Difference between revisions of "CyberShake BBP Integration"

From SCECpedia
Jump to navigationJump to search
Line 570: Line 570:
 
|-
 
|-
 
| M6.85, Santa Monica<br/>src 262, rup 8, rv 1
 
| M6.85, Santa Monica<br/>src 262, rup 8, rv 1
| [[File:Seismogram_USC_262_8_1.grm|400px]]
+
| [[File:Seismogram_USC_262_8_1.png|400px]]
| [[File:Seismogram_USC_262_8_1_site_response.grm|400px]]
+
| [[File:Seismogram_USC_262_8_1_site_response.png|400px]]
| [[File:Seismogram_USC_262_8_1_hf.grm|400px]]
+
| [[File:Seismogram_USC_262_8_1_hf.png|400px]]
| [[File:Seismogram_USC_262_8_1_bbp.grm|400px]]
+
| [[File:Seismogram_USC_262_8_1_bbp.png|400px]]
 
|}
 
|}

Revision as of 06:19, 24 September 2020

This page details the process of integrating CyberShake with the Broadband Platform, so that we can produce stochastic high-frequency seismograms in CyberShake as a complement to deterministic low-frequency seismograms.

We performed a similar integration process for CyberShake 1.4 and CyberShake Study 15.12. This time, we would like to avoid maintaining a separate CyberShake version of the high-frequency stochastic codes. Instead, we would like to invoke the BBP codes from CyberShake, so that as new modifications are made to BBP, CyberShake can use these new codes without requiring the entire integration process again.

Approach

We have identified the following BBP executables, or elements, which are needed in CyberShake:

  • srf2stoch (C)
  • hb_high (Fortran)
  • wcc_getpeak (C)
  • wcc_siteamp14 (C)
  • wcc_tfilter (C)
  • wcc_resamp_arbdt (C)
  • wcc_add (C)
  • integ_diff (C)

As part of the BBP, each of these pieces of code contains a main() function, which typically works as follows:

main() {
  // Parse command-line parameters
  // Open and read input files into input data structure
  // Execute science kernel, populating output data structure
  // Open and write output files from output data structure
}

For CyberShake, we would like to be able to use the science kernels of the BBP elements, but provide CyberShake-specific parameters, pass data structures around in memory between multiple elements, and read from and write to different data formats. To accomplish this, we propose extracting the science kernels from the main() functions and creating subroutines with them, separating the I/O from the scientific calculations. Following this approach, a revised main method would contain:

main() {
  // Open and read input files into input data structure
  science_kernel_subroutine(command-line arguments, input_data_structure, output_data_structure)
  // Open and write output files from output data structure
}

science_kernel_subroutine(command-line arguments, input_data_structure, output_data_structure) {
  // Use input data structure for processing
  // Place results in output data structure
}

Then, in the CyberShake codebase, we would use this function as follows:

// Open and read input files into input data structure
// Allocate memory for output data structure
// Create parameter string with CyberShake-specific parameters for getpar to parse
science_kernel_subroutine(parameter_string, input_data_structure, output_data_structure)
// Do additional processing with output data structure
// Open and write output data structure to file

The intent is that these changes would be pushed out to the BBP codebase and also to Rob Graves, so that future revisions work from this refactored version and are straightforward to integrate with CyberShake.

Filename convention

Since for CyberShake we link the CyberShake code with the subroutine object files, the BBP main methods can't be in the subroutine object files (otherwise we would have multiple main()s), and must be contained in separate files. The convention we will follow is to have all subroutines in <module>_sub.c, and the main method with subroutine prototypes in <module>_main.c. The compiled executable used by BBP will have the name <module>_sub, to distinguish it from the non-refactored executables when testing.

Example

As an example, below we show the proposed modifications for wcc_getpeak. wcc_getpeak doesn't have an output data structure; in the BBP the result is printed, whereas in the subroutine it is returned.

Change Original code Modified code
Subroutine prototype N/A

float wcc_getpeak(int param_string_len, char** param_string, float* seis, struct statdata* head1);

Extract science kernel into subroutine

int main(ac,av) {
...
float *s1, amax;
int i;
...
float max = -1.0e+20;
float min = 1.0e+20;
int inbin = 0;
int outbin = 0;
int keepsign = 0;
float scale = 1.0;
...
s1 = NULL;
s1 = read_wccseis(infile,&head1,s1,inbin);
for(i=0;i<head1.nt;i++)
 {
 if(s1[i] > max)
  max = s1[i];
 if(s1[i] < min)
  min = s1[i];
 }
...

...
s1 = NULL;
s1 = read_wccseis(infile,&head1,s1,inbin);

float peak = wcc_getpeak(ac, av, s1, head1);

printf("%10.2f %13.5e %s\n",head1.edist,peak,head1.stat);
}

float wcc_getpeak(int param_string_len, char** param_string, float* s1, struct statdata* head1) {
 float amax;
 int i;

 float max = -1.0e+20;
 float min = 1.0e+20;
 int keepsign = 0;
 float scale = 1.0;
...
 for(i=0;i<head1->nt;i++)
  {
  if(s1[i] > max)
   max = s1[i];
  if(s1[i] < min)
   min = s1[i];
  }
 ...

Relocate parsing of non-I/O parameters to subroutine

int main(ac,av) {
...
sprintf(infile,"stdin");

setpar(ac,av);
getpar("infile","s",infile);
getpar("inbin","d",&inbin);
getpar("keepsign","d",&keepsign);
getpar("scale","f",&scale);
endpar();

s1 = NULL;
s1 = read_wccseis(infile,&head1,s1,inbin);
...

float wcc_getpeak(int param_string_len, char** param_string, float* s1, struct statdata* head1) {
 float max = -1.0e+20;
 float min = 1.0e+20;
 int keepsign = 0;
 float scale = 1.0;

 setpar(param_string_len, param_string);
 getpar("keepsign","d",&keepsign);
 getpar("scale","f",&scale);
 endpar();

 for(i=0;i<head1->nt;i++)
...

Retain I/O in main function

int main(ac,av)
...
char infile[128];

float max = -1.0e+20;
float min = 1.0e+20;
int inbin = 0;
int outbin = 0;
int keepsign = 0;
float scale = 1.0;

sprintf(infile,"stdin");

setpar(ac,av);
getpar("infile","s",infile);
getpar("inbin","d",&inbin);
getpar("keepsign","d",&keepsign);
getpar("scale","f",&scale);
endpar();

s1 = NULL;
s1 = read_wccseis(infile,&head1,s1,inbin);

for(i=0;i<head1.nt;i++)
...

int main(ac,av)
...
char infile[128];

int inbin = 0;
int outbin = 0;

sprintf(infile,"stdin");

setpar(ac,av);
getpar("infile","s",infile);
getpar("inbin","d",&inbin);
endpar();

s1 = NULL;
s1 = read_wccseis(infile,&head1,s1,inbin);

float peak = wcc_getpeak(ac, av, s1, head1);

printf("%10.2f %13.5e %s\n",head1.edist,peak,head1.stat);

Here is a way CyberShake code could use this modified code:

...
float* seis = malloc(nt*sizeof(float));
fread(fp_in, sizeof(float), nt, seis);
struct statdata head1;
head1.nt = nt;
char** param_string = NULL;
float peak = wcc_getpeak(param_string, 0, seis, &head1);
...

Migration Status

Element Refactored Passes BBP test Called from CyberShake
wcc_getpeak yes yes
wcc_add yes yes
wcc_tfilter yes yes
wcc_resamp_arbdt yes yes
integ_diff yes yes
wcc_siteamp14 yes yes
hb_high yes yes yes
srf2stoch yes yes yes

Verification

hb_high

hb_high is the most difficult code to verify, as it's the most complex.

To assist in verification, we constructed scatter plots comparing the value in CyberShake-calling-BBP to the value in BBP directly, for each point in the acceleration time series.

Our initial results are

CS v BBP scatter s280 r7 rv0 initial.png

Upon further investigation, we uncovered a few issues.

  1. The default value in the BBP for kappa is 0.04, which is what we are using in CyberShake. However, when using the LA Basin velocity model, the default value is overwritten in the BBP and 0.045 is used instead. Once we updated to the right value of kappa, that improved the scatter a bit:
    CS v BBP scatter s280 r7 rv0 kappa.png
  2. In the CyberShake code, the geographic coordinates are rounded to 4 decimal places to agree with the output file that srf2stoch produces. However, this rounding is accomplished by multiplying, adding 0.5, casting to an int as a floor(), and then dividing. This is fine for positive numbers, but for negative numbers int and floor are not equivalent. Casting to an int will truncate towards 0, and floor() will truncate towards negative infinity. The rounding used when C writes files matches using floor(). Once this modification is also made, the seismograms very closely agree:
    CS v BBP scatter s280 r7 rv0 floor.png

A visual comparison looks good:

CS v BBP acc s280 r7 rv0.png

A numerical comparison is also good:

Average absolute difference: 0.000000
Max absolute difference: 0.000005
Average absolute percent difference: 0.024559
Max absolute percent difference: 150.758352
Bins:
0.000000 <= reference amplitude < 0.010000:
	avg diff=0.000000, avg percent diff=0.028306%
0.010000 <= reference amplitude < 0.100000:
	avg diff=0.000000, avg percent diff=0.000988%
0.100000 <= reference amplitude < 1.000000:
	avg diff=0.000002, avg percent diff=0.001870%
1.000000 <= reference amplitude < 10.000000:

Site Response

Visual comparisons for source 280, rupture 7, rv 0:

CS v BBP amp vel s280 r7 rv0.png
CS v BBP amp vel s280 r7 rv0 zoom.png
CS v BBP amp vel s280 r7 rv0 zoom2.png

The scatterplot:

CS v BBP amp vel scatter s280 r7 rv0.png

A numerical comparison looks good:

Average absolute difference: 0.000004
Max absolute difference: 0.000069
Average absolute percent difference: 0.265180
Max absolute percent difference: 265.281916
Bins:
0.000000 <= reference amplitude < 0.010000:
	avg diff=0.000001, avg percent diff=0.297034%
0.010000 <= reference amplitude < 0.100000:
	avg diff=0.000022, avg percent diff=0.076535%
0.100000 <= reference amplitude < 1.000000:
1.000000 <= reference amplitude < 10.000000:
10.000000 <= reference amplitude < 100.000000:
100.000000 <= reference amplitude < 1000.000000:
1000.000000 <= reference amplitude < 10000.000000:

Filtered comparisons

Comparisons should be done using filtered seismograms (2-pass 4th order Butterworth, high-pass at 1 Hz). Below are results for all 10 ruptures after filtering. Each rupture has approximately 60% more rupture surface points than the previous rupture.

The differences in scatter starting with the Garlock rupture can be traced to differences in a division followed by a floor().

For example, a value in the rupt array (which I think is initiation time) is 2.64000e+01. In the CyberShake code, this is represented by 26.3999691; in the BBP code, 26.3999996. The input dt is set to 0.01, but that is represented as 9.99999978E-03. The rupt value divided by dt gives 2639.99707 for CyberShake and 2640.00000 for BBP. Taking the floor, then, gives 2639 for CyberShake and 2640 for BBP. This value is used as an offset into an array, which is then multiplied by another value, so the difference is magnified. Compiling with O0 does not reduce the error.

We seem to get larger/more differences for events with larger ground motions.

Event Seismogram Zoomed Seismogram Scatterplot Numerical comparison
Source 280
Rupture 7
Rup var 0
Tank Canyon, M6.65
CS v BBP amp vel 280 7 0 filtered.png CS v BBP amp vel 280 7 0 filtered zoom.png CS v BBP scatter 280 7 0 filtered.png
12093 of 100000 values above threshold 1e-06
Average absolute difference: 0.000001
Max absolute difference: 0.000012
Average absolute percent difference: 1.488839
Max absolute percent difference: 26.480208
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000001, avg % diff=1.43%
0.01 <= reference amplitude < 0.1:
  avg diff=0.000007, avg % diff=0.0507%
0.1 <= reference amplitude < 1.0:
  None
Source 269
Rupture 8
Rup var 1
Santa Monica, M6.85
CS v BBP amp vel 262 8 1.png CS v BBP amp vel 262 8 1 zoom.png CS v BBP scatter 262 8 1.png
49372 of 100000 values above threshold 1e-06
Average absolute difference: 0.000016
Max absolute difference: 0.001929
Average absolute percent difference: 5.249444
Max absolute percent difference: 210.588523
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000001, avg % diff=5.46%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000166, avg % diff=0.381%
0.10 <= reference amplitude < 1.00:
  avg diff=0.000698, avg % diff=0.334%
1.00 <= reference amplitude < 10.00:
  None
Source 220
Rupture 10
Rup var 2
North Channel, M6.85
CS v BBP amp vel 220 10 2.png CS v BBP amp vel 220 10 2 zoom.png CS v BBP scatter 220 10 2.png
49686 of 100000 values above threshold 1e-06
Average absolute difference: 0.000002
Max absolute difference: 0.000124
Average absolute percent difference: 2.517283
Max absolute percent difference: 25.023028
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000000, avg % diff=2.71%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000018, avg % diff=0.0608%
0.10 <= reference amplitude < 1.00:
  avg diff=0.000080, avg % diff=0.0667%
1.00 <= reference amplitude < 10.00:
  None
Source 269
Rupture 24
Rup var 3
Santa Ynez, M6.95
CS v BBP amp vel 269 24 3.png CS v BBP amp vel 269 24 3 zoom.png CS v BBP scatter 269 24 3.png
49313 of 100000 above threshold 1e-06
Average absolute difference: 0.000001
Max absolute difference: 0.000062
Average absolute percent difference: 6.031340
Max absolute percent difference: 18.205049
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000000, avg percent diff=6.32%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000015, avg percent diff=0.0606%
0.10 <= reference amplitude < 1.00:
  avg diff=0.000061, avg percent diff=0.0578%
1.00 <= reference amplitude < 10.00:
  None
Source 105
Rupture 3
Rup var 4
San Jacinto, M6.95
CS v BBP amp vel 105 3 4.png CS v BBP amp vel 105 3 4 zoom.png CS v BBP scatter 105 3 4.png
16886 of 100000 values above threshold 1e-06
Average absolute difference: 0.000001
Max absolute difference: 0.000019
Average absolute percent difference: 0.202181
Max absolute percent difference: 19.772487
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000001, avg % diff=0.210%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000009, avg % diff=0.0584%
0.10 <= reference amplitude < 1.00:
  None
Source 21
Rupture 2
Rup var 5
Garlock, M7.15
CS v BBP amp vel 21 2 5.png CS v BBP amp vel 21 2 5 zoom.png CS v BBP scatter 21 2 5.png
49823 of 100000 values above threshold 1e-06
Average absolute difference: 0.000031
Max absolute difference: 0.002349
Average absolute percent difference: 6.556178
Max absolute percent difference: 68462.868722
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000008, avg % diff=7.18%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000238, avg % diff=0.961%
0.10 <= reference amplitude < 1.00:
  avg diff=0.000484, avg % diff=0.449%
1.00 <= reference amplitude < 10.00:
  None
Source 100
Rupture 1
Rup var 6
San Jacinto, M7.25
CS v BBP amp vel 100 1 6.png CS v BBP amp vel 100 1 6 zoom.png CS v BBP scatter 100 1 6.png
48228 of 100000 values above threshold 1e-06
Average absolute difference: 0.000019
Max absolute difference: 0.000873
Average absolute percent difference: 21.214481
Max absolute percent difference: 2036.818780
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000005, avg % diff=24.45%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000095, avg % diff=0.298%
0.10 <= reference amplitude < 1.00:
  avg diff=0.000227, avg % diff=0.162%
1.00 <= reference amplitude < 10.00:
  None
Source 10
Rupture 1
Rup var 7
Elsinore, M7.55
CS v BBP amp vel 10 1 7.png CS v BBP amp vel 10 1 7 zoom.png CS v BBP scatter 10 1 7.png
49416 of 100000 values above threshold 1e-06
Average absolute difference: 0.000274
Max absolute difference: 0.018217
Average absolute percent difference: 11.981339
Max absolute percent difference: 12207.828835
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000019, avg % diff=15.68%
0.01 <= reference amplitude < 0.10:
  avg diff=0.000608, avg % diff=1.714%
0.10 <= reference amplitude < 1.00:
  avg diff=0.001434, avg % diff=0.625%
1.00 <= reference amplitude < 10.00:
  avg diff=0.000409, avg % diff=0.0349%
10.00 <= reference amplitude < 100.00:
  None
Source 86
Rupture 0
Rup var 8
San Andreas, M7.65
CS v BBP amp vel 86 0 8.png CS v BBP amp vel 86 0 8 zoom.png CS v BBP scatter 86 0 8.png
50394 of 100000 values above threshold 1.000000e-06
Average absolute difference: 0.002977
Max absolute difference: 0.098694
Average absolute percent difference: 15.380488
Max absolute percent difference: 32075.901501
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000096, avg % diff=22.57%
0.01 <= reference amplitude < 0.10:
  avg diff=0.003266, avg % diff=8.784%
0.10 <= reference amplitude < 1.00:
  avg diff=0.008658, avg % diff=2.653%
1.00 <= reference amplitude < 10.00:
  avg diff=0.010934, avg % diff=0.799%
10.00 <= reference amplitude < 100.00:
  None
Source 128
Rupture 1296
Rup var 9
San Andreas, M8.45
CS v BBP amp vel 128 1296 9.png CS v BBP amp vel 128 1296 9 zoom.png CS v BBP scatter 128 1296 9.png
64646 of 100000 values above threshold 1.000000e-06
Average absolute difference: 0.006450
Max absolute difference: 0.206112
Average absolute percent difference: 19.995406
Max absolute percent difference: 163277.921485
Bins:
1.0e-06 <= reference amplitude < 0.01:
  avg diff=0.000113, avg % diff=31.06%
0.01 <= reference amplitude < 0.10:
  avg diff=0.005178, avg % diff=13.34%
0.10 <= reference amplitude < 1.00:
  avg diff=0.016088, avg % diff=4.456%
1.00 <= reference amplitude < 10.00:
  avg diff=0.019644, avg % diff=1.262%
10.00 <= reference amplitude < 100.00:
  None

