Hack Night - Xarray tutorial - Lvl: basic intro

Author: Sai Nudurupati 19Oct17

Material presented here is extensively mined (copied with permission) from the tutorial (https://github.com/geohackweek/tutorial_contents/blob/master/nDarrays/notebooks/ndarrays_intro.ipynb) presented by Joe Hamman (github: jhamman) at Geohack week (Sep 17) at University of Washington. I have also referred online documentation for 'xarray' (http://xarray.pydata.org/en/stable/)



In [ ]:

    
# Ignore warnings
import warnings; warnings.simplefilter('ignore')



In [ ]:

    
%matplotlib inline

N-dimensional Numpy Arrays



In [ ]:

    
import numpy as np

sst = np.random.random(size=(2, 4, 4))

print sst
# print(sst[:, ::1, ::3])  # indexing



In [ ]:

    
sst.shape



In [ ]:

    
print(sst[:, ::1, ::3])

Pandas dataframe



In [ ]:

    
import pandas as pd

df = pd.read_csv('https://climate.nasa.gov/system/internal_resources/details/original/647_Global_Temperature_Data_File.txt',
                 sep=r"\s*", names=['year', '1yr', '5yr'], index_col='year')

df.loc[1984: ]



In [ ]:

    
df.plot()

The network Common Data Format (netCDF)

collection of self-describing, machine-independent binary data formats and software tools that facilitate the creation, access and sharing of data stored in N-dimensional arrays

Xarray

extends some of core functionality of the Pandas library to N-dimensional arrays
can be used for:
- multidimensional data (e.g. climate data: x, y, z, time)
- structured data on a regular grid
- data in netCDF format
Two main data structures of xarray
- DataArray
- Dataset

DataArray

xarray's implementation of a labeled, multi-dimensional array
the DataArray has these key properties:
- data: Ndimensional array(NumPy or dask)
- dims: dimension names for each axis
- coords: dictionary-like container of arrays that label each point, and
- attrs: ordered dictionary holding metadata

Dataset

xarray's multi=dimensional equivalent of a Pandas Dataframe
dict-like container of DataArray objects with aligned dimensions
Datasets have these key properties:
- dims: dictionary mapping from dimension names to the fixed length of each dimension,
- data_vars: dict-like container of DataArrays corresponding to data variables,
- coords: dictionary-like container of DataArrays intended to label points used in data_vars
- attrs: ordered dictionary holding metadata

Lets look at some datasets

Import xarray library



In [ ]:

    
import xarray as xr

Open a dataset



In [ ]:

    
ds = xr.open_dataset('3B43.20100101.7A.nc')

At this point Python is just scanning the contents of the file. It is not reading the data into its memory.

Dataset properties



In [ ]:

    
ds

Extracting DataArrays from a Dataset



In [ ]:

    
precipitation = ds['pcp']

Indexing this DataArray

traditional or the Matlab/Numpy way



In [ ]:

    
precipitation[0, 0, 0]

using .loc (positional indexing)



In [ ]:

    
ds['pcp'].loc['2010-01-01']

using .isel (refers to seleciton by integer position)



In [ ]:

    
ds['pcp'].isel(time=0, latitude=0, longitude=0)

Plotting data in 2 dimensions



In [ ]:

    
map_data = ds['pcp'].sel(time='2010-01-01')



In [ ]:

    
map_data.plot()

Want to customize it?



In [ ]:

    
import matplotlib.pyplot as plt
map_data.plot(cmap=plt.cm.Blues)
plt.title('Global Precipitation data')
plt.tight_layout()
plt.show()



In [ ]: