def plot_pattcorrs(pcdf, pcdf2=None, rmin=None, axis=None, tftype='seasonal'):
""" plot_pattcorrs(pcdf, axis):
this figure plots range of pattern correlations
as bars, with mean patt corr marker, plus % common
variation annotated.
pcdf: DataFrame of pattern correlations
pcdf2: second set of data to plot
rmin: minimum r value (pattern correlation) that is significant for the data
axis: plot axis
tftype: time frequency type: 'seasonal' or 'monthly'
"""
import math as math
if tftype=='seasonal':
numt=4
seasons=('SON','DJF','MAM','JJA')
else:
print 'ONLY SEASONAL IMPLEMENTED @@ 11/25/14'
numt=12 # monthly
seasons=con.get_mon()
# within ensemble pattern correlation combinations
# number of combinations: n choose k (here k=2)
# n! / (k! *(n-k)!)
ncomb = math.factorial(len(pcdf)) / (2 * math.factorial(len(pcdf)-2) )
# first, create matrix of data to take mean, max, min, etc
pcstack = np.zeros((ncomb,numt)) # ncomb combos x 4 seasons (or 12 mon)
stackii=0
keys=pcdf.keys()
for eii,skey1 in enumerate(keys):
for ineii,skey2 in enumerate(keys[eii+1:len(keys)]):
# stack the pc's so I can take max/min, etc
pcstack[stackii] = pcdf[skey1][skey2]
stackii+=1
mxsea = np.max(pcstack,axis=0)
mnsea = np.min(pcstack,axis=0)
avgsea = np.mean(pcstack,axis=0)
tmpstack=copy.copy(pcstack)
tmpstack[pcstack<0] = 0 # get rid of negative correlations. they are zero for all intents & purposes in this case
pcstacksq = np.power(tmpstack,2) # square the patt corr (the ones that are positive)
avgseasq = np.mean(pcstacksq,axis=0) # average squared patt corr (coef of determination)
if not pcdf2.empty:
ncomb2 = math.factorial(len(pcdf2)) / (2 * math.factorial(len(pcdf2)-2) )
pcstack2 = np.zeros((ncomb2,numt)) # ncomb combos x 4 seasons
stackii=0
keys=pcdf2.keys()
for eii,skey1 in enumerate(keys):
for ineii,skey2 in enumerate(keys[eii+1:len(keys)]):
# stack the pc's so I can take max/min, etc
pcstack2[stackii] = pcdf2[skey1][skey2]
stackii+=1
mxsea2 = np.max(pcstack2,axis=0)
mnsea2 = np.min(pcstack2,axis=0)
avgsea2 = np.mean(pcstack2,axis=0)
tmpstack2=copy.copy(pcstack2)
tmpstack2[pcstack2<0] = 0 # get rid of negative correlations. they are zero for all intents & purposes in this case
pcstacksq2 = np.power(tmpstack2,2) # square the patt corr (the ones that are positive)
avgseasq2 = np.mean(pcstacksq2,axis=0) # average squared patt corr (coef of determination)
# add annual mean
annwgts = np.array((91,90,92,92))/365.
