def run_polarization(args):
# get inputs
c = args[0]
numCluster = args[1]
nproc = args[2]
clust_method = args[3]
type = args[4]
fs = args[5]
dataPath = args[6]
templatePath = args[7]
outPath = args[8]
norm_component = args[9]
MAD = args[10]
norm_thresh = args[11]
xcorr_percent_thresh = args[12]
snipLen = args[13]
winLen = args[14]
slide = args[15]
if norm_component:
outPath = outPath + "normalized_components/"
templatePath = templatePath + type + "_normalized_3D_clustering/" + clust_method + "/"
else:
templatePath = templatePath + type + "_3D_clustering/" + clust_method + "/"
# get window parameters
numSteps = int((snipLen-winLen)/slide)
# set stations and components
chans = ["HHN","HHE","HHZ"]
stations = ["PIG2","PIG4","PIG5"]
#stat_coords = np.array([[-100.748596,-75.016701],[-100.786598,-75.010696],[-100.730904,-75.009201],[-100.723701,-75.020302],[-100.802696,-75.020103]])
stat_coords = np.array([[-100.786598,-75.010696],[-100.723701,-75.020302],[-100.802696,-75.020103]])
plotStat = 'PIG2'
# read imagery data, get coordinate system, convert station coordinates to x and y, and take average station location
file = "/media/Data/Data/PIG/TIF/LC08_L1GT_001113_20131012_20170429_01_T2_B4.TIF"
sat_data = rasterio.open(file)
p2 = Proj(sat_data.crs,preserve_units=False)
p1 = Proj(proj='latlong',preserve_units=False)
[stat_x,stat_y] = transform(p1,p2,stat_coords[:,0],stat_coords[:,1])
avg_stat_x = np.mean(stat_x)
avg_stat_y = np.mean(stat_y)
# set frequency
prefiltFreq = [0.05,1]
freq = [0.05,1]
# load waveforms
#waves = obspy.read(templatePath + type + '_waveforms_' + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + 'Hz.h5')
# load detection times
detFile = h5py.File(templatePath + "detection_times.h5","r")
detTimes = list(detFile["times"])
detFile.close()
# load clustering results
clustFile = h5py.File(templatePath + str(numCluster) + "/" + str(numCluster) + "_cluster_predictions_" + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + "Hz.h5","r")
pred = np.array(list(clustFile["cluster_index"]))
centroids = list(clustFile["centroids"])
clustFile.close()
# read in correlation results for the current cluster
corrFile = h5py.File(templatePath + str(numCluster) + "/centroid" + str(c) + "_correlations_" + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + "Hz.h5","r")
cluster_xcorr_coef = np.array(list(corrFile["corrCoefs"]))
corrFile.close()
# get indices of waveforms in current cluster and events in cluster above xcorr threshold
clusterInd = [i for i, x in enumerate(pred==c) if x]
if MAD:
mad = stats.median_absolute_deviation(abs(cluster_xcorr_coef))
mad = round(mad,2)
xcorr_percent_thresh = mad/max(abs(cluster_xcorr_coef))
print(mad)
print(xcorr_percent_thresh)
n_events = round(xcorr_percent_thresh*len(clusterInd))
else:
n_events = round(xcorr_percent_thresh*len(clusterInd))
threshInd = abs(cluster_xcorr_coef).argsort()[-1*n_events:][::-1]
# make array for storing pca vector sums and storing data to plot
event_index = np.zeros((numSteps*len(threshInd),1),"float64")
all_first_components = np.zeros((numSteps*len(threshInd),2),"float64")
clusterEventsAligned = np.zeros((len(threshInd),snipLen*fs),'float64')
# loop through indices of events in current cluster
for i in range(len(threshInd)):
# make array for storage of pca components and empty obspy stream for storing one trace from each station
first_component_sums = np.zeros((numSteps,2),"float64")
event_stat = obspy.read()
event_stat.clear()
# loop through stations to get one trace from each to find earliest arrival
try:
for stat in range(len(stations)):
# get times bounds for current event and read event
#eventLims = [waves[clusterInd[threshInd[i]]].stations.starttime,waves[clusterInd[threshInd[i]]].stations.starttime + snipLen]
starttime = UTCDateTime(detTimes[clusterInd[threshInd[i]]])
endtime = starttime + snipLen
eventLims = [starttime,endtime]
event_stat += readEvent(dataPath + "MSEED/noIR/",stations[stat],chans[1],eventLims,freq)
# find station with earliest arrival
first_stat = observed_first_arrival(event_stat)
# loop though stations to perform PCA on all windows in the event on each station's data
for stat in range(len(stations)):
# compute pca components for all windows in the event
first_components = compute_pca(dataPath,stations[stat],chans,fs,winLen,slide,numSteps,freq,eventLims)
# correct polarization direction based on first arrival
first_components_corrected = correct_polarization(first_components,stations,first_stat,avg_stat_x,avg_stat_y,stat_x,stat_y)
# sum results (this is vector sum across stations of pca first components for each window)
first_component_sums = first_component_sums + first_components_corrected
# give user output every % complete
if round(i/len(threshInd)*100) > round((i-1)/len(threshInd)*100):
print(str(round(i/len(threshInd)*100)) + " % complete (cluster " + str(c) +")")
# fill results vector
all_first_components[i*numSteps:(i+1)*numSteps,:] = first_component_sums
event_index[i*numSteps:(i+1)*numSteps,1] = clusterInd[threshInd[i]]*np.ones((numSteps,1))
except:
# give output if no data on current station
print("Skipping cluster " + str(c) + " event " + str(i) + " (missing data on " + stations[stat] + ")")
# make plots
fig,ax = plt.subplots(nrows=2,ncols=1,figsize=(8, 8),gridspec_kw={'height_ratios':[1,0.4]})
# do some stuff
corners = np.array([[sat_data.bounds[0],sat_data.bounds[1]], # bottom left
[sat_data.bounds[0],sat_data.bounds[3]], # top left
[sat_data.bounds[2],sat_data.bounds[1]], # bottom right
[sat_data.bounds[2],sat_data.bounds[3]]]) # top right
corners_lon,corners_lat = transform(p2,p1,corners[:,0],corners[:,1])
# plot imagery
show(sat_data,ax=ax[0],cmap="gray")
# handle axes
percent = xcorr_percent_thresh*100
plt.suptitle("Cluster " + str(c) + " Polarizations (top " + str(percent) + "% of events)")
# make array for storage
back_azimuths = np.empty((0,1),"float64")
baz_event_index = np.empty((0,1),"float64")
# plot pca compoments that exceed norm threshold
# be wary- the transformed coordinate system's x-axis is meters north and the y-axis is meters east, so the pca_first_component[~,0] (which is cartesian x) is in [L] east
# and therefore along the transformed y-axis and the pca_first_component[~,1] (which is cartesian y) is in [L] north and therefore along the transformed x-axis
count = 0
for s in range(len(all_first_components)):
# only plot and save results if length of resultant vector has a norm exceeding the threshold
if np.linalg.norm(all_first_components[s,:]) > norm_thresh:
# calculate back azimuths and save in array
baz = compute_baz(all_first_components[s,:])
back_azimuths = np.vstack((back_azimuths,baz))
baz_event_index = np.vstack(event_index[s])
count += 1
# compute histogram of back azimuths
baz_hist,edges = np.histogram(back_azimuths,bins=np.linspace(0,360,37))
# set up colormap
colors = [ cm.plasma(x) for x in np.linspace(0,1,max(baz_hist)+1)]
# plot all rays in 10-degree bins with length proportional to # of windows in that bin
rays = np.zeros((36,2),'float64')
scale = 40000
max_width = 2*np.pi*scale/36
max_width = 8
for i in range(36):
angle = i*10
rays[i,:] = compute_rays(angle)
rayLength = baz_hist[i]/max(baz_hist)*scale
[x,y] = [np.linspace(avg_stat_x,avg_stat_x+rays[i,0]*rayLength,100),
np.linspace(avg_stat_y,avg_stat_y+rays[i,1]*rayLength,100)]
lwidths=np.linspace(0,max_width,100)*rayLength/scale
points = np.array([x, y]).T.reshape(-1, 1, 2)
segments = np.concatenate([points[:-1], points[1:]], axis=1)
lc = LineCollection(segments, linewidths=lwidths,color='maroon',alpha=0.5,zorder=len(all_first_components)*10)
ax[0].add_collection(lc)
# define, transform, and plot lat/lon grid
lat = [-74,-75]
lon = [-98,-100,-102,-104]
x_lab_pos=[]
y_lab_pos=[]
