[3777] | 1 | #!/usr/bin/env python3 |
---|
| 2 | ####################################################################################################### |
---|
[3809] | 3 | ### Python script to output stratification data over time from "restartpem#.nc" files ### |
---|
| 4 | ### and to plot orbital parameters from "obl_ecc_lsp.asc" ### |
---|
[3777] | 5 | ####################################################################################################### |
---|
[3458] | 6 | |
---|
| 7 | import os |
---|
[3771] | 8 | import sys |
---|
[3458] | 9 | import numpy as np |
---|
[3777] | 10 | from glob import glob |
---|
[3458] | 11 | from netCDF4 import Dataset |
---|
| 12 | import matplotlib.pyplot as plt |
---|
[3819] | 13 | from mpl_toolkits.axes_grid1.inset_locator import inset_axes |
---|
| 14 | from matplotlib.colors import LinearSegmentedColormap, LogNorm |
---|
[3777] | 15 | from scipy.interpolate import interp1d |
---|
[3458] | 16 | |
---|
[3777] | 17 | |
---|
[3771] | 18 | def get_user_inputs(): |
---|
[3777] | 19 | """ |
---|
| 20 | Prompt the user for: |
---|
| 21 | - folder_path: directory containing NetCDF files (default: "starts") |
---|
| 22 | - base_name: base filename (default: "restartpem") |
---|
| 23 | - infofile: name of the PEM info file (default: "info_PEM.txt") |
---|
| 24 | Validates existence of folder and infofile before returning. |
---|
| 25 | """ |
---|
| 26 | folder_path = input( |
---|
| 27 | "Enter the folder path containing the NetCDF files " |
---|
| 28 | "(press Enter for default [starts]): " |
---|
| 29 | ).strip() or "starts" |
---|
[3771] | 30 | while not os.path.isdir(folder_path): |
---|
[3777] | 31 | print(f" » \"{folder_path}\" does not exist or is not a directory.") |
---|
| 32 | folder_path = input( |
---|
| 33 | "Enter a valid folder path (press Enter for default [starts]): " |
---|
| 34 | ).strip() or "starts" |
---|
[3458] | 35 | |
---|
[3777] | 36 | base_name = input( |
---|
| 37 | "Enter the base name of the NetCDF files " |
---|
| 38 | "(press Enter for default [restartpem]): " |
---|
| 39 | ).strip() or "restartpem" |
---|
[3458] | 40 | |
---|
[3777] | 41 | infofile = input( |
---|
| 42 | "Enter the name of the PEM info file " |
---|
| 43 | "(press Enter for default [info_PEM.txt]): " |
---|
| 44 | ).strip() or "info_PEM.txt" |
---|
[3771] | 45 | while not os.path.isfile(infofile): |
---|
[3777] | 46 | print(f" » \"{infofile}\" does not exist or is not a file.") |
---|
| 47 | infofile = input( |
---|
| 48 | "Enter a valid PEM info filename (press Enter for default [info_PEM.txt]): " |
---|
| 49 | ).strip() or "info_PEM.txt" |
---|
[3771] | 50 | |
---|
[3809] | 51 | orbfile = input( |
---|
| 52 | "Enter the name of the orbital parameters ASCII file " |
---|
| 53 | "(press Enter for default [obl_ecc_lsp.asc]): " |
---|
| 54 | ).strip() or "obl_ecc_lsp.asc" |
---|
| 55 | while not os.path.isfile(orbfile): |
---|
| 56 | print(f" » \"{orbfile}\" does not exist or is not a file.") |
---|
| 57 | orbfile = input( |
---|
| 58 | "Enter a valid orbital parameters ASCII filename (press Enter for default [obl_ecc_lsp.asc]): " |
---|
| 59 | ).strip() or "info_PEM.txt" |
---|
[3771] | 60 | |
---|
[3809] | 61 | return folder_path, base_name, infofile, orbfile |
---|
[3458] | 62 | |
---|
[3809] | 63 | |
---|
[3777] | 64 | def list_netcdf_files(folder_path, base_name): |
---|
| 65 | """ |
---|
| 66 | List and sort all NetCDF files matching the pattern {base_name}#.nc |
---|
| 67 | in folder_path. Returns a sorted list of full file paths. |
---|
| 68 | """ |
---|
| 69 | pattern = os.path.join(folder_path, f"{base_name}[0-9]*.nc") |
---|
| 70 | all_files = glob(pattern) |
---|
| 71 | if not all_files: |
---|
| 72 | return [] |
---|
[3458] | 73 | |
---|
[3777] | 74 | def extract_index(pathname): |
---|
| 75 | fname = os.path.basename(pathname) |
---|
| 76 | idx_str = fname[len(base_name):-3] |
---|
| 77 | return int(idx_str) if idx_str.isdigit() else float('inf') |
---|
| 78 | |
---|
| 79 | sorted_files = sorted(all_files, key=extract_index) |
---|
| 80 | return sorted_files |
---|
| 81 | |
---|
| 82 | |
---|
| 83 | def open_sample_dataset(file_path): |
---|
| 84 | """ |
---|
| 85 | Open a single NetCDF file and extract: |
---|
| 86 | - ngrid, nslope |
---|
| 87 | - longitude, latitude |
---|
| 88 | Returns (ngrid, nslope, longitude_array, latitude_array). |
---|
| 89 | """ |
---|
| 90 | with Dataset(file_path, 'r') as ds: |
---|
| 91 | ngrid = ds.dimensions['physical_points'].size |
---|
| 92 | nslope = ds.dimensions['nslope'].size |
---|
| 93 | longitude = ds.variables['longitude'][:].copy() |
---|
| 94 | latitude = ds.variables['latitude'][:].copy() |
---|
| 95 | return ngrid, nslope, longitude, latitude |
---|
| 96 | |
---|
| 97 | |
---|
| 98 | def collect_stratification_variables(files, base_name): |
---|
| 99 | """ |
---|
