def compute_seasonal_means_for_each_year(sim_config, season_to_months=None, var_name="", level=0):
"""
For each year so the significance test could be applied
:param sim_config:
:param season_to_months:
:return OrderedDict(season name: mean field for the season)
"""
assert isinstance(sim_config, RunConfig)
if season_to_months is None:
season_to_months = get_default_season_to_months_dict()
season_to_field = OrderedDict()
for season, months in season_to_months.items():
season_to_field[season] = analysis.get_mean_2d_fields_for_months(
path=sim_config.data_path, var_name=var_name, level=level, months=months,
start_year=sim_config.start_year, end_year=sim_config.end_year)
return season_to_field
def plot_corr_fields_and_std(vname="TT",
season_to_months=None,
sim_config=None):
"""
:param vname:
:param season_to_months:
:param sim_config:
"""
bmp_info = analysis.get_basemap_info_from_hdf(
file_path=sim_config.data_path)
# Get Anusplin data managers
obs_path = "/home/huziy/skynet3_rech1/anusplin_links"
pcp_obs_manager = AnuSplinManager(variable="pcp", folder_path=obs_path)
tmax_obs_manager = AnuSplinManager(variable="stmx", folder_path=obs_path)
tmin_obs_manager = AnuSplinManager(variable="stmn", folder_path=obs_path)
season_to_corr = OrderedDict()
season_to_area_mean_corr = OrderedDict()
season_to_area_mean_std_diff = OrderedDict()
season_to_area_mean_obs = OrderedDict()
season_to_area_mean_model = OrderedDict()
season_to_std_diff = OrderedDict()
season_to_years = OrderedDict()
# Convert modelled precip from M/s to mm/day
if vname == "PR":
model_convert_coef = 24 * 3600 * 1000
else:
model_convert_coef = 1
years = None
for season, months in season_to_months.items():
# Obs
if vname == "TT":
years, vals_max = tmax_obs_manager.get_seasonal_fields_interpolated_to(
start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats,
months=months)
_, vals_min = tmin_obs_manager.get_seasonal_fields_interpolated_to(
start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats,
months=months)
seasonal_obs = (years, (vals_min + vals_max) * 0.5)
elif vname == "PR":
seasonal_obs = pcp_obs_manager.get_seasonal_fields_interpolated_to(
start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats,
months=months)
years = seasonal_obs[0]
else:
raise Exception("Unknown variable: {}".format(vname))
season_to_years[season] = years
# Model
seasonal_model = analysis.get_mean_2d_fields_for_months(
path=sim_config.data_path,
var_name=vname,
level=0,
months=months,
start_year=sim_config.start_year,
end_year=sim_config.end_year)
seasonal_model *= model_convert_coef
# 2d fields
season_to_corr[season] = stat_helpers.calculate_correlation_nd(
seasonal_obs[1], seasonal_model)
season_to_corr[season] = np.ma.masked_where(
seasonal_obs[1][0, :, :].mask, season_to_corr[season])
i_arr, j_arr = np.where(~seasonal_obs[1][0, :, :].mask)
season_to_std_diff[season] = np.ma.masked_all_like(
seasonal_obs[1][0, :, :])
season_to_std_diff[season][
i_arr,
j_arr] = (seasonal_model[:, i_arr, j_arr].std(axis=0) -
seasonal_obs[1][:, i_arr, j_arr].std(axis=0)
) / seasonal_obs[1][:, i_arr, j_arr].std(axis=0) * 100
# area averages
season_to_area_mean_obs[season] = seasonal_obs[1][:, i_arr,
j_arr].mean(axis=1)
season_to_area_mean_model[season] = seasonal_model[:, i_arr,
j_arr].mean(axis=1)
season_to_area_mean_corr[season] = np.corrcoef(
season_to_area_mean_obs[season],
season_to_area_mean_model[season])[0, 1]
season_to_area_mean_std_diff[season] = (
season_to_area_mean_model[season].std() -
season_to_area_mean_obs[season].std()
) / season_to_area_mean_obs[season].std() * 100
# plt.figure()
# plt.plot(years, seasonal_obs[1][:, i_arr, j_arr].mean(axis=1), label="obs")
# plt.plot(years, seasonal_model[:, i_arr, j_arr].mean(axis=1), label="model")
# plt.legend()
#
#
# plt.figure()
# im = plt.pcolormesh(season_to_corr[season].T)
# plt.colorbar(im)
# plt.show()
# plotting
plot_utils.apply_plot_params(font_size=8, width_cm=23, height_cm=15)
nrows = 3
gs = gridspec.GridSpec(nrows=nrows,
ncols=len(season_to_months) + 1,
width_ratios=[
1,
] * len(season_to_months) + [
0.05,
])
fig = plt.figure()
xx, yy = bmp_info.get_proj_xy()
clevs_corr = np.arange(-1, 1.1, 0.1)
clevs_std_diff = np.arange(-100, 110, 10)
cmap = "seismic"
cmap_std = "seismic"
cs_corr = None
cs_std = None
for col, season in enumerate(season_to_months):
# Correlation fields
row = 0
ax = fig.add_subplot(gs[row, col])
ax.set_title(season)
if col == 0:
pass
# ax.set_ylabel("Correlation")
cs_corr = bmp_info.basemap.contourf(xx,
yy,
season_to_corr[season],
levels=clevs_corr,
ax=ax,
cmap=cmap)
bmp_info.basemap.drawcoastlines(ax=ax)
row += 1
ax = fig.add_subplot(gs[row, col])
if col == 0:
pass
# ax.set_ylabel(r"$\left( \sigma_{\rm model} - \sigma_{\rm obs.}\right)/\sigma_{\rm obs.} \cdot 100\%$")
cs_std = bmp_info.basemap.contourf(xx,
yy,
season_to_std_diff[season],
levels=clevs_std_diff,
ax=ax,
cmap=cmap_std,
extend="both")
bmp_info.basemap.drawcoastlines(ax=ax)
# area average
row += 1
ax = fig.add_subplot(gs[row, col])
if col == 0:
ax.set_ylabel(varname_to_ts_label[vname])
ax.plot(season_to_years[season],
season_to_area_mean_obs[season],
"k",
label="Obs.")
