Geospatial modelling of geotechnical parameters…..worth the effort? (Part 2)

In my previous post (click here), I highlighted some of the reasons why RQD is probably not the best “parameter” to geostatistically model. So why is RQD being geostatiscally modelled by some practitioners? In my opinion, simply because it’s just about the only geotechnical parameter that is almost universally available (We got RQD!) But does this make it valid to attempt to interpolate it? What kind of result are we hoping to get and how are we going to use it? Importantly, does it justify the effort?

This brings me to “models” and their “usefulness”, given a specific “objective”. A model, whatever form, is just a human construct to better understand complex real world systems, by providing necessary information to make some sort of decision or prediction. Models are precisely wrong, as a statement is undeniable. But they can be roughly right to guide decisions. Models themselves require information, which is just “data” put into context. “Data” can be anything: a number, character, word, graphs, codes, symbol...even RQD!

Keep in mind that geotechnical models can be conceptualized to various degrees of sophistication, for example;

• a summary description of the geotechnical characteristics of the principal domains,
• tabulated/graphed statistics of key parameters within domains,
• a simple 3D geospatial domain model with assigned average values,
• 3D geospatial domain model with stochastic models of structures and other parameters values,
• stochastic 3D models of structures (e.g. discrete fracture networks)
• geospatial model of 3D deterministic large-scale structures,
• geostatistical block models of certain parameters (where appropriate!).

The decision on what type of model to construct is dictated by many factors, including; 

• aims, objectives and end use of model
• level of study and required reliability (e.g. conceptual, FS, operations, etc)
• scale of required models
• time constraints, and most importantly
• availability and quality of data to support selected modelling process

RQD is not really robust scientific “data”, therefore it does it make sense to use it for rigorous geostatistical interpolation. However, because it is almost universally collected it does have a certain utility to identify zones of poor rock mass quality in 3D space.

One of the biggest issues with the validity of interpolating RQD is that, as explained in the previous post, it’s main purpose is to “red flag” highly fractured zones, like faults, shear zones and contacts, which are usually logged in smaller intervals compared to host rock mass. In this case, apart from the smearing effect due to compositing, linearly averaging between points on a larger support (block model) simply doesn’t make sense.

So are geospatial models of RQD still useful?...maybe. As I see it, they can be used as crude heat maps to indicate where “poor” and “good” ground are fuzzily distributed. These heat maps should be used to guide further interrogation of data, further data collection planning and model synthesis, but in general, should not really be fed directly into down-stream engineering calculations. Furthermore, there always is the risk that anyone seeing a colourful 3D RQD model may falsely give it credibility and assume some scientific rigour.

Let me illustrate my point. OK, the image of the pretty coloured contours of RQD on the pit on the previous post was actually created by me. Now, before you start calling me out for being a hypocrite, hear me out…..

I created this “model” for one small distinct “objective” and that was speed to understanding of geological controls on rock mass quality (not just RQD values!). I had been given a database of RQD from mixed geotech and resource drilling. I knew the data was a bit funky. Also, I only had little background on project geology and very little time to get my team to develop a conceptual geotechnical model. So what did I do? After quickly (I mean couple of hours) reviewing the geological context, lithology and alteration models and main stuctures, I threw the “data” at a geological modelling package, whipped up some linear averaging interpolant biased along main perceived anisotropy directions (from my brief appreciation of fault and lithostratigraphic controls) and voila! Pretty pictures indeed. It cannot be called rigorous in any way, shape or form.

Was it scientifically valid? No. But, it was somewhat “useful” for my time compressed objective. On overlaying all other geological data in 3D and a bit of filtering and exploratory stats, I quickly discovered which lithologies were qualitatively “poorer” than others, which faults, and importantly, which contacts potentially controlled rock mass quality. The end result is the title image this article. Yep, that’s the end result….. A rough hand-drawn sketch of my initial interpretation, or hypothesis (one of the many possibles!), to give to my team to explore.

Did it justify the effort? Well yes, for this specific purpose and circumstance only. It was useful to quickly arrive at the starting point to guide the conceptual modelling process. Well, did I use this model for down-stream engineering? Definitely no. I understood that it was not a rigorous model. I understand the issues with RQD as “data” and issues with interpolating RQD, and it definitely was not presented in any report!

Geostatistical interpolation of RQD is difficult to scientifically justify from a number of perspectives. However, geospatial RQD models can be still useful under limited conditions and purposes, and perhaps to demonstrate what could be possible if we have sufficient amounts of geotechnical data. Unfortunately, RQD is the wrong variable.

In my next post, I’ll explore some other geotechnical variables that may better suited to geostatistical and other forms of “modelling”.