tools/gis_utils.py

import xarray as xr
import numpy as np
from shapely.geometry import Point
from shapely.ops import transform
import pyproj
from typing import  Dict
from pathlib import Path
import re
from shapely.wkt import loads
import json
import pandas as pd
import geopandas as gpd


def _load_field_mapping(consortium, consortium_folder_name: str, sensors_df: pd.DataFrame) -> Dict:
    """

        Load field geometries from geojson files in data/01-raw/{consortium_folder_name}/field_shapes.

        Each file must be named {field_name}_irri.geojson.

        The mapping is done if field_name (lowercase) appears inside datastream_name.lower().

        Returns {sensor_name: geometry_wkt}.

    """

    base_dir = Path(f"data/01_raw/{consortium_folder_name}/fields_shapes")
    if not base_dir.exists():
        raise FileNotFoundError(f"Folder not found: {base_dir}")

    result = {}

    # lowercase names for matching
    sensor_names = sensors_df["datastream_name"].tolist()
    sensor_names_lower = [s.lower() for s in sensor_names]

    # iterate over all geojson field files
    for geojson_path in base_dir.glob("*.geojson"):

        stem = geojson_path.stem.lower()

        # different extraction rules per consortium
        if consortium == "consortium0":
            field_name = stem.replace("_irri", "")
        elif consortium == "consortium1":
            field_name = stem  # use full stem
        else:
            field_name = stem  # fallback

        gdf = gpd.read_file(geojson_path)
        if gdf.empty:
            continue

        geom_wkt = gdf.geometry.iloc[0].wkt

        for sensor_name, sensor_lower in zip(sensor_names, sensor_names_lower):
            if field_name in sensor_lower:  # this now works for both cases
                result[sensor_name] = geom_wkt

    return result

def _infer_dataset_crs(ds: xr.Dataset) -> str:
    """Infer CRS from dataset coordinates."""
    # check for common CRS attributes
    if 'crs' in ds.attrs:
        return ds.attrs['crs']

    # check coordinate ranges to infer CRS
    x_range = float(ds.x.max() - ds.x.min())
    y_range = float(ds.y.max() - ds.y.min())

    # if ranges are small (< 1), likely lat/lon (WGS84)
    if x_range < 1 and y_range < 1:
        return "EPSG:4326"

    # else, assume UTM zone 32N (common for Northern Italy/Trento area)
    return "EPSG:32632"


# todo: instead of closest maybe change to be contained/within the pixel only.
def _get_closest_pixel_indices(ds: xr.Dataset, lat: float, lon: float,

                               transformer: pyproj.Transformer) -> tuple:
    """Get indices of closest pixel to given location."""
    # transform coordinates
    x_proj, y_proj = transformer.transform(lon, lat)

    # find nearest pixel indices
    x_idx = int(np.argmin(np.abs(ds.x.values - x_proj)))
    y_idx = int(np.argmin(np.abs(ds.y.values - y_proj)))

    return y_idx, x_idx


def _get_field_pixel_indices(ds: xr.Dataset,

                             field_geom,

                             transformer: pyproj.Transformer) -> tuple:
    """Get indices of all pixels within field boundary."""

    # clean and make geometry work
    '''

    field_geom = __swap_coordinates(field_geom)

    print("Geometry to parse:", field_geom)'''

    # Load geometry from the swapped WKT string
    field_geom = loads(field_geom)

    # transform field geometry to dataset CRS
    field_proj = transform(transformer.transform, field_geom)

    # obtain pixel bounds from dataset
    y_coords = ds.y.values
    x_coords = ds.x.values

    # create meshgrid of pixel centers
    x_grid, y_grid = np.meshgrid(x_coords, y_coords)

    # check which pixels fall within the field
    y_indices = []
    x_indices = []

    for i in range(len(y_coords)):
        for j in range(len(x_coords)):
            pixel_point = Point(x_grid[i, j], y_grid[i, j])
            if field_proj.contains(pixel_point):
                y_indices.append(i)
                x_indices.append(j)

    return y_indices, x_indices


def __swap_coordinates(geometry):
    """

    Swaps x and y coordinates in a geometry string for both POLYGON and MULTIPOLYGON.

