def plot_scatter(est, ref):
import matplotlib.pyplot as plt
from scipy import stats
maske = ~np.isnan(est) & ~np.isnan(ref)
slope, intercept, r_value, p_value, std_err = stats.linregress(est[maske], ref[maske])
line = slope * est +intercept
from pcc import skill_score
SS = skill_score(est,ref,th=0)
plt.scatter(est, ref, label='Reflectivity [dBZ]', color='grey', alpha=0.6)
text = ('f(x) = ' + str(round(slope,3)) + 'x + ' + str(round(intercept,3)) +
'\nCorr: ' + str(round(r_value,3)) + r'$\pm$: '+ str(round(std_err,3))+
'\nN: '+ str(int(SS['N']))+
'\nHit: ' + str(round(SS['H']/SS['N'],3)*100)+'%'+
'\nMiss: ' + str(round(SS['M']/SS['N'],3)*100)+'%'+
'\nFalse: ' + str(round(SS['F']/SS['N'],3)*100)+'%'+
'\nCnegative: ' + str(round(SS['C']/SS['N'],3)*100)+'%'+
'\nHR: ' + str(round(SS['HR'],3))+
'\nPOD: ' + str(round(SS['POD'],3))+
'\nFAR: ' + str(round(SS['FAR'],3))+
'\nBID: ' + str(round(SS['BID'],3))+
'\nHSS: ' + str(round(SS['HSS'],3))+
'\nBias: '+ str(round(SS['bias'],3))+
'\nRMSE: '+ str(round(SS['RMSE'],3))
)
plt.annotate(text, xy=(0.01, 0.99), xycoords='axes fraction', fontsize=10,
horizontalalignment='left', verticalalignment='top')
t1 = np.linspace(0,70,70)
plt.plot(t1,t1,'k-')
plt.plot(t1, t1*slope + intercept, 'r-', lw=3 ,label='Regression')
plt.plot(t1, t1*slope + (intercept+5), 'r-.', lw=1.5 ,label=r'Reg $\pm$ 5 mdBZ')
plt.plot(t1, t1*slope + (intercept-5), 'r-.', lw=1.5 )
plt.plot(np.nanmean(est),np.nanmean(ref), 'ob', lw = 4,label='Mean')
plt.legend(loc='lower right', fontsize=10, scatterpoints= 1, numpoints=1, shadow=True)
plt.xlim(0,70)
plt.ylim(0,70)
A = np.copy(rrr)
B = np.ma.masked_invalid(gprof_pp_b)
#A[A<TH_rain] = 18
#B[B<TH_rain] = 18
ref = np.copy(rrr)
est = np.ma.masked_invalid(gprof_pp_b)
mask = ~np.isnan(B) & ~np.isnan(A)
slope, intercept, r_value, p_value, std_err = stats.linregress(
B[mask], A[mask])
line = slope * B + intercept
from pcc import skill_score
RR = skill_score(est, ref, 0.001)
from pcc import plot_score
plot_score(est, ref, RR)
#plt.scatter(B[mask],A[mask], label='RR [mm/h]')
plt.plot(B, line, 'r-')
#maxAB = np.nanmax([np.nanmax(A[mask]),np.nanmax(B[mask])])
#plt.xlim(0,maxAB + 1)
#plt.ylim(0,maxAB + 1)
plt.legend(loc='upper center',
bbox_to_anchor=(0.5, 1.1),
ncol=2,
fancybox=True,
shadow=True,
fontsize='small',
title="Slope: " + str(round(slope, 3)) + ', Intercept: ' +
top='off',
labelbottom='off',
right='off',
left='off',
labelleft='off')
plt.xlim(-420,390)
plt.ylim(-4700, -3700)
ax4 = fig.add_subplot(224, aspect='equal')#------------------------------------
maske = ~np.isnan(ggg) & ~np.isnan(rrr)
slope, intercept, r_value, p_value, std_err = stats.linregress(ggg[maske], rrr[maske])
line = slope * ggg +intercept
from pcc import skill_score
SS = skill_score(ggg,rrr,th=0)
plt.scatter(ggg, rrr, label='RR [mm/h]', color='grey', alpha=0.6)
text = ('f(x) = ' + str(round(slope,3)) + 'x + ' + str(round(intercept,3)) +
'\nCorr: ' + str(round(r_value,3)) + r'$\pm$: '+ str(round(std_err,3))+
