effect of seting the number_of_logic_tree_samples a none zero value

[Parisa]

Dear Developers,

I want to run an event based hazard job for 100,000 years

investigation_time = 1

ses_per_logic_tree_path = 100,000

I have a logic tree with 4 branches of GMPE for each TectonicRegionType.

In order to prevent getting a Memory Error, I set number_of_logic_tree_samples = 1

I have three questions in this regard:

1) when number_of_logic_tree_samples = 1, does it mean that for each event and each site, OQ will pick only one of the 4 GMPEs, regarding their weights? the ones with higher weights are more probable to get selected?

2) if we have different PoEs, the selected GMPE may be different for a particular event-site combination?

3) if for example number_of_logic_tree_samples = 2, should we simply average the results of two realizations?

Thank you

Parisa

[Michele Simionato]

On Saturday, July 7, 2018 at 6:21:40 PM UTC+2, Parisa wrote:

Dear Developers,

I want to run an event based hazard job for 100,000 years

investigation_time = 1

ses_per_logic_tree_path = 100,000

I have a logic tree with 4 branches of GMPE for each TectonicRegionType.

In order to prevent getting a Memory Error, I set number_of_logic_tree_samples = 1

I have three questions in this regard:

1) when number_of_logic_tree_samples = 1, does it mean that for each event and each site, OQ will pick only one of the 4 GMPEs, regarding their weights? the ones with higher weights are more probable to get selected?

Yes.

 

2) if we have different PoEs, the selected GMPE may be different for a particular event-site combination?

It is not clear what you mean. The PoEs used for the hazard maps/hazard spectra? In that case no, the selected GMPE

with number_of_logic_tree_samples=1 is the same for all PoEs.

  

3) if for example number_of_logic_tree_samples = 2, should we simply average the results of two realizations?

If you have number_of_logic_tree_samples = 2 and mean_hazard_curves=true the engine will compute the mean

by averaging the two realizations, taking the weights into account,

HTH,

               Michele

 

[Parisa]

Thank you very much.

About my 3rd question, it is not exported using --eo and I could access it only through OQ WebUI.

and two more question on "Scenario" hazard calculation:

- I'm not sure what "ground motion values" are changing in each Monte Carlo simulation when we specify a value for number_of_ground_motion_fields. The result of simulations are very variant for different "eid"s, even when I have defined any parameter in my rupture model or site model completely certain.

- how can we access the mean resulting gmv for each "sid"? not the gmv for each "rlzi-sid-eid" separately.

[Michele Simionato]

On Tuesday, July 10, 2018 at 4:05:11 PM UTC+2, Parisa wrote:

Thank you very much.

About my 3rd question, it is not exported using --eo and I could access it only through OQ WebUI.

and two more question on "Scenario" hazard calculation:

- I'm not sure what "ground motion values" are changing in each Monte Carlo simulation when we specify a value for number_of_ground_motion_fields. The result of simulations are very variant for different "eid"s, even when I have defined any parameter in my rupture model or site model completely certain.

- how can we access the mean resulting gmv for each "sid"? not the gmv for each "rlzi-sid-eid" separately.

The values of the GMFs changes a lot for different events. That's normal and good. This is also why the scientists here are very much against producing any kind of mean ground motion field, because we do not want the user to get the impression that there is single good mean field. All the GMFs must be taken into account. Having said so, you can disable the variability in the lognormal distribution of the GMFs by setting truncation_level = 0 and you can consider that as the mean, but then you are defeating the purpose of the scenario calculator which is to compute the distribution of the GMFs and therefore of the losses.

     Michele

 

[Parisa]

Thank you for the information, but actually I need a single final PGA map (gmf) for a single scenario. It is kind of impossible for me to consider all 1000 gmfs for my purpose. that is the reason I averaged the gmvs of different realizations.

I repeated the calculations with smaller values as truncation_level. The mean values are not the same or even close. the smaller truncation_level results in lower PGAs in my case. the difference is about 0.1g.
so in such cases, which one would you recommend?

- mean gmv of 1000 rlz with truncation_level = 3

- 1 rlz with truncaition_level = 0.001   (resulted in generally lower PGAs)

(I got an AssertationError when setting truncaition_level = 0)