ax.plot(season_to_years[season],
season_to_area_mean_model[season],
"r",
label="Mod.")
ax.set_title(r"$r=" +
"{:.2f}, ".format(season_to_area_mean_corr[season]) +
r" \varepsilon_{\sigma} = " +
"{:.1f}".format(season_to_area_mean_std_diff[season]) +
r" \% $")
plt.setp(ax.xaxis.get_majorticklabels(), rotation=70)
assert isinstance(ax, Axes)
ax.xaxis.set_minor_locator(MultipleLocator())
ax.grid()
if col == 0:
ax.legend(loc="upper left")
# add colorbars
col = len(season_to_months)
row = 0
ax = fig.add_subplot(gs[row, col])
plt.colorbar(cs_corr, cax=ax)
row += 1
ax = fig.add_subplot(gs[row, col])
plt.colorbar(cs_std, cax=ax)
from crcm5.analyse_hdf import common_plot_params
fig.tight_layout()
fig.savefig(os.path.join(
img_folder,
"{}_{}-{}_corr_std.png".format(vname, sim_config.start_year,
sim_config.end_year)),
bbox_inches="tight",
dpi=common_plot_params.FIG_SAVE_DPI)
plt.close(fig)
def compare(paths=None, path_to_control_data=None, control_label="",
labels=None, varnames=None, levels=None, months_of_interest=None,
start_year=None, end_year=None):
"""
Comparing 2D fields
:param paths: paths to the simulation results
:param varnames:
:param labels: Display name for each simulation (number of labels should
be equal to the number of paths)
:param path_to_control_data: the path with which the comparison done i.e. a in the following
formula
delta = (x - a)/a * 100%
generates one image file per variable (in the folder images_for_lake-river_paper):
compare_varname_<control_label>_<label1>_..._<labeln>_startyear_endyear.png
"""
# get coordinate data (assumes that all the variables and runs have the same coordinates)
lons2d, lats2d, basemap = analysis.get_basemap_from_hdf(file_path=path_to_control_data)
x, y = basemap(lons2d, lats2d)
lake_fraction = analysis.get_array_from_file(path=path_to_control_data, var_name="lake_fraction")
if lake_fraction is None:
lake_fraction = np.zeros(lons2d.shape)
ncolors = 10
# +1 to include white
diff_cmap = cm.get_cmap("RdBu_r", ncolors + 1)
for var_name, level in zip(varnames, levels):
sfmt = infovar.get_colorbar_formatter(var_name)
control_means = analysis.get_mean_2d_fields_for_months(path=path_to_control_data, var_name=var_name,
months=months_of_interest,
start_year=start_year, end_year=end_year,
level=level)
control_mean = np.mean(control_means, axis=0)
fig = plt.figure()
assert isinstance(fig, Figure)
gs = gridspec.GridSpec(2, len(paths) + 1, wspace=0.5)
# plot the control
ax = fig.add_subplot(gs[0, 0])
assert isinstance(ax, Axes)
ax.set_title("{0}".format(control_label))
ax.set_ylabel("Mean: $X_{0}$")
to_plot = infovar.get_to_plot(var_name, control_mean,
lake_fraction=lake_fraction, mask_oceans=True, lons=lons2d, lats=lats2d)
# determine colorabr extent and spacing
field_cmap, field_norm = infovar.get_colormap_and_norm_for(var_name, to_plot, ncolors=ncolors)
basemap.pcolormesh(x, y, to_plot, cmap=field_cmap, norm=field_norm)
cb = basemap.colorbar(format=sfmt)
assert isinstance(cb, Colorbar)
# cb.ax.set_ylabel(infovar.get_units(var_name))
units = infovar.get_units(var_name)
info = "Variable:" \
"\n{0}" \
"\nPeriod: {1}-{2}" \
"\nMonths: {3}" \
"\nUnits: {4}"
info = info.format(infovar.get_long_name(var_name), start_year, end_year,
",".join([datetime(2001, m, 1).strftime("%b") for m in months_of_interest]), units)
ax.annotate(info, xy=(0.1, 0.3), xycoords="figure fraction")
sel_axes = [ax]
for the_path, the_label, column in zip(paths, labels, list(range(1, len(paths) + 1))):
means_for_years = analysis.get_mean_2d_fields_for_months(path=the_path, var_name=var_name,
months=months_of_interest,
start_year=start_year, end_year=end_year)
the_mean = np.mean(means_for_years, axis=0)
# plot the mean value
ax = fig.add_subplot(gs[0, column])
sel_axes.append(ax)
ax.set_title("{0}".format(the_label))
to_plot = infovar.get_to_plot(var_name, the_mean, lake_fraction=lake_fraction,
mask_oceans=True, lons=lons2d, lats=lats2d)
basemap.pcolormesh(x, y, to_plot, cmap=field_cmap, norm=field_norm)
ax.set_ylabel("Mean: $X_{0}$".format(column))
cb = basemap.colorbar(format=sfmt)
# cb.ax.set_ylabel(infovar.get_units(var_name))
# plot the difference
ax = fig.add_subplot(gs[1, column])
sel_axes.append(ax)
ax.set_ylabel("$X_{0} - X_0$".format(column))
# #Mask only if the previous plot (means) is masked
thediff = the_mean - control_mean
if hasattr(to_plot, "mask"):
to_plot = np.ma.masked_where(to_plot.mask, thediff)
else:
to_plot = thediff
if var_name == "PR": # convert to mm/day
to_plot = infovar.get_to_plot(var_name, to_plot, mask_oceans=False)
