Kakhovka dam break
Simulating a current event with limited data available.
On June 6th 2023 part of the Kakhovka dam was destroyed and left a breach in the retaining construction. The reservoir, estimated to hold a total of about 18'000'000'000 m3 of water discharged uncontrolled into the Dnipro river bed.
Two simulations of the dam break were conducted. They both included the entire reservoir up to Zaporizhzhia and down to the black sea with different breach sizes. The terrain date used was from the Shuttle Radar Topography Mission which provides data with a 1 arc-second sampling. Effective cell size of the model was about 25m. Which led to a relatively small model of 14576x12498 or about 182 million cells. Actual wet cells during the simulation did not exceed 20 million cells. The model covered approx. 350km of reach and 12'500 km2 water surface. Simulated time was 7 days and run time were between 6 and 13 hours on styx-GPU.
uncertainty & assumptions
Information about the size of the dam breach was limited in the hours after the incidient and could only be attaind from imagery. There is unknown bathymetry both upstream and downstream of the dam, which is not represented in the SRTM data. Due to known heights on the border of the reservoir as well as the Dnipro river below the dam, assumptions about the bathymetry were made.
The resulting estimated capacity of about 18'000'000'000 m3 was coherent with other sources. According to the US Department of Agriculture, the water level in the reservoir was about 17 m above sea level in May.
Validation of the data is limited. Several sites provide good before/after satellite imagery from MAXAR satellites and there are a lot of images from the flooded regions. These however are mostly hard to localise without regional knowledge. Generally speaking, there seems to be a reasonably good match between simulation and actual event. Notably for the town of Krynki there is a good match, whereas the simulation unterestimated flooding in the surrounding forested areas. This is most likely due to the land cover and subsequent inaccuracy of the terrain data.
Many of the ground photos of Kherson could not be located, some arial shots indicate a reasonable accuarcy there as well.
validation part 2
Sentinel 2 data of the Hola Prystan region shows a generally good match to the simulation. There is some overestimation in the outlines to be seen. With relatively small elevation differences in the whole valley, small differences in height can make for big differences in extent. Unfotunately cloud cover in the days after the event doesn't allow for a comparison in other areas of interest.
A quick and dirty approach to a current event was applied. The model was not calibrated, rough estimates about the bathymetry and nature of the breach had to be made. Satellite data was relatively coarse and not filtered for actuall terrain but included vegetation.
Nontheless relatively accurate results could be obtained rapidly, within several hours of the event.
Further validation has to be done. As the event is completely documented, more conclusions about the actual accuracy can be drawn.
More accurate terrain data as well as validated flow depth data could help improve the model in the future.