Fixed long standing bug in the vulnerability functions

[Michele Simionato]

With the nightly builds of today we have fixed a long standing bug in

the vulnerability functions, a bug which has been lurking in the code

since the first implementation of the risk calculators.

From the point of view of a regular user the bug is minor but

it changes (slightly) the risk numbers produced by the engine,

so it is being communicated here, for the sake of transparency.

If you are using the scenario_damage or classical_damage calculator

you are not affected by the bug, since such calculators rely on

fragility functions, not on vulnerability functions.

If your vulnerability functions have coefficients of variation

which are all zero you are not affected by this bug.

If your vulnerability functions have nonzero coefficients of

variation, you may be affected by this bug --or you may be not--,

depending on whether or not the calculation is producing ground motion

fields with values below the minimum intensity level in the

vulnerability function --or not--.

If there are sites outside of the maximum distance, for such sites the

ground motion field will be zero, thus below the minimum intensity

level, which is normally greater than zero: your calculation will most

likely be affected by the bug.

For more details, please refer to the comments in the associated bug

report: https://bugs.launchpad.net/oq-engine/+bug/1459926++

Effect: If the asset_correlation parameter has not been used for the

calculation, or if it has been set to zero, the statistical effect of

the change in the risk numbers is equivalent to simply changing the

random seed. However, if your asset_correlation parameter was set to a

nonzero value and if your calculation is affected by the bug as

explained above, the risk numbers generated by the engine were

incorrect, and this bug fix implements the correction for the error.

The next release of the engine (OpenQuake 1.5) will include this

fix, but you can have it now if you install the nightly builds.

This is important only for people doing scenario risk or event based

risk calculations with nonzero asset correlation and nonzero coefficients of

correlation. If you are such an user please tell us, since we are

not aware of anybody using this feature except the GEM staff.

That's all,

     Michele Simionato on behalf of the OpenQuake Team

[Clotaire Michel]

I am using these features, although I did not start my sensitivity analysis yet, so an update will solve the issue without big problems