print annwgts
#fig,axs = plt.subplots(1,1)
#fig.set_size_inches(6,4) # more squat
ax = axis #[0]
legtpl=()
wi=0.1 # width of bar
incr=0.3 # how much to shift in b/w the 2 sets of data
xxsea2=np.arange(1,6)
xboxmin=xxsea2-.2
if rmin!=None:
ax.axhspan(-1*rmin,rmin,color='orange',alpha=.3) # shade where corr is NOT significant
fillcol='0.7'
fillcol2=ccm.get_linecolor('dodgerblue')
for boxii in range(0,4): # loop through seasons
# shaded bars/boxes
ax.fill_between((xboxmin[boxii],xboxmin[boxii]+wi),mnsea[boxii],mxsea[boxii],color=fillcol, alpha=.5)
if not pcdf2.empty:
ax.fill_between((xboxmin[boxii]+incr,xboxmin[boxii]+incr+wi),mnsea2[boxii], mxsea2[boxii],color=fillcol2, alpha=.5)
# markers
ax.plot(xboxmin[boxii]+wi/2.,avgsea[boxii],color='k',marker='_',linestyle='none',markersize=15)#,alpha=.7)
if not pcdf2.empty:
ax.plot(xboxmin[boxii]+wi/2.+incr,avgsea2[boxii],color='b',marker='_',linestyle='none',markersize=15)#,alpha=.7)
# mean values (text)
val = '$%.0f$'%(avgseasq[boxii]*100) # @@ square the corrs before taking mean
ax.annotate(val+'%', xy=(xboxmin[boxii]+wi/2. -.09, .95), xycoords='data')
if not pcdf2.empty:
val = '$%.0f$'%(avgseasq2[boxii]*100) # @@ square the corrs before taking mean
ax.annotate(val+'%', xy=(xboxmin[boxii]+wi/2.+incr -.06, .95), xycoords='data')
boxii=boxii+1
# add annual mean -----------
annwgtst = np.tile(annwgts,(len(pcdf),1)) # tile
# each ens w/ each other
# pcstacksq is ncomb x numt
annwgtst = np.tile(annwgts,(ncomb,1)) # tile
ann = np.average(pcstack,weights=annwgts,axis=1) # ann mean per patt corr
annmax = np.max(ann)
annmin = np.min(ann)
avgann = np.mean(ann)
annsq = np.average(pcstacksq,weights=annwgts,axis=1) # ann mean per patt corr
avgannsq = np.mean(annsq)
if not pcdf2.empty:
# add annual mean
annwgtst = np.tile(annwgts,(len(pcdf2),1)) # tile
# each ens w/ each other
# pcstacksq is ncomb x numt
annwgtst = np.tile(annwgts,(ncomb2,1)) # tile
ann = np.average(pcstack2,weights=annwgts,axis=1) # ann mean per patt corr
annmax2 = np.max(ann)
annmin2 = np.min(ann)
avgann2 = np.mean(ann)
annsq2 = np.average(pcstacksq2,weights=annwgts,axis=1) # ann mean per patt corr
avgannsq2 = np.mean(annsq2)
# plot annual mean markers
ax.fill_between((xboxmin[boxii],xboxmin[boxii]+wi),annmin,annmax,color=fillcol,alpha=.5)
if not pcdf2.empty:
ax.fill_between((xboxmin[boxii]+incr,xboxmin[boxii]+incr+wi),annmin2,annmax2,color=fillcol2,alpha=.5)
ax.plot(xboxmin[boxii]+wi/2.,avgann,color='k',marker='_',linestyle='none',markersize=15)#,alpha=.9)
if not pcdf2.empty:
ax.plot(xboxmin[boxii]+wi/2.+incr,avgann2,color='b',marker='_',linestyle='none',markersize=15)#,alpha=.9)
val = '$%.0f$'%(avgannsq*100) # @@ square the corrs before taking mean
ax.annotate(val+'%', xy=(xboxmin[boxii]+wi/2. -.09, .95), xycoords='data')
if not pcdf2.empty:
val = '$%.0f$'%(avgannsq2*100) # @@ square the corrs before taking mean
ax.annotate(val+'%', xy=(xboxmin[boxii]+wi/2.+incr -.06, .95), xycoords='data')
ax.set_ylabel('Pattern Correlation')
ax.set_xlim((.5,5.5))
ax.set_xticks(xxsea2)
ax.set_xticklabels((seasons)+('ANN',))
ax.set_ylim((0,1))
ax.grid(True)