| 100 | Scan all files to collect: |
---|
| 101 | - variable names for each stratification property |
---|
| 102 | - max number of strata (max_nb_str) |
---|
| 103 | - global min base elevation and max top elevation |
---|
| 104 | Returns: |
---|
| 105 | - var_info: dict mapping each property_name -> sorted list of var names |
---|
| 106 | - max_nb_str: int |
---|
| 107 | - min_base_elev: float |
---|
| 108 | - max_top_elev: float |
---|
| 109 | """ |
---|
[3458] | 110 | max_nb_str = 0 |
---|
[3777] | 111 | min_base_elev = np.inf |
---|
| 112 | max_top_elev = -np.inf |
---|
[3771] | 113 | |
---|
[3777] | 114 | property_markers = { |
---|
| 115 | 'heights': 'stratif_slope', # "..._top_elevation" |
---|
| 116 | 'co2_ice': 'h_co2ice', |
---|
| 117 | 'h2o_ice': 'h_h2oice', |
---|
| 118 | 'dust': 'h_dust', |
---|
| 119 | 'pore': 'h_pore', |
---|
[3790] | 120 | 'pore_ice': 'poreice_volfrac' |
---|
[3777] | 121 | } |
---|
| 122 | var_info = {prop: set() for prop in property_markers} |
---|
[3458] | 123 | |
---|
[3777] | 124 | for file_path in files: |
---|
| 125 | with Dataset(file_path, 'r') as ds: |
---|
| 126 | if 'nb_str_max' in ds.dimensions: |
---|
| 127 | max_nb_str = max(max_nb_str, ds.dimensions['nb_str_max'].size) |
---|
[3458] | 128 | |
---|
[3777] | 129 | nslope = ds.dimensions['nslope'].size |
---|
| 130 | for k in range(1, nslope + 1): |
---|
| 131 | var_name = f"stratif_slope{k:02d}_top_elevation" |
---|
| 132 | if var_name in ds.variables: |
---|
| 133 | arr = ds.variables[var_name][:] |
---|
| 134 | min_base_elev = min(min_base_elev, np.min(arr)) |
---|
| 135 | max_top_elev = max(max_top_elev, np.max(arr)) |
---|
| 136 | var_info['heights'].add(var_name) |
---|
[3458] | 137 | |
---|
[3777] | 138 | for full_var in ds.variables: |
---|
| 139 | for prop, marker in property_markers.items(): |
---|
| 140 | if (marker in full_var) and prop != 'heights': |
---|
| 141 | var_info[prop].add(full_var) |
---|
[3458] | 142 | |
---|
[3777] | 143 | for prop in var_info: |
---|
| 144 | var_info[prop] = sorted(var_info[prop]) |
---|
[3458] | 145 | |
---|
[3777] | 146 | return var_info, max_nb_str, min_base_elev, max_top_elev |
---|
[3458] | 147 | |
---|
| 148 | |
---|
[3777] | 149 | def load_full_datasets(files): |
---|
| 150 | """ |
---|
| 151 | Open all NetCDF files and return a list of Dataset objects. |
---|
| 152 | (They should be closed by the caller after use.) |
---|
| 153 | """ |
---|
| 154 | return [Dataset(fp, 'r') for fp in files] |
---|
| 155 | |
---|
| 156 | |
---|
| 157 | def extract_stratification_data(datasets, var_info, ngrid, nslope, max_nb_str): |
---|
| 158 | """ |
---|
| 159 | Build: |
---|
| 160 | - heights_data[t_idx][isl] = 2D array (ngrid, n_strata_current) of top_elevations. |
---|
| 161 | - raw_prop_arrays[prop] = 4D array (ngrid, ntime, nslope, max_nb_str) of per-strata values. |
---|
| 162 | Returns: |
---|
| 163 | - heights_data: list (ntime) of lists (nslope) of 2D arrays |
---|
| 164 | - raw_prop_arrays: dict mapping each property_name -> 4D array |
---|
| 165 | - ntime: number of time steps (files) |
---|
| 166 | """ |
---|
| 167 | ntime = len(datasets) |
---|
| 168 | |
---|
| 169 | heights_data = [ |
---|
| 170 | [None for _ in range(nslope)] |
---|
| 171 | for _ in range(ntime) |
---|
| 172 | ] |
---|
| 173 | for t_idx, ds in enumerate(datasets): |
---|
| 174 | for var_name in var_info['heights']: |
---|
| 175 | slope_idx = int(var_name.split("slope")[1].split("_")[0]) - 1 |
---|
| 176 | if 0 <= slope_idx < nslope: |
---|
| 177 | raw = ds.variables[var_name][0, :, :] # (n_strata, ngrid) |
---|
| 178 | heights_data[t_idx][slope_idx] = raw.T # (ngrid, n_strata) |
---|
| 179 | |
---|
| 180 | raw_prop_arrays = {} |
---|
| 181 | for prop in var_info: |
---|
| 182 | if prop == 'heights': |
---|
| 183 | continue |
---|
| 184 | raw_prop_arrays[prop] = np.zeros((ngrid, ntime, nslope, max_nb_str), dtype=np.float32) |
---|
| 185 | |
---|
| 186 | def slope_index_from_var(vname): |
---|
| 187 | return int(vname.split("slope")[1].split("_")[0]) - 1 |
---|
| 188 | |
---|
| 189 | for prop in raw_prop_arrays: |
---|
| 190 | slope_map = {} |
---|
| 191 | for vname in var_info[prop]: |
---|
| 192 | isl = slope_index_from_var(vname) |
---|
| 193 | if 0 <= isl < nslope: |
---|
| 194 | slope_map[isl] = vname |
---|
| 195 | |
---|
| 196 | arr = raw_prop_arrays[prop] |
---|
| 197 | for t_idx, ds in enumerate(datasets): |
---|
| 198 | for isl, var_name in slope_map.items(): |
---|
| 199 | raw = ds.variables[var_name][0, :, :] # (n_strata, ngrid) |
---|
| 200 | n_strata_current = raw.shape[0] |
---|
| 201 | arr[:, t_idx, isl, :n_strata_current] = raw.T |
---|
| 202 | |
---|
| 203 | return heights_data, raw_prop_arrays, ntime |
---|
| 204 | |
---|
| 205 | |
---|
| 206 | def normalize_to_fractions(raw_prop_arrays): |
---|
| 207 | """ |
---|
| 208 | Given raw_prop_arrays for 'co2_ice', 'h2o_ice', 'dust', 'pore' (in meters), |