vmin = np.ma.min(to_plot)
vmax = np.ma.max(to_plot)
d = max(abs(vmin), abs(vmax))
vmin = -d
vmax = d
field_norm, bounds, vmn_nice, vmx_nice = infovar.get_boundary_norm(vmin, vmax, diff_cmap.N,
exclude_zero=False)
basemap.pcolormesh(x, y, to_plot, cmap=diff_cmap, norm=field_norm, vmin=vmn_nice, vmax=vmx_nice)
cb = basemap.colorbar(format=sfmt)
t, pval = ttest_ind(means_for_years, control_means, axis=0)
sig = pval < 0.1
basemap.contourf(x, y, sig.astype(int), nlevels=2, hatches=["+", None], colors="none")
# cb.ax.set_ylabel(infovar.get_units(var_name))
# plot coastlines
for the_ax in sel_axes:
basemap.drawcoastlines(ax=the_ax, linewidth=common_plot_params.COASTLINE_WIDTH)
# depends on the compared simulations and the months of interest
fig_file_name = "compare_{0}_{1}_{2}_months-{3}.jpeg".format(var_name, control_label,
"_".join(labels),
"-".join([str(m) for m in months_of_interest]))
figpath = os.path.join(images_folder, fig_file_name)
fig.savefig(figpath, dpi=cpp.FIG_SAVE_DPI, bbox_inches="tight")
plt.close(fig)
def plot_control_and_differences_in_one_panel_for_all_seasons_for_all_vars(
varnames=None, levels=None,
season_to_months=None,
start_year=None,
end_year=None):
season_list = list(season_to_months.keys())
pvalue_max = 0.1
# crcm5-r vs crcm5-hcd-r
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r_spinup.hdf"
# control_label = "CRCM5-R"
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r_spinup2.hdf", ]
# labels = ["CRCM5-HCD-R"]
# crcm5-hcd-rl vs crcm5-hcd-r
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r_spinup2.hdf"
# control_label = "CRCM5-HCD-R"
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf", ]
# labels = ["CRCM5-HCD-RL"]
# compare simulations with and without interflow
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf"
# control_label = "CRCM5-HCD-RL"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_do_not_discard_small.hdf", ]
# labels = ["CRCM5-HCD-RL-INTFL"]
# very high hydr cond
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_do_not_discard_small.hdf"
# control_label = "CRCM5-HCD-RL-INTFL"
##
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_sani-10000.hdf", ]
# labels = ["CRCM5-HCD-RL-INTFL-sani=10000"]
# Interflow effect
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf"
# control_label = "CRCM5-HCD-RL"
# ##
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_spinup_ITFS.hdf5", ]
# labels = ["ITFS"]
# total lake effect
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r.hdf5"
# control_label = "CRCM5-NL"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5", ]
# labels = ["CRCM5-L2", ]
# lake effect (lake-atm interactions)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r.hdf5"
# control_label = "CRCM5-R"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r.hdf5", ]
# labels = ["CRCM5-HCD-R", ]
# lake effect (lake-river interactions)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r.hdf5"
# control_label = "CRCM5-L1"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5", ]
# labels = ["CRCM5-HCD-L2", ]
# interflow effect ()
control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5"
control_label = "CRCM5-L2"
paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS.hdf5", ]
labels = ["CRCM5-L2I", ]
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS_avoid_truncation1979-1989.hdf5", ]
# labels = ["CRCM5-HCD-RL-INTFb", ]
# interflow effect (avoid truncation and bigger slopes)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS.hdf5"
# control_label = "CRCM5-HCD-RL-INTF"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS_avoid_truncation1979-1989.hdf5", ]
# labels = ["CRCM5-HCD-RL-INTF-improved", ]
#
row_labels = [
r"{} vs {}".format(s, control_label) for s in labels
]
print(labels)
# varnames = ["QQ", ]
# levels = [None, ]
assert len(levels) == len(varnames)
lons2d, lats2d, basemap = analysis.get_basemap_from_hdf(file_path=control_path)
x, y = basemap(lons2d, lats2d)
# save the domain properties for reuse
domain_props = DomainProperties()
domain_props.basemap = basemap
domain_props.lons2d = lons2d
domain_props.lats2d = lats2d
domain_props.x = x
domain_props.y = y
lake_fraction = analysis.get_array_from_file(path=control_path, var_name=infovar.HDF_LAKE_FRACTION_NAME)
dpth_to_bedrock = analysis.get_array_from_file(path=control_path, var_name=infovar.HDF_DEPTH_TO_BEDROCK_NAME)
assert dpth_to_bedrock is not None
if lake_fraction is None:
lake_fraction = np.zeros(lons2d.shape)
ncolors = 10
# +1 to include white
diff_cmap = cm.get_cmap("RdBu", ncolors + 1)