## darkolivegreen3 = np.array([162, 205, 90])/255.
## darkseagreen = np.array([143, 188, 143])/255.
## darkseagreen4 = np.array([105, 139, 105])/255.
## dodgerblue = np.array([30, 144, 255])/255.
## orangered4 = np.array([139, 37, 0])/255.
## lightsteelblue3 = np.array([162, 181, 205])/255. # more grey looking
## lightsteelblue4 = np.array([110, 123, 139])/255. # more grey looking
## steelblue3 = np.array([79, 148, 205])/255. # more blue looking
## steelblue4 = np.array([54, 100, 139])/255. # more blue looking
## colordict = {'kemctl1r1': darkseagreen, 'kemctl1r2': darkseagreen4, 'kemctl1r3': lightsteelblue3,
## 'kemctl1r4': lightsteelblue4, 'kemctl1r5': steelblue4, 'kemctl1ens': dodgerblue,
## 'kemctl1': orangered4, 'kemhadctl': darkolivegreen3 }
colordict = {'r1': ccm.get_linecolor('warm1'), #'firebrick'),
'r4': ccm.get_linecolor('warm2'), #'firebrick1'),
'r3': ccm.get_linecolor('warm3'), #'yelloworange'),#'chocolate1'),
'r5': ccm.get_linecolor('warm4'), #'darkyellow') #'skyblue'), #yelloworange'),
'r2': ccm.get_linecolor('warm5'), #'steelblue3'), #'darkgoldenrod1'),
'ens': ccm.get_linecolor('magenta'),
'': ccm.get_linecolor('mediumblue'), #'mediumpurple1'), #darkyellow'), # @@ empty key?
'kemhad': ccm.get_linecolor('deepskyblue')}
# Load data into dicts
# order ens simulations in order of most ice loss in melt season to least. Then ens mean, PERT2, observations if requested
## sims = bcasename+'r1', bcasename+'r4', bcasename+'r3', bcasename+'r5', bcasename+'r2', bcasename+'ens',bcasename
## if addobs:
## sims = sims + ('kemhadctl',)
fig,axs = plt.subplots(1,1)
fig.set_size_inches(6,4) # more squat
ax = axs #[0]
legtpl=()
wi=0.1 # width of bar
incr=0.3 # how much to shift in b/w the 2 sets of data
xxsea2=np.arange(1,6)
xboxmin=xxsea2-.2
ax.axhspan(-1*rmin,rmin,color='orange',alpha=.3) # shade where corr is NOT significant
fillcol='0.7'
fillcol2=ccm.get_linecolor('dodgerblue')
for boxii in range(0,4): # loop through seasons
# shaded bars/boxes
ax.fill_between((xboxmin[boxii],xboxmin[boxii]+wi),ensminsea[boxii],ensmaxsea[boxii],color=fillcol)
ax.fill_between((xboxmin[boxii]+incr,xboxmin[boxii]+incr+wi),mnsea[boxii], mxsea[boxii],color=fillcol2, alpha=.5)
# markers
ax.plot(xboxmin[boxii]+wi/2.,ensmeansea[boxii],color='k',marker='_',linestyle='none',markersize=15)#,alpha=.9)
ax.plot(xboxmin[boxii]+wi/2.+incr,avgsea[boxii],color='b',marker='_',linestyle='none',markersize=15)#,alpha=.7)
# mean values (text)
#val = '$%.0f$'%(ensmeansea2[boxii]*ensmeansea2[boxii]*100)
val = '$%.0f$'%(ensmeanseasq[boxii]*100)# @@ square the corrs before taking mean
#print val
#print ensmeansea2[boxii]*ensmeansea2[boxii]*100
ax.annotate(val+'%', xy=(xboxmin[boxii]-.07, .95), xycoords='data')
# dblob should have regional means *with* time dimension
# want to test sig different mean and variance b/w ctl and pert
allstats=sh.calc_runstats(dblob,sims, seas=seasons,siglevel=siglevel)