---|
[3809] | 209 | normalize each set of strata so that the sum of those four = 1 per cell. |
---|
[3777] | 210 | Returns: |
---|
| 211 | - frac_arrays: dict mapping same keys -> 4D arrays of fractions (0..1). |
---|
| 212 | """ |
---|
| 213 | co2 = raw_prop_arrays['co2_ice'] |
---|
| 214 | h2o = raw_prop_arrays['h2o_ice'] |
---|
| 215 | dust = raw_prop_arrays['dust'] |
---|
| 216 | pore = raw_prop_arrays['pore'] |
---|
| 217 | |
---|
| 218 | total = co2 + h2o + dust + pore |
---|
| 219 | mask = total > 0.0 |
---|
| 220 | |
---|
| 221 | frac_co2 = np.zeros_like(co2, dtype=np.float32) |
---|
| 222 | frac_h2o = np.zeros_like(h2o, dtype=np.float32) |
---|
| 223 | frac_dust = np.zeros_like(dust, dtype=np.float32) |
---|
| 224 | frac_pore = np.zeros_like(pore, dtype=np.float32) |
---|
| 225 | |
---|
| 226 | frac_co2[mask] = co2[mask] / total[mask] |
---|
| 227 | frac_h2o[mask] = h2o[mask] / total[mask] |
---|
| 228 | frac_dust[mask] = dust[mask] / total[mask] |
---|
| 229 | frac_pore[mask] = pore[mask] / total[mask] |
---|
| 230 | |
---|
| 231 | return { |
---|
| 232 | 'co2_ice': frac_co2, |
---|
| 233 | 'h2o_ice': frac_h2o, |
---|
| 234 | 'dust': frac_dust, |
---|
| 235 | 'pore': frac_pore |
---|
| 236 | } |
---|
| 237 | |
---|
| 238 | |
---|
| 239 | def read_infofile(file_name): |
---|
| 240 | """ |
---|
| 241 | Reads "info_PEM.txt". Expects: |
---|
[3809] | 242 | - First line: parameters where the 3rd value is martian_to_earth conversion factor. |
---|
| 243 | - Each subsequent line: floats where first value is simulation timestamp (in Mars years). |
---|
| 244 | Returns: |
---|
| 245 | - date_time: 1D numpy array of timestamps (Mars years) |
---|
| 246 | - martian_to_earth: float conversion factor |
---|
[3777] | 247 | """ |
---|
| 248 | date_time = [] |
---|
| 249 | with open(file_name, 'r') as fp: |
---|
[3809] | 250 | first = fp.readline().split() |
---|
| 251 | martian_to_earth = float(first[2]) |
---|
[3777] | 252 | for line in fp: |
---|
| 253 | parts = line.strip().split() |
---|
| 254 | if not parts: |
---|
| 255 | continue |
---|
| 256 | try: |
---|
| 257 | date_time.append(float(parts[0])) |
---|
| 258 | except ValueError: |
---|
| 259 | continue |
---|
[3809] | 260 | return np.array(date_time, dtype=np.float64), martian_to_earth |
---|
[3777] | 261 | |
---|
| 262 | |
---|
| 263 | def get_yes_no_input(prompt: str) -> bool: |
---|
| 264 | """ |
---|
| 265 | Prompt the user with a yes/no question. Returns True for yes, False for no. |
---|
| 266 | """ |
---|
[3771] | 267 | while True: |
---|
[3777] | 268 | choice = input(f"{prompt} (y/n): ").strip().lower() |
---|
| 269 | if choice in ['y', 'yes']: |
---|
| 270 | return True |
---|
| 271 | elif choice in ['n', 'no']: |
---|
| 272 | return False |
---|
| 273 | else: |
---|
| 274 | print("Please respond with y or n.") |
---|
| 275 | |
---|
| 276 | |
---|
| 277 | def prompt_discretization_step(max_top_elev): |
---|
| 278 | """ |
---|
| 279 | Prompt for a positive float dz such that 0 < dz <= max_top_elev. |
---|
| 280 | """ |
---|
| 281 | while True: |
---|
| 282 | entry = input( |
---|
| 283 | "Enter the discretization step of the reference grid for the elevation [m]: " |
---|
| 284 | ).strip() |
---|
[3771] | 285 | try: |
---|
[3777] | 286 | dz = float(entry) |
---|
[3771] | 287 | if dz <= 0: |
---|
[3777] | 288 | print(" » Discretization step must be strictly positive!") |
---|
[3771] | 289 | continue |
---|
[3777] | 290 | if dz > max_top_elev: |
---|
| 291 | print( |
---|
| 292 | f" » {dz:.3e} m is greater than the maximum top elevation " |
---|
| 293 | f"({max_top_elev:.3e} m). Please enter a smaller value." |
---|
| 294 | ) |
---|
[3771] | 295 | continue |
---|
[3777] | 296 | return dz |
---|
[3771] | 297 | except ValueError: |
---|
[3777] | 298 | print(" » Invalid numeric value. Please try again.") |
---|
[3771] | 299 | |
---|
[3458] | 300 | |
---|
[3777] | 301 | def interpolate_data_on_refgrid( |
---|
| 302 | heights_data, |
---|
| 303 | prop_arrays, |
---|
| 304 | min_base_for_interp, |
---|
| 305 | max_top_elev, |
---|
| 306 | dz, |
---|
| 307 | exclude_sub=False |
---|
| 308 | ): |
---|
| 309 | """ |
---|
[3809] | 310 | Build a reference elevation grid and interpolate strata fractions onto it. |
---|
[3458] | 311 | |
---|
[3777] | 312 | Returns: |
---|
| 313 | - ref_grid: 1D array of elevations (nz,) |
---|
[3809] | 314 | - gridded_data: dict mapping each property_name to 4D array |
---|
| 315 | (ngrid, ntime, nslope, nz) with interpolated fractions. |
---|
| 316 | - top_index: 3D array (ngrid, ntime, nslope) of ints: |
---|
| 317 | number of levels covered by the topmost stratum. |
---|
[3777] | 318 | """ |
---|
| 319 | if exclude_sub and (dz > max_top_elev): |
---|
| 320 | ref_grid = np.array([0.0, max_top_elev], dtype=np.float32) |
---|
| 321 | else: |
---|
