1. Seismicity: prediction or forecasting?
Steve Spottiswoode
SANIRE September 2009
300 files; 300 Mb; 5 mines; lines code

2. Structure of talk Prediction vs forecasting
Views on short-term prediction
Test for useful predictability of large events
Seismic data
Removing Type “A” events
Stacking of fore- and after-shocks
Seismicity rates
Number of events
Large events appear to have more foreshocks: why?
Most hopeful events
Apparent volume
Conclusions

3. Prediction vs forecasting
These words are commonly used interchangeably.
US seismologists (e.g. Jordan, 2009) make a clear distinction between forecasting and prediction:
Prediction refers to short-term warnings with a high probability (and hopefully a high probability of success).
Forecasting can apply at all times and places and describes a low probability of occurrence,
Mines would like successful prediction of large and damaging seismic events or rockbursts.
Wikipedea: “A prediction is a statement or claim that a particular event will occur in the future in more certain terms than a forecast.”

4. California earthquake forecasting, updated hourly
based on:
50-year history
and aftershocks

5. Views on short-term prediction
Mines expect short-term prediction:
“…a local production manager put it bluntly: they want to know where, when and how large the next potentially damaging tremor will be.”
Van Aswegen (2007)
More views 

6. Views on short-term prediction
“This (seismic) information is often only made available by mines after injuries and deaths and cannot be relied on because it always indicates that there were no major warning (sic) before the main incidents that injure and claims lives.”
Presidential Mine Health and Safety Audit (DME, 2009)
Is it the data or the value of the data for warnings that is unreliable?

7. Testing for useful predictability of large events
We want to see whether
- the event rate or character changes before large events
and
this change does not commonly occur before small events.
If both are true, then useful prediction is possible.
If either is false, then useful prediction is not possible.
I will test these over time scales of minutes to two weeks.

8. Data from five mines 2 Carbon Leader Reef (CLR)
1 Ventersdorp Contact Reef (VCR)
1 Vaal Reefs (VR)
1 Platinum (Plat)

9. Removing Type “A” events
Small & close together in space and time.
Mines are locating numerous Type “A” events
Ebrahim-Trollope (1998): Fracturing (vs Structural)
Finnie et al (2000): Development blasts
Richardson & Jordan (2001 & 2002): Fracture dominated
Spottiswoode and Linzer (2005): Development blasts
More on Type “A”

10. Time sequence of individual Type “A” events
Clusters of 10 events and more locate in different places to large events
Many events in succession
More on Type “A”

11. Groups of Type “A” events migrate
Mid point of clusters migrate at 0.77 m/s
Use only M>-0.8 events to avoid Type “A” events

12. Stacking fore- and after-shocks
Main shocks are “stacked” at zero time and distance, with locations of fore and after shocks expressed as differences in time and location
Consider all events within 200m of main shocks.
Foreshocks at negative time and aftershocks at positive time

13. Event rate ~Constant rate Aftershocks Blasting time Fore shocks
Quieter after big
Drie; # evs; 14 days & 48 hours; M>2.5
Lots of graphs 

14. Event rate over three weeks
Drie; # evs; 14 days & 48 hours; M>2.5

15. Event rate over three days
Drie; # evs; 14 days & 48 hours; M>2.5
Fore shocks

16. Apparent Volume

17. Q: Why do large events appear to have more and bigger foreshocks
Q: Why do large events appear to have more and bigger foreshocks? A: Small aftershocks are under-reported
Only events A & B had two fore shocks within prvious 24 hours
Mathematically, there is a simple symmetry between foreshocks and aftershocks.

18. Recent & previous seismicity: Details for “anomalous” events follow
Were any individual large events preceeded by accelerated seismicity rate?
Drie; values; 14 days & 48 hours; M > 2.5
Recent & previous seismicity: Details for “anomalous” events follow

19. Drie; values; 14 days & 48 hours; M > 2.5

20. Conclusions
Using data from current mine networks, changes in seismicity before large events are the same as changes before small events, on time scales of up to two weeks.
Small events are numerous and can occur at any time.
Use of seismic data for estimating daily changes in seismic hazard therefore appears to be of little use.

21. Spottiswoode retirement fund
Acknowledgements
Discussions with many people:
Martin Pretorius, Kevin Brentley & team, Gerrie van Aswegen, Gerhardt Hoffmann, Aleksander Mendecki, Ray Durrheim, Shaun Murphy, Kevin Riemer, Frans Castelyn, Shana Ebrahim-Trollope, Stan Lasocki
Data from mines
Sponsored by
Spottiswoode retirement fund