# Do the plotting for each variable
fig = plt.figure()
assert isinstance(fig, Figure)
# plot the control data
ncols = len(season_list) + 1 # +1 is for the colorbar
gs = gridspec.GridSpec(len(varnames), ncols, width_ratios=[1.0, ] * (ncols - 1) + [0.07], top=0.95)
lev_width_3d = np.ones(dpth_to_bedrock.shape + infovar.soil_layer_widths_26_to_60.shape)
lev_width_3d *= infovar.soil_layer_widths_26_to_60[np.newaxis, np.newaxis, :]
lev_bot_3d = lev_width_3d.cumsum(axis=2)
correction = -lev_bot_3d + dpth_to_bedrock[:, :, np.newaxis]
# Apply the correction only at points where the layer bottom is lower than
# the bedrock
lev_width_3d[correction < 0] += correction[correction < 0]
lev_width_3d[lev_width_3d < 0] = 0
# plot the plots one file per variable
for var_name, level, the_row in zip(varnames, levels, list(range(len(varnames)))):
sfmt = infovar.get_colorbar_formatter(var_name)
season_to_control_mean = {}
label_to_season_to_difference = {}
label_to_season_to_significance = {}
try:
# Calculate the difference for each season, and save the results to dictionaries
# to access later when plotting
for season, months_of_interest in season_to_months.items():
print("working on season: {0}".format(season))
control_means = analysis.get_mean_2d_fields_for_months(path=control_path, var_name=var_name,
months=months_of_interest,
start_year=start_year, end_year=end_year,
level=level)
control_mean = np.mean(control_means, axis=0)
control_mean = infovar.get_to_plot(var_name, control_mean,
lake_fraction=domain_props.lake_fraction,
lons=lons2d, lats=lats2d, level_width_m=lev_width_3d[:, :, level])
# multiply by the number of days in a season for PR and TRAF to convert them into mm from mm/day
if var_name in ["PR", "TRAF", "TDRA"]:
control_mean *= get_num_days(months_of_interest)
infovar.change_units_to(varnames=[var_name, ], new_units=r"${\rm mm}$")
season_to_control_mean[season] = control_mean
print("calculated mean from {0}".format(control_path))
# calculate the difference for each simulation
for the_path, the_label in zip(paths, row_labels):
modified_means = analysis.get_mean_2d_fields_for_months(path=the_path, var_name=var_name,
months=months_of_interest,
start_year=start_year, end_year=end_year,
level=level)
tval, pval = ttest_ind(modified_means, control_means, axis=0, equal_var=False)
significance = ((pval <= pvalue_max) & (~control_mean.mask)).astype(int)
print("pval ranges: {} to {}".format(pval.min(), pval.max()))
modified_mean = np.mean(modified_means, axis=0)
if the_label not in label_to_season_to_difference:
label_to_season_to_difference[the_label] = OrderedDict()
label_to_season_to_significance[the_label] = OrderedDict()
modified_mean = infovar.get_to_plot(var_name, modified_mean,
lake_fraction=domain_props.lake_fraction, lons=lons2d,
lats=lats2d, level_width_m=lev_width_3d[:, :, level])
# multiply by the number of days in a season for PR and TRAF to convert them into mm from mm/day
if var_name in ["PR", "TRAF", "TDRA"]:
modified_mean *= get_num_days(months_of_interest)
diff_vals = modified_mean - control_mean
print("diff ranges: min: {0}; max: {1}".format(diff_vals.min(), diff_vals.max()))
label_to_season_to_difference[the_label][season] = diff_vals
label_to_season_to_significance[the_label][season] = significance
print("Calculated mean and diff from {0}".format(the_path))
except NoSuchNodeError:
print("Could not find {0}, skipping...".format(var_name))
continue
for the_label, data in label_to_season_to_difference.items():
axes = []
for col in range(ncols):
axes.append(fig.add_subplot(gs[the_row, col]))
# Set season titles
if the_row == 0:
for the_season, ax in zip(season_list, axes):
ax.set_title(the_season)
_plot_row(axes, data, the_label, var_name, increments=True, domain_props=domain_props,
season_list=season_list, significance=label_to_season_to_significance[the_label])
var_label = infovar.get_long_display_label_for_var(var_name)
if var_name in ["I1"]:
var_label = "{}\n{} layer".format(var_label, ordinal(level + 1))
axes[0].set_ylabel(var_label)
fig.suptitle("({}) vs ({})".format(labels[0], control_label), font_properties=FontProperties(weight="bold"))
folderpath = os.path.join(images_folder, "seasonal_mean_maps/{0}_vs_{1}_for_{2}_{3}-{4}".format(
"_".join(labels), control_label, "-".join(list(season_to_months.keys())), start_year, end_year))
if not os.path.isdir(folderpath):
os.mkdir(folderpath)
imname = "{0}_{1}.png".format("-".join(varnames), "_".join(labels + [control_label]))
impath = os.path.join(folderpath, imname)
fig.savefig(impath, bbox_inches="tight")
def compare(paths=None,