line = np.linspace(corners_lat[0]+1,corners_lat[2]-1,100)
for l in lon:
line_x,line_y = transform(p1,p2,np.linspace(l,l,100),line)
ax[0].plot(line_x,line_y,linestyle='--',linewidth=1,c='gray',alpha=1)
y_lab_pos.append(line_y[np.argmin(np.abs(line_x-corners[0,0]))])
line = np.linspace(corners_lon[0]-2,corners_lon[1]+1,100)
for l in lat:
line_x,line_y = transform(p1,p2,line,np.linspace(l,l,100))
ax[0].plot(line_x,line_y,linestyle='--',linewidth=1,c='gray',alpha=1)
x_lab_pos.append(line_x[np.argmin(np.abs(line_y-corners[0,1]))])
ax[0].set_xlim([corners[0,0],corners[2,0]])
ax[0].set_ylim([corners[0,1],corners[1,1]])
ax[0].set_xticks(x_lab_pos)
ax[0].set_xticklabels(labels=[str(lat[0]) + '$^\circ$',str(lat[1]) + '$^\circ$'])
ax[0].set_xlabel("Latitude")
ax[0].set_yticks(y_lab_pos)
ax[0].set_yticklabels(labels=[str(lon[0]) + '$^\circ$',str(lon[1]) + '$^\circ$',str(lon[2]) + '$^\circ$',str(lon[3]) + '$^\circ$'])
ax[0].set_ylabel("Longitude")
# colors
k1 = plt.rcParams['axes.prop_cycle'].by_key()['color'][0]
k2 = plt.rcParams['axes.prop_cycle'].by_key()['color'][1]
# plot ice front
front_x = [-1.644e6,-1.64e6,-1.638e6,-1.626e6,-1.611e6,-1.6095e6,-1.6055e6,-1.6038e6,-1.598e6,-1.6005e6,-1.6e6,-1.595e6]
front_y = [-3.34e5,-3.33e5,-3.44e5,-3.445e5,-3.475e5,-3.43e5,-3.4e5,-3.413e5,-3.356e5,-3.32e5,-3.289e5,-3.29e5]
ax[0].plot(front_x, front_y,c=k1,zorder=len(all_first_components)*10)
# plot rift
rift1_x = [-1.63e6,-1.6233e6,-1.6132e6,-1.6027e6]
rift1_y = [-3.255e5,-3.237e5,-3.236e5,-3.281e5]
ax[0].plot(rift1_x,rift1_y,c=k2,zorder=len(all_first_components)*10)
rift2_x = [-1.63e6,-1.6232e6,-1.6132e6]
rift2_y = [-3.28e5,-3.2706e5,-3.236e5]
ax[0].plot(rift2_x,rift2_y,c=k2,zorder=len(all_first_components)*10)
# plot station locations
ax[0].scatter(stat_x,stat_y,marker="^",c='black',zorder=count*10)
# add north arrow
ax[0].arrow(avg_stat_x-65000,avg_stat_y+70000,-10000,0,width = 500,head_width=3000,head_length=3000,fc="k", ec="k",zorder=len(all_first_components)*10)
ax[0].text(avg_stat_x-74000,avg_stat_y+73000,"N",size="large",zorder=len(all_first_components)*10)
# add distance scale
ax[0].plot(np.linspace(avg_stat_x-60000,avg_stat_x-80000,10),np.ones(10)*avg_stat_y-30000,c="k")
ax[0].text(avg_stat_x-82000,avg_stat_y-26000,"20 km",size="medium")
# plot events and centroid
waveFile = h5py.File(templatePath + str(numCluster) + "/aligned_cluster" + str(c) + "_waveform_matrix_" + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + "Hz.h5","r")
alignedWaves = np.array(list(waveFile["waveforms"]))
waveFile.close()
alignedWave_fs = 2
t = np.linspace(0,snipLen*3,(snipLen*alignedWave_fs+1)*3)
maxAmps = np.zeros((len(threshInd),1),'float64')
for w in range(len(threshInd)):
ax[1].plot(t,alignedWaves[threshInd[w]],'k',alpha=0.1)
maxAmps[w] = np.max(np.abs(alignedWaves[threshInd[w]]))
ax[1].plot(t,centroids[c].ravel()/np.max(abs(centroids[c].ravel()))*np.median(maxAmps))
ax[1].set_ylim([-4*np.median(maxAmps),4*np.median(maxAmps)])
ax[1].title.set_text('Centroid and Events Used in Polarization Analysis')
ax[1].ticklabel_format(style='sci', axis='y',scilimits=(0,0))
ax[1].set_xlabel("Time (seconds)")
ax[1].set_ylabel("Velocity (m/s)")
ax[1].set_xlim([t[0],t[-1]])
clusterChans = ["HHZ","HHN","HHE"]
ax[1].set_xticks([0,snipLen/2,snipLen,snipLen*3/2,snipLen*2,snipLen*5/2,snipLen*3])
xPos = [snipLen,snipLen*2]
for xc in xPos:
ax[1].axvline(x=xc,color='k',linestyle='--')
ax[1].set_xticklabels(['0','250\n'+clusterChans[0],'500 0 ','250\n'+clusterChans[1],'500 0 ','250\n' + clusterChans[2],'500'])
ax[1].grid(linestyle=":")
ax[1].grid()
plt.tight_layout()
#plt.show()
plt.savefig(outPath + "win_len_" + str(winLen) + "/norm>" + str(norm_thresh) + "/top_" + str(percent) + "%/" + str(numCluster)+ "/cluster_" + str(c) + "_polarizations.png")
plt.close()
# save actual backazimuth data
outFile = h5py.File(outPath + "win_len_" + str(winLen) + "/norm>" + str(norm_thresh) + "/top_" + str(percent) + "%/" + str(numCluster) + "/cluster_" + str(c) + "_backazimuths.h5","w")