'\nN: '+ str(int(SS['N']))+
'\nHit: ' + str(SS['H'])+
'\nMiss: ' + str(SS['M'])+
'\nFalse: ' + str(SS['F'])+
'\nCnegative: ' + str(SS['C'])+
'\nHR: ' + str(round(SS['HR'],3))+
'\nPOD: ' + str(round(SS['POD'],3))+
'\nFAR: ' + str(round(SS['FAR'],3))+
'\nBID: ' + str(round(SS['BID'],3))+
'\nHSS: ' + str(round(SS['HSS'],3))+
labelbottom='off',
right='off',
left='off',
labelleft='off')
plt.xlim(-420, 390)
plt.ylim(-4700, -3700)
ax4 = fig.add_subplot(
224, aspect='equal') #------------------------------------
try:
maske = ~np.isnan(ggg) & ~np.isnan(rrr)
slope, intercept, r_value, p_value, std_err = stats.linregress(
ggg[maske], rrr[maske])
SS = skill_score(ggg, rrr, th=TH_ref)
ax4.scatter(ggg,
rrr,
label='Reflectivity [dBZ]',
color='grey',
alpha=0.6)
r_value_s, p_value_s = stats.spearmanr(ggg[maske], rrr[maske])
text = ('f(x) = ' + str(round(slope, 3)) + 'x + ' +
str(round(intercept, 3)) + '\nCorr: ' +
str(round(r_value, 3)) + r'$\pm$ ' +
str(round(std_err, 3)) + '\nN: ' + str(int(SS['N'])) +
'\nHit: ' + str(SS['H']) + '\nMiss: ' + str(SS['M']) +
'\nFalse: ' + str(SS['F']) + '\nCnegative: ' +
labelbottom='off',
right='off',
left='off',
labelleft='off')
plt.xlim(-420,390)
plt.ylim(-4700, -3700)
ax4 = fig.add_subplot(224, aspect='equal')#------------------------------------
maske = ~np.isnan(ggg) & ~np.isnan(rrr)
slope, intercept, r_value, p_value, std_err = stats.linregress(ggg[maske], rrr[maske])
line = slope * ggg +intercept
from pcc import skill_score
SS = skill_score(ggg,rrr,th=TH_ref)
ax4.scatter(ggg, rrr, label='Reflectivity [dBZ]', color='grey', alpha=0.6)
r_value_s, p_value_s = stats.spearmanr(ggg[maske],rrr[maske])
text = ('f(x) = ' + str(round(slope,3)) + 'x + ' + str(round(intercept,3)) +
'\nCorr: ' + str(round(r_value,3)) + r'$\pm$: '+ str(round(std_err,3))+
'\nN: '+ str(int(SS['N']))+
'\nHit: ' + str(SS['H'])+
'\nMiss: ' + str(SS['M'])+
'\nFalse: ' + str(SS['F'])+
'\nCnegative: ' + str(SS['C'])+
'\nHR: ' + str(round(SS['HR'],3))+
'\nPOD: ' + str(round(SS['POD'],3))+
'\nFAR: ' + str(round(SS['FAR'],3))+
A = np.copy(rrr)
B = np.ma.masked_invalid(gprof_pp_b)
#A[A<TH_rain] = 18
#B[B<TH_rain] = 18
ref = np.copy(rrr)
est = np.ma.masked_invalid(gprof_pp_b)
mask = ~np.isnan(B) & ~np.isnan(A)
slope, intercept, r_value, p_value, std_err = stats.linregress(B[mask], A[mask])
line = slope*B+intercept
from pcc import skill_score
RR = skill_score(est,ref,0.001)
from pcc import plot_score
plot_score(est,ref,RR)
#plt.scatter(B[mask],A[mask], label='RR [mm/h]')
plt.plot(B,line,'r-')
#maxAB = np.nanmax([np.nanmax(A[mask]),np.nanmax(B[mask])])
#plt.xlim(0,maxAB + 1)
#plt.ylim(0,maxAB + 1)
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.1), ncol=2, fancybox=True, shadow=True,
fontsize='small', title= "Slope: " + str(round(slope,3))
+ ', Intercept: '+ str(round(intercept,3)) + "\n Correlation: " +
str(round(r_value,3)) + ', Std_err: '+ str(round(std_err,3)))
plt.xlabel("DPR Ref [dbz]")
plt.ylabel("RADOLAN Ref [dbz]")
plt.title(" .")