path_to_control_data=None,
control_label="",
labels=None,
varnames=None,
levels=None,
months_of_interest=None,
start_year=None,
end_year=None):
"""
Comparing 2D fields
:param paths: paths to the simulation results
:param varnames:
:param labels: Display name for each simulation (number of labels should
be equal to the number of paths)
:param path_to_control_data: the path with which the comparison done i.e. a in the following
formula
delta = (x - a)/a * 100%
generates one image file per variable (in the folder images_for_lake-river_paper):
compare_varname_<control_label>_<label1>_..._<labeln>_startyear_endyear.png
"""
# get coordinate data (assumes that all the variables and runs have the same coordinates)
lons2d, lats2d, basemap = analysis.get_basemap_from_hdf(
file_path=path_to_control_data)
x, y = basemap(lons2d, lats2d)
lake_fraction = analysis.get_array_from_file(path=path_to_control_data,
var_name="lake_fraction")
if lake_fraction is None:
lake_fraction = np.zeros(lons2d.shape)
ncolors = 10
# +1 to include white
diff_cmap = cm.get_cmap("RdBu_r", ncolors + 1)
for var_name, level in zip(varnames, levels):
sfmt = infovar.get_colorbar_formatter(var_name)
control_means = analysis.get_mean_2d_fields_for_months(
path=path_to_control_data,
var_name=var_name,
months=months_of_interest,
start_year=start_year,
end_year=end_year,
level=level)
control_mean = np.mean(control_means, axis=0)
fig = plt.figure()
assert isinstance(fig, Figure)
gs = gridspec.GridSpec(2, len(paths) + 1, wspace=0.5)
# plot the control
ax = fig.add_subplot(gs[0, 0])
assert isinstance(ax, Axes)
ax.set_title("{0}".format(control_label))
ax.set_ylabel("Mean: $X_{0}$")
to_plot = infovar.get_to_plot(var_name,
control_mean,
lake_fraction=lake_fraction,
mask_oceans=True,
lons=lons2d,
lats=lats2d)
# determine colorabr extent and spacing
field_cmap, field_norm = infovar.get_colormap_and_norm_for(
var_name, to_plot, ncolors=ncolors)
basemap.pcolormesh(x, y, to_plot, cmap=field_cmap, norm=field_norm)
cb = basemap.colorbar(format=sfmt)
assert isinstance(cb, Colorbar)
# cb.ax.set_ylabel(infovar.get_units(var_name))
units = infovar.get_units(var_name)
info = "Variable:" \
"\n{0}" \
"\nPeriod: {1}-{2}" \
"\nMonths: {3}" \
"\nUnits: {4}"
info = info.format(
infovar.get_long_name(var_name), start_year, end_year, ",".join([
datetime(2001, m, 1).strftime("%b") for m in months_of_interest
]), units)
ax.annotate(info, xy=(0.1, 0.3), xycoords="figure fraction")
sel_axes = [ax]
for the_path, the_label, column in zip(paths, labels,
list(range(1,
len(paths) + 1))):
means_for_years = analysis.get_mean_2d_fields_for_months(
path=the_path,
var_name=var_name,
months=months_of_interest,
start_year=start_year,
end_year=end_year)
the_mean = np.mean(means_for_years, axis=0)
# plot the mean value
ax = fig.add_subplot(gs[0, column])
sel_axes.append(ax)
ax.set_title("{0}".format(the_label))
to_plot = infovar.get_to_plot(var_name,
the_mean,
lake_fraction=lake_fraction,
mask_oceans=True,
lons=lons2d,
lats=lats2d)
basemap.pcolormesh(x, y, to_plot, cmap=field_cmap, norm=field_norm)
ax.set_ylabel("Mean: $X_{0}$".format(column))
cb = basemap.colorbar(format=sfmt)
# cb.ax.set_ylabel(infovar.get_units(var_name))
# plot the difference
ax = fig.add_subplot(gs[1, column])
sel_axes.append(ax)
ax.set_ylabel("$X_{0} - X_0$".format(column))
# #Mask only if the previous plot (means) is masked
thediff = the_mean - control_mean
if hasattr(to_plot, "mask"):
to_plot = np.ma.masked_where(to_plot.mask, thediff)
else:
to_plot = thediff
if var_name == "PR": # convert to mm/day
to_plot = infovar.get_to_plot(var_name,
to_plot,
mask_oceans=False)
vmin = np.ma.min(to_plot)
vmax = np.ma.max(to_plot)
d = max(abs(vmin), abs(vmax))
vmin = -d
vmax = d
field_norm, bounds, vmn_nice, vmx_nice = infovar.get_boundary_norm(
vmin, vmax, diff_cmap.N, exclude_zero=False)
basemap.pcolormesh(x,
y,
to_plot,
cmap=diff_cmap,
norm=field_norm,
vmin=vmn_nice,
vmax=vmx_nice)
cb = basemap.colorbar(format=sfmt)
t, pval = ttest_ind(means_for_years, control_means, axis=0)
sig = pval < 0.1
basemap.contourf(x,
y,
sig.astype(int),
nlevels=2,
hatches=["+", None],
colors="none")
# cb.ax.set_ylabel(infovar.get_units(var_name))
# plot coastlines
for the_ax in sel_axes:
basemap.drawcoastlines(
ax=the_ax, linewidth=common_plot_params.COASTLINE_WIDTH)
# depends on the compared simulations and the months of interest
fig_file_name = "compare_{0}_{1}_{2}_months-{3}.jpeg".format(
var_name, control_label, "_".join(labels),
"-".join([str(m) for m in months_of_interest]))
