GIS with Python and IPython

Part II: Working with Rasters

Set-up our environment as before

Let's import the packages we will use and set the paths for outputs.

In [1]:
# Let's import pandas and some other basic packages we will use 
from __future__ import division
import pandas as pd
import numpy as np
import os, sys

# GIS packages
import geopandas as gpd
from geopandas.tools import overlay
from shapely.geometry import Polygon, Point
import georasters as gr
# Alias for Geopandas
gp = gpd

# Plotting
import matplotlib as mpl
import seaborn as sns
# Setup seaborn
sns.set()

# Mapping
import geoplot as gplt
import geoplot.crs as gcrs
import mapclassify as mc
import textwrap

%pylab --no-import-all
%matplotlib inline
Using matplotlib backend: <object object at 0x190835a20>
%pylab is deprecated, use %matplotlib inline and import the required libraries.
Populating the interactive namespace from numpy and matplotlib
In [2]:
# Functions for plotting
def center_wrap(text, cwidth=32, **kw):
    '''Center Text (to be used in legend)'''
    lines = text
    #lines = textwrap.wrap(text, **kw)
    return "\n".join(line.center(cwidth) for line in lines)

def MyChoropleth(mydf, myfile='', myvar='',
                  mylegend='',
                  k=5,
                  extent=[-180, -90, 180, 90],
                  bbox_to_anchor=(0.2, 0.5),
                  edgecolor='white', facecolor='lightgray',
                  scheme='FisherJenks', bins=None, pct=None,
                  legend_labels=None,
                  save=True,
                  percent=False,
                  cmap='Reds',
                  **kwargs):
    # Chloropleth
    # Color scheme
    if scheme=='EqualInterval':
        scheme = mc.EqualInterval(mydf[myvar], k=k)
    elif scheme=='Quantiles':
        scheme = mc.Quantiles(mydf[myvar], k=k)
    elif scheme=='BoxPlot':
        scheme = mc.BoxPlot(mydf[myvar], k=k)
    elif scheme=='FisherJenks':
        scheme = mc.FisherJenks(mydf[myvar], k=k)
    elif scheme=='FisherJenksSampled':
        scheme = mc.FisherJenksSampled(mydf[myvar], k=k)
    elif scheme=='HeadTailBreaks':
        scheme = mc.HeadTailBreaks(mydf[myvar], k=k)
    elif scheme=='JenksCaspall':
        scheme = mc.JenksCaspall(mydf[myvar], k=k)
    elif scheme=='JenksCaspallForced':
        scheme = mc.JenksCaspallForced(mydf[myvar], k=k)
    elif scheme=='JenksCaspallSampled':
        scheme = mc.JenksCaspallSampled(mydf[myvar], k=k)
    elif scheme=='KClassifiers':
        scheme = mc.KClassifiers(mydf[myvar], k=k)
    elif scheme=='Percentiles':
        scheme = mc.Percentiles(mydf[myvar], pct=pct)
    elif scheme=='UserDefined':
        scheme = mc.UserDefined(mydf[myvar], bins=bins)
    
    if legend_labels is None:
        # Format legend
        upper_bounds = scheme.bins
        # get and format all bounds
        bounds = []
        for index, upper_bound in enumerate(upper_bounds):
            if index == 0:
                lower_bound = mydf[myvar].min()
            else:
                lower_bound = upper_bounds[index-1]
            # format the numerical legend here
            if percent:
                bound = f'{lower_bound:.0%} - {upper_bound:.0%}'
            else:
                bound = f'{float(lower_bound):,.0f} - {float(upper_bound):,.0f}'
            bounds.append(bound)
        legend_labels = bounds
    #Plot
    ax = gplt.choropleth(
        mydf, hue=myvar, projection=gcrs.PlateCarree(central_longitude=0.0, globe=None),
        edgecolor='white', linewidth=1,
        cmap=cmap, legend=True,
        scheme=scheme,
        legend_kwargs={'bbox_to_anchor': bbox_to_anchor,
                       'frameon': True,
                       'title':mylegend,
                       },
        legend_labels = legend_labels,
        figsize=(24, 16),
        rasterized=True,
    )
    gplt.polyplot(
        countries, projection=gcrs.PlateCarree(central_longitude=0.0, globe=None),
        edgecolor=edgecolor, facecolor=facecolor,
        ax=ax,
        rasterized=True,
        extent=extent,
    )
    if save:
        plt.savefig(pathgraphs + myfile + '_' + myvar +'.pdf', dpi=300, bbox_inches='tight')
        plt.savefig(pathgraphs + myfile + '_' + myvar +'.png', dpi=300, bbox_inches='tight')
    pass
In [3]:
# Paths
pathout = './data/'

if not os.path.exists(pathout):
    os.mkdir(pathout)
    
pathgraphs = './graphs/'
if not os.path.exists(pathgraphs):
    os.mkdir(pathgraphs)

