BERATools 0.2.0__py3-none-any.whl → 0.2.2__py3-none-any.whl

beratools/core/algo_common.py

@@ -0,0 +1,497 @@
+"""
+Copyright (C) 2025 Applied Geospatial Research Group.
+
+This script is licensed under the GNU General Public License v3.0.
+See <https://gnu.org/licenses/gpl-3.0> for full license details.
+
+Author: Richard Zeng
+
+Description:
+    This script is part of the BERA Tools.
+    Webpage: https://github.com/appliedgrg/beratools
+
+    The purpose of this script is to provide common algorithms
+    and utility functions/classes.
+"""
+import math
+import tempfile
+from pathlib import Path
+
+import geopandas as gpd
+import numpy as np
+import pyproj
+import rasterio
+import shapely
+import shapely.affinity as sh_aff
+import shapely.geometry as sh_geom
+import shapely.ops as sh_ops
+import skimage.graph as sk_graph
+from osgeo import gdal
+from scipy import ndimage
+
+import beratools.core.algo_cost as algo_cost
+import beratools.core.constants as bt_const
+
+DISTANCE_THRESHOLD = 2  # 1 meter for intersection neighborhood
+
+def process_single_item(cls_obj):
+    """
+    Process a class object for universal multiprocessing.
+
+    Args:
+        cls_obj: Class object to be processed
+
+    Returns:
+        cls_obj: Class object after processing
+
+    """
+    cls_obj.compute()
+    return cls_obj
+
+def read_geospatial_file(file_path, layer=None):
+    """
+    Read a geospatial file, clean the geometries and return a GeoDataFrame.
+
+    Args:
+        file_path (str): The path to the geospatial file (e.g., .shp, .gpkg).
+        layer (str, optional): The specific layer to read if the file is
+        multi-layered (e.g., GeoPackage).
+
+    Returns:
+        GeoDataFrame: The cleaned GeoDataFrame containing the data from the file
+        with valid geometries only.
+        None: If there is an error reading the file or layer.
+
+    """
+    try:
+        if layer is None:
+            # Read the file without specifying a layer
+            gdf = gpd.read_file(file_path)
+        else:
+            # Read the file with the specified layer
+            gdf = gpd.read_file(file_path, layer=layer)
+
+        # Clean the geometries in the GeoDataFrame
+        gdf = clean_geometries(gdf)
+        gdf["BT_UID"] = range(len(gdf))  # assign temporary UID
+        return gdf
+
+    except Exception as e:
+        print(f"Error reading file {file_path}: {e}")
+        return None
+
+def has_multilinestring(gdf):
+    """Check if any geometry is a MultiLineString."""
+    # Filter out None values (invalid geometries) from the GeoDataFrame
+    valid_geometries = gdf.geometry
+    return any(isinstance(geom, sh_geom.MultiLineString) for geom in valid_geometries)
+
+def clean_geometries(gdf):
+    """
+    Remove rows with invalid, None, or empty geometries from the GeoDataFrame.
+
+    Args:
+        gdf (GeoDataFrame): The GeoDataFrame to clean.
+
+    Returns:
+        GeoDataFrame: The cleaned GeoDataFrame with valid, non-null,
+        and non-empty geometries.
+
+    """
+    # Remove rows where the geometry is invalid, None, or empty
+    gdf = gdf[gdf.geometry.is_valid]  # Only keep valid geometries
+    gdf = gdf[~gdf.geometry.isna()]  # Remove rows with None geometries
+    gdf = gdf[
+        gdf.geometry.apply(lambda geom: not geom.is_empty)
+    ]  # Remove empty geometries
+    return gdf
+
+def clean_line_geometries(line_gdf):
+    """Clean line geometries in the GeoDataFrame."""
+    if line_gdf is None:
+        return line_gdf
+    
+    if line_gdf.empty:
+        return line_gdf
+    
+    line_gdf = line_gdf[
+        ~line_gdf.geometry.isna()
+        & ~line_gdf.geometry.is_empty
+    ]
+    line_gdf = line_gdf[line_gdf.geometry.length > bt_const.SMALL_BUFFER]
+    return line_gdf
+
+def prepare_lines_gdf(file_path, layer=None, proc_segments=True):
+    """
+    Split lines at vertices or return original rows.
+
+    It handles for MultiLineString.
+
+    """
+    # Check if there are any MultiLineString geometries
+    gdf = read_geospatial_file(file_path, layer=layer)
+
+    # Explode MultiLineStrings into individual LineStrings
+    if has_multilinestring(gdf):
+        gdf = gdf.explode(index_parts=False)
+
+    split_gdf_list = []
+
+    for row in gdf.itertuples(index=False):  # Use itertuples to iterate
+        line = row.geometry  # Access geometry directly via the named tuple
+
+        # If proc_segment is True, split the line at vertices
+        if proc_segments:
+            coords = list(line.coords)  # Extract the list of coordinates (vertices)
+
+            # For each LineString, split the line into segments by the vertices
+            for i in range(len(coords) - 1):
+                segment = sh_geom.LineString([coords[i], coords[i + 1]])
+
+                # Copy over all non-geometry columns (excluding 'geometry')
+                attributes = {
+                    col: getattr(row, col) for col in gdf.columns if col != "geometry"
+                }
+                single_row_gdf = gpd.GeoDataFrame(
+                    [attributes], geometry=[segment], crs=gdf.crs
+                )
+                split_gdf_list.append(single_row_gdf)
+
+        else:
+            # If not proc_segment, add the original row as a single-row GeoDataFrame
+            attributes = {
+                col: getattr(row, col) for col in gdf.columns if col != "geometry"
+            }
+            single_row_gdf = gpd.GeoDataFrame(
+                [attributes], geometry=[line], crs=gdf.crs
+            )
+            split_gdf_list.append(single_row_gdf)
+
+    return split_gdf_list
+
+
+# TODO use function from common
+def morph_raster(corridor_thresh, canopy_raster, exp_shk_cell, cell_size_x):
+    # Process: Stamp CC and Max Line Width
+    temp1 = corridor_thresh + canopy_raster
+    raster_class = np.ma.where(temp1 == 0, 1, 0).data
+
+    if exp_shk_cell > 0 and cell_size_x < 1:
+        # Process: Expand
+        # FLM original Expand equivalent
+        cell_size = int(exp_shk_cell * 2 + 1)
+        expanded = ndimage.grey_dilation(
+            raster_class, size=(cell_size, cell_size)
+        )
+
+        # Process: Shrink
+        # FLM original Shrink equivalent
+        file_shrink = ndimage.grey_erosion(
+            expanded, size=(cell_size, cell_size)
+        )
+
+    else:
+        if bt_const.BT_DEBUGGING:
+            print("No Expand And Shrink cell performed.")
+        file_shrink = raster_class
+
+    # Process: Boundary Clean
+    clean_raster = ndimage.gaussian_filter(file_shrink, sigma=0, mode="nearest")
+
+    return clean_raster
+
+
+def closest_point_to_line(point, line):
+    if not line:
+        return None
+
+    pt = line.interpolate(line.project(sh_geom.Point(point)))
+    return pt
+
+
+def line_coord_list(line):
+    point_list = []
+    try:
+        for point in list(line.coords):  # loops through every point in a line
+            # loops through every vertex of every segment
+            if point:  # adds all the vertices to segment_list, which creates an array
+                point_list.append(sh_geom.Point(point[0], point[1]))
+    except Exception as e:
+        print(e)
+
+    return point_list
+
+
+def intersection_of_lines(line_1, line_2):
+    """
+    Only LINESTRING is dealt with for now.
+
+    Args:
+    line_1 :
+    line_2 :
+
+    Returns:
+    sh_geom.Point: intersection point
+
+    """
+    # intersection collection, may contain points and lines
+    inter = None
+    if line_1 and line_2:
+        inter = line_1.intersection(line_2)
+
+    # TODO: intersection may return GeometryCollection, LineString or MultiLineString
+    if inter:
+        if (
+            type(inter) is sh_geom.GeometryCollection
+            or type(inter) is sh_geom.LineString
+            or type(inter) is sh_geom.MultiLineString
+        ):
+            return inter.centroid
+
+    return inter
+
+def get_angle(line, vertex_index):
+    """
+    Calculate the angle of the first or last segment.