fpvaldf=pd.DataFrame(allstats['fpval']) # pval of f statistic for variance significance
fpvaldft=fpvaldf.transpose()
tpvaldf=pd.DataFrame(allstats['tpval']) # pval of t statistic for mean significance
tpvaldft=tpvaldf.transpose()
import cccmacmaps as ccm
if plottype=='calcregunccascade':
effdof=False
col=('0.3',ccm.get_linecolor('firebrick'))
if field=='st' and region in ('eurasia','eurasiamori'):
#xlab = '$\Delta$ Eurasia SAT ($^\circ$C)'
xlab = '($^\circ$C)'
elif field=='st' and region=='polcap60':
#xlab = '$\Delta$ >60$^\circ$N SAT ($^\circ$C)'
xlab = '($^\circ$C)'
else:
xlab=None
for sea in seasons:
seas=(sea,)
xlims=None
fig,axs = plt.subplots(2,1)
fig.set_size_inches(4,8)
fig.set_frameon(False)
casename + '_r' + str(eii) + 'i1p1_185001-201212.nc'
orignatdt[eii] = cnc.getNCvar(fname,field,timesel=origsel,seas=season)
orignatdf=pd.DataFrame(orignatdt,index=origyrs)
pastnatorig = orignatdf.loc[subyrs]
presnatorig = orignatdf.loc[subyrs2]
(natorigtstat,natorigpv) = cutl.ttest_ind(presnatorig,pastnatorig,axis=0)
if dopct:
diffnatorig = (presnatorig.mean(axis=0) - pastnatorig.mean(axis=0)) / pastnatorig.mean(axis=0)*100
else:
diffnatorig = presnatorig.mean(axis=0) - pastnatorig.mean(axis=0)
# TIMESERIES
firebrick=ccm.get_linecolor('firebrick')
hcol=ccm.get_linecolor('darkolivegreen3')
hcolline=ccm.get_linecolor('darkolivegreen3')#'darkseagreen4')
natcol=ccm.get_linecolor('steelblue3')
natcolline=ccm.get_linecolor('steelblue3')#4')
fig,axs=plt.subplots(1,1)
axs.plot(years,allnatdf,color=natcol,alpha=.5)
axs.plot(years,allflddf,color='r',alpha=.3)
axs.plot(origyrs,orignatdf,color=natcol,linewidth=2)
axs.plot(origyrs,origdf,color='brown',linewidth=2)
axs.plot(hadyrs,haddf,color='b',linewidth=2)
axs.plot(nsidcyrs,nsidcdf,color='g',linewidth=2)
axs.set_title(field + ' ' + str(season))
fieldr2='tas'; ncfieldr2='tas'; compr2='Amon';
regionr2='eurasiamori'; rstr2='Eur SAT'; rstr2long='Eurasian SAT'; runits2='$^\circ$C'; rkey2='eursat'
convr2=1
aconvr2=1 # for agcm sims
sear2='DJF'
sear='DJF'
timeselc='1979-01-01,1989-12-31'
timeselp='2002-01-01,2012-12-31'
lecol='red'
ocol='green'
picol = '0.8'
acol=ccm.get_linecolor('paperblue')
when='14:51:28.762886'; styearsR = [ 8., 7., 2., 8., 8.] # variable SIC styears
styearsE=[ 4., 1., 7., 3., 1.]; styearsN=[1.] #when for these: 17:01:16.908687
def load_pifield(fdict,seas,conv=1,subsampyrs=11,numsamp=50,
styear=None,anomyears=None,local=False,detrend=True,
verb=False,addcyc=False):
""" TAKEN FROM load_canesmfield() in canesm_LE_composite.py
loads subsampled CGCM data from specified simulation
(assumes a long control, e.g. piControl)
number of total chunks will be determined by length of sim data
and numyrs (ntime / numyrs). First the simulation is chunked into
#rdf=df.ix[:,'R1':'R5']
#edf=df.ix[:,'E1':'E5']
#regressdf = rdf.append(edf) # didn't append like I expected. inserted NaNs in places
# this will work:
allens=['R1','R2','R3','R4','R5','E1','E2','E3','E4','E5'] # has to be an array!
