This page provides an introduction to use the SNAP Java API from Python.
SNAP implementation language is Java and therefore SNAP's "native" API is a Java API. According to the SNAP architecture, the reference SNAP Java API documentation of the two SNAP sub-systems SNAP Engine and SNAP Desktop nearly fully applies to use from Python as well.
It is possible to use the SNAP Java API from Python and there are basically two different ways to achieve this:
- Use an standard Python (CPython) installation
For either way, it is possible to call SNAP code from your Python programs/scripts and to extend SNAP by plugins written in Python. However, the two ways have pros and cons which should be considered before you make a decision which you want to follow.
Use the standard Python (CPython) approach, if
- you require using the Python scientific extension libraries such as numpy, scipy, matplotlib, etc.;
- you already have CPython code and you want to incorporate SNAP functions;
- you plan to implement a fast data processor plugin in Python;
- you do not plan to develop SNAP Desktop user interface extensions;
- you do not require full portability on all platforms;
- your code has (or will have) dependencies on a lot of non-standard libraries.
With the standard Python approach extension of SNAP is currently limited to raster data processor (
Use the Jython approach, if
- you plan to develop SNAP Desktop user interface extensions;
- you require full portability on all platforms;
- do not require efficient array/raster data processing as provided by numpy and the like (because Jython does yet not well support these);.
Once you made your decision which approach to take, you can head to the dedicated section below.
This approach is explained on the Jython Approach page.
Standard Python (CPython) Approach
With this approach you can use a standard Python (CPython) interpreter installed on your computer (SNAP does not include a CPython interpreter.) The supported versions are Python 2.7, 3.3 to 3.6 64-bit (Linux + Darwin) and both 32-bit and 64-bit (Windows) as well as Anaconda distributions.
Please note that you must use a 32-bit Python if your SNAP installation is 32-bit and accordingly use a 64-bit Python if your SNAP installation is 64-bit.
SNAP provides the Python module
snappy which allows you to access the SNAP Java API from Python.
snappy requires either a SNAP installation or a SNAP build.
For how to configure Python for SNAP please see Configure Python to use the SNAP-Python (snappy) interface
Examples of SNAP API usage from Python
The following first example reads some raster data and displays/stores raster data as an RGB image:
from snappy import ProductIO
import numpy as np
import matplotlib.pyplot as plt
p = ProductIO.readProduct('snappy/testdata/MER_FRS_L1B_SUBSET.dim')
rad13 = p.getBand('radiance_13')
w = rad13.getRasterWidth()
h = rad13.getRasterHeight()
rad13_data = np.zeros(w * h, np.float32)
rad13.readPixels(0, 0, w, h, rad13_data)
rad13_data.shape = h, w
imgplot = plt.imshow(rad13_data)
Due to numerous writers implemented in the SNAP Engine reading and writing data in snappy is relatively simple. The general syntax is:
from snappy import ProductIO
p = ProductIO.readProduct('snappy/testdata/MER_FRS_L1B_SUBSET.dim') # read product
ProductIO.writeProduct(p, '<your/out/directory>', '<write format>') # write product
Get and Set
It makes sense to start exploring the capabilities of snappy by reading a product (see above: snappy 1st Contact) and try out the methods and fields it contains. Calling get on an object in snappy (e.g.
p.getRasterHeight()) returns either an Integer or String value or an object. Upon the latter, you can again normally call get again. This might again return an object with fields to return.
In the other direction, you may also set fields. For instance, if you called get and received a String value, you may set the same field of this object with a String.
rad13 = p.getBand('radiance_13')
Sometimes you might only be interested in “copying” a field from one object to another, let’s say from a source band to a target band. Then you may simply call get and set in one line without interacting with the return of get:
target_band = target_product.getBand('some_output_band_based_on_rad13')
target_band.setNoDataValue(rad13.getNoDataValue()) # no data value from the source Band object rad13
Often, snappy returns Integers (e.g. p.getSceneRasterWidth()) or Strings (e.g. p.getName()):
name = p.getName()
width = p.getSceneRasterWidth()
band_names = p.getBandNames()
However, you might sometimes walk into Java objects that appear foreign to a Python user. These can e.g. be objects such as the return of:
In this case, you simply convert this to a Python list by calling:
bands = list(p.getBandNames())
>>> ['Oa01_reflectance', 'Oa02_reflectance', 'Oa03_reflectance', 'Oa04_reflectance']
>>> <class 'list'>
In other cases, using str()converts a Java String to a Python String. For instance, you might be interested in the ‘autogrouping’ of the product. It defines how the bands are grouped in the product. The return of:
This can be easily converted to a Python String:
autogrouping = str(p.getAutoGrouping())
>>> <class 'str'>
You may set the value by:
p.setAutoGrouping('<a String correctly formatted to be recognized as autogrouping>')
Processing in snappy
Snappy generally offers to ways how to process data:
Option A is suited when you aim at using only SNAP Engine Operators.
Option B is suited when you aim at doing custom computations for which you need to read data into Python numpy arrays.
Both options can, of course, occur in one workflow.