+
+    # TODO: use np.arctan2 instead of np.arctan
+
+    Args:
+    line: LineString
+    end_index: 0 or -1 of the line vertices. Consider the multipart.
+
+    """
+    pts = line_coord_list(line)
+
+    if vertex_index == 0:
+        pt_1 = pts[0]
+        pt_2 = pts[1]
+    elif vertex_index == -1:
+        pt_1 = pts[-1]
+        pt_2 = pts[-2]
+
+    delta_x = pt_2.x - pt_1.x
+    delta_y = pt_2.y - pt_1.y
+    if np.isclose(pt_1.x, pt_2.x):
+        angle = np.pi / 2
+        if delta_y > 0:
+            angle = np.pi / 2
+        elif delta_y < 0:
+            angle = -np.pi / 2
+    else:
+        angle = np.arctan(delta_y / delta_x)
+
+        # arctan is in range [-pi/2, pi/2], regulate all angles to [[-pi/2, 3*pi/2]]
+        if delta_x < 0:
+            angle += np.pi  # the second or fourth quadrant
+
+    return angle
+
+def points_are_close(pt1, pt2):
+    if (
+        abs(pt1.x - pt2.x) < DISTANCE_THRESHOLD
+        and abs(pt1.y - pt2.y) < DISTANCE_THRESHOLD
+    ):
+        return True
+    else:
+        return False
+
+def generate_raster_footprint(in_raster, latlon=True):
+    inter_img = "image_overview.tif"
+
+    src_ds = gdal.Open(in_raster)
+    width, height = src_ds.RasterXSize, src_ds.RasterYSize
+    src_crs = src_ds.GetSpatialRef().ExportToWkt()
+
+    geom = None
+    with tempfile.TemporaryDirectory() as tmp_folder:
+        if bt_const.BT_DEBUGGING:
+            print("Temporary folder: {}".format(tmp_folder))
+
+        if max(width, height) <= 1024:
+            inter_img = in_raster
+        else:
+            if width >= height:
+                options = gdal.TranslateOptions(width=1024, height=0)
+            else:
+                options = gdal.TranslateOptions(width=0, height=1024)
+
+            inter_img = Path(tmp_folder).joinpath(inter_img).as_posix()
+            gdal.Translate(inter_img, src_ds, options=options)
+
+            shapes = gdal.Footprint(None, inter_img, dstSRS=src_crs, format="GeoJSON")
+            target_feat = shapes["features"][0]
+            geom = sh_geom.shape(target_feat["geometry"])
+
+    if latlon:
+        out_crs = pyproj.CRS("EPSG:4326")
+        transformer = pyproj.Transformer.from_crs(pyproj.CRS(src_crs), out_crs)
+
+        geom = sh_ops.transform(transformer.transform, geom)
+
+    return geom
+
+def save_raster_to_file(in_raster_mem, in_meta, out_raster_file):
+    """
+    Save raster matrix in memory to file.
+
+    Args:
+        in_raster_mem: numpy raster
+        in_meta: input meta
+        out_raster_file: output raster file
+
+    """
+    with rasterio.open(out_raster_file, "w", **in_meta) as dest:
+        dest.write(in_raster_mem, indexes=1)
+
+def generate_perpendicular_line_precise(points, offset=20):
+    """
+    Generate a perpendicular line to the input line at the given point.
+
+    Args:
+        points (list[Point]): The points where to generate the perpendicular lines.
+        offset (float): The length of the perpendicular line.
+
+    Returns:
+        shapely.geometry.LineString: The generated perpendicular line.
