Fatal Road Crashes: Part 2

Sun 28 June 2020

Filed under Cartopy

Lethal Roads, V2¶

Introduction¶

Overview¶

A little while ago, I did blog posts on:

Fatal Road Crash Locations
Australian 1Km Population Grid

Just for fun, I decided the combine both datasets. You would expect there to be more fatal crashs where there are more people. I wanted to take population density into account to find out Queensland's most dangerous roads

Implementation¶

Notebook magic commands¶

In [146]:

%matplotlib inline

In [147]:

%load_ext watermark

The watermark extension is already loaded. To reload it, use:
  %reload_ext watermark

In [148]:

%load_ext lab_black

The lab_black extension is already loaded. To reload it, use:
  %reload_ext lab_black

Notebook imports¶

In [149]:

# all imports should go here

import pandas as pd
import sys
import os
import subprocess
import datetime
import platform
import datetime

import cartopy.crs as ccrs
from cartopy.io.img_tiles import GoogleTiles
from cartopy.io.img_tiles import OSM
import cartopy.feature as cfeature
from cartopy.mpl.gridliner import (
    LONGITUDE_FORMATTER,
    LATITUDE_FORMATTER,
)
from cartopy.io import shapereader

from osgeo import gdal, osr

import matplotlib.pyplot as plt
import matplotlib.colors as colors
from matplotlib.patches import Polygon
from matplotlib.collections import PatchCollection
import matplotlib.cm as cm

import numpy as np
import math

import geopandas as gpd

Map Projections¶

Introduction¶

To recap, we can't use my standard PlateCarree map projection, because in dealing with population density, we want an Equal-Area projection. The Australian Bureau of Statistics (ABS) used Albers_Conic_Equal_Area for their data grid.

To show the difference, I plot the world in Plate Carree, then get the ABS TGeoTIFF file, extract and print the projection details, and plot the world in that projection.

In [150]:

fig = plt.figure(figsize=(20, 20))
ax = fig.add_subplot(
    1,
    1,
    1,
    projection=ccrs.PlateCarree(central_longitude=150),
)

ax.set_global()
ax.coastlines()


plt.title('World in Plate Carree', {'fontsize': 20}, pad=40)
plt.show()

No description has been provided for this image

Read Dataset¶

Read the ABS Population Density dataset, extract and display projection details

In [151]:

pop = gdal.Open(
    'd:\\qml\\Australian_Population_Grid_2011.tif'
)
data = pop.ReadAsArray()
gt = pop.GetGeoTransform()
proj = pop.GetProjection()

inproj = osr.SpatialReference()
inproj.ImportFromWkt(proj)

print("[ RASTER BAND COUNT ]: ", pop.RasterCount)
cols = pop.RasterXSize
print('cols = ', cols)
rows = pop.RasterYSize
print(' rows = ', rows)
bands = pop.RasterCount
print('bands = ', bands)
driver = pop.GetDriver().LongName
print('driver =', driver)

print('MetaData = ', pop.GetMetadata())

print(inproj)

[ RASTER BAND COUNT ]:  1
cols =  6382
 rows =  3821
bands =  1
driver = GeoTIFF
MetaData =  {'AREA_OR_POINT': 'Area', 'DataType': 'Generic'}
PROJCS["GDA_1994_Albers",
    GEOGCS["GCS_GDA_1994",
        DATUM["Geocentric_Datum_of_Australia_1994",
            SPHEROID["GRS_1980",6378137,298.257222101,
                AUTHORITY["EPSG","7019"]],
            AUTHORITY["EPSG","6283"]],
        PRIMEM["Greenwich",0],
        UNIT["degree",0.0174532925199433]],
    PROJECTION["Albers_Conic_Equal_Area"],
    PARAMETER["standard_parallel_1",-18],
    PARAMETER["standard_parallel_2",-36],
    PARAMETER["latitude_of_center",0],
    PARAMETER["longitude_of_center",132],
    PARAMETER["false_easting",0],
    PARAMETER["false_northing",0],
    UNIT["metre",1,
        AUTHORITY["EPSG","9001"]]]

In [152]:

proj_wkt = pop.GetProjection()
print(proj_wkt)

