invert4geom.Inversion.optimize_inversion_zref_density_contrast_windowed

Inversion.optimize_inversion_zref_density_contrast_windowed(n_trials, constraints_df, window_width, window_overlap=0.0, window_buffer=0.0, min_constraints=1, merge_method='spline', merge_damping=None, zref_limits=None, density_contrast_limits=None, starting_topography=None, starting_topography_kwargs=None, regional_grav_kwargs=None, fname=None, progressbar=True, **kwargs)

Find spatially variable optimal zref and / or density contrast values by running independent zref / density contrast optimizations within windows of the model region, and merging the per-window optimal values into grids used in a final inversion.

The model region is tiled with square windows of width window_width, optionally overlapping by a fraction window_overlap of their width. For each window, the gravity data, model, and constraint points are cropped to the window (padded by window_buffer to reduce edge effects) and a standard optimization (optimize_inversion_zref_density_contrast) is run to find the window’s optimal parameter values, scored with the constraint points falling inside the (unpadded) window. Windows containing fewer than min_constraints constraint points are skipped, as are windows where the optimization fails. The optimal values of all successful windows, located at the window centers, are then merged into continuous grids by interpolation (merge_method), clipped to the range of the per-window values. A final inversion is then run over the full region using these spatially variable zref and / or density contrast grids.

K-folds cross-validation (lists of constraint dataframes) is not supported; provide a single dataframe of constraint points.

Parameters:

• n_trials (int) – number of optimization trials to run within each window.

• constraints_df (DataFrame) – constraint points used for scoring, with columns for the two horizontal coordinates and “upward”.

• window_width (float) – width of the square windows, in meters (or degrees for tesseroid models).

• window_overlap (float) – fraction (0 to <1) of the window width which adjacent windows overlap by, by default 0.0

• window_buffer (float) – width of a buffer zone of gravity data included around each window to reduce edge effects, must be a multiple of the grid spacing. Constraint points within the buffer are not used for scoring, by default 0.0

• min_constraints (int) – minimum number of constraint points which must fall within a window for it to be included, by default 1

• merge_method (str) – method used to interpolate the per-window optimal values into grids, either “spline” (verde.Spline) or “nearest” (verde.KNeighbors), by default “spline”

• merge_damping (float | None) – damping parameter for the merging spline, by default None

• zref_limits (tuple[float, float] | None) – upper and lower limits for the reference level, in meters, by default None

• density_contrast_limits (tuple[float, float] | None) – upper and lower limits for the density contrast, in kg/m^-3, by default None

• starting_topography (Dataset | None) – a starting topography model, cropped to each window, by default None

• starting_topography_kwargs (dict[str, Any] | None) – kwargs passed to utils.create_topography, same as in optimize_inversion_zref_density_contrast, by default None

• regional_grav_kwargs (dict[str, Any] | None) – kwargs passed to DatasetAccessorInvert4Geom.regional_separation. Any dataframes (e.g. training constraints) are automatically cropped to each window, by default None

• fname (str | None) – root file name for saving results, each window’s results are saved to <fname>_window_<i> and the final inversion to <fname>.pickle, by default a random name.

• progressbar (bool) – show a progress bar over the windows, by default True

• **kwargs (Any) – additional kwargs passed to each window’s optimize_inversion_zref_density_contrast call.

Returns:

Inversion object with the final inversion results, the merged zref and density contrast grids as attributes of the model attribute, and the per-window results in the windowed_optimization_df attribute.

Return type:

Inversion