The main goal of this library is to generalize methods that work in lower dimensions to higher-dimensional data.
Multi-dimensional data often arise as spatio-temporal datacubes,
e.g. climate data or time series of geospatial satellite data.
Many data analysis methods are designed to work on single images
or time series at a single point.
nd
makes it easy to broadcast these methods across a whole dataset,
adding additional features such as automatic parallelization.
Examples include
- pixelwise change detection algorithms
- reprojection between coordinate systems
- machine learning algorithms
nd
is built on xarray
.
Internally, all data are passed around as xarray
Datasets and all provided methods expect this format as inputs.
An xarray.Dataset
is essentially a Python representation of the NetCDF file format and as such easily reads/writes NetCDF files.
nd
is making heavy use of the xarray
and rasterio
libraries.
The GDAL library is only used via rasterio
as a compatibility layer to enable reading supported file formats.
nd.open_dataset
may be used to read any NetCDF file or any GDAL-readable file into an xarray.Dataset
.
Read the Documentation for detailed user guides.
You can also have a look at these two example notebooks:
pip install nd
It is recommended that you have GDAL available before installation and also make sure to have the correct environment variable set:
export GDAL_DATA=$(gdal-config --datadir)
Note that the following algorithms require the libgsl-dev
C library to be installed:
nd.change.OmnibusTest
xarray
provides all data structures required for dealing with n
-dimensional data in Python. nd
explicitly does not aim to add additional data structures or file formats.
Rather, the aim is to bring the various corners of the scientific ecosystem in Python closer together.
As such, nd
adds functionality to more seamlessly integrate libraries like xarray
, rasterio
, scikit-learn
, etc.
For example:
-
nd
allows to reproject an entire multivariate and multi-temporal dataset between different coordinate systems by wrappingrasterio
methods. -
nd
provides a wrapper forscikit-learn
estimators to easily apply classification algorithms to raster data.
Additionally, nd
contains a growing library of algorithms that are especially useful for spatio-temporal datacubes, for example:
-
change detection algorithms
-
spatio-temporal filters
Since xarray
is our library of choice for representing geospatial raster data, this is also an attempt to promote the use of xarray
and the NetCDF file format in the Earth Observation community.
NetCDF (specifically NetCDF-4) is a highly efficient file format that was built on top of HDF5. It is capable of random access which ties in with indexing and slicing in numpy
.
Because slices of a large dataset can be accessed independently, it becomes feasible to handle larger-than-memory file sizes. NetCDF-4 also supports data compression using zlib
. Random access capability for compressed data is maintained through data chunking.
Furthermore, NetCDF is designed to be fully self-descriptive. Crucially, it has a concept of named dimensions and coordinates, can store units and arbitrary metadata.
For feature requests and bug reports please submit an issue on the Github repository.