PROJCS["GDA_1994_Albers",GEOGCS["GCS_GDA_1994",DATUM["Geocentric_Datum_of_Australia_1994",SPHEROID["GRS_1980",6378137,298.257222101,AUTHORITY["EPSG","7019"]],AUTHORITY["EPSG","6283"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433]],PROJECTION["Albers_Conic_Equal_Area"],PARAMETER["standard_parallel_1",-18],PARAMETER["standard_parallel_2",-36],PARAMETER["latitude_of_center",0],PARAMETER["longitude_of_center",132],PARAMETER["false_easting",0],PARAMETER["false_northing",0],UNIT["metre",1,AUTHORITY["EPSG","9001"]]]

In [153]:

proj = osr.SpatialReference()
proj.ImportFromWkt(proj_wkt)
print(proj)

PROJCS["GDA_1994_Albers",
    GEOGCS["GCS_GDA_1994",
        DATUM["Geocentric_Datum_of_Australia_1994",
            SPHEROID["GRS_1980",6378137,298.257222101,
                AUTHORITY["EPSG","7019"]],
            AUTHORITY["EPSG","6283"]],
        PRIMEM["Greenwich",0],
        UNIT["degree",0.0174532925199433]],
    PROJECTION["Albers_Conic_Equal_Area"],
    PARAMETER["standard_parallel_1",-18],
    PARAMETER["standard_parallel_2",-36],
    PARAMETER["latitude_of_center",0],
    PARAMETER["longitude_of_center",132],
    PARAMETER["false_easting",0],
    PARAMETER["false_northing",0],
    UNIT["metre",1,
        AUTHORITY["EPSG","9001"]]]

In [154]:

extent_tif = (
    gt[0],
    gt[0] + pop.RasterXSize * gt[1],
    gt[3] + pop.RasterYSize * gt[5],
    gt[3],
)

print(pop.RasterXSize, pop.RasterYSize, extent_tif)

6382 3821 (-3853000.0, 2529000.0, -4835000.0, -1014000.0)

Create Geospatial Environment¶

Create a Cartopy projection that matches the ABS projection

In [155]:

albo = ccrs.AlbersEqualArea(
    central_latitude=0,
    false_easting=0,
    false_northing=0,
    central_longitude=132,
    standard_parallels=(-18, -36),
)

Plot example map¶

Plot the world in our chosen projection

In [156]:

fig = plt.figure(figsize=(20, 20))
ax = fig.add_subplot(1, 1, 1, projection=albo)

ax.set_global()
ax.coastlines()

# plt.imshow(data, extent=extent_tif, transform=albo)
plt.title(
    'World in Albers Equal Area (centered on Australia)',
    {'fontsize': 30},
    pad=40,
)

plt.show()

Population Density¶

We start by plotting at the 1Km grid resolution

Helper Functions¶

Define a function to draw the Queensland map background for our data, to be constant across different visualizations

In [157]:

def draw_queensland(projection=None):
    '''
    draw_queensland:  draw a map of Queensland in the nominated projection
    
    Parameters:
    projection: Cartopy ccrs, defaults to None
    
    Returns:
    (Figure object, Axes object) tuple
    
    Notes:
    Extent of QLD hardcoded to include some of neighbouring states for context
    
    Size of map hard coded to be 20, 20
    
    We draw coastlines, lat/lon gridlines, and state borders
    
    The coastlines are fairly coarse, but are not the main focus of the map, so OK
    '''
    fig = plt.figure(figsize=(20, 20))
    ax = fig.add_subplot(1, 1, 1, projection=projection)

    ax.set_extent(
        (137.4, 154.2, -30, -9.8), crs=ccrs.PlateCarree()
    )

    ax.coastlines(color='black', zorder=5)

    # draw lat lon grid
    ax.gridlines()

    # Create a feature for States/Admin 1 regions at 1:10m from Natural Earth
    states_provinces = cfeature.NaturalEarthFeature(
        category='cultural',
        name='admin_1_states_provinces_lines',
        scale='10m',
        facecolor='none',
    )

    # plot state border
    SOURCE = 'Natural Earth'
    LICENSE = 'public domain'
    ax.add_feature(
        states_provinces, edgecolor='gray', zorder=5
    )

    return fig, ax


# end draw_queensland

Plot the Population Density¶

This map is largely a failure, because most of the population density is located inside a few pixels. Queensland (and Australia) is a big empty place!

