Integrating external data from a CSV

• Products used: ls8_sr

Keywords data used; landsat 8, external data, data methods; interpolation, data used; csv, intertidal, coastal, water

Background

It is often useful to combine external data (e.g. from a CSV or other dataset) with data loaded from Digital Earth Africa. For example, we may want to combine data from a tide or river guage with satellite data to determine water levels at the exact moment each satellite observation was made. This can allow us to manipulate and extract satellite data and obtain additional insights using data from our external source.

Description

This example notebook loads in a time series of external tide modelling data from a CSV, and combines it with satellite data loaded from Digital Earth Africa. This workflow could be applied to any external time series dataset (e.g. river guages, tide guagues, rainfall measurements etc). The notebook demonstrates how to:

1. Load a time series of Landsat 8 data

2. Load in external time series data from a CSV

3. Convert this data to an xarray.Dataset and link it to each satellite observation by interpolating values at each satellite timestep

4. Add this new data as a variable in our satellite dataset, and use this to filter satellite imagery to high and low tide imagery

5. Demonstrate how to swap dimensions between time and the new tide_height variable

---

Getting started

To run this analysis, run all the cells in the notebook, starting with the “Load packages” cell.

Load packages

%matplotlib inline

import datacube
import pandas as pd
import matplotlib.pyplot as plt

from deafrica_tools.datahandling import load_ard, mostcommon_crs

pd.plotting.register_matplotlib_converters()

Connect to the datacube

dc = datacube.Datacube(app='Integrating_external_data')

Loading satellite data

First we load in three years of Landsat 8 data. We will use Msasani Bay in Tanzania for this demonstration, as we have a CSV of tide height data for this location that we wish to combine with satellite data. We load a single band nir which clearly differentiates between water (low values) and land (higher values). This will let us verify that we can use external tide data to identify low and high tide satellite observations.

lat, lon = -6.745, 39.261
buffer = 0.06

# Create a reusable query
query = {
    'x': (lon-buffer, lon+buffer),
    'y': (lat+buffer, lat-buffer),
    'time': ('2015-01-01', '2017-12-31'),
    'measurements': ['nir'],
    'resolution': (-30, 30)
}

# Identify the most common projection system in the input query
output_crs = mostcommon_crs(dc=dc, product='ls8_sr', query=query)

# Load Landsat 8 data
ds = load_ard(dc=dc,
              products=['ls8_sr'],
              output_crs=output_crs,
              align=(15, 15),
              group_by='solar_day',
              **query)

print(ds)

Using pixel quality parameters for USGS Collection 2
Finding datasets
    ls8_sr
Applying pixel quality/cloud mask
Re-scaling Landsat C2 data
Loading 65 time steps
<xarray.Dataset>
Dimensions:      (time: 65, y: 443, x: 443)
Coordinates:
  * time         (time) datetime64[ns] 2015-01-07T07:32:04.385652 ... 2017-12...
  * y            (y) float64 -7.389e+05 -7.39e+05 ... -7.522e+05 -7.522e+05
  * x            (x) float64 5.222e+05 5.222e+05 ... 5.354e+05 5.355e+05
    spatial_ref  int32 32637
Data variables:
    nir          (time, y, x) float32 nan nan nan ... 0.006003 0.006718 0.006443
Attributes:
    crs:           EPSG:32637
    grid_mapping:  spatial_ref

Integrating external data

Load in a CSV of external data with timestamps

In the code below, we aim to take a CSV file of external data (in this case, half-hourly tide heights for a location in Msasani Bay, Tanzania for a five year period between 2014 and 2018 generated using the OTPS TPXO8 tidal model), and link this data back to our satellite data timeseries.

We can load the existing tide height data using the pandas module which we imported earlier. The data has a column of time values, which we will set as the index column (roughly equivelent to a dimension in xarray).

tide_data = pd.read_csv('../Supplementary_data/Integrating_external_data/msasani_bay_-6.749102_39.261819_2014-2018_tides.csv',
                          parse_dates=['time'],
                          index_col='time')

tide_data.head()

	tide_height
time
2014-01-01 00:00:00	1.798
2014-01-01 00:30:00	1.885
2014-01-01 01:00:00	1.867
2014-01-01 01:30:00	1.743
2014-01-01 02:00:00	1.521

Link external data to satellite data

Now that we have the tide height data, we need to estimate the tide height for each of our satellite images. We can do this by interpolating between the data points we do have (hourly measurements) to get an estimated tide height for the exact moment each satellite image was taken.