right='off',
left='off',
labelleft='off')
plt.xlim(-420, 390)
plt.ylim(-4700, -3700)
ax4 = fig.add_subplot(
224, aspect='equal') #------------------------------------
maske = ~np.isnan(ggg) & ~np.isnan(rrr)
slope, intercept, r_value, p_value, std_err = stats.linregress(
ggg[maske], rrr[maske])
line = slope * ggg + intercept
from pcc import skill_score
SS = skill_score(ggg, rrr, th=0)
plt.scatter(ggg, rrr, label='RR [mm/h]', color='grey', alpha=0.6)
text = ('f(x) = ' + str(round(slope, 3)) + 'x + ' +
str(round(intercept, 3)) + '\nCorr: ' + str(round(r_value, 3)) +
r'$\pm$: ' + str(round(std_err, 3)) + '\nN: ' + str(int(SS['N'])) +
'\nHit: ' + str(SS['H']) + '\nMiss: ' + str(SS['M']) +
'\nFalse: ' + str(SS['F']) + '\nCnegative: ' + str(SS['C']) +
'\nHR: ' + str(round(SS['HR'], 3)) + '\nPOD: ' +
str(round(SS['POD'], 3)) + '\nFAR: ' + str(round(SS['FAR'], 3)) +
'\nBID: ' + str(round(SS['BID'], 3)) + '\nHSS: ' +
str(round(SS['HSS'], 3)) + '\nBias: ' + str(round(SS['bias'], 3)) +
'\nRMSE: ' + str(round(SS['RMSE'], 3)))
ax4.annotate(text,
def validation_plot_log(data1, data2, th_ss):
"""
Function:
Plot for the validation of two datasets
Input:
data1, data2 ::: Input Data
Output:
Validation PLot
"""
# Todo: Schoener machen!
#mini = np.nanmin([np.nanmin(data1), np.nanmin(data2)]) - 5.0
mini = 0.0
maxi = np.nanmax([np.nanmax(data1), np.nanmax(data2)]) + 5.0
print 'LIMITS: ', mini, maxi
cd1 = 'blue'
cd2 = 'green'
from scipy import stats, linspace
fig = plt.figure(figsize=(14,14))
ax1 = fig.add_subplot(223, aspect='auto')#---------------------------------------------------------------
maske = ~np.isnan(data1) & ~np.isnan(data2)
slope, intercept, r_value, p_value, std_err = stats.linregress(data1[maske], data2[maske])
line = slope * data1 +intercept
slope2, intercept2, r_value2, p_value2, std_err2 = stats.linregress(data2[maske], data1[maske])
line2 = slope2 * data2 +intercept2
from pcc import skill_score
SS = skill_score(data1,data2,th=th_ss)
ax1.scatter(data1, data2, label='RR in mm/h', color='grey', alpha=0.6)
#plt.boxplot(data1[maske],vert=0)
#plt.boxplot(data2[maske],vert=1)
plt.yscale('log')
plt.xscale('log')
r_value_s, p_value_s = stats.spearmanr(data1[maske],data2[maske])
text = ('f(x) = ' + str(round(slope,3)) + 'x + ' + str(round(intercept,3)) +
'\nCorr: ' + str(round(r_value,3)) + r'$\pm$: '+ str(round(std_err,3))+
'\nN: '+ str(int(SS['N']))+
'\nHit: ' + str(SS['H'])+
'\nMiss: ' + str(SS['M'])+
'\nFalse: ' + str(SS['F'])+
'\nCnegative: ' + str(SS['C'])+
'\nHR: ' + str(round(SS['HR'],3))+
'\nPOD: ' + str(round(SS['POD'],3))+
'\nFAR: ' + str(round(SS['FAR'],3))+
'\nBID: ' + str(round(SS['BID'],3))+
'\nHSS: ' + str(round(SS['HSS'],3))+
'\nBias: '+ str(round(SS['bias'],3))+
'\nRMSE: '+ str(round(SS['RMSE'],3))+
'\nCorrS:' + str(round(r_value_s,3))
)
ax1.annotate(text, xy=(0.01, 0.99), xycoords='axes fraction', fontsize=10,
horizontalalignment='left', verticalalignment='top')
#t1 = linspace(0,50,5000)