Low-frequency site response

We performed low-frequency site response comparisons using the same set of ruptures. This resulted in an average absolute difference of ~2e-5. In all but the smallest value bin, percent differences are between 0.05 and 0.0002%. These results are several orders of magnitude closer than the high-frequency results, and are close enough for our purposes.

Merging code

The merging code low-pass filters the low-frequency seismogram (after site response has been added), resamples it at the same dt as the high frequency seismogram, combines the two, and then processes to obtain IM values. We ran our tests using the standard set of rupture variations and found that the average absolute difference was around 1e-6. In all but the smallest value bin, percent differences are less than 0.005%. These results are better than either the high-frequency or site response results, and are good enough for our purposes.

Changes from Study 15.12

We pioneered the general concept of performing stochastic calculations alongside deterministic ones in CyberShake_1.4 and then futher developed it in CyberShake_Study_15.12. For this integration effort, we are going to make a few changes from the Study 15.12 work.

  • Since the low-frequency content of the stochastic seismograms should be ignored, we are moving the HF filtering from the MergeIM step into HFSim so that we aren't tempted to use the stochastic output of that code unfiltered.
  • We have decided to low-pass filter the deterministic seismograms before merging. This will be done in the MergeIM stage.

Results

Below are example seismograms from each step of the process (deterministic, deterministic + site response, stochastic, and merged) for 5 different ruptures.

Event Deterministic Deterministic + site response Stochastic Merged
M6.85, Santa Monica
src 262, rup 8, rv 1
Seismogram USC 262 8 1.png Seismogram USC 262 8 1 site response.png Seismogram USC 262 8 1 hf.png Seismogram USC 262 8 1 bbp.png