def test_completeOMIReproduction_coarsegrid(tmp_path): errors = [] # Calculate EOFs raw_olr = olr.load_noaa_interpolated_olr(olr_data_filename) shorter_olr = olr.restrict_time_coverage(raw_olr, np.datetime64('1979-01-01'), np.datetime64('2012-12-31')) coarse_lat = np.arange(-20., 20.1, 8.0) coarse_long = np.arange(0., 359.9, 20.0) interpolated_olr = olr.interpolate_spatial_grid(shorter_olr, coarse_lat, coarse_long) eofs = omi.calc_eofs_from_olr(interpolated_olr, sign_doy1reference=True, interpolate_eofs=True, strict_leap_year_treatment=True) eofs.save_all_eofs_to_npzfile( tmp_path / "test_completeOMIReproduction_coarsegrid_EOFs.npz") # Validate EOFs against mjoindices own reference (results should be equal) mjoindices_reference_eofs = eof.restore_all_eofs_from_npzfile( mjoindices_reference_eofs_filename_coarsegrid) for idx, target_eof in enumerate(eofs.eof_list): if not mjoindices_reference_eofs.eof_list[idx].close(target_eof): errors.append( "mjoindices-reference-validation: EOF data at index %i is incorrect" % idx) # Calculate PCs raw_olr = olr.load_noaa_interpolated_olr(olr_data_filename) pcs = omi.calculate_pcs_from_olr(raw_olr, eofs, np.datetime64("1979-01-01"), np.datetime64("2018-08-28"), use_quick_temporal_filter=False) pc_temp_file = tmp_path / "test_completeOMIReproduction_coarsegrid_PCs.txt" pcs.save_pcs_to_txt_file(pc_temp_file) # Validate PCs against mjoindices own reference (results should be equal) # Reload pcs instead of using the calculated ones, because the saving routine has truncated some decimals of the # reference values. So do the same with the testing target pcs. pcs = pc.load_pcs_from_txt_file(pc_temp_file) mjoindices_reference_pcs = pc.load_pcs_from_txt_file( mjoindices_reference_pcs_filename_coarsegrid) if not np.all(mjoindices_reference_pcs.time == pcs.time): errors.append( "mjoindices-reference-validation: Dates of PCs do not match.") if not np.allclose(mjoindices_reference_pcs.pc1, pcs.pc1): errors.append( "mjoindices-reference-validation: PC1 values do not match.") if not np.allclose(mjoindices_reference_pcs.pc2, pcs.pc2): errors.append( "mjoindices-reference-validation: PC2 values do not match.") assert not errors, "errors occurred:\n{}".format("\n".join(errors))
def test_save_all_eofs_to_npzfile(tmp_path): filename = tmp_path / "test.npz" doys = tools.doy_list() lat = np.array([-10., 0., 10.]) long = np.array([0., 5.]) explained_variances = np.array([np.arange(1, doys.size + 1, 1) + 111, np.arange(1, doys.size + 1, 1) + 222, np.arange(1, doys.size + 1, 1) + 333, np.arange(1, doys.size + 1, 1) + 444, np.arange(1, doys.size + 1, 1) + 555, np.arange(1, doys.size + 1, 1) + 666]) eigenvalues = np.array([np.arange(1, doys.size + 1, 1) + 1111, np.arange(1, doys.size + 1, 1) + 2222, np.arange(1, doys.size + 1, 1) + 3333, np.arange(1, doys.size + 1, 1) + 4444, np.arange(1, doys.size + 1, 1) + 5555, np.arange(1, doys.size + 1, 1) + 6666]) no_obs = doys * 5 eofs = [] for doy in doys: eof1 = np.array([1, 2, 3, 4, 5, 6]) * doy eof2 = np.array([10, 20, 30, 40, 50, 60]) * doy eofs.append(eof.EOFData(lat, long, eof1, eof2, explained_variances=np.squeeze(explained_variances[:, doy - 1]), eigenvalues=np.squeeze(eigenvalues[:, doy - 1]), no_observations=no_obs[doy - 1])) target = eof.EOFDataForAllDOYs(eofs) target.save_all_eofs_to_npzfile(filename) errors = [] target_reloaded = eof.restore_all_eofs_from_npzfile(filename) if not target_reloaded.eof_list == eofs: errors.append("List of EOFData objects incorrect") if not np.all(target_reloaded.lat == lat): errors.append("Lat is incorrect") if not np.all(target_reloaded.long == long): errors.append("Long is incorrect") if not target_reloaded.eofdata_for_doy(1) == eofs[0]: errors.append("Sample EOF data is incorrect") assert not errors, "errors occurred:\n{}".format("\n".join(errors))
