def test_signchange(self):
ts = pd.Series([-2., -1., 1., 2., 3], index=np.arange(5))
sc = utils.signchange(ts)
assert_array_equal(sc, [0, 0, 1, 0, 0])
ts = pd.Series([-2., -1., 1., 2., 3][::-1], index=np.arange(5))
sc = utils.signchange(ts)
assert_array_equal(sc, [0, 0, 0, 1, 0])
def test_signchange(self):
ts = pd.Series([-2., -1., 1., 2., 3], index=np.arange(5))
sc = utils.signchange(ts)
assert_array_equal(sc, [0, 0, 1, 0, 0])
ts = pd.Series([-2., -1., 1., 2., 3][::-1], index=np.arange(5))
sc = utils.signchange(ts)
assert_array_equal(sc, [0, 0, 0, 1, 0])
def t_star_from_refmb(gdir, mbdf):
"""Computes the t* for the glacier, given a series of MB measurements.
Could be multiprocessed but its probably not necessary.
Parameters
----------
gdirs: the list of oggm.GlacierDirectory objects where to write the data.
mbdf: a pd.Series containing the observed MB data indexed by year
Returns
-------
A dict: {t_star:[], bias:[], bias_std:[], prcp_fac:float}
"""
# Only divide 0, we believe the original RGI entities to be the ref...
years, temp_yr_ts, prcp_yr_ts = mb_yearly_climate_on_glacier(gdir, 1.,
div_id=0)
# which years to look at
selind = np.searchsorted(years, mbdf.index)
temp_yr_ts = temp_yr_ts[selind]
prcp_yr_ts = prcp_yr_ts[selind]
temp_yr = np.mean(temp_yr_ts)
prcp_yr = np.mean(prcp_yr_ts)
# Average oberved mass-balance
ref_mb = np.mean(mbdf)
ref_mb_std = np.std(mbdf)
# Average mass-balance per mu and fac
mu_yr_clim_df = gdir.read_pickle('mu_candidates', div_id=0)
odf = pd.DataFrame(index=mu_yr_clim_df.columns)
out = dict()
for prcp_fac in mu_yr_clim_df:
mu_yr_clim = mu_yr_clim_df[prcp_fac]
nmu = len(mu_yr_clim)
mb_per_mu = prcp_yr * prcp_fac - mu_yr_clim * temp_yr
mbts_per_mu = np.atleast_2d(prcp_yr_ts * prcp_fac).repeat(nmu, 0) - \
np.atleast_2d(mu_yr_clim).T * \
np.atleast_2d(temp_yr_ts).repeat(nmu, 0)
std_per_mu = mb_per_mu*0 + np.std(mbts_per_mu, axis=1)
# Diff to reference
diff = (mb_per_mu - ref_mb).dropna()
diff_std = (std_per_mu - ref_mb_std).dropna()
signchange = utils.signchange(diff)
# If sign changes save them
# TODO: ideas to do this better:
# - apply a smooth (take noise into account)
# - take longer stable periods into account
# these stuffs could be proven better (or not) with cross-val
pchange = np.where(signchange == 1)[0]
years = diff.index
diff = diff.values
diff_std = diff_std.values
if len(pchange) > 0:
t_stars = []
bias = []
std_bias = []
for p in pchange:
# Check the side with the smallest bias
ide = p-1 if np.abs(diff[p-1]) < np.abs(diff[p]) else p
if years[ide] not in t_stars:
t_stars.append(years[ide])
bias.append(diff[ide])
std_bias.append(diff_std[ide])
else:
amin = np.argmin(np.abs(diff))
t_stars = [years[amin]]
bias = [diff[amin]]
std_bias = [diff_std[amin]]
# (prcp_fac, t_stars, bias, std_bias)
odf.loc[prcp_fac, 'avg_bias'] = np.mean(bias)
odf.loc[prcp_fac, 'avg_std_bias'] = np.mean(std_bias)
odf.loc[prcp_fac, 'n_tstar'] = len(std_bias)
out[prcp_fac] = {'t_star': t_stars, 'bias': bias, 'std_bias': std_bias,
'prcp_fac': prcp_fac}
# write
gdir.write_pickle(odf, 'prcp_fac_optim')
# we take the closest result and see later if it needs cleverer handling
amin = np.argmin(np.abs(odf.avg_std_bias)) # this gives back an index!
return out[amin]
def t_star_from_refmb(gdir, mbdf):
"""Computes the t* for the glacier, given a series of MB measurements.
