search

NRCan / RICorDE

Gridded water depth estimates from flood inundation
Other
3 stars 1 forks source link

Underestimation as a result of quartile. #2

Open roescob opened 1 year ago

roescob commented 1 year ago

Is your feature request related to a problem? Please describe. I encountered underestimation under a shallow area of the Fort Mcmurray downtown area.  image

Following the documentation I encountered that this was caused by the third quartile (q3) being forced to a lower than actual depth.

Describe the solution you'd like Solution: Increasing the default maximum value allowed for upper bound (cap) of RICorDE_params_default.ini from 7m to 8m solved this issue. image

Describe alternatives you've considered The parameters for the statistical analysis are not always obvious. Giving users a simple way to not apply them might be helpful.

Additional context This was experienced during v1.0.1

cefect commented 1 year ago

Thank you for the comment. I guess this is an issue with the phase 2 filtering of the edge sampled HAND values (through ceiling/floor quartile forcing?) Can you send the histogram of the pre-filtered HAND values? I'm not sure removing the filtering all together is a good idea... but maybe we can think of a more intelligent way of filtering.

roescob commented 1 year ago

image Sampled HAND values within the initial extent.

cefect commented 1 year ago

I think the issue is that this flood was so exceptionally deep. The floods considered during development were more on the order of 2-5m, so 7m was a reasonable upper bound. Can you also post the Hydrograph we have for fort Mac? (Time vs water depth)

roescob commented 1 year ago

I had them previously collected from https://wateroffice.ec.gc.ca/map/index_e.html.

There were 3 available stations: water level fort mcmurray 2020.xlsx

Beware of data gaps

cefect commented 1 year ago

Interesting as these show max water levels ~ 6m. I guess the gauge was knocked out after this?

roescob commented 1 year ago

For sure