Initial Example -- Caloric Suitability Index

Let's download a raster with interesting data so we can visualize and analyze it. Caloric Suitability Index CSI provides estimates for the potential calories that can be produced in any location using various crops.

For rasters we can use georasters or rasterio or various other tools. georasters is simple to use and has many functions that are useful to (social) scientists. It tries to do for rasters what geopandas does for geometries. Although mostly limited to what I have needed, it is expanding slowly to incorporate other uses.

Next we will use it to download a raster in GeoTiff format from the Caloric Suitability Index CSI website. Since the data is made available via Google Driive, we will also learn how to use GD's API to download data. Once we have the datra we will imprt it as a GeoRaster, which is simply a masked numpy array with associated geographical information. We can use many functions and properties of the GeoRaster to analyze our data. Moreover, since it is based on numpy's MaskedArray object, any funtion that works on numpy arrays can be used on a GeoRaster.

Download and Import Caloric Suitability Data (CSI)

Let's download the maximum pre- and post1500 CSI data, i.e. the maximum amount of calories that can be potentially preduced in a location with the crops available pre- and post-1500. See the CSI website or the associated papers (e.g., Galor and Özak (2015,2016) for the construction and properties of the data).

Download

In [5]:
# Import GD API python package
from google_drive_downloader import GoogleDriveDownloader as gdd

# Check whether files have been already downloaded
# Otherwise download
if os.path.exists(pathout + 'pre1500MaxCalories.tif')==False:
    gdd.download_file_from_google_drive(file_id='0By-h7HPv1NhVR1BTX0V6eUdmTW8', resourcekey='0-7-oOUj8ldKwWSmnieI4oog', dest_path=pathout+'pre1500MaxCalories.tif')
if os.path.exists(pathout + 'post1500MaxCalories.tif')==False:
    gdd.download_file_from_google_drive(file_id='0By-h7HPv1NhVamdlWEtSSlpKOTA', resourcekey='0-nWBun0NiYSnYDCH_N2tr-w', dest_path=pathout+'post1500MaxCalories.tif')

Import Rasters

In [6]:
pre1500 = gr.from_file(pathout + 'pre1500MaxCalories.tif')
post1500 = gr.from_file(pathout + 'post1500MaxCalories.tif')

Plot Rasters

We can plot this data easily using georasters.

In [7]:
pre1500.plot()
Out[7]:
<AxesSubplot:>

Not very nice looking, but provides the basic information we may want. Of course we can improve using a few extra commands. Let's start by choosing a colormap and also normalizing the data. You can choose among the many colormaps provided by matplotlib.

In [8]:
myraster = pre1500
cmap = plt.cm.YlGn
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
In [9]:
fig = plt.figure(figsize=(15,10),  dpi=300, facecolor='w', edgecolor='k')
plt.matshow(pre1500.raster, cmap=cmap, norm=norm, rasterized=True)
plt.xticks([])
plt.yticks([])
plt.show()
            
<Figure size 4500x3000 with 0 Axes>

Let's add a colorbar and improve the figure a bit. Then expoprt it for using in our slides or paper.