The code below generates an xarray.Dataset with the same number of timesteps as the original satellite data. The dataset has a single variable tide_height that gives the estimated tide hight for each observation:

# First, we convert the data to an xarray dataset so we can analyse it
# in the same way as our satellite data
tide_data_xr = tide_data.to_xarray()

# We want to convert our hourly tide heights to estimates of exactly how
# high the tide was at the time that each satellite image was taken. To
# do this, we can use `.interp` to 'interpolate' a tide height for each
# satellite timestamp:
satellite_tideheights = tide_data_xr.interp(time=ds.time)

# Print the output xarray object
print(satellite_tideheights)

<xarray.Dataset>
Dimensions:      (time: 65)
Coordinates:
  * time         (time) datetime64[ns] 2015-01-07T07:32:04.385652 ... 2017-12...
    spatial_ref  int32 32637
Data variables:
    tide_height  (time) float64 -1.401 -1.382 -0.7692 ... 0.4866 -0.2992 -0.3579

Now we have our tide heights in xarray format that matches our satellite data, we can add this back into our satellite dataset as a new variable:

# We then want to put these values back into the Landsat dataset so
# that each image has an estimated tide height:
ds['tide_height'] = satellite_tideheights.tide_height

We can now plot tide heights for our satellite data:

# Plot the resulting tide heights for each Landsat image:
ds.tide_height.plot()
plt.show()

Filtering by external data

Now that we have a new variable tide_height in our dataset, we can use xarray indexing methods to manipulate our data using tide heights (e.g. filter by tide to select low or high tide images):

# Select a subset of low tide images (tide less than -1.0 m)
low_tide_ds = ds.sel(time = ds.tide_height < -1.0)

# Select a subset of high tide images (tide greater than 0.1 m)
high_tide_ds = ds.sel(time = ds.tide_height > 0.1)

We can plot the median of the filtered images for low and high tide to examine the difference between them. Taking the median allows us to view a cloud-free image.

For the low-tide images:

# Compute the median of all low tide images to produce a single cloud free image.
low_tide_composite = low_tide_ds.median(dim='time', skipna=True, keep_attrs=False)

# Plot the median low tide image
low_tide_composite.nir.plot(robust=True)
plt.title('Low tide')
plt.show()

and for the high-tide images:

# Compute the median of all high tide images to produce a single cloud free image.
high_tide_composite = high_tide_ds.median(dim='time', skipna=True, keep_attrs=False)

# Plot the median high tide image
high_tide_composite.nir.plot(robust=True)
plt.title('High tide')
plt.show()

Note that many tidal flat areas in the low-tide image (blue-green) now appear inundated by water (dark blue/purple) in the high-tide image.

Swapping dimensions based on external data

By default, xarray uses time as one of the main dimensions in the dataset (in addition to x and y). Now that we have a new tide_height variable, we can change this to be an actual dimension in the dataset in place of time. This enables additional more advanced operations, such as calculating rolling statistics or aggregating by tide_heights.

In the example below, you can see that the dataset now has three dimensions (tide_height, x and y). The dimension time is no longer a dimension in the dataset.

print(ds.swap_dims({'time': 'tide_height'}))

<xarray.Dataset>
Dimensions:      (tide_height: 65, y: 443, x: 443)
Coordinates:
    time         (tide_height) datetime64[ns] 2015-01-07T07:32:04.385652 ... ...
  * y            (y) float64 -7.389e+05 -7.39e+05 ... -7.522e+05 -7.522e+05
  * x            (x) float64 5.222e+05 5.222e+05 ... 5.354e+05 5.355e+05
    spatial_ref  int32 32637
  * tide_height  (tide_height) float64 -1.401 -1.382 -0.7692 ... -0.2992 -0.3579
Data variables:
    nir          (tide_height, y, x) float32 nan nan nan ... 0.006718 0.006443
Attributes:
    crs:           EPSG:32637
    grid_mapping:  spatial_ref

---

Additional information

License: The code in this notebook is licensed under the Apache License, Version 2.0. Digital Earth Africa data is licensed under the Creative Commons by Attribution 4.0 license.

Contact: If you need assistance, please post a question on the Open Data Cube Slack channel or on the GIS Stack Exchange using the open-data-cube tag (you can view previously asked questions here). If you would like to report an issue with this notebook, you can file one on Github.

Compatible datacube version:

print(datacube.__version__)

1.8.7

Last Tested:

from datetime import datetime
datetime.today().strftime('%Y-%m-%d')

'2022-06-21'