In [158]:

fig, ax = draw_queensland(projection=albo)

mycmap = colors.LinearSegmentedColormap.from_list(
    'mycmap', ['white', 'red']
)

pcm = plt.imshow(
    data,
    extent=extent_tif,
    transform=albo,
    cmap=mycmap,
    zorder=1,
    alpha=1,
)

cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n1Km Grid Square')


plt.title(
    'Australian Population Grid, Albers Equal Area, Linear Scale',
    {'fontsize': 30},
    pad=40,
)


plt.show()

Log Scales¶

Using a Log scale for the population density helps a little (brings out the low population density areas better)

In [159]:

fig, ax = draw_queensland(projection=albo)


mycmap = colors.LinearSegmentedColormap.from_list(
    'mycmap', ['white', 'red']
)

pcm = plt.imshow(
    data,
    extent=extent_tif,
    transform=albo,
    cmap=mycmap,
    norm=colors.LogNorm(
        vmin=data.min() + 1, vmax=data.max()
    ),
    zorder=1,
    alpha=1,
)

cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n1Km Grid Square')


plt.title(
    'Australian Population Grid, Albers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)


plt.show()

Data Aggregation¶

We can get a better visualization (for our purposes) by aggregation to give a 100 Km grid

In [160]:

data.shape

Out[160]:

(3821, 6382)

In [161]:

data[0:3800, 0:6300].shape

Out[161]:

(3800, 6300)

pandas magic to collapse a big array into a smaller one

In [162]:

def shrink(data, rows, cols):
    '''
    shrink: reduce an array by aggregating in x and y dimensions to give an array of the specified size
    
    Parameters:
    data: input numpy array
    rows: number of rows in resized and aggregated array
    cols: number of columns in resized and aggregated array
    
    Returns:
    Resized array
    
    Example:
    a = np.array(
    [
        [1, 2, 3, 4, 5, 6],
        [5, 6, 7, 8, 7, 8],
        [9, 10, 11, 12, 13, 14],
        [13, 14, 15, 16, 17, 18],
    ]
            )

    shrink(a, 2, 3)

    array([[14, 22, 26],
           [46, 54, 62]])
    
    
    '''
    return (
        data.reshape(
            rows,
            data.shape[0] // rows,
            cols,
            data.shape[1] // cols,
        )
        .sum(axis=1)
        .sum(axis=2)
    )


# end if

Collapse our fine grid

In [163]:

# collapse 1km grid to 100 km grid

data2 = data[0:3800, 0:6300]
data3 = shrink(data2, 38, 63)

Plot our coarse grid. We can see the population hugging the coast, and the big mining towns inland

In [164]:

fig, ax = draw_queensland(projection=albo)

my_cmap = plt.cm.get_cmap('terrain')
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 1

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data3_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step
    * data[0:3800, 0:6300].shape[1]
    * down_scale_factor,
    extent_tif[3]
    + (y_step)
    * data[0:3800, 0:6300].shape[0]
    * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    data3,
    extent=data3_ext,
    transform=albo,
    cmap=my_cmap,
    norm=colors.LogNorm(),
    zorder=1,
    alpha=0.5,
    vmin=100,
)

cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n100Km Grid Square')

# draw lat lon grid
ax.gridlines()

plt.title(
    'Australian Population Grid 100Km, Albers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)

plt.show()

Medium Grid¶

We can also do the same for a 10 Km grid

In [165]:

# collapse 1km grid to 10 km grid

data2 = data[0:3800, 0:6300]
data4 = shrink(data2, 380, 630)

In [166]:

fig, ax = draw_queensland(projection=albo)

my_cmap = plt.cm.get_cmap('terrain')
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 10

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data4_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step * data4.shape[1] * down_scale_factor,
    extent_tif[3]
    + (y_step) * data4.shape[0] * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    data4,
    extent=data4_ext,
    transform=albo,
    cmap=my_cmap,
    norm=colors.LogNorm(),
    zorder=1,
    alpha=0.5,
    vmin=10,
)

cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n10Km Grid Square')