[3809] | 322 | ref_grid = np.arange(min_base_for_interp, max_top_elev + dz/2, dz) |
---|
[3777] | 323 | nz = len(ref_grid) |
---|
| 324 | print(f"> Number of reference grid points = {nz}") |
---|
[3458] | 325 | |
---|
[3777] | 326 | sample_prop = next(iter(prop_arrays.values())) |
---|
[3809] | 327 | ngrid, ntime, nslope, max_nb_str = sample_prop.shape |
---|
[3458] | 328 | |
---|
[3777] | 329 | gridded_data = { |
---|
| 330 | prop: np.full((ngrid, ntime, nslope, nz), -1.0, dtype=np.float32) |
---|
| 331 | for prop in prop_arrays |
---|
| 332 | } |
---|
| 333 | top_index = np.zeros((ngrid, ntime, nslope), dtype=np.int32) |
---|
| 334 | |
---|
| 335 | for ig in range(ngrid): |
---|
| 336 | for t_idx in range(ntime): |
---|
| 337 | for isl in range(nslope): |
---|
| 338 | h_mat = heights_data[t_idx][isl] |
---|
| 339 | if h_mat is None: |
---|
| 340 | continue |
---|
| 341 | |
---|
[3809] | 342 | raw_h = h_mat[ig, :] |
---|
[3777] | 343 | h_all = np.full((max_nb_str,), np.nan, dtype=np.float32) |
---|
| 344 | n_strata_current = raw_h.shape[0] |
---|
| 345 | h_all[:n_strata_current] = raw_h |
---|
| 346 | |
---|
| 347 | if exclude_sub: |
---|
| 348 | epsilon = 1e-6 |
---|
| 349 | valid_mask = (h_all >= -epsilon) |
---|
| 350 | else: |
---|
| 351 | valid_mask = (~np.isnan(h_all)) & (h_all != 0.0) |
---|
| 352 | |
---|
| 353 | if not np.any(valid_mask): |
---|
| 354 | continue |
---|
| 355 | |
---|
| 356 | h_valid = h_all[valid_mask] |
---|
| 357 | top_h = np.max(h_valid) |
---|
| 358 | i_zmax = np.searchsorted(ref_grid, top_h, side='right') |
---|
| 359 | top_index[ig, t_idx, isl] = i_zmax |
---|
| 360 | if i_zmax == 0: |
---|
| 361 | continue |
---|
| 362 | |
---|
| 363 | for prop, arr in prop_arrays.items(): |
---|
[3809] | 364 | prop_profile_all = arr[ig, t_idx, isl, :] |
---|
| 365 | prop_profile = prop_profile_all[valid_mask] |
---|
[3777] | 366 | if prop_profile.size == 0: |
---|
| 367 | continue |
---|
| 368 | |
---|
| 369 | f_interp = interp1d( |
---|
| 370 | h_valid, |
---|
| 371 | prop_profile, |
---|
| 372 | kind='next', |
---|
| 373 | bounds_error=False, |
---|
| 374 | fill_value=-1.0 |
---|
| 375 | ) |
---|
| 376 | gridded_data[prop][ig, t_idx, isl, :i_zmax] = f_interp(ref_grid[:i_zmax]) |
---|
| 377 | |
---|
| 378 | return ref_grid, gridded_data, top_index |
---|
| 379 | |
---|
| 380 | |
---|
| 381 | def plot_stratification_over_time( |
---|
| 382 | gridded_data, |
---|
| 383 | ref_grid, |
---|
| 384 | top_index, |
---|
| 385 | heights_data, |
---|
| 386 | date_time, |
---|
| 387 | exclude_sub=False, |
---|
| 388 | output_folder="." |
---|
| 389 | ): |
---|
| 390 | """ |
---|
[3809] | 391 | For each grid point and slope, generate a 2×2 figure of: |
---|
[3777] | 392 | - CO2 ice fraction |
---|
| 393 | - H2O ice fraction |
---|
| 394 | - Dust fraction |
---|
| 395 | - Pore fraction |
---|
| 396 | """ |
---|
| 397 | prop_names = ['co2_ice', 'h2o_ice', 'dust', 'pore'] |
---|
| 398 | titles = ["CO2 ice", "H2O ice", "Dust", "Pore"] |
---|
[3792] | 399 | cmap = plt.get_cmap('turbo').copy() |
---|
[3777] | 400 | cmap.set_under('white') |
---|
| 401 | vmin, vmax = 0.0, 1.0 |
---|
| 402 | |
---|
| 403 | sample_prop = next(iter(gridded_data.values())) |
---|
| 404 | ngrid, ntime, nslope, nz = sample_prop.shape |
---|
| 405 | |
---|
| 406 | if exclude_sub: |
---|
| 407 | positive_indices = np.where(ref_grid >= 0.0)[0] |
---|
| 408 | sub_ref_grid = ref_grid[positive_indices] |
---|
| 409 | else: |
---|
| 410 | positive_indices = np.arange(nz) |
---|
| 411 | sub_ref_grid = ref_grid |
---|
| 412 | |
---|
| 413 | for ig in range(ngrid): |
---|
| 414 | for isl in range(nslope): |
---|
| 415 | fig, axes = plt.subplots(2, 2, figsize=(10, 8)) |
---|
| 416 | fig.suptitle( |
---|
[3809] | 417 | f"Content variation over time for (Grid point {ig+1}, Slope {isl+1})", |
---|
[3819] | 418 | fontsize=14, |
---|
| 419 | fontweight='bold' |
---|
[3777] | 420 | ) |
---|
| 421 | |
---|
[3809] | 422 | # Precompute valid stratum tops per time |
---|
[3777] | 423 | valid_tops_per_time = [] |
---|
| 424 | for t_idx in range(ntime): |
---|
[3809] | 425 | raw_h = heights_data[t_idx][isl][ig, :] |
---|
[3777] | 426 | h_all = raw_h[~np.isnan(raw_h)] |
---|
| 427 | if exclude_sub: |
---|
| 428 | h_all = h_all[h_all >= 0.0] |
---|
| 429 | valid_tops_per_time.append(np.unique(h_all)) |
---|
| 430 | |
---|
| 431 | for idx, prop in enumerate(prop_names): |
---|
| 432 | ax = axes.flat[idx] |
---|
[3809] | 433 | data_3d = gridded_data[prop][ig, :, isl, :] |
---|
| 434 | mat_full = data_3d.T |
---|
| 435 | mat = mat_full[positive_indices, :].copy() |
---|
[3777] | 436 | mat[mat < 0.0] = np.nan |
---|
| 437 | |
---|
[3809] | 438 | # Mask above top stratum |
---|
[3777] | 439 | for t_idx in range(ntime): |
---|
| 440 | i_zmax = top_index[ig, t_idx, isl] |
---|
| 441 | if i_zmax <= positive_indices[0]: |
---|