figpath = os.path.join(images_folder, fig_file_name)
fig.savefig(figpath, dpi=cpp.FIG_SAVE_DPI, bbox_inches="tight")
plt.close(fig)
def plot_control_and_differences_in_one_panel_for_all_seasons_for_all_vars(
varnames=None,
levels=None,
season_to_months=None,
start_year=None,
end_year=None):
season_list = list(season_to_months.keys())
pvalue_max = 0.1
# crcm5-r vs crcm5-hcd-r
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r_spinup.hdf"
# control_label = "CRCM5-R"
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r_spinup2.hdf", ]
# labels = ["CRCM5-HCD-R"]
# crcm5-hcd-rl vs crcm5-hcd-r
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r_spinup2.hdf"
# control_label = "CRCM5-HCD-R"
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf", ]
# labels = ["CRCM5-HCD-RL"]
# compare simulations with and without interflow
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf"
# control_label = "CRCM5-HCD-RL"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_do_not_discard_small.hdf", ]
# labels = ["CRCM5-HCD-RL-INTFL"]
# very high hydr cond
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_do_not_discard_small.hdf"
# control_label = "CRCM5-HCD-RL-INTFL"
##
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_sani-10000.hdf", ]
# labels = ["CRCM5-HCD-RL-INTFL-sani=10000"]
# Interflow effect
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl_spinup.hdf"
# control_label = "CRCM5-HCD-RL"
# ##
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_spinup_ITFS.hdf5", ]
# labels = ["ITFS"]
# total lake effect
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r.hdf5"
# control_label = "CRCM5-NL"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5", ]
# labels = ["CRCM5-L2", ]
# lake effect (lake-atm interactions)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-r.hdf5"
# control_label = "CRCM5-R"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r.hdf5", ]
# labels = ["CRCM5-HCD-R", ]
# lake effect (lake-river interactions)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-r.hdf5"
# control_label = "CRCM5-L1"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5", ]
# labels = ["CRCM5-HCD-L2", ]
# interflow effect ()
control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl.hdf5"
control_label = "CRCM5-L2"
paths = [
"/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS.hdf5",
]
labels = [
"CRCM5-L2I",
]
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS_avoid_truncation1979-1989.hdf5", ]
# labels = ["CRCM5-HCD-RL-INTFb", ]
# interflow effect (avoid truncation and bigger slopes)
# control_path = "/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS.hdf5"
# control_label = "CRCM5-HCD-RL-INTF"
#
# paths = ["/skynet3_rech1/huziy/hdf_store/quebec_0.1_crcm5-hcd-rl-intfl_ITFS_avoid_truncation1979-1989.hdf5", ]
# labels = ["CRCM5-HCD-RL-INTF-improved", ]
#
row_labels = [r"{} vs {}".format(s, control_label) for s in labels]
print(labels)
# varnames = ["QQ", ]
# levels = [None, ]
assert len(levels) == len(varnames)
lons2d, lats2d, basemap = analysis.get_basemap_from_hdf(
file_path=control_path)
x, y = basemap(lons2d, lats2d)
# save the domain properties for reuse
domain_props = DomainProperties()
domain_props.basemap = basemap
domain_props.lons2d = lons2d
domain_props.lats2d = lats2d
domain_props.x = x
domain_props.y = y
lake_fraction = analysis.get_array_from_file(
path=control_path, var_name=infovar.HDF_LAKE_FRACTION_NAME)
dpth_to_bedrock = analysis.get_array_from_file(
path=control_path, var_name=infovar.HDF_DEPTH_TO_BEDROCK_NAME)
assert dpth_to_bedrock is not None
if lake_fraction is None:
lake_fraction = np.zeros(lons2d.shape)
ncolors = 10
# +1 to include white
diff_cmap = cm.get_cmap("RdBu", ncolors + 1)
# Do the plotting for each variable
fig = plt.figure()
assert isinstance(fig, Figure)
# plot the control data
ncols = len(season_list) + 1 # +1 is for the colorbar
gs = gridspec.GridSpec(len(varnames),
ncols,
width_ratios=[
1.0,
] * (ncols - 1) + [0.07],
top=0.95)
lev_width_3d = np.ones(dpth_to_bedrock.shape +
infovar.soil_layer_widths_26_to_60.shape)
lev_width_3d *= infovar.soil_layer_widths_26_to_60[np.newaxis,
np.newaxis, :]
lev_bot_3d = lev_width_3d.cumsum(axis=2)
correction = -lev_bot_3d + dpth_to_bedrock[:, :, np.newaxis]
# Apply the correction only at points where the layer bottom is lower than
# the bedrock
lev_width_3d[correction < 0] += correction[correction < 0]
lev_width_3d[lev_width_3d < 0] = 0
# plot the plots one file per variable
for var_name, level, the_row in zip(varnames, levels,
list(range(len(varnames)))):