t1=10 ** np.linspace(np.log10(10**-2), np.log10(10**2), 5000)
plt.plot(t1,t1,'k-')
#plt.plot(np.log10(t1),np.log10(t1),'k-')
#plt.plot(t1,t1*slope+intercept,label='RADOLAN', color='green', lw=1.5, alpha=0.5)
#plt.plot(t1*slope2+intercept2,t1,label='GPM', color='blue', lw=1.5, alpha=0.5)
plt.legend(loc='lower right', fontsize=10, scatterpoints= 1, numpoints=1, shadow=True)
#plt.xlim(mini,maxi)
#plt.ylim(mini,maxi)
plt.xlim(10**-2,10**2)
plt.ylim(10**-2,10**2)
plt.xlabel('GPM DPR')
plt.ylabel('RADOLAN')
plt.grid()
ax2 = fig.add_subplot(221, aspect='auto')#-----------------------------------------------------------------------------
st = 50
counts1, bins1, patches1 = plt.hist(data1[maske], bins=10 ** np.linspace(np.log10(10**-2), np.log10(10**2), st), alpha=0.5,
color=cd2, label='GPM DPR')
plt.yscale('log')
plt.xscale('log')
counts2, bins2, patches2 =plt.hist(data2[maske], bins=10 ** np.linspace(np.log10(10**-2), np.log10(10**2), st),
alpha=0.9, edgecolor='black',
facecolor="None", label='RADOLAN')
plt.yscale('log')
plt.xscale('log')
#plt.xlim(mini, maxi)
plt.ylabel('g_frequency in #')
plt.grid(color=cd2)
plt.legend(loc='upper right')
plt.xlim(10**-2,10**2)
ax3 = fig.add_subplot(224, aspect='auto')#-------------------------------------------------------------------------------
counts2, bins2, patches2 =plt.hist(data2[maske], bins=10 ** np.linspace(np.log10(10**-2), np.log10(10**2), st),orientation='horizontal',
alpha=0.5, color=cd1, label='RADOLAN')
plt.yscale('log')
plt.xscale('log')
counts1, bins1, patches1 = plt.hist(data1[maske], bins=10 ** np.linspace(np.log10(10**-2), np.log10(10**2), st), alpha=0.9, edgecolor='black',
facecolor="None",orientation='horizontal', label='GPM')
plt.yscale('log')
plt.xscale('log')
plt.xlabel('r_frequency in #')
#plt.ylim(mini,maxi)
plt.grid(color=cd1)
plt.legend(loc='upper right')
plt.ylim(10**-2,10**2)
ax4 = fig.add_subplot(222, aspect='auto')#---------------------------------------------------------------------------------
bin_centers1 = np.mean(zip(bins1[:-1], bins1[1:]), axis=1)
bin_centers2 = np.mean(zip(bins2[:-1], bins2[1:]), axis=1)
ax4.plot(bin_centers1,counts1.cumsum(),color=cd2 ,ls='-', lw=2,alpha=0.5, label='GPM')
ax4.plot(bin_centers2,counts2.cumsum(),color=cd1, ls='-', lw=2, alpha=0.5, label='RADOLAN')
#plt.yscale('log')
plt.xscale('log')
maske = ~np.isnan(counts1) & ~np.isnan(counts2)
slope2, intercept2, r_value2, p_value2, std_err2 = stats.linregress(counts1[maske], counts2[maske])
r_value_s, p_value_s = stats.spearmanr(counts1[maske], counts2[maske])
plt.ylabel('frequency in #')
plt.xlabel('RR in mm/h')
tit = 'Corr: '+ str(round(r_value2,3)) + r'$\pm$' + str(round(std_err2,3)) + '\n SCorr: '\
+ str(round(r_value_s,3)) + r'$\pm$' + str(round(p_value_s,3))
plt.legend(loc='lower right')#, title=tit)
#plt.xlim(10**-2,10**3)
plt.grid()
def validation_plot(data1, data2, th_ss):
"""
Function:
Plot for the validation of two datasets
Input:
data1, data2 ::: Input Data
Output:
Validation PLot
"""
