38.4.1 hypoDD example

We will illustrate the use of HYPODD with an example data set from the Norwegian National Seismic network. The data consist of a swarm of event occurring mostly in 2016. All data and parameter files are available in the test data set under WOR/hypodd. The original data consists of 388 events located in an area of about 10km in diameter, see Figure 38.2.

Figure 38.2: Dataset for testing HYPODD. To the left is seen the epicentres and to the right the depth profile taken from SW to NE as outlined on the left figure. There are 388 events.

Despite that two stations (NBB13 and NBB15) are within 5-10 km of the swarm, the epicentre pattern is diffuse. Likewise the depths, where many are at the unlikely depth of zero km. Some testing was done to obtain a best solution in terms of number of events remaining and RMS of travel time residuals, see Figure 38.3. The corresponding parameter file is listed below.

Figure 38.3: Results of the hypodd inversion with best parameters. The number of events have been reduced to 248. The red color indicates the original solutions and the green the hypodd solutions.

It is seen that the epicenters are now in 3 groups with the southernmost being the most clear. The depths are now much more clustered with the southernmost group the most concentrated. Nearly all zero depth events have been removed or relocated.

Now some of the parameters were then varied, one by one, to see the effect on the results.

Separation distance

One of the more sensitive parameters is the event separation distance (WDCT). For the reference solution, 2 km was used for the 2. set of parameters and 5 for the first set. However with such a small cluster, one could think that 1km would be more reasonable so this was tested for the 2. set of parameters, see Figure 38.2 for the result.

Figure 38.4: Testing with event separation distance of 1 km. There are 158 events.

The clustering has not changed much but much less results are obtained. A test was also made with 4km (276 events), however clustering was much less clear. So it seems that the original 2km separation is ok. A test was also made to increase WDCT to 10km in first data set, but results were worse.

Number of iterations

The number of iterations should be enough that the RMS does not decrease anymore.

Figure 38.5: The number of iteration has been changed from 8+8 to 4+4. The number of events is 268.

With fewer iterations, there were more events, but the clustering is less clear. More iterations were also tested, but with no significant difference. In this data set 8+8 iterations seems ok.

Maximum distance

In the standard parameter set the maximum distance it is set to 200km. Considering that far stations might have less clear arrivals than near station, using in shorter distance might improve the results. Testing with 80k max distance is seen in Figure 38.6. Only 4 iterations were needed in the 2. parameter set.

Figure 38.6: Testing with maximum distance of 80 km instead of 200 km. There are 231 events.

Clearly the clustering is less clear than when using standard parameters so larger distances should be used for this data set.

Importance of nearest station

Station NBB13 is very close to the southern cluster so a test was made weighting out the station (with NOR2DD).

Figure 38.7: Results without using station NBB13. There were 212 events.

The clustering is now less clear so one near stations can change results significantly. But the results are still better with HYPODD than using standard locations. NBB13 has readings for 292 events out of the original 388. However of the 248 events with the standard parameters, 225 events had station NBB13 so the staion is clearly important for the best solutions.

Other tests: Using only P resulted in fewer events (232) with more scatter, maybe not surprising. The outlier parameter WRCT did not affect the results much provided they were keeps within 'reasonable' limits as in our example.

These tests are mostly valid for the current data set, other data sets might require other parameters. But the test should give an indication of the importance of the different parameters. In many cases there is a trade-off between number of events which can be relocated and the demand for stricter parameters. For a better understanding, read the hypodd manual and paper.

Hypodd.inp file for the best solution.

*--- input file selection
* cross correlation diff times:
*catalog P diff times:
* event file:
* station file:
*--- output file selection
* original locations:
* relocations:
* station information:
* residual information:
* source paramater information:
*--- data type selection: 
* IDAT:  0 = synthetics; 1= cross corr; 2= catalog; 3= cross & cat 
* IPHA: 1= P; 2= S; 3= P&S
* DIST:max dist [km] between cluster centroid and station 
    2     3      200
*--- event clustering:
* OBSCC:    min # of obs/pair for crosstime data (0= no clustering)
* OBSCT:    min # of obs/pair for network data (0= no clustering)
     0     0        
*--- solution control:
* ISTART:  	1 = from single source; 2 = from network sources
* ISOLV:	1 = SVD, 2=lsqr
* NSET:      	number of sets of iteration with specifications following
    2        2      2 
*--- data weighting and re-weighting: 
* NITER: 		last iteration to used the following weights
* WTCCP, WTCCS:		weight cross P, S 
* WTCTP, WTCTS:		weight catalog P, S 
* WRCC, WRCT:		residual threshold in sec for cross, catalog data 
* WDCC, WDCT:  		max dist [km] between cross, catalog linked pairs10:49 AM 24/02/2020
* DAMP:    		damping (for lsqr only) 
*       ---  CROSS DATA ----- ----CATALOG DATA ----
   8    1      -9   -9   -9   1.00   0.5   5     5   20
   8    1      -9   -9   -9   1.00   0.2   2     2   20
*--- 1D model:
* NLAY:		number of model layers  
* RATIO:	vp/vs ratio 
* TOP:		depths of top of layer (km) 
* VEL: 		layer velocities (km/s)
   6     1.74
* TOP 
0.0 12.0 23.0 31.0 50.0 80.0
6.2 6.6 7.1 8.05 8.25  8.5
*--- event selection:
* CID: 	cluster to be relocated (0 = all)
* ID:	cuspids of event to be relocated (8 per line)
* CID    
* ID