def metrics ( station, datespan, forecasts, doPlot=False ):
"""
Computes a full set of time series comparison metrics
over a specified dateSpan
for a given CO-OPS station 7-digit IDs (str)
and the list of model forecasts.
Returns mx and fx - lists of metrics and corresponding lead times (hrs)
"""
mx = [] # metrics
fx = [] # lead time, in hrs
if doPlot:
import matplotlib
matplotlib.use('Agg',warn=False)
import matplotlib.pyplot as plt
data = coops.getData(station, datespan)
if len(data['values']):
if doPlot:
plt.figure(figsize=(20,4.5))
for model in forecasts:
#model = getPointsWaterlevel (ncFile)
fcst_time = model['time'][60]
lead_time = 1./3600*(fcst_time-datespan[0]).total_seconds()
fx.append(lead_time)
counter = 0
nst = np.nan
for s in model['stations']:
if station in s:
nst = counter
break
counter += 1
# Project
pdates, pdata, pmodel = \
interp.projectTimeSeries (data['dates'], data['values'],
model['time'], model['zeta'][:,nst],
refStepMinutes=6)
#Run metrics
metrics = valstat.metrics (pdata, pmodel, pdates)
mx.append(metrics)
#Plot data and model forecasts (optional)
if doPlot:
plt.plot(pdates, pdata, c='green')
plt.plot(pdates, pmodel, c=np.random.rand(3,))
plt.title('station ' + station)
plt.savefig('ts-metrics-' + station + '.png')
return mx, np.array(fx)
def plot_estofs_timeseries(obs_dates,
obs_values,
mod_dates,
mod_values,
figFile=None,
stationName='',
htp_dates=None,
htp_vals=None,
daterange=None,
ylim=[-2.0, 2.0]):
"""
Project obs and model onto the same timeline;
Compute RMSD
"""
#Sort by date
obs_dates = np.array(obs_dates)
obs_values = np.array(obs_values)
ind = np.argsort(obs_dates)
obs_dates = obs_dates[ind]
obs_values = obs_values[ind]
# Remove nans
ind = np.logical_not(np.isnan(obs_values))
obs_dates = obs_dates[ind]
obs_values = obs_values[ind]
#Sort by date
mod_dates = np.array(mod_dates)
mod_values = np.array(mod_values)
ind = np.argsort(mod_dates)
mod_dates = mod_dates[ind]
mod_values = mod_values[ind]
#Rid of mask
if hasattr(mod_values, 'mask'):
ind = ~mod_values.mask
mod_dates = mod_dates[ind]
mod_values = mod_values[ind]
# Remove nans
ind = np.logical_not(np.isnan(mod_values))
mod_dates = mod_dates[ind]
mod_values = mod_values[ind]
refStepMinutes = 6
refDates, obsValProj, modValProj = \
interp.projectTimeSeries(obs_dates, obs_values,
mod_dates, mod_values,
refStepMinutes)
rmsd = valstat.rms(obsValProj - modValProj)
N = len(obsValProj)
if figFile:
print '[info]: creating a plot ', figFile
plt.figure(figsize=(20, 4.5))
if htp_dates is not None:
plt.plot(htp_dates, htp_vals, color='c', label='ASTRON. TIDE')
plt.plot(obs_dates, obs_values, '.', color='g', label='OBSERVED')
plt.plot(mod_dates,
mod_values,
'.',
color='b',
label='STORM TIDE FCST')
try:
peak_val = np.nanmax(mod_values)
peak_ft = 3.28 * peak_val
peak_dat = mod_dates[np.argmax(mod_values)]
plt.plot(peak_dat, peak_val, 'bo')
plt.text(peak_dat,
1.05 * peak_val,
str(np.round(peak_val, 1)) + "m (" +
str(np.round(peak_ft, 1)) + "ft) MSL",
color='b')
plt.plot([obs_dates[0], mod_dates[-1]], [peak_val, peak_val],
'--b')
plt.plot([peak_dat, peak_dat], [ylim[0], ylim[1]], '--b')
peak_str = str(peak_dat.hour).zfill(2) + ':' + str(
peak_dat.minute).zfill(2)
plt.text(peak_dat, ylim[0], peak_str, color='b')
#Attempt to post-correct
lastdata = np.where(mod_dates == valstat.nearest(
mod_dates, obs_dates[-1])[0])[0][0]
offset = obs_values[-1] - mod_values[lastdata]
offset_ft = 3.28 * offset
plt.plot(mod_dates[lastdata:],
offset + mod_values[lastdata:],
color='gray',
label='POST-CORRECTED')
plt.plot(peak_dat, offset + peak_val, 'o', color='gray')
plt.text(peak_dat, 1.05*(offset+peak_val), \
str(np.round(offset+peak_val,1)) + "m (" + str(np.round(peak_ft+offset_ft,1)) +"ft) MSL", color='gray')
except:
pass
plt.ylim(ylim)
if daterange is not None:
plt.xlim([daterange[0], mod_dates[-1]])
#if rmsd > 0:
#plt.title(stationName+', RMSD=' +str(rmsd)+' meters')
#else:
plt.title(stationName)
plt.legend(bbox_to_anchor=(0.9, 0.35))
plt.grid()
plt.xlabel('DATE UTC')
plt.ylabel('WL, meters LMSL')
plt.savefig(figFile)
#plt.close()
return {'rmsd': rmsd, 'N': N}