# Todo: Schoener machen!
mini = np.nanmin([np.nanmin(data1), np.nanmin(data2)]) - 5.0
maxi = np.nanmax([np.nanmax(data1), np.nanmax(data2)]) + 5.0
print 'Max data1: ', np.nanmax(data1)
print 'Max data2: ', np.nanmax(data2)
cd1 = 'blue'
cd2 = 'green'
print '________________________ Maxi:', maxi
from scipy import stats, linspace
fig = plt.figure(figsize=(14,14))
ax1 = fig.add_subplot(223, aspect='auto')#------------------------------------
maske = ~np.isnan(data1) & ~np.isnan(data2)
slope, intercept, r_value, p_value, std_err = stats.linregress(data1[maske], data2[maske])
line = slope * data1 +intercept
slope2, intercept2, r_value2, p_value2, std_err2 = stats.linregress(data2[maske], data1[maske])
line2 = slope2 * data2 +intercept2
from pcc import skill_score
SS = skill_score(data1,data2,th=th_ss)
ax1.scatter(data1, data2, label='Reflectivity [dBZ]', color='grey', alpha=0.6)
r_value_s, p_value_s = stats.spearmanr(data1[maske],data2[maske])
text = ('f(x) = ' + str(round(slope,3)) + 'x + ' + str(round(intercept,3)) +
'\nCorr: ' + str(round(r_value,3)) + r'$\pm$: '+ str(round(std_err,3))+
'\nN: '+ str(int(SS['N']))+
'\nHit: ' + str(SS['H'])+
'\nMiss: ' + str(SS['M'])+
'\nFalse: ' + str(SS['F'])+
'\nCnegative: ' + str(SS['C'])+
'\nHR: ' + str(round(SS['HR'],3))+
'\nPOD: ' + str(round(SS['POD'],3))+
'\nFAR: ' + str(round(SS['FAR'],3))+
'\nBID: ' + str(round(SS['BID'],3))+
'\nHSS: ' + str(round(SS['HSS'],3))+
'\nBias: '+ str(round(SS['bias'],3))+
'\nRMSE: '+ str(round(SS['RMSE'],3))+
'\nCorrS:' + str(round(r_value_s,3))
)
ax1.annotate(text, xy=(0.01, 0.99), xycoords='axes fraction', fontsize=10,
horizontalalignment='left', verticalalignment='top')
t1 = linspace(mini,maxi,50)
plt.plot(t1,t1,'k-')
#plt.plot(t1, t1*slope + intercept, 'r-', lw=3 ,label='Regression')
#plt.plot(t1, t1*slope + (intercept+5), 'r-.', lw=1.5 ,label=r'Reg $\pm$ 5 mdBZ')
#plt.plot(t1, t1*slope + (intercept-5), 'r-.', lw=1.5 )
plt.plot(t1,t1*slope + intercept,label='RADOLAN', color='green', lw=1.5, alpha=0.5)
plt.plot(t1*slope2 + intercept2,t1,label='GPM', color='blue', lw=1.5, alpha=0.5)
#plt.plot(np.nanmean(data1),np.nanmean(data2), 'ob', lw = 4,label='Mean')
#plt.plot(np.nanmedian(data1),np.nanmedian(data2), 'vb', lw = 4,label='Median')
#import matplotlib as mpl
#mean = [ np.nanmean(data1),np.nanmean(data2)]
#width = np.nanstd(data1)
#height = np.nanstd(data2)
#angle = 0
#ell = mpl.patches.Ellipse(xy=mean, width=width, height=height,
# angle=180+angle, color='blue', alpha=0.8,
# fill=False, ls='--', label='Std')
#ax1.add_patch(ell)
plt.legend(loc='lower right', fontsize=10, scatterpoints= 1, numpoints=1, shadow=True)
#plt.scatter(data1, data2, alpha=0.5)
plt.xlim(mini,maxi)
plt.ylim(mini,maxi)
plt.xlabel('GPM DPR')
plt.ylabel('RADOLAN')
plt.grid()
#plt.colorbar(shrink=1)
ax2 = fig.add_subplot(221, aspect='auto')#------------------------------------
counts1, bins1, patches1 = plt.hist(data1[maske], bins=int(maxi), alpha=0.5,
color=cd2, label='GPM DPR')
counts2, bins2, patches2 =plt.hist(data2[maske], bins=int(maxi),
alpha=0.9, edgecolor='black',
facecolor="None", label='RADOLAN')
plt.xlim(mini, maxi)