In [10]:
myraster = pre1500
cmap = plt.cm.YlOrRd
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
ax = myraster.plot(figsize=(15,10), cmap=cmap, norm=norm, rasterized=True)
plt.xticks([])
plt.yticks([])
plt.title('')
ax = plt.gca()
ax.set_aspect(1)
# create axes instance for colorbar on bottom. 
ax = plt.gca()
pos = ax.get_position() 
l, b, w, h = pos.bounds 
cax = plt.axes([l+.3, b+0.03, .3, 0.01]) 
# draw colorbar on bottom. 
cb = mpl.colorbar.ColorbarBase(cax, cmap=cmap, norm=norm, spacing='proportional', orientation='horizontal')
cax.set_title('Maximum Calories Pre-1500')
plt.savefig(pathgraphs + 'pre1500MaxCalories.pdf', dpi=150, bbox_inches='tight')
plt.show()
In [11]:
myraster = post1500
cmap = plt.cm.YlOrRd
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
ax = myraster.plot(figsize=(15,10), cmap=cmap, norm=norm, rasterized=True)
plt.xticks([])
plt.yticks([])
plt.title('')
ax = plt.gca()
ax.set_aspect(1)
# create axes instance for colorbar on bottom. 
ax = plt.gca()
pos = ax.get_position() 
l, b, w, h = pos.bounds 
cax = plt.axes([l+.3, b+0.03, .3, 0.01]) 
# draw colorbar on bottom. 
cb = mpl.colorbar.ColorbarBase(cax, cmap=cmap, norm=norm, spacing='proportional', orientation='horizontal')
cax.set_title('Maximum Calories Post-1500')
plt.savefig(pathgraphs + 'post1500MaxCalories.pdf', dpi=150, bbox_inches='tight')
plt.show()

Not bad! Now let us add country borders so we can visualize a little bit better. We need to import a shapefile with the country borders. Let's use the same source as in the previous notebook.

In [12]:
import requests
import io

#headers = {'User-Agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_10_1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/39.0.2171.95 Safari/537.36'}
headers = {'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36', 'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8'}

url = 'https://naturalearth.s3.amazonaws.com/10m_cultural/ne_10m_admin_0_countries.zip'
r = requests.get(url, headers=headers)
countries = gp.read_file(io.BytesIO(r.content))
#countries = gpd.read_file('https://www.naturalearthdata.com/http//www.naturalearthdata.com/download/10m/cultural/ne_10m_admin_0_countries.zip')

Let's remove Antarctica so we do not plot it.

In [13]:
countries = countries.query("CONTINENT!='Antarctica'")
In [14]:
myraster = pre1500
cmap = plt.cm.YlGn
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
df3 = countries.copy()
df3.geometry = countries.boundary
df3['fake'] = 0
plt.figure(figsize=(15,10))
plt.xticks([])
plt.yticks([])
plt.title('')
ax =plt.gca()
ax.set_aspect(1)
img_extent = (myraster.xmin, myraster.xmax, myraster.ymin, myraster.ymax)
ax.imshow(myraster.raster, norm=norm, origin='upper',extent=img_extent, cmap=cmap, interpolation='bilinear', aspect=1)
df3.plot(ax=ax, color='black', edgecolor='k', linewidth=0.5, rasterized=True)
# create axes instance for colorbar on bottom. 
ax = plt.gca()
pos = ax.get_position() 
l, b, w, h = pos.bounds 
cax = plt.axes([l+.3, b+0.05, .3, 0.01]) 
# draw colorbar on bottom. 
cb = mpl.colorbar.ColorbarBase(cax, cmap=cmap, norm=norm, spacing='proportional', orientation='horizontal')
cax.set_title('Maximum Calories Pre-1500')
plt.savefig(pathgraphs + 'pre1500MaxCaloriesBorders.pdf', dpi=150, bbox_inches='tight')
plt.show()
In [15]:
myraster = post1500
cmap = plt.cm.YlOrRd
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
df3 = countries.copy()
df3.geometry = countries.boundary
df3['fake'] = 0
plt.figure(figsize=(15,10))
plt.xticks([])
plt.yticks([])
plt.title('')
ax =plt.gca()
ax.set_aspect(1)
img_extent = (myraster.xmin, myraster.xmax, myraster.ymin, myraster.ymax)
ax.imshow(myraster.raster, norm=norm, origin='upper',extent=img_extent, cmap=cmap, interpolation='bilinear', aspect=1)
df3.plot(ax=ax, color='black', edgecolor='k', linewidth=0.5, rasterized=True)
# create axes instance for colorbar on bottom. 
ax = plt.gca()
pos = ax.get_position() 
l, b, w, h = pos.bounds 
cax = plt.axes([l+.3, b+0.1, .3, 0.01]) 
# draw colorbar on bottom. 
cb = mpl.colorbar.ColorbarBase(cax, cmap=cmap, norm=norm, spacing='proportional', orientation='horizontal')
cax.set_title('Maximum Calories Post-1500')
plt.savefig(pathgraphs + 'post1500MaxCaloriesBorders.pdf', dpi=150, bbox_inches='tight')
plt.show()

Analyze some properties of these rasters

Let's start by looking at some properties of these rasters and of the data they hold.