rens=['R1','R2','R3','R4','R5']
eens=['E1','E2','E3','E4','E5']
sub = df.loc[:,allens]
subx=sub.ix[0,:]
suby=sub.ix[1,:]
mm, bb, rval, pval, std_err = sp.stats.linregress(subx,suby)
#sube = df.loc[0,['E1','E2','E3','E4','E5']]
firebrick=ccm.get_linecolor('firebrick')
printtofile=True
fig,ax = plt.subplots(1)
fig.set_size_inches(6,5)
rr = plt.scatter(df.filter(regex='R').values[0],df.filter(regex='R').values[1],
color='0.3',marker='o',s=8**2,alpha=0.7)
ee = plt.scatter(sub.ix[0,5:],sub.ix[1,5:],
color=firebrick,marker='o',s=8**2,alpha=0.7)
#plt.scatter(df['HAD'].values[0],df['HAD'].values[1],color=cd['HAD'],marker='s',s=8**2)
#ns=plt.scatter(df['NSIDC'].values[0],df['NSIDC'].values[1],color=cd['NSIDC'],marker='o',s=8**2,alpha=0.7) # @@@@ add back?
ns=plt.scatter(df['NSIDC'].values[0],df['NSIDC'].values[1],color='green',marker='o',s=8**2,alpha=0.7)
#plt.scatter(df['ENS'].values[0],df['ENS'].values[1],color=cd['ENS'],marker='s',s=10**2)
#plt.scatter(df['ENSE'].values[0],df['ENSE'].values[1],color=cd['ENSE'],marker='o',s=10**2)
axylims = ax.get_ylim()
rmse = np.sqrt(np.square(flddiffdt[skey]))
rmseclim = np.sqrt(np.square(flddiffclimdt[skey]))
rmsedt[skey] = cutl.calc_regmean(rmse, lat, lon, region)
rmseclimdt[skey] = cutl.calc_regmean(rmseclim, lat, lon, region)
annrmseclimdt[skey] = cutl.annualize_monthlyts(rmseclimdt[skey])
climrmsedt[skey], rmsestd = cutl.climatologize(rmsedt[skey])
# <codecell>
from matplotlib import gridspec
colors = ("b", "g", "m", "y", "c", "k", "r", ccm.get_linecolor("mediumpurple4"))
cdict = {
"iga": "0.5",
"gregory_2xco2": "0.3",
"kel11": "b",
"kel09": "g",
"kel14": "m",
"kel15": "y",
"kel17": "c",
"kel18": "k",
"kel20": "r",
"kel24": ccm.get_linecolor("limegreen"),
"kel10": "b",
"kel12": "g",
"kel16": "m",
"kel19": "y",
montrenddf = pd.DataFrame(montrenddt)
mndf = pd.DataFrame(monmndt)
mxdf = pd.DataFrame(monmxdt)
avgdf = pd.DataFrame(monavgdt)
superensdf=pd.DataFrame(data=superslopes,index=mons,columns=superkeys[0])
canesm=superensdf.CanESM2
# plot seasonal trends for all ensemble members
hmontrenddf = pd.DataFrame(hmontrenddt,index=(1,))
nmontrenddf = pd.DataFrame(nmontrenddt,index=(1,))
superensdf.plot(color='k',alpha=0.5,legend=False)
plt.plot(canesm,color='r',linewidth=3)
plt.plot(hmontrenddf.values[0],'green',linewidth=3) # not sure why but this has an array w/in an array....
plt.plot(nmontrenddf.values[0],color=ccm.get_linecolor('dodgerblue'),linewidth=3)# not sure why but this has an array w/in an array....
plt.title('SIE 1979-2012 trends (/yr)')
plt.xlabel('Month')
plt.xlim((0,11))
if printtofile:
plt.savefig('SIE_CMIP5_allmemstrend_seacycle_1979-2012_wCanESM2.pdf')
#superensdf.hist()
# min / max histograms each month
fig,axs = plt.subplots(3,4)
fig.set_size_inches(12,9)
for aii,ax in enumerate(axs.flat):
mon = mons[aii]
ax.hist(mndf[mon],color='.5',alpha=0.5)