plt.ylabel('g_frequency in #')
plt.grid(color=cd2)
plt.legend(loc='upper right')
ax3 = fig.add_subplot(224, aspect='auto')#------------------------------------
counts2, bins2, patches2 =plt.hist(data2[maske], bins=int(maxi),orientation='horizontal',
alpha=0.5, color=cd1, label='RADOLAN')
counts1, bins1, patches1 = plt.hist(data1[maske], bins=int(maxi), alpha=0.9, edgecolor='black',
facecolor="None",orientation='horizontal', label='GPM')
plt.xlabel('r_frequency in #')
plt.ylim(mini,maxi)
plt.grid(color=cd1)
plt.legend(loc='upper right')
ax4 = fig.add_subplot(222, aspect='auto')#------------------------------------
bin_centers1 = np.mean(zip(bins1[:-1], bins1[1:]), axis=1)
bin_centers2 = np.mean(zip(bins2[:-1], bins2[1:]), axis=1)
ax4.plot(counts1.cumsum(),color=cd2 ,ls='-', lw=2,alpha=0.5, label='GPM')
ax4.plot(counts2.cumsum(),color=cd1, ls='-', lw=2, alpha=0.5, label='RADOLAN')
maske = ~np.isnan(counts1) & ~np.isnan(counts2)
slope2, intercept2, r_value2, p_value2, std_err2 = stats.linregress(counts1[maske], counts2[maske])
r_value_s, p_value_s = stats.spearmanr(counts1[maske], counts2[maske])
plt.ylabel('frequency in #')
plt.xlabel('Reflectivity in dBz')
tit = 'Corr: '+ str(round(r_value2,3)) + r'$\pm$' + str(round(std_err2,3)) + '\n SCorr: '\
+ str(round(r_value_s,3)) + r'$\pm$' + str(round(p_value_s,3))
plt.legend(loc='lower right')#, title=tit)
plt.grid()
plt.ylim(0,maxAB + 1)
plt.legend(loc='upper center', bbox_to_anchor=(0.5, 1.1), ncol=2, fancybox=True, shadow=True,
fontsize='small', title= "Slope: " + str(round(slope,3))
+ ', Intercept: '+ str(round(intercept,3)) + "\n Correlation: " +
str(round(r_value,3)) + ', Std_err: '+ str(round(std_err,3)))
plt.xlabel(GPMI_name[ip])
plt.ylabel("RADOLAN RR [mm/h]")
plt.title(" .")
plt.grid(True)
plt.tight_layout()
plt.show()
###############################################################################
import pcc
R = pcc.skill_score(est,ref,0.1)
pcc.plot_score(est,ref,R)
plt.xlabel(GPMI_name[ip])
plt.ylabel("RADOLAN RR [mm/h]")
#plt.yscale('log')
#plt.xscale('log')
plt.title(" .")
plt.tight_layout()
plt.show()
'''
plt.hist(A[mask],bins=200, color='red', alpha=0.4, label='RADOLAN interpoliert')
plt.hist(B[mask], bins=200, color='blue', alpha=0.4, label='GPROF')
shadow=True,
fontsize='small',
title="Slope: " + str(round(slope, 3)) + ', Intercept: ' +
str(round(intercept, 3)) + "\n Correlation: " +
str(round(r_value, 3)) + ', Std_err: ' + str(round(std_err, 3)))
plt.xlabel(GPMI_name[ip])
plt.ylabel("RADOLAN RR [mm/h]")
plt.title(" .")
plt.grid(True)
plt.tight_layout()
plt.show()
###############################################################################
import pcc
R = pcc.skill_score(est, ref, 0.1)
pcc.plot_score(est, ref, R)
plt.xlabel(GPMI_name[ip])
plt.ylabel("RADOLAN RR [mm/h]")
#plt.yscale('log')
#plt.xscale('log')
plt.title(" .")
plt.tight_layout()
plt.show()
'''
plt.hist(A[mask],bins=200, color='red', alpha=0.4, label='RADOLAN interpoliert')
plt.hist(B[mask], bins=200, color='blue', alpha=0.4, label='GPROF')
#pdf = stats.norm.pdf(sorted(B[mask]), B[mask], B[mask])
#plt.plot(sorted(B[mask]), pdf)
plt.xlabel("RR [mm/h]")