    Handles both outer rings and inner rings (holes) in polygons.



    Parameters:

    - geometry (str): The geometry string



    Returns:

    - str: The geometry string with swapped coordinates.

    """
    geometry = geometry.strip()
    geometry_type = geometry.split("(")[0].strip()

    def swap_coordinate_pairs(coord_string):
        """Helper function to swap coordinate pairs in a string."""
        coordinates = re.findall(r"([-+]?\d*\.\d+|\d+)\s([-+]?\d*\.\d+|\d+)", coord_string)
        return ",".join(f"{y} {x}" for x, y in coordinates)

    if geometry_type == 'POLYGON':
        # handle single polygon with possible inner rings
        rings = re.findall(r"\(([-+\d\s.,]+)\)", geometry)
        swapped_rings = [swap_coordinate_pairs(ring) for ring in rings]
        swapped_geometry = f"POLYGON (({'),('.join(swapped_rings)}))"

    elif geometry_type == 'MULTIPOLYGON':
        # split the geometry into individual polygons
        polygons = re.findall(r"\(\(([-+\d\s.,()]+?)\)\)", geometry)

        swapped_polygons = []
        for polygon in polygons:
            # split polygon into rings (first is outer, rest are inner)
            rings = polygon.split('),(')

            # swap coordinates for each ring
            swapped_rings = [swap_coordinate_pairs(ring) for ring in rings]

            # reconst. the polygon with all its rings
            if len(rings) > 1:
                # if there are inner rings, join them with ),(
                swapped_polygons.append(f"(({'),('.join(swapped_rings)}))")
            else:
                # if there's only outer ring
                swapped_polygons.append(f"(({swapped_rings[0]}))")

        # re-build the MULTIPOLYGON
        swapped_geometry = f"MULTIPOLYGON ({','.join(swapped_polygons)})"

    else:
        raise ValueError(f"Unsupported geometry type: {geometry_type}")

    # val the swapped geometry
    try:
        swapped_geom = loads(swapped_geometry)
        if not swapped_geom.is_valid:
            print(f"Warning: Invalid geometry after swap: {swapped_geom.wkt}")
    except Exception as e:
        raise ValueError(f"Invalid geometry after swapping coordinates: {str(e)}")

    return swapped_geometry



# this function is just to get what fields are sensors part of
# todo: this implementation proves there are problems with current shapes in the services. No full mapping exists!
def get_fields_4_sensors(fields_json, locations_df):
    # --- Load JSON file ---
    with open(fields_json, "r") as f:
        fields_data = json.load(f)

    # --- Build list of field geometries ---
    field_geoms = []
    field_ids = []

    for fdict in fields_data:
        field_geom = fdict["feature"]
        field_geom = __swap_coordinates(field_geom)
        print("Geometry to parse:", field_geom)
        geom = loads(field_geom)
        field_geoms.append(geom)
        field_ids.append(fdict["field_id"])

    # create fields gdf
    fields_gdf = gpd.GeoDataFrame({"field_id": field_ids, "geometry": field_geoms}, crs="EPSG:4326")

    # convert to gdf crossing x and y
    devices_gdf = gpd.GeoDataFrame(
        locations_df,
        geometry=gpd.points_from_xy(locations_df.y, locations_df.x),
        crs="EPSG:4326"
    )

    # spatial joint points in polygons!
    joined = gpd.sjoin(devices_gdf, fields_gdf, how="left", predicate="within")

    # agg field IDs per device
    device_to_fields = (
        joined.groupby("datastream_id")["field_id"]
        .apply(lambda x: list(x.dropna().unique()))
        .reindex(locations_df["datastream_id"])
        .reset_index()
    )

    # merge back to original devices df
    locations_with_fields = locations_df.merge(device_to_fields, on="datastream_id")

    print(locations_with_fields.head())

    return locations_with_fields