The size of the rasters

In [16]:
print('The size of the pre-1500 raster is', pre1500.shape)
print('The size of the post-1500 raster is', post1500.shape)
The size of the pre-1500 raster is (2084, 4320)
The size of the post-1500 raster is (2160, 4320)

The geographical information of each raster

This includes the range of latitude and longitude, the size of their cells, projection, etc.

In [17]:
print('The geographical properties of the pre-1500 raster are', pre1500.geot)
print('The minimum latitude is', pre1500.ymin, 'and the maximum latitude is', pre1500.ymax, 'for the pre-1500 raster.')
print('The minimum longitude is', pre1500.xmin, 'and the maximum longitude is', pre1500.xmax, 'for the pre-1500 raster.')
print('The cells of the raster are', np.abs(pre1500.y_cell_size), 'degrees north-south', 'and', np.abs(pre1500.x_cell_size), 'east-west.')
The geographical properties of the pre-1500 raster are (-180.0, 0.08333333333333333, 0.0, 83.66666666666667, 0.0, -0.08333333333333333)
The minimum latitude is -89.99999999999999 and the maximum latitude is 83.66666666666667 for the pre-1500 raster.
The minimum longitude is -180.0 and the maximum longitude is 180.0 for the pre-1500 raster.
The cells of the raster are 0.08333333333333333 degrees north-south and 0.08333333333333333 east-west.

Basic statistics of underlying data

In [18]:
print('Average CSI in the world pre-1500 is', pre1500.mean())
print('Median CSI in the world pre-1500 is', pre1500.median())
print('Maximum CSI in the world pre-1500 is', pre1500.max())
print('Minimum CSI in the world pre-1500 is', pre1500.min())
print('Standard deviation of CSI in the world pre-1500 is', pre1500.std())
Average CSI in the world pre-1500 is 3625.1296518179392
Median CSI in the world pre-1500 is 1417.5
Maximum CSI in the world pre-1500 is 19454.5
Minimum CSI in the world pre-1500 is 0.0
Standard deviation of CSI in the world pre-1500 is 4382.756839216608

Exercise

Compute the same information for the post-1500 raster.

Basic computation on rasters

We can now perform various computation using our rasters. E.g., find the difefrence in CSI in each location due to the Columbian Exchange.

In [19]:
post1500-pre1500
---------------------------------------------------------------------------
RasterGeoTWarning                         Traceback (most recent call last)
Input In [19], in <cell line: 1>()
----> 1 post1500-pre1500

File ~/anaconda3/envs/EconGrowthUG/lib/python3.9/site-packages/georasters/georasters.py:436, in GeoRaster.__sub__(self, other)
    435 def __sub__(self, other):
--> 436     return self+other.__neg__()

File ~/anaconda3/envs/EconGrowthUG/lib/python3.9/site-packages/georasters/georasters.py:420, in GeoRaster.__add__(self, other)
    418 if isinstance(other, GeoRaster):
    419     if self.geot != other.geot:
--> 420         raise RasterGeoTWarning("Rasters do not have same geotransform. \
    421                                 If needed first create union or allign them.")
    422     if self.nodata_value == other.nodata_value:
    423         ndv = self.nodata_value

RasterGeoTWarning: Rasters do not have same geotransform.                                         If needed first create union or allign them.

So, why does it fail? In this case, basically because the rasters do not have the same size (see above). More generally, it could be that the rasters do not have the same geographical settings. So, we must first make them amenable for each other. Luckily, georasters has a functioon for that.

In [20]:
(pre1500n, post1500n) = gr.align_georasters(pre1500, post1500)

Let's see that now the new georasters align correctly and so can be used for analysis.