# draw lat lon grid
ax.gridlines()

plt.title(
    'Australian Population Grid 10Km, Albers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)

plt.show()

In [ ]:

Fatal Crash Locations¶

So now we have our population grids sorted, we will use the coarse grid to look at the relation between population density, and fatal crash rate

Read Crash Location Dataset¶

In [167]:

crash_path = 'D:\\QLDCrashLocations\\data\\locations.csv'
crashes = pd.read_csv(
    crash_path, dtype={'Loc_Post_Code': str}
)

In [168]:

fatal = crashes[crashes["Crash_Severity"] == 'Fatal']

In [169]:

fatal.head(1)

Out[169]:

	Crash_Ref_Number	Crash_Severity	Crash_Year	Crash_Month	Crash_Day_Of_Week	Crash_Hour	Crash_Nature	Crash_Type	Crash_Longitude_GDA94	Crash_Latitude_GDA94	...	Count_Casualty_MedicallyTreated	Count_Casualty_MinorInjury	Count_Casualty_Total	Count_Unit_Car	Count_Unit_Motorcycle_Moped	Count_Unit_Truck	Count_Unit_Bus	Count_Unit_Bicycle	Count_Unit_Pedestrian	Count_Unit_Other
18	19	Fatal	2001	January	Thursday	9	Hit pedestrian	Hit pedestrian	151.934354	-27.566384	...	0	0	1	0	0	1	0	0	1	0

1 rows × 52 columns

Transform Crash Location Coordinates¶

We transform then crash location lat/lon coordinates into the Albers Equal Area coordinate system

In [170]:

albo_xy = albo.transform_points(
    ccrs.PlateCarree(),
    np.array(fatal['Crash_Longitude_GDA94']),
    np.array(fatal['Crash_Latitude_GDA94']),
)

In [171]:

albo_xy.shape

Out[171]:

(4761, 3)

Map the Fatal Crash Locations¶

Because of the coarse coastlines we chose to use, we don't draw the off-shore islands, so there seem to be crashes at sea (not dataset defects).

In [172]:

fig, ax = draw_queensland(projection=albo)

ax.plot(
    albo_xy[:, 0], albo_xy[:, 1], '+', label='Fatal Crashes'
)


ax.legend()

plt.show()

Combined Visualization¶

We combine the visualization of the two datasets. Note we have a colorbar and a legend

Subjectively, we can see more crashes in areas of high population density

In [173]:

fig, ax = draw_queensland(projection=albo)

my_cmap = plt.cm.get_cmap('terrain')
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 100

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data3_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step * data3.shape[1] * down_scale_factor,
    extent_tif[3]
    + (y_step) * data3.shape[0] * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    data3,
    extent=data3_ext,
    transform=albo,
    cmap=my_cmap,
    norm=colors.LogNorm(),
    zorder=1,
    alpha=0.5,
    vmin=100,
)


# plot fatal crashes
ax.plot(
    albo_xy[:, 0],
    albo_xy[:, 1],
    'r+',
    label='Fatal Crashes',
)


cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n100Km Grid Square')

plt.title(
    'QLD Fatal Crash Locations, Australian Population Grid 100Km, \nAlbers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)

ax.legend()

plt.show()

We can repeat the mapping for our medium population grid, but it dosen't add much to the understanding

In [174]:

fig, ax = draw_queensland(projection=albo)

my_cmap = plt.cm.get_cmap('terrain')
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 100

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data4_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step * data3.shape[1] * down_scale_factor,
    extent_tif[3]
    + (y_step) * data3.shape[0] * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    data4,
    extent=data4_ext,
    transform=albo,
    cmap=my_cmap,
    norm=colors.LogNorm(),
    zorder=1,
    alpha=0.5,
    vmin=10,
)


# plot fatal crashes
ax.plot(
    albo_xy[:, 0],
    albo_xy[:, 1],
    'r+',
    label='Fatal Crashes',
)


cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)
cbar.ax.set_title('Population per \n100Km Grid Square')

plt.title(
    'QLD Fatal Crash Locations, Australian Population Grid 100Km, \nAlbers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)

ax.legend()

plt.show()