| 442 | mat[:, t_idx] = np.nan |
---|
| 443 | else: |
---|
| 444 | count_z = np.count_nonzero(positive_indices < i_zmax) |
---|
| 445 | mat[count_z:, t_idx] = np.nan |
---|
| 446 | |
---|
| 447 | im = ax.pcolormesh( |
---|
| 448 | date_time, |
---|
| 449 | sub_ref_grid, |
---|
| 450 | mat, |
---|
| 451 | cmap=cmap, |
---|
| 452 | shading='auto', |
---|
| 453 | vmin=vmin, |
---|
| 454 | vmax=vmax |
---|
| 455 | ) |
---|
| 456 | ax.set_title(titles[idx], fontsize=12) |
---|
[3809] | 457 | ax.set_xlabel("Time (Mars years)") |
---|
[3777] | 458 | ax.set_ylabel("Elevation (m)") |
---|
| 459 | |
---|
| 460 | fig.subplots_adjust(right=0.88) |
---|
[3819] | 461 | fig.tight_layout(rect=[0, 0, 0.88, 1.0]) |
---|
[3777] | 462 | cbar_ax = fig.add_axes([0.90, 0.15, 0.02, 0.7]) |
---|
[3809] | 463 | fig.colorbar(im, cax=cbar_ax, orientation='vertical', label="Content") |
---|
[3777] | 464 | |
---|
| 465 | fname = os.path.join( |
---|
| 466 | output_folder, f"layering_evolution_ig{ig+1}_is{isl+1}.png" |
---|
| 467 | ) |
---|
| 468 | fig.savefig(fname, dpi=150) |
---|
| 469 | |
---|
| 470 | |
---|
[3819] | 471 | def plot_stratification_rgb_over_time( |
---|
| 472 | gridded_data, |
---|
| 473 | ref_grid, |
---|
| 474 | top_index, |
---|
| 475 | heights_data, |
---|
| 476 | date_time, |
---|
| 477 | exclude_sub=False, |
---|
| 478 | output_folder="." |
---|
| 479 | ): |
---|
| 480 | """ |
---|
| 481 | Plot stratification over time colored using RGB ternary mix of H2O ice (blue), CO2 ice (violet), and dust (orange). |
---|
| 482 | Includes a triangular legend showing the mix proportions. |
---|
| 483 | """ |
---|
| 484 | |
---|
| 485 | # Define constant colors |
---|
| 486 | violet = np.array([255, 0, 255], dtype=float) / 255 |
---|
| 487 | blue = np.array([0, 0, 255], dtype=float) / 255 |
---|
| 488 | orange = np.array([255, 165, 0], dtype=float) / 255 |
---|
| 489 | |
---|
| 490 | # Prepare elevation mask |
---|
| 491 | mask_elev = (ref_grid >= 0.0) if exclude_sub else np.ones_like(ref_grid, dtype=bool) |
---|
| 492 | elev = ref_grid[mask_elev] |
---|
| 493 | |
---|
| 494 | # Generate legend image once |
---|
| 495 | res = 300 |
---|
| 496 | u = np.linspace(0, 1, res) |
---|
| 497 | v = np.linspace(0, np.sqrt(3)/2, res) |
---|
| 498 | X, Y = np.meshgrid(u, v) |
---|
| 499 | V_bary = 2 * Y / np.sqrt(3) |
---|
| 500 | U_bary = X - 0.5 * V_bary |
---|
| 501 | W_bary = 1 - U_bary - V_bary |
---|
| 502 | mask_triangle = (U_bary >= 0) & (V_bary >= 0) & (W_bary >= 0) |
---|
| 503 | |
---|
| 504 | legend_rgb = ( |
---|
| 505 | U_bary[..., None] * violet |
---|
| 506 | + V_bary[..., None] * orange |
---|
| 507 | + W_bary[..., None] * blue |
---|
| 508 | ) |
---|
| 509 | legend_rgb = np.clip(legend_rgb, 0.0, 1.0) |
---|
| 510 | legend_rgba = np.zeros((res, res, 4)) |
---|
| 511 | legend_rgba[..., :3] = legend_rgb |
---|
| 512 | legend_rgba[..., 3] = mask_triangle.astype(float) |
---|
| 513 | |
---|
| 514 | # Loop over grid and slope |
---|
| 515 | h2o = gridded_data['h2o_ice'] |
---|
| 516 | co2 = gridded_data['co2_ice'] |
---|
| 517 | dust = gridded_data['dust'] |
---|
| 518 | ngrid, ntime, nslope, nz = h2o.shape |
---|
| 519 | |
---|
| 520 | for ig in range(ngrid): |
---|
| 521 | for isl in range(nslope): |
---|
| 522 | # Compute RGB stratification over time |
---|
| 523 | rgb = np.ones((nz, ntime, 3), dtype=float) |
---|
| 524 | for t in range(ntime): |
---|
| 525 | mask_z = np.arange(nz) < top_index[ig, t, isl] |
---|
| 526 | if not mask_z.any(): |
---|
| 527 | continue |
---|
| 528 | cH2O = np.clip(h2o[ig, t, isl, mask_z], 0, None) |
---|
| 529 | cCO2 = np.clip(co2[ig, t, isl, mask_z], 0, None) |
---|
| 530 | cDust = np.clip(dust[ig, t, isl, mask_z], 0, None) |
---|
| 531 | total = cH2O + cCO2 + cDust |
---|
| 532 | total[total == 0] = 1.0 |
---|
| 533 | fH2O = cH2O / total |
---|
| 534 | fCO2 = cCO2 / total |
---|
| 535 | fDust = cDust / total |
---|
| 536 | mix = ( |
---|
| 537 | np.outer(fH2O, blue) |
---|
| 538 | + np.outer(fCO2, violet) |
---|
| 539 | + np.outer(fDust, orange) |
---|
| 540 | ) |
---|
| 541 | mix = np.clip(mix, 0.0, 1.0) |
---|
| 542 | rgb[mask_z, t, :] = mix |
---|
| 543 | |
---|
| 544 | display_rgb = rgb[mask_elev, :, :] |
---|
| 545 | |
---|
| 546 | # Create figure with legend |
---|
| 547 | fig, (ax_main, ax_leg) = plt.subplots( |
---|
| 548 | 1, 2, figsize=(12, 5), dpi=200, |
---|
| 549 | gridspec_kw={'width_ratios': [5, 1]} |
---|
| 550 | ) |
---|
| 551 | |
---|
| 552 | # Main stratification panel |
---|
| 553 | ax_main.imshow( |
---|
| 554 | display_rgb, |
---|
| 555 | aspect='auto', |
---|
| 556 | extent=[date_time[0], date_time[-1], elev.min(), elev.max()], |
---|
| 557 | interpolation='nearest', |
---|
| 558 | origin='lower' |
---|
| 559 | ) |
---|
| 560 | ax_main.set_facecolor('white') |
---|
| 561 | ax_main.set_title(f"Ternary mix over time (Grid point {ig+1}, Slope {isl+1})", fontweight='bold') |
