a thoughtful web.
Good ideas and conversation. No ads, no tracking.   Login or Take a Tour!
comment
DWol  ·  2464 days ago  ·  link  ·    ·  parent  ·  post: Pubski: June 28, 2017

Went to the store to buy chicken, was confronted by a lot of commotion outside - policemen on the ground shouting at colleagues of theirs running around on the roof. A group of bystanders started to form around a lady who said that there had been an attempted armed robbery - moving from a cellphone shop to into the grocery store itself. Unfortunately for these geniuses the area's police station is right across the road so at least 4 were apparently caught after a shootout (!).

___

On a different note, had some interesting work-related ideas and I thought it might be interesting to share:

So a big part of my work involves modelling mining site water balances. One of the most important factors from a risk perspective is the effect of rainfall on dam storage levels. Both in the sense that you can have too little water (halting or otherwise affecting production) or too much water (Bento Rodrigues).

Luckily, rainfall is one thing there is an abundance of data on for long time spans and across a broad spatial scale i.e. finding historical time-series data for whatever area you are interested in is not hard. The question is then: how best to take into account the change in rainfall over time from a modelling/simulation perspective?

On the simpler end you have the approach I took last year in my preliminary research - average out historical data into two annual groups, a wet and dry season. This is a bit too simple however. The next step is to add a bit of spice by propagating the variance of each seasonal value through the model (probably Monte Carlo?). That way you can see the sensitivity of the outputs to the input as well.

I've however been looking at some work by others that seems to hint that we need to go further. In particular, the effect of this kind of (hourly/daily/monthly) variation is felt dynamically - the water-related processes can hardly be assumed to be at steady-state. So now we need to set up a dynamic model of the process.

The last spanner in the works is that, in many places, there are climatic oscillations which act on scales bigger than a year (e.g. El NiƱos and the like). The effect of this is to cause more incidences of droughts and "floods" than what would be expected by chance, if chance were defined by the distribution of values historically (according to these guys).

Their analysis was purely historical one, looking into it from the euphemistic perspective of portfolio risk... for me it raised some important questions as to how to incorporate this in a predictive model to evaluate processes in the design stage, or help current operations to adapt to un-envisioned risks. Today I read about a Markov chain-based model that incorporates the chance of switching from e.g. a wetter-than-usual to a dryer-than-usual rainfall histogram based on the historical tendencies.

I think this stuff has some wider implementation possibilities - many ore bodies also have this kind of dual character. Complex mineralisations can have you switching from a low-sulphide to a high-sulphide ore. If these are just averaged out, you lose a lot of important insights into how, maybe, the downstream flotation is affected or even the potential for acid rock drainage impacts. So a lot of interesting work to be done in this space!