Fatal Crash Density¶

We can count the fatal crashes within each grid square of our coarse grid, to give us crash density. Note increasing row number runs south

In [175]:

data3_ext

Out[175]:

[-3853000.0, 2447000.0, -4814000.0, -1014000.0]

In [176]:

crash_rate = np.zeros(data3.shape)

# get the fatal crashes per 100 km grid square
for x, y in zip(albo_xy[:, 0], albo_xy[:, 1]):
    row = 37 - int(
        (y - data3_ext[2]) // (down_scale_factor * 1000)
    )
    col = int(
        (x - data3_ext[0]) // (down_scale_factor * 1000)
    )
    crash_rate[row, col] = crash_rate[row, col] + 1
# end for

In [177]:

data3.shape, down_scale_factor

Out[177]:

((38, 63), 100)

Plot the Crash Density¶

In [178]:

fig, ax = draw_queensland(projection=albo)

my_cmap = colors.LinearSegmentedColormap.from_list(
    'mycmap', ['white', 'red']
)
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 100

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data3_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step * data3.shape[1] * down_scale_factor,
    extent_tif[3]
    + (y_step) * data3.shape[0] * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    crash_rate,
    extent=data3_ext,
    transform=albo,
    cmap=my_cmap,
    norm=colors.LogNorm(),
    zorder=1,
    alpha=0.5,
    vmin=1,
)


# plot fatal crashes
ax.plot(albo_xy[:, 0], albo_xy[:, 1], 'k+', alpha=0.2)


cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)

plt.title(
    'QLD Fatal Crash Count, Australian Population Grid 100Km, \nAlbers Equal Area, Log Scale',
    {'fontsize': 30},
    pad=40,
)


plt.show()

The crash density hugs the coast as we expect. The crash density in Cape York is a little puzzling

Crash Rate per Capita¶

So now we will get the crash rate per person for each coarse grid square. Because there are so many grid squares with no population, we have to check for this

In [179]:

crash_rate_pc = np.zeros(data3.shape)

for row in range(data3.shape[0]):
    for col in range(data3.shape[1]):
        if data3[row, col] == 0:
            crash_rate_pc[row, col] = 0
        else:
            crash_rate_pc[row, col] = (
                crash_rate[row, col] / data3[row, col]
            )
        # end if

    # end for
# end for

Get the maximum crash rate per person

In [180]:

np.max(np.max(crash_rate_pc))

Out[180]:

0.39089221905003857

Final Visualization¶

We get a partially surprising (to me) result. It is well known that outback country roads have a higher crash rate than could be expected by population density (people driving on unfamiliar roads, poorly maintained roads, driving tired on long trips, driving with stock on the road), but I didn't expect Cape York to be so bad. Birdsville maybe I can guess the reasons for (Birdsville Races draws huge crowds with high alcohol consumption). Maybe Cape York being so remote means that the long time-to-hospital cause more crashes to be fatal.

It might be an interesting sociology / data science project to take up.

In [181]:

fig, ax = draw_queensland(projection=albo)


my_cmap = colors.LinearSegmentedColormap.from_list(
    'mycmap', ['white', 'red']
)
my_cmap.set_under('grey', alpha=0.0)

# because we are truncating the data array to show, we must recompute the
# map extent it covers
down_scale_factor = 100

# original x and y step (we know it is = 1km, but get value from GEOTIF object for generality)
x_step = gt[1]
y_step = gt[5]

data3_ext = [
    extent_tif[0],
    extent_tif[0]
    + x_step * data3.shape[1] * down_scale_factor,
    extent_tif[3]
    + (y_step) * data3.shape[0] * down_scale_factor,
    extent_tif[3],
]

pcm = plt.imshow(
    crash_rate_pc,
    extent=data3_ext,
    transform=albo,
    cmap=my_cmap,
    # norm=colors.LogNorm(),
    zorder=1,
    alpha=0.9,
    vmin=0,
)


# plot fatal crashes
ax.plot(albo_xy[:, 0], albo_xy[:, 1], 'k+', alpha=0.2)


cbar = fig.colorbar(
    pcm, ax=ax, shrink=0.5, pad=0.1, extend='min'
)

plt.title(
    'QLD Fatal Crash Per Capita, Australian Population Grid 100Km, \nAlbers Equal Area, Linear Scale',
    {'fontsize': 30},
    pad=40,
)

plt.show()