In [21]:
print('The size of the pre-1500 raster is', pre1500n.shape)
print('The size of the post-1500 raster is', post1500n.shape)
The size of the pre-1500 raster is (2084, 4320)
The size of the post-1500 raster is (2084, 4320)
In [22]:
print('The geographical properties of the new pre-1500 raster are', pre1500n.geot)
print('The minimum latitude is', pre1500n.ymin, 'and the maximum latitude is', pre1500n.ymax, 'for the new pre-1500 raster.')
print('The minimum longitude is', pre1500n.xmin, 'and the maximum longitude is', pre1500n.xmax, 'for the new pre-1500 raster.')
print('The cells of the new pre-1500 raster are', np.abs(pre1500n.y_cell_size), 'degrees north-south', 'and', np.abs(pre1500n.x_cell_size), 'east-west.')
The geographical properties of the new pre-1500 raster are (-180.0, 0.08333333333333333, 0.0, 83.66666666666667, 0.0, -0.08333333333333333)
The minimum latitude is -89.99999999999999 and the maximum latitude is 83.66666666666667 for the new pre-1500 raster.
The minimum longitude is -180.0 and the maximum longitude is 180.0 for the new pre-1500 raster.
The cells of the new pre-1500 raster are 0.08333333333333333 degrees north-south and 0.08333333333333333 east-west.
In [23]:
print('The geographical properties of the new post-1500 raster are', post1500n.geot)
print('The minimum latitude is', post1500n.ymin, 'and the maximum latitude is', post1500n.ymax, 'for the new post-1500 raster.')
print('The minimum longitude is', post1500n.xmin, 'and the maximum longitude is', post1500n.xmax, 'for the new post-1500 raster.')
print('The cells of the new post-1500 raster are', np.abs(post1500n.y_cell_size), 'degrees north-south', 'and', np.abs(post1500n.x_cell_size), 'east-west.')
The geographical properties of the new post-1500 raster are (-180.0, 0.08333333333333333, 0.0, 83.66666666666667, 0.0, -0.08333333333333333)
The minimum latitude is -89.99999999999999 and the maximum latitude is 83.66666666666667 for the new post-1500 raster.
The minimum longitude is -180.0 and the maximum longitude is 180.0 for the new post-1500 raster.
The cells of the new post-1500 raster are 0.08333333333333333 degrees north-south and 0.08333333333333333 east-west.

The new rasters have the same sizes and the same geographical settings. Yay! Now, let's compute the change in CSI due to the Columbian Exchange.

In [24]:
colex = post1500n - pre1500n

Let's find some stats on the change in CSI

In [25]:
print('Average change in CSI in the world pre-1500 is', colex.mean())
print('Median change in CSI in the world pre-1500 is', colex.median())
print('Maximum change in CSI in the world pre-1500 is', colex.max())
print('Minimum change in CSI in the world pre-1500 is', colex.min())
print('Standard deviation of change in CSI in the world pre-1500 is', colex.std())
Average change in CSI in the world pre-1500 is 866.3604811874781
Median change in CSI in the world pre-1500 is 0.0
Maximum change in CSI in the world pre-1500 is 16169.150390625
Minimum change in CSI in the world pre-1500 is 0.0
Standard deviation of change in CSI in the world pre-1500 is 1848.7616487285682

Let's plot the change so we can see which regions in the world gained the most.

In [26]:
myraster = colex
cmap = plt.cm.YlGn
norm = mpl.colors.Normalize(vmin=myraster.min(), vmax=myraster.max())
df3 = countries.copy()
df3.geometry = countries.boundary
df3['fake'] = 0
plt.figure(figsize=(15,10))
plt.xticks([])
plt.yticks([])
plt.title('')
ax =plt.gca()
ax.set_aspect(1)
img_extent = (myraster.xmin, myraster.xmax, myraster.ymin, myraster.ymax)
ax.imshow(myraster.raster, norm=norm, origin='upper',extent=img_extent, cmap=cmap, interpolation='bilinear', aspect=1)
df3.plot(ax=ax, color='black', edgecolor='k', linewidth=0.5, rasterized=True)
# create axes instance for colorbar on bottom. 
ax = plt.gca()
pos = ax.get_position() 
l, b, w, h = pos.bounds 
cax = plt.axes([l+.3, b+0.05, .3, 0.01]) 
# draw colorbar on bottom. 
cb = mpl.colorbar.ColorbarBase(cax, cmap=cmap, norm=norm, spacing='proportional', orientation='horizontal')
cax.set_title('Difference in Calories Pre- vs Post-1500')
plt.savefig(pathgraphs + 'ColExMaxCaloriesBorders.pdf', dpi=150, bbox_inches='tight')
plt.show()