+
+    """
+    # Compute the angle of the line
+    if len(points) not in [2, 3]:
+        return None
+    
+    center = points[1]
+    perp_line = None
+
+    if len(points) == 2:
+        head = points[0]
+        tail = points[1]
+
+        delta_x = head.x - tail.x
+        delta_y = head.y - tail.y
+        angle = 0.0
+
+        if math.isclose(delta_x, 0.0):
+            angle = math.pi / 2
+        else:
+            angle = math.atan(delta_y / delta_x)
+
+        start = [center.x + offset / 2.0, center.y]
+        end = [center.x - offset / 2.0, center.y]
+        line = sh_geom.LineString([start, end])
+        perp_line = sh_aff.rotate(
+            line, angle + math.pi / 2.0, origin=center, use_radians=True
+        )
+    elif len(points) == 3:
+        head = points[0]
+        tail = points[2]
+
+        angle_1 = _line_angle(center, head)
+        angle_2 = _line_angle(center, tail)
+        angle_diff = (angle_2 - angle_1) / 2.0
+        head_new = sh_geom.Point(
+            center.x + offset / 2.0 * math.cos(angle_1),
+            center.y + offset / 2.0 * math.sin(angle_1),
+        )
+        if head.has_z:
+            head_new = shapely.force_3d(head_new)
+        try:
+            perp_seg_1 = sh_geom.LineString([center, head_new])
+            perp_seg_1 = sh_aff.rotate(
+                perp_seg_1, angle_diff, origin=center, use_radians=True
+            )
+            perp_seg_2 = sh_aff.rotate(
+                perp_seg_1, math.pi, origin=center, use_radians=True
+            )
+            perp_line = sh_geom.LineString(
+                [list(perp_seg_1.coords)[1], list(perp_seg_2.coords)[1]]
+            )
+        except Exception as e:
+            print(e)
+
+    return perp_line
+
+
+def _line_angle(point_1, point_2):
+    """
+    Calculate the angle of the line.
+
+    Args:
+        point_1, point_2: start and end points of shapely line
+
+    """
+    delta_y = point_2.y - point_1.y
+    delta_x = point_2.x - point_1.x
+
+    angle = math.atan2(delta_y, delta_x)
+    return angle
+
+def corridor_raster(
+    raster_clip, out_meta, source, destination, cell_size, corridor_threshold
+):
+    """
+    Calculate corridor raster.
+
+    Args:
+        raster_clip (raster):
+        out_meta : raster file meta
+        source (list of point tuple(s)): start point in row/col
+        destination (list of point tuple(s)): end point in row/col
+        cell_size (tuple): (cell_size_x, cell_size_y)
+        corridor_threshold (double)
+
+    Returns:
+    corridor raster
+    
+    """
+    try:
+        # change all nan to BT_NODATA_COST for workaround
+        if len(raster_clip.shape) > 2:
+            raster_clip = np.squeeze(raster_clip, axis=0)
+
+        algo_cost.remove_nan_from_array_refactor(raster_clip)
+
+        # generate the cost raster to source point
+        mcp_source = sk_graph.MCP_Geometric(raster_clip, sampling=cell_size)
+        source_cost_acc = mcp_source.find_costs(source)[0]
+        del mcp_source
+
+        # # # generate the cost raster to destination point
+        mcp_dest = sk_graph.MCP_Geometric(raster_clip, sampling=cell_size)
+        dest_cost_acc = mcp_dest.find_costs(destination)[0]
+
+        # Generate corridor
+        corridor = source_cost_acc + dest_cost_acc
+        corridor = np.ma.masked_invalid(corridor)
+
+        # Calculate minimum value of corridor raster
+        if np.ma.min(corridor) is not None:
+            corr_min = float(np.ma.min(corridor))
+        else:
+            corr_min = 0.5
+
+        # normalize corridor raster by deducting corr_min
+        corridor_norm = corridor - corr_min
+        corridor_thresh_cl = np.ma.where(corridor_norm >= corridor_threshold, 1.0, 0.0)
+
+    except Exception as e:
+        print(e)
+        print("corridor_raster: Exception occurred.")