---|
| 562 | ax_main.set_xlabel("Time (Mars years)") |
---|
| 563 | ax_main.set_ylabel("Elevation (m)") |
---|
| 564 | |
---|
| 565 | # Legend panel |
---|
| 566 | ax_leg.imshow( |
---|
| 567 | legend_rgba, |
---|
| 568 | extent=[0, 1, 0, np.sqrt(3)/2], |
---|
| 569 | origin='lower', |
---|
| 570 | interpolation='nearest' |
---|
| 571 | ) |
---|
| 572 | |
---|
| 573 | # Draw triangle border |
---|
| 574 | triangle = np.array([[0, 0], [1, 0], [0.5, np.sqrt(3)/2], [0, 0]]) |
---|
| 575 | ax_leg.plot(triangle[:, 0], triangle[:, 1], 'k-', linewidth=1) |
---|
| 576 | |
---|
| 577 | # Dashed gridlines |
---|
| 578 | ticks = np.linspace(0.25, 0.75, 3) |
---|
| 579 | for f in ticks: |
---|
| 580 | ax_leg.plot([1 - f, 0.5 * (1 - f)], [0, (1 - f)*np.sqrt(3)/2], '--', color='k', linewidth=0.5) |
---|
| 581 | ax_leg.plot([f, f + 0.5 * (1 - f)], [0, (1 - f)*np.sqrt(3)/2], '--', color='k', linewidth=0.5) |
---|
| 582 | y = (np.sqrt(3)/2) * f |
---|
| 583 | ax_leg.plot([0.5 * f, 1 - 0.5 * f], [y, y], '--', color='k', linewidth=0.5) |
---|
| 584 | |
---|
| 585 | # Legend labels |
---|
| 586 | ax_leg.text(0, -0.05, 'H2O ice', ha='center', va='top', fontsize=8) |
---|
| 587 | ax_leg.text(1, -0.05, 'CO2 ice', ha='center', va='top', fontsize=8) |
---|
| 588 | ax_leg.text(0.5, np.sqrt(3)/2 + 0.05, 'Dust', ha='center', va='bottom', fontsize=8) |
---|
| 589 | ax_leg.axis('off') |
---|
| 590 | |
---|
| 591 | plt.tight_layout() |
---|
| 592 | |
---|
| 593 | # Save figure |
---|
| 594 | fname = os.path.join(output_folder, f"layering_rgb_evolution_ig{ig+1}_is{isl+1}.png") |
---|
| 595 | fig.savefig(fname, dpi=150, bbox_inches='tight') |
---|
| 596 | |
---|
| 597 | |
---|
| 598 | def plot_dust_to_ice_ratio_over_time( |
---|
| 599 | gridded_data, |
---|
| 600 | ref_grid, |
---|
| 601 | top_index, |
---|
| 602 | heights_data, |
---|
| 603 | date_time, |
---|
| 604 | exclude_sub=False, |
---|
| 605 | output_folder="." |
---|
| 606 | ): |
---|
| 607 | """ |
---|
| 608 | Plot the dust-to-ice ratio in the stratification over time, |
---|
| 609 | using a blue-to-orange colormap: |
---|
| 610 | - blue: ice-dominated (low dust-to-ice ratio) |
---|
| 611 | - orange: dust-dominated (high dust-to-ice ratio) |
---|
| 612 | """ |
---|
| 613 | h2o = gridded_data['h2o_ice'] |
---|
| 614 | dust = gridded_data['dust'] |
---|
| 615 | ngrid, ntime, nslope, nz = h2o.shape |
---|
| 616 | |
---|
| 617 | # Elevation mask |
---|
| 618 | if exclude_sub: |
---|
| 619 | elevation_mask = (ref_grid >= 0.0) |
---|
| 620 | elev = ref_grid[elevation_mask] |
---|
| 621 | else: |
---|
| 622 | elevation_mask = np.ones_like(ref_grid, dtype=bool) |
---|
| 623 | elev = ref_grid |
---|
| 624 | |
---|
| 625 | # Define custom blue-to-orange colormap |
---|
| 626 | blue = np.array([0, 0, 255], dtype=float) / 255 |
---|
| 627 | orange = np.array([255, 165, 0], dtype=float) / 255 |
---|
| 628 | custom_cmap = LinearSegmentedColormap.from_list('BlueOrange', [blue, orange], N=256) |
---|
| 629 | |
---|
| 630 | # Log‑ratio bounds and small epsilon to avoid log(0) |
---|
| 631 | vmin, vmax = -2, 1 |
---|
| 632 | epsilon = 1e-6 |
---|
| 633 | |
---|
| 634 | # Loop over grids and slopes |
---|
| 635 | for ig in range(ngrid): |
---|
| 636 | for isl in range(nslope): |
---|
| 637 | log_ratio_array = np.full((nz, ntime), np.nan, dtype=np.float32) |
---|
| 638 | |
---|
| 639 | # Compute log10(dust/ice) profile at each time step |
---|
| 640 | for t in range(ntime): |
---|
| 641 | zmax = top_index[ig, t, isl] |
---|
| 642 | if zmax <= 0: |
---|
| 643 | continue |
---|
| 644 | |
---|
| 645 | h2o_profile = np.clip(h2o[ig, t, isl, :zmax], 0, None) |
---|
| 646 | dust_profile = np.clip(dust[ig, t, isl, :zmax], 0, None) |
---|
| 647 | |
---|
| 648 | with np.errstate(divide='ignore', invalid='ignore'): |
---|
| 649 | ratio_profile = np.where( |
---|
| 650 | h2o_profile > 0, |
---|
| 651 | dust_profile / h2o_profile, |
---|
| 652 | 10**(vmax + 1) |
---|
| 653 | ) |
---|
| 654 | log_ratio = np.log10(ratio_profile + epsilon) |
---|
| 655 | log_ratio = np.clip(log_ratio, vmin, vmax) |
---|
| 656 | |
---|
| 657 | log_ratio_array[:zmax, t] = log_ratio |
---|
| 658 | |
---|
| 659 | # Convert back to linear ratio and apply elevation mask |
---|
| 660 | ratio_array = 10**log_ratio_array |
---|
| 661 | ratio_display = ratio_array[elevation_mask, :] |
---|
| 662 | |
---|
| 663 | # Plot |
---|
| 664 | fig, ax = plt.subplots(figsize=(8, 6), dpi=150) |
---|
| 665 | im = ax.imshow( |
---|
| 666 | ratio_display, |
---|
| 667 | aspect='auto', |
---|
| 668 | extent=[date_time[0], date_time[-1], elev.min(), elev.max()], |
---|
| 669 | origin='lower', |
---|
| 670 | interpolation='nearest', |
---|
| 671 | cmap='managua_r', |
---|
| 672 | norm=LogNorm(vmin=10**vmin, vmax=10**vmax) |
---|
| 673 | ) |
---|
| 674 | |
---|