outFile.create_dataset("backazimuths",data=back_azimuths)
outFile.create_dataset("index",data=baz_event_index)
outFile.close()
event_stat += readEvent(dataPath + "MSEED/noIR/", stats[stat], chans[1],
eventLims, freq)
# find station with earliest arrival
first_stat = observed_first_arrival(event_stat)
# loop though stations to perform PCA on all windows in the event on each station's data
for stat in range(len(stats)):
# compute pca components for all windows in the event
first_components = compute_pca(dataPath, stats[stat], chans, fs, winLen,
slide, numSteps, freq, eventLims)
# correct polarization direction based on first arrival
first_components_corrected = correct_polarization(first_components, stats,
first_stat, avg_stat_x,
avg_stat_y, stat_x,
stat_y)
# sum results (this is vector sum across stations of pca first components for each window)
first_component_sums = first_component_sums + first_components_corrected
# fill results vector
all_first_components = first_component_sums
# make plots
#fig,ax = plt.subplots(nrows=2,ncols=1,figsize=(8, 8),gridspec_kw={'height_ratios':[1,0.4]})
fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(8, 8))
# do some stuff
corners = np.array([
[sat_data.bounds[0], sat_data.bounds[1]], # bottom left
def run_polarization(args):
# get inputs
c = args[0]
numCluster = args[1]
nproc = args[2]
clust_method = args[3]
type = args[4]
fs = args[5]
dataPath = args[6]
templatePath = args[7]
outPath = args[8]
norm_component = args[9]
MAD = args[10]
norm_thresh = args[11]
xcorr_percent_thresh = args[12]
snipLen = args[13]
winLen = args[14]
slide = args[15]
if norm_component:
outPath = outPath + "normalized_components/"
templatePath = templatePath + type + "_normalized_3D_clustering/" + clust_method + "/"
else:
templatePath = templatePath + type + "_3D_clustering/" + clust_method + "/"
# get window parameters
numSteps = int((snipLen-winLen)/slide)
# set stations and components
chans = ["HHN","HHE","HHZ"]
stations = ["PIG2","PIG4","PIG5"]
#stat_coords = np.array([[-100.748596,-75.016701],[-100.786598,-75.010696],[-100.730904,-75.009201],[-100.723701,-75.020302],[-100.802696,-75.020103]])
stat_coords = np.array([[-100.786598,-75.010696],[-100.723701,-75.020302],[-100.802696,-75.020103]])
# read imagery data, get coordinate system, convert station coordinates to x and y, and take average station location
file = "/media/Data/Data/PIG/TIF/LC08_L1GT_001113_20131012_20170429_01_T2_B4.TIF"
sat_data = rasterio.open(file)
p2 = Proj(sat_data.crs,preserve_units=False)
p1 = Proj(proj='latlong',preserve_units=False)
[stat_x,stat_y] = transform(p1,p2,stat_coords[:,0],stat_coords[:,1])
avg_stat_x = np.mean(stat_x)
avg_stat_y = np.mean(stat_y)
# set frequency
prefiltFreq = [0.05,1]
freq = [0.05,1]
# load waveforms
#waves = obspy.read(templatePath + type + '_waveforms_' + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + 'Hz.h5')
# load detection times
detFile = h5py.File(templatePath + "detection_times.h5","r")
detTimes = list(detFile["times"])
detFile.close()
# load clustering results
clustFile = h5py.File(templatePath + str(numCluster) + "/" + str(numCluster) + "_cluster_predictions_" + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + "Hz.h5","r")
pred = np.array(list(clustFile["cluster_index"]))
centroids = list(clustFile["centroids"])
clustFile.close()
# read in correlation results for the current cluster
corrFile = h5py.File(templatePath + str(numCluster) + "/centroid" + str(c) + "_correlations_" + str(prefiltFreq[0]) + "-" + str(prefiltFreq[1]) + "Hz.h5","r")
cluster_xcorr_coef = np.array(list(corrFile["corrCoefs"]))
corrFile.close()
# get indices of waveforms in current cluster and events in cluster above xcorr threshold
clusterInd = [i for i, x in enumerate(pred==c) if x]
if MAD:
mad = stats.median_absolute_deviation(abs(cluster_xcorr_coef))
mad = round(mad,2)
xcorr_percent_thresh = mad/max(abs(cluster_xcorr_coef))
print(mad)
print(xcorr_percent_thresh)
n_events = round(xcorr_percent_thresh*len(clusterInd))
else:
n_events = round(xcorr_percent_thresh*len(clusterInd))
threshInd = abs(cluster_xcorr_coef).argsort()[-1*n_events:][::-1]
# make array for storing pca vector sums and storing data to plot
event_index = np.zeros((numSteps*len(threshInd),1),"float64")
all_first_components = np.zeros((numSteps*len(threshInd),2),"float64")
clusterEventsAligned = np.zeros((len(threshInd),snipLen*fs),'float64')
# loop through indices of events in current cluster
for i in range(len(threshInd)):
# make array for storage of pca components and empty obspy stream for storing one trace from each station
first_component_sums = np.zeros((numSteps,2),"float64")
event_stat = obspy.read()
event_stat.clear()
# loop through stations to get one trace from each to find earliest arrival
#try:
for stat in range(len(stations)):
# get times bounds for current event and read event
#eventLims = [waves[clusterInd[threshInd[i]]].stations.starttime,waves[clusterInd[threshInd[i]]].stations.starttime + snipLen]
starttime = UTCDateTime(detTimes[clusterInd[threshInd[i]]])
endtime = starttime + snipLen
eventLims = [starttime,endtime]
event_stat += readEvent(dataPath + "MSEED/noIR/",stations[stat],chans[1],eventLims,freq)
# find station with earliest arrival
first_stat = observed_first_arrival(event_stat)
# loop though stations to perform PCA on all windows in the event on each station's data
for stat in range(len(stations)):
# compute pca components for all windows in the event
first_components = compute_pca(dataPath,stations[stat],chans,fs,winLen,slide,numSteps,freq,eventLims)
# correct polarization direction based on first arrival
first_components_corrected = correct_polarization(first_components,stations,first_stat,avg_stat_x,avg_stat_y,stat_x,stat_y)
# sum results (this is vector sum across stations of pca first components for each window)
first_component_sums = first_component_sums + first_components_corrected
# give user output every % complete
if round(i/len(threshInd)*100) > round((i-1)/len(threshInd)*100):
print(str(round(i/len(threshInd)*100)) + " % complete (cluster " + str(c) +")")
# fill results vector
all_first_components[i*numSteps:(i+1)*numSteps,:] = first_component_sums
event_index[i*numSteps:(i+1)*numSteps,1] = clusterInd[threshInd[i]]*np.ones((numSteps,1))
#except:
# give output if no data on current station
#print("Skipping cluster " + str(c) + " event " + str(i) + " (missing data on " + stations[stat] + ")")
# make array for storage
back_azimuths = np.empty((0,1),"float64")
baz_event_index = np.empty((0,1),"float64")
# plot pca compoments that exceed norm threshold
# be wary- the transformed coordinate system's x-axis is meters north and the y-axis is meters east, so the pca_first_component[~,0] (which is cartesian x) is in [L] east
# and therefore along the transformed y-axis and the pca_first_component[~,1] (which is cartesian y) is in [L] north and therefore along the transformed x-axis
for s in range(len(all_first_components)):
# only plot and save results if length of resultant vector has a norm exceeding the threshold
if np.linalg.norm(all_first_components[s,:]) > norm_thresh:
# calculate back azimuths and save in array
baz = compute_baz(all_first_components[s,:])
back_azimuths = np.vstack((back_azimuths,baz))
baz_event_index = np.vstack(event_index[s])
# save actual backazimuth data
outFile = h5py.File(outPath + "win_len_" + str(winLen) + "/norm>" + str(norm_thresh) + "/top_" + str(percent) + "%/" + str(numCluster) + "/cluster_" + str(c) + "_backazimuths.h5","w")
outFile.create_dataset("backazimuths",data=back_azimuths)
outFile.create_dataset("index",data=baz_event_index)
outFile.close()