+        return None
+
+    return corridor_thresh_cl
+
+def remove_holes(geom):
+    if geom.geom_type == "Polygon":
+        if geom.interiors:
+            return sh_geom.Polygon(geom.exterior)
+        return geom
+    elif geom.geom_type == "MultiPolygon":
+        new_polygons = []
+        for polygon in geom.geoms:  # Iterate through MultiPolygon
+            if polygon.interiors:
+                new_polygons.append(sh_geom.Polygon(polygon.exterior))
+            else:
+                new_polygons.append(polygon)
+        return sh_geom.MultiPolygon(new_polygons)
+    return geom  # Return other geometry types as is

beratools/core/algo_cost.py

@@ -0,0 +1,192 @@
+"""
+Copyright (C) 2025 Applied Geospatial Research Group.
+
+This script is licensed under the GNU General Public License v3.0.
+See <https://gnu.org/licenses/gpl-3.0> for full license details.
+
+Author: Richard Zeng
+
+Description:
+    This script is part of the BERA Tools.
+    Webpage: https://github.com/appliedgrg/beratools
+
+    This file hosts cost raster related functions.
+"""
+import numpy as np
+import scipy
+
+import beratools.core.constants as bt_const
+
+
+def cost_raster(
+    in_raster,
+    meta,
+    tree_radius=2.5,
+    canopy_ht_threshold=2.5,
+    max_line_dist=2.5,
+    canopy_avoid=0.4,
+    cost_raster_exponent=1.5,
+):
+    """
+    General version of cost_raster.
+
+    To be merged later: variables and consistent nodata solution
+
+    """
+    if len(in_raster.shape) > 2:
+        in_raster = np.squeeze(in_raster, axis=0)
+    
+    # regulate canopy_avoid between 0 and 1
+    avoidance = max(0, min(1, canopy_avoid))
+    cell_x, cell_y = meta["transform"][0], -meta["transform"][4]
+
+    kernel_radius = int(tree_radius / cell_x)
+    kernel = circle_kernel_refactor(2 * kernel_radius + 1, kernel_radius)
+    dyn_canopy_ndarray = dyn_np_cc_map(in_raster, canopy_ht_threshold)
+
+    cc_std, cc_mean = cost_focal_stats(dyn_canopy_ndarray, kernel)
+    cc_smooth = cost_norm_dist_transform(
+        dyn_canopy_ndarray, max_line_dist, [cell_x, cell_y]
+    )
+
+    cost_clip = dyn_np_cost_raster_refactor(
+        dyn_canopy_ndarray, cc_mean, cc_std, cc_smooth, avoidance, cost_raster_exponent
+    )
+
+    # TODO use nan or BT_DATA?
+    cost_clip[in_raster == bt_const.BT_NODATA] = np.nan
+    dyn_canopy_ndarray[in_raster == bt_const.BT_NODATA] = np.nan
+
+    return cost_clip, dyn_canopy_ndarray
+
+def remove_nan_from_array_refactor(matrix, replacement_value=bt_const.BT_NODATA_COST):
+    # Use boolean indexing to replace nan values
+    matrix[np.isnan(matrix)] = replacement_value
+
+def dyn_np_cc_map(in_chm, canopy_ht_threshold):
+    """
+    Create a new canopy raster.
+
+    MaskedArray based on the threshold comparison of in_chm (canopy height model) 
+    with canopy_ht_threshold. It assigns 1.0 where the condition is True (canopy) 
+    and 0.0 where the condition is False (non-canopy).
+
+    """
+    canopy_ndarray = np.ma.where(in_chm >= canopy_ht_threshold, 1.0, 0.0).astype(float)
+    return canopy_ndarray
+
+def cost_focal_stats(canopy_ndarray, kernel):
+    mask = canopy_ndarray.mask
+    in_ndarray = np.ma.where(mask, np.nan, canopy_ndarray)
+
+    # Function to compute mean and standard deviation
+    def calc_mean(arr):
+        # Check if the array is empty or full of NaNs
+        if arr.size == 0 or np.all(np.isnan(arr)):
+            return np.nan  # Or any other value you'd prefer for empty arrays
+        return np.nanmean(arr)
+
+    def calc_std(arr):
+        # Check if the array is empty or full of NaNs
+        if arr.size == 0 or np.all(np.isnan(arr)):
+            return np.nan  # Or any other placeholder you prefer
+        return np.nanstd(arr)
+
+    # Apply the generic_filter function to compute mean and std
+    mean_array = scipy.ndimage.generic_filter(
+        in_ndarray, calc_mean, footprint=kernel, mode="nearest"
+    )
+    std_array = scipy.ndimage.generic_filter(
+        in_ndarray, calc_std, footprint=kernel, mode="nearest"
+    )
+
+    return std_array, mean_array
+
+def cost_norm_dist_transform(canopy_ndarray, max_line_dist, sampling):
+    """Compute a distance-based cost map based on the proximity of valid data points."""