sfmt = infovar.get_colorbar_formatter(var_name)
season_to_control_mean = {}
label_to_season_to_difference = {}
label_to_season_to_significance = {}
try:
# Calculate the difference for each season, and save the results to dictionaries
# to access later when plotting
for season, months_of_interest in season_to_months.items():
print("working on season: {0}".format(season))
control_means = analysis.get_mean_2d_fields_for_months(
path=control_path,
var_name=var_name,
months=months_of_interest,
start_year=start_year,
end_year=end_year,
level=level)
control_mean = np.mean(control_means, axis=0)
control_mean = infovar.get_to_plot(
var_name,
control_mean,
lake_fraction=domain_props.lake_fraction,
lons=lons2d,
lats=lats2d,
level_width_m=lev_width_3d[:, :, level])
# multiply by the number of days in a season for PR and TRAF to convert them into mm from mm/day
if var_name in ["PR", "TRAF", "TDRA"]:
control_mean *= get_num_days(months_of_interest)
infovar.change_units_to(varnames=[
var_name,
],
new_units=r"${\rm mm}$")
season_to_control_mean[season] = control_mean
print("calculated mean from {0}".format(control_path))
# calculate the difference for each simulation
for the_path, the_label in zip(paths, row_labels):
modified_means = analysis.get_mean_2d_fields_for_months(
path=the_path,
var_name=var_name,
months=months_of_interest,
start_year=start_year,
end_year=end_year,
level=level)
tval, pval = ttest_ind(modified_means,
control_means,
axis=0,
equal_var=False)
significance = ((pval <= pvalue_max) &
(~control_mean.mask)).astype(int)
print("pval ranges: {} to {}".format(
pval.min(), pval.max()))
modified_mean = np.mean(modified_means, axis=0)
if the_label not in label_to_season_to_difference:
label_to_season_to_difference[the_label] = OrderedDict(
)
label_to_season_to_significance[
the_label] = OrderedDict()
modified_mean = infovar.get_to_plot(
var_name,
modified_mean,
lake_fraction=domain_props.lake_fraction,
lons=lons2d,
lats=lats2d,
level_width_m=lev_width_3d[:, :, level])
# multiply by the number of days in a season for PR and TRAF to convert them into mm from mm/day
if var_name in ["PR", "TRAF", "TDRA"]:
modified_mean *= get_num_days(months_of_interest)
diff_vals = modified_mean - control_mean
print("diff ranges: min: {0}; max: {1}".format(
diff_vals.min(), diff_vals.max()))
label_to_season_to_difference[the_label][
season] = diff_vals
label_to_season_to_significance[the_label][
season] = significance
print("Calculated mean and diff from {0}".format(the_path))
except NoSuchNodeError:
print("Could not find {0}, skipping...".format(var_name))
continue
for the_label, data in label_to_season_to_difference.items():
axes = []
for col in range(ncols):
axes.append(fig.add_subplot(gs[the_row, col]))
# Set season titles
if the_row == 0:
for the_season, ax in zip(season_list, axes):
ax.set_title(the_season)
_plot_row(axes,
data,
the_label,
var_name,
increments=True,
domain_props=domain_props,
season_list=season_list,
significance=label_to_season_to_significance[the_label])
var_label = infovar.get_long_display_label_for_var(var_name)
if var_name in ["I1"]:
var_label = "{}\n{} layer".format(var_label,
ordinal(level + 1))
axes[0].set_ylabel(var_label)
fig.suptitle("({}) vs ({})".format(labels[0], control_label),
font_properties=FontProperties(weight="bold"))
folderpath = os.path.join(
images_folder, "seasonal_mean_maps/{0}_vs_{1}_for_{2}_{3}-{4}".format(
"_".join(labels), control_label,
"-".join(list(season_to_months.keys())), start_year, end_year))
if not os.path.isdir(folderpath):
os.mkdir(folderpath)
imname = "{0}_{1}.png".format("-".join(varnames),
"_".join(labels + [control_label]))
impath = os.path.join(folderpath, imname)
fig.savefig(impath, bbox_inches="tight")
def plot_corr_fields_and_std(vname="TT", season_to_months=None, sim_config=None):
"""
:param vname:
:param season_to_months:
:param sim_config:
"""
bmp_info = analysis.get_basemap_info_from_hdf(file_path=sim_config.data_path)
# Get Anusplin data managers
obs_path = "/home/huziy/skynet3_rech1/anusplin_links"
pcp_obs_manager = AnuSplinManager(variable="pcp", folder_path=obs_path)
tmax_obs_manager = AnuSplinManager(variable="stmx", folder_path=obs_path)
tmin_obs_manager = AnuSplinManager(variable="stmn", folder_path=obs_path)
season_to_corr = OrderedDict()
season_to_area_mean_corr = OrderedDict()
season_to_area_mean_std_diff = OrderedDict()
season_to_area_mean_obs = OrderedDict()
season_to_area_mean_model = OrderedDict()
season_to_std_diff = OrderedDict()
season_to_years = OrderedDict()
# Convert modelled precip from M/s to mm/day
if vname == "PR":
