step. You would need very very small time step
Check your Courant number.
I would say there 2 main issues. 1, the model may become unstable and crash. 2 after the simulation take a look at the results and the courant. Sometimes it finishes the simulation, but the results are either overestimated or underestimated (you can see that profiles are not smooth

Gera said...

Thank you for the reply.
I think the new "Courant cond. controlled timestep" feature in hecras can be useful in this case.