+    # Convert masked array to a regular array and fill the masked areas with np.nan
+    in_ndarray = canopy_ndarray.filled(np.nan)
+    
+    # Compute the Euclidean distance transform (edt) where the valid values are
+    euc_dist_array = scipy.ndimage.distance_transform_edt(
+        np.logical_not(np.isnan(in_ndarray)), sampling=sampling
+    )
+    
+    # Apply the mask back to set the distances to np.nan
+    euc_dist_array[canopy_ndarray.mask] = np.nan
+    
+    # Calculate the smoothness (cost) array
+    normalized_cost = float(max_line_dist) - euc_dist_array
+    normalized_cost[normalized_cost <= 0.0] = 0.0
+    smooth_cost_array = normalized_cost / float(max_line_dist)
+
+    return smooth_cost_array
+
+def dyn_np_cost_raster_refactor(
+    canopy_ndarray, cc_mean, cc_std, cc_smooth, avoidance, cost_raster_exponent
+):
+    # Calculate the lower and upper bounds for canopy cover (mean ± std deviation)
+    lower_bound = cc_mean - cc_std
+    upper_bound = cc_mean + cc_std
+
+    # Calculate the ratio between the lower and upper bounds
+    ratio_lower_upper = np.divide(
+        lower_bound,
+        upper_bound,
+        where=upper_bound != 0,
+        out=np.zeros(lower_bound.shape, dtype=float),
+    )
+
+    # Normalize the ratio to a scale between 0 and 1
+    normalized_ratio = (1 + ratio_lower_upper) / 2
+
+    # Adjust where the sum of mean and std deviation is less than or equal to zero
+    adjusted_cover = cc_mean + cc_std
+    adjusted_ratio = np.where(adjusted_cover <= 0, 0, normalized_ratio)
+
+    # Combine canopy cover ratio with smoothing, weighted by avoidance factor
+    weighted_cover = adjusted_ratio * (1 - avoidance) + (cc_smooth * avoidance)
+
+    # Final cost modification based on canopy presence (masked by canopy_ndarray)
+    final_cost = np.where(canopy_ndarray.data == 1, 1, weighted_cover)
+
+    # Apply the exponential transformation to the cost values
+    exponent_cost = np.exp(final_cost)
+
+    # Raise the cost to the specified exponent
+    result_cost_raster = np.power(exponent_cost, float(cost_raster_exponent))
+
+    return result_cost_raster
+
+def circle_kernel_refactor(size, radius):
+    """
+    Create a circular kernel using Scipy.
+
+    Args:
+    size : kernel size
+    radius : radius of the circle
+
+    Returns:
+    kernel (ndarray): A circular kernel.
+
+    Examples:
+    kernel_scipy = create_circle_kernel_scipy(17, 8)
+    will replicate xarray-spatial kernel
+    cell_x = 0.3
+    cell_y = 0.3
+    tree_radius = 2.5
+    convolution.circle_kernel(cell_x, cell_y, tree_radius)
+
+    """
+    # Create grid points (mesh)
+    y, x = np.ogrid[:size, :size]
+
+    # Center of the kernel
+    center_x, center_y = (size - 1) / 2, (size - 1) / 2
+
+    # Calculate the distance from the center
+    distance = np.sqrt((x - center_x) ** 2 + (y - center_y) ** 2)
+
+    # Create a circular kernel
+    kernel = distance <= radius
+    return kernel.astype(float)