rasterstats 0.9.0

Statistics

By default, the zonal_stats function will return the following statistics

• min

• max

• mean

• count

Optionally, these statistics are also available

• sum

• std

• median

• majority

• minority

• unique

• range

• nodata

• percentile (see note below for details)

You can specify the statistics to calculate using the stats argument:

>>> stats = zonal_stats("tests/data/polygons.shp",
                         "tests/data/elevation.tif",
                         stats=['min', 'max', 'median', 'majority', 'sum'])

>>> # also takes space-delimited string
>>> stats = zonal_stats("tests/data/polygons.shp",
                         "tests/data/elevation.tif",
                         stats="min max median majority sum")

Note that certain statistics (majority, minority, and unique) require significantly more processing due to expensive counting of unique occurences for each pixel value.

You can also use a percentile statistic by specifying percentile_<q> where <q> can be a floating point number between 0 and 100.

User-defined Statistics

You can define your own aggregate functions using the add_stats argument. This is a dictionary with the name(s) of your statistic as keys and the function(s) as values. For example, to reimplement the mean statistic:

from __future__ import division
import numpy as np

def mymean(x):
    return np.ma.mean(x)

then use it in your zonal_stats call like so:

stats = zonal_stats(vector, raster, add_stats={'mymean':mymean})

Specifying Geometries

In addition to the basic usage above, rasterstats supports other mechanisms of specifying vector geometries.

It integrates with other python objects that support the geo_interface (e.g. Fiona, Shapely, ArcPy, PyShp, GeoDjango):

>>> import fiona

>>> # an iterable of objects with geo_interface
>>> lyr = fiona.open('/path/to/vector.shp')
>>> features = (x for x in lyr if x['properties']['state'] == 'CT')
>>> zonal_stats(features, '/path/to/elevation.tif')
...

>>> # a single object with a geo_interface
>>> lyr = fiona.open('/path/to/vector.shp')
>>> zonal_stats(lyr.next(), '/path/to/elevation.tif')
...

Or by using with geometries in “Well-Known” formats:

>>> zonal_stats('POINT(-124 42)', '/path/to/elevation.tif')
...

Feature Properties

By default, an __fid__ property is added to each feature’s results. None of the other feature attributes/proprties are copied over unless copy_properties is set to True:

>>> stats = zonal_stats("tests/data/polygons.shp",
                         "tests/data/elevation.tif"
                         copy_properties=True)

>>> stats[0].has_key('name')  # name field from original shapefile is retained
True

Rasterization Strategy

There is no right or wrong way to rasterize a vector. The default strategy is to include all pixels along the line render path (for lines), or cells where the center point is within the polygon (for polygons). Alternatively, you can opt for the all_touched strategy which rasterizes the geometry by including all pixels that it touches. You can enable this specifying:

>>> zonal_stats(..., all_touched=True)

The figure above illustrates the difference; the default all_touched=False is on the left while the all_touched=True option is on the right. Both approaches are valid and there are tradeoffs to consider. Using the default rasterizer may miss polygons that are smaller than your cell size resulting in None stats for those geometries. Using the all_touched strategy includes many cells along the edges that may not be representative of the geometry and may give severly biased results in some cases.

Working with categorical rasters

You can treat rasters as categorical (i.e. raster values represent discrete classes) if you’re only interested in the counts of unique pixel values.

For example, you may have a raster vegetation dataset and want to summarize vegetation by polygon. Statistics such as mean, median, sum, etc. don’t make much sense in this context (What’s the sum of oak + grassland?).

The polygon below is comprised of 12 pixels of oak (raster value 32) and 78 pixels of grassland (raster value 33):

>>> zonal_stats(lyr.next(), '/path/to/vegetation.tif', categorical=True)
[{'__fid__': 1, 32: 12, 33: 78}]

rasterstats will report using the pixel values as keys. To associate the pixel values with their appropriate meaning (for example oak instead of 32), you can use a category_map:

>>> cmap = {32: 'oak', 33: 'grassland'}
>>> zonal_stats(lyr.next(), '/path/to/vegetation.tif',
                categorical=True, category_map=cmap)
[{'__fid__': 1, 'oak': 12, 'grassland': 78}]

“Mini-Rasters”

Internally, we create a masked raster dataset for each feature in order to calculate statistics. Optionally, we can include these data in the output of zonal_stats using the raster_out argument:

stats = zonal_stats(vector, raster, raster_out=True)

Which gives us three additional keys for each feature:

mini_raster     | Numpy ndarray
mini_raster_GT  | Six-tuple defining the geotransform (GDAL ordering)
mini_raster_NDV | Nodata value in the returned array

Keep in mind that having ndarrays in your stats dictionary means it is more difficult to serialize to json and other text formats.