python - How to calculate the shape from lon/lat coordinates and draw it with plotly

Question

I have a plotly based map where I am showing several coordinates in a Mapbox scatter plot. Now I want to get a shape from these coordinates and draw a shape on the map.

The coordinates are available as a pandas.Series. Below an extract of the coordinates.

0    [[51.795, 3.363], [51.79483333333334, 3.363], ...
1    [[51.42536, 2.622246666666667], [51.4256883333 ...

How can I get a shape for these coordinates which boundaries are the outmost coordinates of the cluster?

question from:https://stackoverflow.com/questions/65647292/how-to-calculate-the-shape-from-lon-lat-coordinates-and-draw-it-with-plotly

Answer 1

In geometry, there are basically two main concepts Convex hull and Alpha shape (also called Concave hull) to get the shape of a finite set of points. The following picture should visually describe the differences between them [2].

In case you want to have the convex hull you can use scipy.spatial.ConvexHull [4].

Please find below a basic example of how to calculate the convex hull for some randomly generated coordinates with scipy.spatial.ConvexHull and to draw the shape with plotly.

import plotly.express as px
import numpy as np

from scipy.spatial import ConvexHull

def get_random_coordinate(): 
    return [ np.random.uniform(54.9, 56.2), np.random.uniform(2.3, 3.2) ]

points = [ get_random_coordinate() for _ in range(20) ]

def get_convex_hull():
    hull = ConvexHull(points)
    return [ [hull.points[visible_facet][1],hull.points[visible_facet][0]] for visible_facet in hull.vertices ]

convex_hull_points = get_convex_hull()
    
fig = px.scatter_mapbox(points, lat=0, lon=1, zoom=6)
fig.update_layout(
    mapbox = {
        'style': "open-street-map",
        'layers': [
            {
            'source': {
                'type': "FeatureCollection",
                'features': [{
                    'type': "Feature",
                    'geometry': {
                        'type': "MultiPolygon",
                        'coordinates': [[convex_hull_points]] 
                    }
                }]
            },
            'type': "fill", 'below': "traces", 'color': "#FF00AA", 'opacity': 0.5
            }
        ]}
    )
fig.show()

Below the final result.

As alternatives to scipy you can also check geopandas and alphashape.