model_convert_coef = 24 * 3600 * 1000
else:
model_convert_coef = 1
years = None
for season, months in season_to_months.items():
# Obs
if vname == "TT":
years, vals_max = tmax_obs_manager.get_seasonal_fields_interpolated_to(start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats,
months=months)
_, vals_min = tmin_obs_manager.get_seasonal_fields_interpolated_to(start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats, months=months)
seasonal_obs = (years, (vals_min + vals_max) * 0.5)
elif vname == "PR":
seasonal_obs = pcp_obs_manager.get_seasonal_fields_interpolated_to(start_year=sim_config.start_year,
end_year=sim_config.end_year,
lons_target=bmp_info.lons,
lats_target=bmp_info.lats,
months=months)
years = seasonal_obs[0]
else:
raise Exception("Unknown variable: {}".format(vname))
season_to_years[season] = years
# Model
seasonal_model = analysis.get_mean_2d_fields_for_months(path=sim_config.data_path, var_name=vname, level=0,
months=months, start_year=sim_config.start_year, end_year=sim_config.end_year)
seasonal_model *= model_convert_coef
# 2d fields
season_to_corr[season] = stat_helpers.calculate_correlation_nd(seasonal_obs[1], seasonal_model)
season_to_corr[season] = np.ma.masked_where(seasonal_obs[1][0, :, :].mask, season_to_corr[season])
i_arr, j_arr = np.where(~seasonal_obs[1][0, :, :].mask)
season_to_std_diff[season] = np.ma.masked_all_like(seasonal_obs[1][0, :, :])
season_to_std_diff[season][i_arr, j_arr] = (seasonal_model[:, i_arr, j_arr].std(axis=0) - seasonal_obs[1][:, i_arr, j_arr].std(axis=0)) / seasonal_obs[1][:, i_arr, j_arr].std(axis=0) * 100
# area averages
season_to_area_mean_obs[season] = seasonal_obs[1][:, i_arr, j_arr].mean(axis=1)
season_to_area_mean_model[season] = seasonal_model[:, i_arr, j_arr].mean(axis=1)
season_to_area_mean_corr[season] = np.corrcoef(season_to_area_mean_obs[season], season_to_area_mean_model[season])[0, 1]
season_to_area_mean_std_diff[season] = (season_to_area_mean_model[season].std() - season_to_area_mean_obs[season].std()) / season_to_area_mean_obs[season].std() * 100
# plt.figure()
# plt.plot(years, seasonal_obs[1][:, i_arr, j_arr].mean(axis=1), label="obs")
# plt.plot(years, seasonal_model[:, i_arr, j_arr].mean(axis=1), label="model")
# plt.legend()
#
#
# plt.figure()
# im = plt.pcolormesh(season_to_corr[season].T)
# plt.colorbar(im)
# plt.show()
# plotting
plot_utils.apply_plot_params(font_size=8, width_cm=23, height_cm=15)
nrows = 3
gs = gridspec.GridSpec(nrows=nrows, ncols=len(season_to_months) + 1, width_ratios=[1, ] * len(season_to_months) + [0.05, ])
fig = plt.figure()
xx, yy = bmp_info.get_proj_xy()
clevs_corr = np.arange(-1, 1.1, 0.1)
clevs_std_diff = np.arange(-100, 110, 10)
cmap = "seismic"
cmap_std = "seismic"
cs_corr = None
cs_std = None
for col, season in enumerate(season_to_months):
# Correlation fields
row = 0
ax = fig.add_subplot(gs[row, col])
ax.set_title(season)
if col == 0:
pass
# ax.set_ylabel("Correlation")
cs_corr = bmp_info.basemap.contourf(xx, yy, season_to_corr[season], levels=clevs_corr, ax=ax, cmap=cmap)
bmp_info.basemap.drawcoastlines(ax=ax)
row += 1
ax = fig.add_subplot(gs[row, col])
if col == 0:
pass
# ax.set_ylabel(r"$\left( \sigma_{\rm model} - \sigma_{\rm obs.}\right)/\sigma_{\rm obs.} \cdot 100\%$")
cs_std = bmp_info.basemap.contourf(xx, yy, season_to_std_diff[season], levels=clevs_std_diff, ax=ax, cmap=cmap_std, extend="both")
bmp_info.basemap.drawcoastlines(ax=ax)
# area average
row += 1
ax = fig.add_subplot(gs[row, col])
if col == 0:
ax.set_ylabel(varname_to_ts_label[vname])
ax.plot(season_to_years[season], season_to_area_mean_obs[season], "k", label="Obs.")
ax.plot(season_to_years[season], season_to_area_mean_model[season], "r", label="Mod.")
ax.set_title(r"$r=" + "{:.2f}, ".format(season_to_area_mean_corr[season]) + r" \varepsilon_{\sigma} = " + "{:.1f}".format(season_to_area_mean_std_diff[season]) + r" \% $")
plt.setp(ax.xaxis.get_majorticklabels(), rotation=70)
assert isinstance(ax, Axes)
ax.xaxis.set_minor_locator(MultipleLocator())
ax.grid()
if col == 0:
ax.legend(loc="upper left")
# add colorbars
col = len(season_to_months)
row = 0
ax = fig.add_subplot(gs[row, col])
plt.colorbar(cs_corr, cax=ax)
row += 1
ax = fig.add_subplot(gs[row, col])
plt.colorbar(cs_std, cax=ax)
from crcm5.analyse_hdf import common_plot_params
fig.tight_layout()
fig.savefig(os.path.join(img_folder, "{}_{}-{}_corr_std.png".format(vname, sim_config.start_year, sim_config.end_year)), bbox_inches="tight", dpi=common_plot_params.FIG_SAVE_DPI)
plt.close(fig)