Statistical Analysis¶

I thought I would test my perception that population density and fatal crash density were correlated

Create two arrays for non-zero population and crash density data pairs, and do a quick-and-dirty plot

In [182]:

density = []
crashs = []
for row in range(data3.shape[0]):
    for col in range(data3.shape[1]):
        if (crash_rate[row, col] > 0) and (
            data3[row, col] > 0
        ):
            density.append(data3[row, col])
            crashs.append(crash_rate[row, col])
        # end if
    # end for
# end for

_ = plt.plot(density, crashs, 'o')

Do a less quick and dirty plot

In [183]:

fig = plt.figure(figsize=(20, 20))
ax = fig.add_subplot(1, 1, 1)


# ax.loglog()
ax.grid(axis='both')
ax.set_xlabel(
    'Population Density (people per 100Km grid square)'
)
ax.set_ylabel('Fatal Crashes per 100Km grid square')

ax.plot(density, crashs, 'o')

Out[183]:

[<matplotlib.lines.Line2D at 0x259c0037f98>]

Do the same plot but with log-log scales

In [184]:

fig = plt.figure(figsize=(20, 20))
ax = fig.add_subplot(1, 1, 1)


ax.loglog()
ax.grid(axis='both')
ax.set_xlabel(
    'Population Density (people per 100Km grid square)'
)
ax.set_ylabel('Fatal Crashes per 100Km grid square')

ax.plot(density, crashs, 'o')

Out[184]:

[<matplotlib.lines.Line2D at 0x259b3f65b70>]

Visualize with Seaborn¶

We use Seaborn to visualize the linear relation between population and crash density

In [185]:

import seaborn as sns

df_dict = {'density': (density), 'crashs': (crashs)}
df = pd.DataFrame(df_dict)

sns.regplot('density', 'crashs', data=df)

D:\Anaconda3\envs\ac5-py37\lib\site-packages\scipy\stats\stats.py:1713: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.
  return np.add.reduce(sorted[indexer] * weights, axis=axis) / sumval

Out[185]:

<matplotlib.axes._subplots.AxesSubplot at 0x259b2dba438>

Visualize via Seaborn the log-log relationship

In [186]:

df_dict = {
    'density': np.log10(density),
    'crashs': np.log10(crashs),
}
df = pd.DataFrame(df_dict)

sns.regplot('density', 'crashs', data=df)

Out[186]:

<matplotlib.axes._subplots.AxesSubplot at 0x259b2ea8710>

Linear Regression¶

We use OLS to find that the relation between population and crash density is unlikely to have been observed by chance

In [187]:

from statsmodels.formula.api import ols

df_dict = {'density': density, 'crashs': crashs}
df = pd.DataFrame(df_dict)

res1 = ols('crashs ~ density', data=df).fit()
res1.summary()

Out[187]:

OLS Regression Results
Dep. Variable:	crashs	R-squared:	0.932
Model:	OLS	Adj. R-squared:	0.932
Method:	Least Squares	F-statistic:	1946.
Date:	Sun, 21 Jun 2020	Prob (F-statistic):	2.18e-84
Time:	20:03:12	Log-Likelihood:	-676.10
No. Observations:	143	AIC:	1356.
Df Residuals:	141	BIC:	1362.
Df Model:	1
Covariance Type:	nonrobust

	coef	std err	t	P>\|t\|	[0.025	0.975]
Intercept	10.1650	2.363	4.302	0.000	5.494	14.836
density	0.0007	1.65e-05	44.110	0.000	0.001	0.001

Omnibus:	96.913	Durbin-Watson:	1.331
Prob(Omnibus):	0.000	Jarque-Bera (JB):	1035.411
Skew:	2.192	Prob(JB):	1.46e-225
Kurtosis:	15.432	Cond. No.	1.47e+05

Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
[2] The condition number is large, 1.47e+05. This might indicate that there are
strong multicollinearity or other numerical problems.