Option A: Process a product using a SNAP Engine Operator and write the target product
SNAP Operators are available in snappy via
GPF.createProduct(). Its first parameter is a String denoting the name of the Operator as denoted in the Engine and available via GPT. If you have added GPT to your environment variables, you may call GPT from cmd in order to check out the available Operators, their description and parameters. In snappy, we provide the parameters through the second parameter of
GPF.createProduct(). This parameter is a Java Hashmap, an object that is equivalent to a Python dictionary. The parameters must be named exactly with the String parameter name provided in GPT.
# 1. Imports
from snappy import ProductIO
from snappy import GPF
from snappy import Hashmap
# 2. Fill parameter Hashmap
parameters = Hashmap()
# 3. Call Operator
operator_name = 'Resample'
target_product = GPF.createProduct(operator_name, parameters, p)
# 4. Write Product
# Write target Product with ProductIO:
write_format = 'BEAM-DIMAP' # in this case write as BEAM-DIMAP
ProductIO.writeProduct(target_product , <'your/out/directory'>, write_format)
# Alternative solution: Computations are faster when using GPF to write the product instead of ProductIO:
incremental = false # most writer don't support the incremental writing mode (update exsiting file), except BEAM-DIMAP.
GPF.writeProduct(target_product , File(<'your/out/directory'>), write_format, incremental, ProgressMonitor.NULL)
Instead of the NULL progress monitor, you can use a different monitor if you like to receive progress messages on the command line.
PWPM = jpy.get_type('com.bc.ceres.core.PrintWriterProgressMonitor')
JavaSystem = jpy.get_type('java.lang.System')
monitor = PWPM(JavaSystem.out)
pm = createProgressMonitor()
GPF.writeProduct(target_product , File(<'your/out/directory'>), write_format, incremental, pm)
Option B: Process a product using custom data computations in Python
Using an implemented Operator might not be enough in cases where you aim to implement your own computation. The methods
writePixels() help you to retrieve the necessary for the computation and save the result. We recommend to completely set up your target product in your script before computation starts. As in the examples above, p is still our source Product.
# 1. Imports
from snappy import ProductUtils
width = p.getSceneRasterWidth() # often the target product dimensions are the same as the source product dimensions
height = p.getSceneRasterHeight()
target_product = snappy.Product('My target product', 'The type of my target product', width, height)
# 2. Optional: Copy or set meta information
# It is also possible to target specific fields. Just one example:
target_product.setDescription('Product containing very valuable output bands')
# 3. Set product writer
# Set the writer with the write_format defined above (here: 'BEAM-DIMAP'):
# 4. Add and configure target Bands
# Now, you could copy bands form the source product if you are interested in writing them to the target product as well. Check out ProductUtils.copyBand() regarding this task.
# Before starting our computations, we must create the computed bands of our target Product:
band_name = 'an_output_band_name'
target_band = target_product.addBand(band_name, snappy.ProductData.TYPE_FLOAT32)
# further configure the created band:
nodata_value = p.getBand(<'source_band_name'>).getNoDataValue()
# You could set values of other fields, some might be important for creating an output suiting your expectations.
# 5. Write header
# All the structure and meta information we just added to the target_product are still in memory. Hence, we must write its header before writing data. The single argument of writeHeader() is
# the absolute path to the expected product without file extension.
# The last String of this path is the target Product name as it is being written.
The target Product is ready for data to be written to it. Check out snappy examples on Github in order to know how to use
writePixels() for reading data tiles, rows, columns or single pixels into numpy arrays and write them from output arrays into the respective band of the target Product.
It is worth to note that the order of the parameters width and height is switched from what Python users are familiar with from packages like numpy.
After having created the target product we could now adjust its metadata, e.g. by copying it from the source product using ProductUtils. This is an optional step.
Further code examples
More example code of how to use the SNAP API in Python can be found in
<snappy-dir>/examples. There is also a directory
<snappy-dir>/testdata with a single EO test data product (
*.dim) in it which you can pass as argument to the various examples. Please try
$ cd <snappy-dir>/examples
$ <python-exe> snappy_ndvi.py ../testdata/MER_FRS_L1B_SUBSET.dim
ProductIOPlugInManager = snappy.jpy.get_type('org.esa.snap.framework.dataio.ProductIOPlugInManager')
it = ProductIOPlugInManager.getInstance().getAllReaderPlugIns()
Also in the snap-examples
repository are some more example snippets to find.
Python Operator Plugins
Python can be used extend SNAP by new raster data processor plugins, i.e. operators. Operators can be executed from the command-line, or invoked from the SNAP Desktop GUI, and be used as nodes in processing XML graphs. In the future, the SNAP development team will open more extension points for Python developers.
Example code for an operator plugin version of the
snappy_ndvi.py example from above is provided on GitHub: https://github.com/senbox-org/snap-examples/tree/master/snap-engine-python-operator
How to Configure SNAP to pick up the build output automatically
etc folder in the SNAP installation directory. Inside this directory you will find the
snap.conf file. Change the access right of it so that you are allowed to make changes to it. There you will find the
extra_clusters property. Specify the path, to the cluster folder of the build output directory.
Ensure to remove the '#' character at the beginning of the line.
Now when you start SNAP the build output is automatically used by SNAP and you can test the latest builds.