# Calculate the eofs. In the postprocessing, the signs of the EOFs are adjusted and the EOF in a period # around DOY 300 are replaced by an interpolation see Kiladis, 2014). # The switch strict_leap_year_treatment has major implications only for the EOFs calculated for DOY 366 and causes only # minor differences for the other DOYs. While the results for setting strict_leap_year_treatment=False are closer to the # original values, the calculation strict_leap_year_treatment=True is somewhat more stringently implemented using # built-in datetime functionality. # See documentation of mjoindices.tools.find_doy_ranges_in_dates() for details. eofs = omi.calc_eofs_from_olr(interpolated_olr, sign_doy1reference=True, interpolate_eofs=True, strict_leap_year_treatment=False) eofs.save_all_eofs_to_npzfile(eofnpzfile) # ### Some diagnostic plots to evaluate the calculated EOFs. # Load precalculated EOFs first. eofs = eof.restore_all_eofs_from_npzfile(eofnpzfile) # Check the explained variance by the EOFs. Values are lower by about a factor of 2 than in Kiladis (2014), # which is correct! fig = eof.plot_explained_variance_for_all_doys(eofs) fig.show() fig.savefig(fig_dir / "EOFs_ExplainedVariance.png") # Check details of the EOF pair for a particular DOY in the following. doy = 50 # Plot EOFs for this DOY. fig = eof.plot_individual_eof_map(eofs.eofdata_for_doy(doy), doy) fig.show() fig.savefig(fig_dir / "EOF_Sample.png") # Plot the explained variance for the first 10 EOFs of this DOY to check the drop of explained variances after EOF2. fig = eof.plot_individual_explained_variance_all_eofs(
def test_completeOMIReproduction(tmp_path): errors = [] # Calculate EOFs raw_olr = olr.load_noaa_interpolated_olr(olr_data_filename) shorter_olr = olr.restrict_time_coverage(raw_olr, np.datetime64('1979-01-01'), np.datetime64('2012-12-31')) interpolated_olr = olr.interpolate_spatial_grid_to_original(shorter_olr) eofs = omi.calc_eofs_from_olr(interpolated_olr, sign_doy1reference=True, interpolate_eofs=True, strict_leap_year_treatment=False) eofs.save_all_eofs_to_npzfile(tmp_path / "test_completeOMIReproduction_EOFs.npz") # Validate EOFs against original (results are inexact but close) orig_eofs = eof.load_all_original_eofs_from_directory( originalOMIDataDirname) corr_1, diff_mean_1, diff_std_1, diff_abs_percent68_1, diff_abs_percent95_1, diff_abs_percent99_1 = mjoindices.evaluation_tools.calc_comparison_stats_for_eofs_all_doys( orig_eofs, eofs, 1, exclude_doy366=False, percentage=False, do_print=False) if not np.all(corr_1 > 0.994): errors.append( "original-validation: Correlation for EOF1 at least for one DOY too low!" ) if not np.all(diff_abs_percent99_1 < 0.0084): errors.append( "original-validation: 99% percentile for EOF1 at least for one DOY too high!" ) if not np.all(diff_abs_percent68_1 < 0.0018): errors.append( "original-validation: 68% percentile for EOF1 at least for one DOY too high!" ) corr_2, diff_mean_2, diff_std_2, diff_abs_percent68_2, diff_abs_percent95_2, diff_abs_percent99_2 = mjoindices.evaluation_tools.calc_comparison_stats_for_eofs_all_doys( orig_eofs, eofs, 2, exclude_doy366=False, percentage=False, do_print=False) if not np.all(corr_2 > 0.993): errors.append( "original-validation: Correlation for EOF2 at least for one DOY too low!" ) if not np.all(diff_abs_percent99_2 < 0.0065): errors.append( "original-validation: 99% percentile for EOF2 at least for one DOY too high!" ) if not np.all(diff_abs_percent68_2 < 0.0018): errors.append( "original-validation: 68% percentile for EOF2 at least for one DOY too high!" ) # Validate explained variance against original (results are inexact but close) orig_explained_variance_1, orig_explained_variance_2 = mjoindices.evaluation_tools.load_omi_explained_variance( original_omi_explained_variance_file) corr_var1, diff_mean_var1, diff_std_var1, diff_vec_var1, diff_abs_percent68_var1, diff_abs_percent95_var1, diff_abs_percent99_var1 = mjoindices.evaluation_tools.calc_comparison_stats_for_explained_variance( orig_explained_variance_1, eofs.explained_variance1_for_all_doys(), do_print=False, exclude_doy366=False) if not diff_std_var1 < 0.0008: errors.append( "original-validation: Std.Dev. of the difference of both explained variances for EOF1 is to too high!" ) if not diff_abs_percent99_var1 < 0.0017: errors.append( "original-validation: 99% percentile of the difference of both explained variances for EOF1 is to too high!" ) if not diff_abs_percent68_var1 < 0.0009: errors.append( "original-validation: 68% percentile of the difference of both explained variances for EOF1 is to too high!" ) corr_var2, diff_mean_var2, diff_std_var2, diff_vec_var2, diff_abs_percent68_var2, diff_abs_percent95_var2, diff_abs_percent99_var2 = mjoindices.evaluation_tools.calc_comparison_stats_for_explained_variance( orig_explained_variance_2, eofs.explained_variance2_for_all_doys(), do_print=False, exclude_doy366=False) if not diff_std_var2 < 0.0008: errors.append( "original-validation: Std.Dev. of the difference of both explained variances for EOF2 is to too high!" ) if not diff_abs_percent99_var2 < 0.0018: errors.append( "original-validation: 99% percentile of the difference of both explained variances for EOF2 is to too high!" ) if not diff_abs_percent68_var2 < 0.001: errors.append( "original-validation: 68% percentile of the difference of both explained variances for EOF2 is to too high!" ) # Validate EOFs against mjoindices own reference (results should be equal) mjoindices_reference_eofs = eof.restore_all_eofs_from_npzfile( mjoindices_reference_eofs_filename) for idx, target_eof in enumerate(eofs.eof_list): if not mjoindices_reference_eofs.eof_list[idx].close(target_eof): errors.append( "mjoindices-reference-validation: EOF data at index %i is incorrect" % idx) # Calculate PCs raw_olr = olr.load_noaa_interpolated_olr(olr_data_filename) pcs = omi.calculate_pcs_from_olr(raw_olr, eofs, np.datetime64("1979-01-01"), np.datetime64("2018-08-28"), use_quick_temporal_filter=False) pc_temp_file = tmp_path / "test_completeOMIReproduction_PCs.txt" pcs.save_pcs_to_txt_file(pc_temp_file) # Validate PCs against original (results are inexact but close) orig_pcs = pc.load_original_pcs_from_txt_file(origOMIPCsFilename) # Reload pcs instead of using the calculated ones, because the saving routine has truncated some decimals of the # reference values. So do the same with the testing target