In [188]:

_ = plt.plot(df['density'], res1.predict(), 'g-')
_ = plt.plot(df['density'], df['crashs'], 'r+')

In [189]:

_ = plt.plot(res1.predict(), res1.resid, 'o')
_ = plt.axhline(0, color='gray')

Crash Rate¶

Finally we look at the crash rate per capita, as a function of population density

I think the results are inconclusive, but maybe as density goes up, and you are more likely to be close to a hospital, fatal crashes per person decreases.

In [190]:

rate_pc = [a / b for a, b in zip(crashs, density)]

In [191]:

fig = plt.figure(figsize=(20, 20))
ax = fig.add_subplot(1, 1, 1)


ax.loglog()
ax.grid(axis='both')
ax.set_ylabel(
    'Crash Rate (crash / persons in 100Km grid square)'
)
ax.set_xlabel('Persons per 100Km grid square')

ax.plot(density, rate_pc, 'o')

Out[191]:

[<matplotlib.lines.Line2D at 0x259c018b5c0>]

Conclusions / Summary¶

Some interesting and thought provoking findings can be discovered by poking around in, and combining, public datasets.

Reproducibility¶

Notebook version status¶

In [192]:

theNotebook = '2020-06-10-dc-RoadCrashes2.ipynb'

In [193]:

# show info to support reproducibility


def python_env_name():
    envs = subprocess.check_output(
        'conda env list'
    ).splitlines()
    # get unicode version of binary subprocess output
    envu = [x.decode('ascii') for x in envs]
    active_env = list(
        filter(lambda s: '*' in str(s), envu)
    )[0]
    env_name = str(active_env).split()[0]
    return env_name


# end python_env_name

print('python version : ' + sys.version)
print('python environment :', python_env_name())


print('current wkg dir: ' + os.getcwd())
print('Notebook name: ' + theNotebook)
print(
    'Notebook run at: '
    + str(datetime.datetime.now())
    + ' local time'
)
print(
    'Notebook run at: '
    + str(datetime.datetime.utcnow())
    + ' UTC'
)
print('Notebook run on: ' + platform.platform())

python version : 3.7.1 (default, Dec 10 2018, 22:54:23) [MSC v.1915 64 bit (AMD64)]
python environment : ac5-py37
current wkg dir: C:\Users\donrc\Documents\JupyterNotebooks\RoadCrashes2NotebookProject\develop
Notebook name: 2020-06-10-dc-RoadCrashes2.ipynb
Notebook run at: 2020-06-21 20:03:34.908312 local time
Notebook run at: 2020-06-21 10:03:34.909315 UTC
Notebook run on: Windows-10-10.0.18362-SP0

In [194]:

%watermark

2020-06-21T20:03:34+10:00

CPython 3.7.1
IPython 7.2.0

compiler   : MSC v.1915 64 bit (AMD64)
system     : Windows
release    : 10
machine    : AMD64
processor  : Intel64 Family 6 Model 94 Stepping 3, GenuineIntel
CPU cores  : 8
interpreter: 64bit

In [195]:

%watermark -h -iv

pandas     1.0.0
matplotlib 3.1.1
osgeo.gdal 2.3.3
cartopy    0.17.0
numpy      1.15.4
geopandas  0.5.1
osgeo      2.3.3
platform   1.0.8
seaborn    0.9.0
host name: DESKTOP-SODFUN6

In [ ]:

Fatal Road Crashes: Part 2

Lethal Roads, V2¶

Introduction¶

Overview¶

Implementation¶

Notebook magic commands¶

Notebook imports¶

Map Projections¶

Introduction¶

Read Dataset¶

Create Geospatial Environment¶

Plot example map¶

Population Density¶

Helper Functions¶

Plot the Population Density¶

Log Scales¶

Data Aggregation¶

Medium Grid¶

Fatal Crash Locations¶

Read Crash Location Dataset¶

Transform Crash Location Coordinates¶

Map the Fatal Crash Locations¶

Combined Visualization¶

Fatal Crash Density¶

Plot the Crash Density¶

Crash Rate per Capita¶

Final Visualization¶

Statistical Analysis¶

Visualize with Seaborn¶

Linear Regression¶

Crash Rate¶

Conclusions / Summary¶

Reproducibility¶

Notebook version status¶

Comments

About