| 675 | # Add colorbar with simplified ratio labels |
---|
| 676 | cbar = fig.colorbar(im, ax=ax, orientation='vertical') |
---|
| 677 | cbar.set_label('Dust / H₂O ice (ratio)') |
---|
| 678 | |
---|
| 679 | # Define custom ticks and labels |
---|
| 680 | ticks = [1e-2, 1e-1, 1, 1e1] |
---|
| 681 | labels = ['1:100', '1:10', '1:1', '10:1'] |
---|
| 682 | cbar.set_ticks(ticks) |
---|
| 683 | cbar.set_ticklabels(labels) |
---|
| 684 | |
---|
| 685 | # Save figure |
---|
| 686 | plt.tight_layout() |
---|
| 687 | fname = os.path.join( |
---|
| 688 | output_folder, |
---|
| 689 | f"dust_to_ice_ratio_grid{ig+1}_slope{isl+1}.png" |
---|
| 690 | ) |
---|
| 691 | fig.savefig(fname, dpi=150) |
---|
| 692 | |
---|
| 693 | |
---|
[3809] | 694 | def plot_strata_count_and_total_height(heights_data, date_time, output_folder="."): |
---|
| 695 | """ |
---|
| 696 | For each grid point and slope, plot: |
---|
| 697 | - Number of strata vs time |
---|
| 698 | - Total deposit height vs time |
---|
| 699 | """ |
---|
| 700 | ntime = len(heights_data) |
---|
| 701 | nslope = len(heights_data[0]) |
---|
| 702 | ngrid = heights_data[0][0].shape[0] |
---|
| 703 | |
---|
| 704 | for ig in range(ngrid): |
---|
| 705 | for isl in range(nslope): |
---|
| 706 | n_strata_t = np.zeros(ntime, dtype=int) |
---|
| 707 | total_height_t = np.zeros(ntime, dtype=float) |
---|
| 708 | |
---|
| 709 | for t_idx in range(ntime): |
---|
| 710 | h_mat = heights_data[t_idx][isl] |
---|
| 711 | raw_h = h_mat[ig, :] |
---|
| 712 | valid_mask = (~np.isnan(raw_h)) & (raw_h != 0.0) |
---|
| 713 | if np.any(valid_mask): |
---|
| 714 | h_valid = raw_h[valid_mask] |
---|
| 715 | n_strata_t[t_idx] = h_valid.size |
---|
| 716 | total_height_t[t_idx] = np.max(h_valid) |
---|
| 717 | |
---|
| 718 | fig, axes = plt.subplots(2, 1, figsize=(8, 6), sharex=True) |
---|
| 719 | fig.suptitle( |
---|
| 720 | f"Strata count & total height over time for (Grid point {ig+1}, Slope {isl+1})", |
---|
[3819] | 721 | fontsize=14, |
---|
| 722 | fontweight='bold' |
---|
[3809] | 723 | ) |
---|
| 724 | |
---|
| 725 | axes[0].plot(date_time, n_strata_t, marker='+', linestyle='-') |
---|
| 726 | axes[0].set_ylabel("Number of strata") |
---|
| 727 | axes[0].grid(True) |
---|
| 728 | |
---|
| 729 | axes[1].plot(date_time, total_height_t, marker='+', linestyle='-') |
---|
| 730 | axes[1].set_xlabel("Time (Mars years)") |
---|
| 731 | axes[1].set_ylabel("Total height (m)") |
---|
| 732 | axes[1].grid(True) |
---|
| 733 | |
---|
| 734 | fig.tight_layout(rect=[0, 0, 1, 0.95]) |
---|
| 735 | fname = os.path.join( |
---|
| 736 | output_folder, f"strata_count_height_ig{ig+1}_is{isl+1}.png" |
---|
| 737 | ) |
---|
| 738 | fig.savefig(fname, dpi=150) |
---|
| 739 | |
---|
| 740 | |
---|
| 741 | def read_orbital_data(orb_file, martian_to_earth): |
---|
| 742 | """ |
---|
| 743 | Read the .asc file containing obliquity, eccentricity and Ls p. |
---|
| 744 | Columns: |
---|
| 745 | 0 = time in thousand Martian years |
---|
| 746 | 1 = obliquity (deg) |
---|
| 747 | 2 = eccentricity |
---|
| 748 | 3 = Ls p (deg) |
---|
| 749 | Converts times to Earth years. |
---|
| 750 | """ |
---|
| 751 | data = np.loadtxt(orb_file) |
---|
| 752 | dates_mka = data[:, 0] |
---|
| 753 | dates_yr = dates_mka * 1e3 / martian_to_earth |
---|
| 754 | obliquity = data[:, 1] |
---|
| 755 | eccentricity = data[:, 2] |
---|
| 756 | lsp = data[:, 3] |
---|
| 757 | return dates_yr, obliquity, eccentricity, lsp |
---|
| 758 | |
---|
| 759 | |
---|
| 760 | def plot_orbital_parameters(infofile, orb_file, date_time, output_folder="."): |
---|
| 761 | """ |
---|
| 762 | Plot the evolution of obliquity, eccentricity and Ls p |
---|
| 763 | versus simulated time. |
---|
| 764 | """ |
---|
| 765 | # Read conversion factor from infofile |
---|
| 766 | _, martian_to_earth = read_infofile(infofile) |
---|
| 767 | |
---|
| 768 | # Read orbital data |
---|
| 769 | dates_yr, obl, ecc, lsp = read_orbital_data(orb_file, martian_to_earth) |
---|
| 770 | |
---|
| 771 | # Interpolate orbital parameters at simulation dates (date_time) |
---|
| 772 | obl_interp = interp1d(dates_yr, obl, kind='linear', bounds_error=False, fill_value="extrapolate")(date_time) |
---|
| 773 | ecc_interp = interp1d(dates_yr, ecc, kind='linear', bounds_error=False, fill_value="extrapolate")(date_time) |
---|
| 774 | lsp_interp = interp1d(dates_yr, lsp, kind='linear', bounds_error=False, fill_value="extrapolate")(date_time) |
---|
| 775 | |
---|
| 776 | # Plot |
---|
| 777 | fig, axes = plt.subplots(3, 1, figsize=(8, 10), sharex=True) |
---|
[3819] | 778 | fig.suptitle("Orbital Parameters vs Simulated Time", fontsize=14, fontweight='bold') |
---|
[3809] | 779 | |
---|
| 780 | axes[0].plot(date_time, obl_interp, 'r+', linestyle='-') |
---|
| 781 | axes[0].set_ylabel("Obliquity (°)") |
---|
| 782 | axes[0].grid(True) |
---|
| 783 | |
---|