Could be multiprocessed but its probably not necessary.
Parameters
----------
gdirs: the list of oggm.GlacierDirectory objects where to write the data.
mbdf: a pd.Series containing the observed MB data indexed by year
Returns:
--------
(t_star, bias, prcp_fac): lists of t*, their associated bias and a prcp fac
"""
# Only divide 0, we believe the original RGI entities to be the ref...
years, temp_yr_ts, prcp_yr_ts = mb_yearly_climate_on_glacier(gdir, 1.,
div_id=0)
# which years to look at
selind = np.searchsorted(years, mbdf.index)
temp_yr_ts = temp_yr_ts[selind]
prcp_yr_ts = prcp_yr_ts[selind]
temp_yr = np.mean(temp_yr_ts)
prcp_yr = np.mean(prcp_yr_ts)
# Average oberved mass-balance
ref_mb = np.mean(mbdf)
ref_mb_std = np.std(mbdf)
# Average mass-balance per mu and fac
mu_yr_clim_df = gdir.read_pickle('mu_candidates', div_id=0)
odf = pd.DataFrame(index=mu_yr_clim_df.columns)
out = dict()
for prcp_fac in mu_yr_clim_df:
mu_yr_clim = mu_yr_clim_df[prcp_fac]
nmu = len(mu_yr_clim)
mb_per_mu = prcp_yr * prcp_fac - mu_yr_clim * temp_yr
mbts_per_mu = np.atleast_2d(prcp_yr_ts * prcp_fac).repeat(nmu, 0) - \
np.atleast_2d(mu_yr_clim).T * \
np.atleast_2d(temp_yr_ts).repeat(nmu, 0)
std_per_mu = mb_per_mu*0 + np.std(mbts_per_mu, axis=1)
# Diff to reference
diff = (mb_per_mu - ref_mb).dropna()
diff_std = (std_per_mu - ref_mb_std).dropna()
signchange = utils.signchange(diff)
# If sign changes save them
# TODO: ideas to do this better:
# - apply a smooth (take noise into account)
# - take longer stable periods into account
# these stuffs could be proven better (or not) with cross-val
pchange = np.where(signchange == 1)[0]
years = diff.index
diff = diff.values
diff_std = diff_std.values
if len(pchange) > 0:
t_stars = []
bias = []
std_bias = []
for p in pchange:
# Check the side with the smallest bias
ide = p-1 if np.abs(diff[p-1]) < np.abs(diff[p]) else p
if years[ide] not in t_stars:
t_stars.append(years[ide])
bias.append(diff[ide])
std_bias.append(diff_std[ide])
else:
amin = np.argmin(np.abs(diff))
t_stars = [years[amin]]
bias = [diff[amin]]
std_bias = [diff_std[amin]]
# (prcp_fac, t_stars, bias, std_bias)
odf.loc[prcp_fac, 'avg_bias'] = np.mean(bias)
odf.loc[prcp_fac, 'avg_std_bias'] = np.mean(std_bias)
odf.loc[prcp_fac, 'n_tstar'] = len(std_bias)
out[prcp_fac] = t_stars, bias, prcp_fac
# write
gdir.write_pickle(odf, 'prcp_fac_optim')
# we take the closest result and see later if it needs cleverer handling
amin = np.argmin(np.abs(odf.avg_std_bias)) # this gives back an index!
return out[amin]