pcs. pcs = pc.load_pcs_from_txt_file(pc_temp_file) tempa, tempb, corr_pc1, diff_mean_pc1, diff_std_pc1, diff_ts_abs_pc1, diff_abs_percent68_pc1, diff_abs_percent95_pc1, diff_abs_percent99_pc1 = mjoindices.evaluation_tools.calc_timeseries_agreement( orig_pcs.pc1, orig_pcs.time, pcs.pc1, pcs.time, exclude_doy366=False, do_print=False) if not corr_pc1 > 0.998: errors.append( "original-validation: Correlation for PC1 timeseries is to too low!" ) if not diff_std_pc1 < 0.0458: errors.append( "original-validation: Std.Dev. of the difference of both PC1 timeseries is to too high!" ) if not diff_abs_percent99_pc1 < 0.157: errors.append( "original-validation: 99% percentile of the difference of both PC1 timeseries is to too high!" ) if not diff_abs_percent68_pc1 < 0.0327: errors.append( "original-validation: 68% percentile of the difference of both PC1 timeseries is to too high!" ) tempa, tempb, corr_pc2, diff_mean_pc2, diff_std_pc2, diff_ts_abs_pc2, diff_abs_percent68_pc2, diff_abs_percent95_pc2, diff_abs_percent99_pc2 = mjoindices.evaluation_tools.calc_timeseries_agreement( orig_pcs.pc2, orig_pcs.time, pcs.pc2, pcs.time, exclude_doy366=False, do_print=False) if not corr_pc2 > 0.998: errors.append( "original-validation: Correlation for PC2 timeseries is to too low!" ) if not diff_std_pc2 < 0.0488: errors.append( "original-validation: Std.Dev. of the difference of both PC2 timeseries is to too high!" ) if not diff_abs_percent99_pc2 < 0.1704: errors.append( "original-validation: 99% percentile of the difference of both PC2 timeseries is to too high!" ) if not diff_abs_percent68_pc2 < 0.0353: errors.append( "original-validation: 68% percentile of the difference of both PC2 timeseries is to too high!" ) strength = np.sqrt(np.square(pcs.pc1) + np.square(pcs.pc2)) orig_strength = np.sqrt(np.square(orig_pcs.pc1) + np.square(orig_pcs.pc2)) tempa, tempb, corr_strength, diff_mean_strength, diff_std_strength, diff_ts_abs_strength, diff_abs_percent68_strength, diff_abs_percent95_strength, diff_abs_percent99_strength = mjoindices.evaluation_tools.calc_timeseries_agreement( orig_strength, orig_pcs.time, strength, pcs.time, exclude_doy366=False, do_print=False) if not corr_strength > 0.9998: errors.append( "original-validation: Correlation for strength timeseries is to too low!" ) if not diff_std_strength < 0.0103: errors.append( "original-validation: Std.Dev. of the difference of both strength timeseries is to too high!" ) if not diff_abs_percent99_strength < 0.0341: errors.append( "original-validation: 99% percentile of the difference of both strength timeseries is to too high!" ) if not diff_abs_percent68_strength < 0.0079: errors.append( "original-validation: 68% percentile of the difference of both strength timeseries is to too high!" ) # Validate PCs against mjoindices own reference (results should be equal) mjoindices_reference_pcs = pc.load_pcs_from_txt_file( mjoindices_reference_pcs_filename) if not np.all(mjoindices_reference_pcs.time == pcs.time): errors.append( "mjoindices-reference-validation: Dates of PCs do not match.") if not np.allclose(mjoindices_reference_pcs.pc1, pcs.pc1): errors.append( "mjoindices-reference-validation: PC1 values do not match.") if not np.allclose(mjoindices_reference_pcs.pc2, pcs.pc2): errors.append( "mjoindices-reference-validation: PC2 values do not match.") assert not errors, "errors occurred:\n{}".format("\n".join(errors))