| 784 | axes[1].plot(date_time, ecc_interp, 'b+', linestyle='-') |
---|
| 785 | axes[1].set_ylabel("Eccentricity") |
---|
| 786 | axes[1].grid(True) |
---|
| 787 | |
---|
| 788 | axes[2].plot(date_time, lsp_interp, 'g+', linestyle='-') |
---|
| 789 | axes[2].set_ylabel("Ls p (°)") |
---|
| 790 | axes[2].set_xlabel("Time (Mars years)") |
---|
| 791 | axes[2].grid(True) |
---|
| 792 | |
---|
| 793 | plt.tight_layout(rect=[0, 0, 1, 0.96]) |
---|
| 794 | fname = os.path.join(output_folder, "orbital_parameters.png") |
---|
| 795 | fig.savefig(fname, dpi=150) |
---|
| 796 | |
---|
| 797 | |
---|
[3777] | 798 | def main(): |
---|
| 799 | # 1) Get user inputs |
---|
[3809] | 800 | folder_path, base_name, infofile, orbfile = get_user_inputs() |
---|
[3777] | 801 | |
---|
| 802 | # 2) List and verify NetCDF files |
---|
| 803 | files = list_netcdf_files(folder_path, base_name) |
---|
| 804 | if not files: |
---|
[3809] | 805 | print(f"No NetCDF files named \"{base_name}#.nc\" found in \"{folder_path}\".") |
---|
[3777] | 806 | sys.exit(1) |
---|
[3809] | 807 | print(f"> Found {len(files)} NetCDF file(s).") |
---|
[3777] | 808 | |
---|
[3809] | 809 | # 3) Open one sample to get grid dimensions & coordinates |
---|
[3777] | 810 | sample_file = files[0] |
---|
| 811 | ngrid, nslope, longitude, latitude = open_sample_dataset(sample_file) |
---|
[3809] | 812 | print(f"> ngrid = {ngrid}, nslope = {nslope}") |
---|
[3777] | 813 | |
---|
[3809] | 814 | # 4) Collect variable info + global min/max elevations |
---|
| 815 | var_info, max_nb_str, min_base_elev, max_top_elev = collect_stratification_variables(files, base_name) |
---|
| 816 | print(f"> max strata per slope = {max_nb_str}") |
---|
| 817 | print(f"> min base elev = {min_base_elev:.3f} m, max top elev = {max_top_elev:.3f} m") |
---|
[3777] | 818 | |
---|
[3809] | 819 | # 5) Load full datasets |
---|
[3777] | 820 | datasets = load_full_datasets(files) |
---|
| 821 | |
---|
[3809] | 822 | # 6) Extract stratification data |
---|
| 823 | heights_data, raw_prop_arrays, ntime = extract_stratification_data(datasets, var_info, ngrid, nslope, max_nb_str) |
---|
[3777] | 824 | |
---|
[3809] | 825 | # 7) Close datasets |
---|
[3777] | 826 | for ds in datasets: |
---|
| 827 | ds.close() |
---|
| 828 | |
---|
[3809] | 829 | # 8) Normalize to fractions |
---|
[3777] | 830 | frac_arrays = normalize_to_fractions(raw_prop_arrays) |
---|
| 831 | |
---|
[3809] | 832 | # 9) Ask whether to include subsurface |
---|
[3777] | 833 | show_subsurface = get_yes_no_input("Show subsurface layers?") |
---|
| 834 | exclude_sub = not show_subsurface |
---|
| 835 | if exclude_sub: |
---|
| 836 | min_base_for_interp = 0.0 |
---|
[3809] | 837 | print("> Interpolating only elevations >= 0 m (surface strata).") |
---|
[3777] | 838 | else: |
---|
| 839 | min_base_for_interp = min_base_elev |
---|
[3809] | 840 | print(f"> Interpolating full depth down to {min_base_elev:.3f} m.") |
---|
[3777] | 841 | |
---|
[3809] | 842 | # 10) Prompt discretization step |
---|
[3777] | 843 | dz = prompt_discretization_step(max_top_elev) |
---|
| 844 | |
---|
[3809] | 845 | # 11) Build reference grid and interpolate |
---|
[3777] | 846 | ref_grid, gridded_data, top_index = interpolate_data_on_refgrid( |
---|
[3809] | 847 | heights_data, frac_arrays, min_base_for_interp, max_top_elev, dz, exclude_sub=exclude_sub |
---|
[3777] | 848 | ) |
---|
| 849 | |
---|
[3809] | 850 | # 12) Read timestamps and conversion factor from infofile |
---|
| 851 | date_time, martian_to_earth = read_infofile(infofile) |
---|
[3777] | 852 | if date_time.size != ntime: |
---|
[3809] | 853 | print(f"Warning: {date_time.size} timestamps vs {ntime} NetCDF files.") |
---|
[3777] | 854 | |
---|
[3819] | 855 | # 13) Plot stratification data over time |
---|
[3777] | 856 | plot_stratification_over_time( |
---|
[3809] | 857 | gridded_data, ref_grid, top_index, heights_data, date_time, |
---|
| 858 | exclude_sub=exclude_sub, output_folder="." |
---|
[3777] | 859 | ) |
---|
[3819] | 860 | plot_stratification_rgb_over_time( |
---|
| 861 | gridded_data, ref_grid, top_index, heights_data, date_time, |
---|
| 862 | exclude_sub=exclude_sub, output_folder="." |
---|
| 863 | ) |
---|
| 864 | plot_dust_to_ice_ratio_over_time( |
---|
| 865 | gridded_data, ref_grid, top_index, heights_data, date_time, |
---|
| 866 | exclude_sub=exclude_sub, output_folder="." |
---|
| 867 | ) |
---|
[3809] | 868 | plot_strata_count_and_total_height(heights_data, date_time, output_folder=".") |
---|
[3777] | 869 | |
---|
[3819] | 870 | # 14) Plot orbital parameters |
---|
[3809] | 871 | plot_orbital_parameters(infofile, orbfile, date_time, output_folder=".") |
---|
| 872 | |
---|
[3819] | 873 | # 15) Show all figures |
---|
[3809] | 874 | plt.show() |
---|
| 875 | |
---|
| 876 | |
---|
[3777] | 877 | if __name__ == "__main__": |
---|
| 878 | main() |
---|
| 879 | |
---|