Skip to content

Latest commit

 

History

History
110 lines (72 loc) · 6.53 KB

README.md

File metadata and controls

110 lines (72 loc) · 6.53 KB
Raw

eDNA-metadata

A Jupyter notebook tool for reproducibly gathering sample site metadata through the Google Earth Engine API.

Input: a CSV file with the coordinates of your environmental DNA sample locations and a list of all the datasets you want to sample

Output: a CSV file with the value calculated from each dataset for each sample site

Usage

You will need a table of sample site locations in CSV format with columns id, latitude, and longitude (see sample_input.csv).

Decide which datasets you want to use. You can browse them in the Google Earth Engine data catalog. To access the dataset you will need its code snippet, which is displayed under 'Earth Engine Snippet' on the dataset's page.

Getting started

You should run the notebook in a conda environment as is recommended for both Jupyter Notebook and Google Earth Engine.

Install conda if you don't have it already, following the instructions here.

Make sure you are in the eDNA-metadata/ directory. Create a new conda environment from the included environment.yml file, which will install all the necessary packages:

conda env create -f environment.yml

You should now have a new conda environment called ee. Activate it:

conda activate ee

Optionally, install jupyter notebook extensions, which adds some nice features like collapsing sections:

conda install -c conda-forge jupyter_contrib_nbextensions

You will need to do a one-time authentication which links Earth Engine to your Google account. Run:

earthengine authenticate

and follow the resulting instructions.

Run Jupyter notebook within the environment:

jupyter notebook

Make sure that the environment is activated before running jupyter notebook. If it's not, you could end up running the version of Jupyter notebook installed in your base environment, which won't have access to the packages you just installed.

When the Jupyter notebook tab opens, open metadata.ipynb, and go through the sections of the notebook to set up your datasets.

Raster datasets (Earth Engine Image datasets)

To load an ee.Image dataset, use the RasterDataset class in metadata.ipynb. For example, suppose we are interested in getting data of the soil pH at surface level from the OpenLandMap Soil pH in H20 dataset:

Instantiate a RasterDataset for it, using the desired band (b0 is the pH at 0 cm depth; other bands are for other depths), and giving it a descriptive name (optional):

RasterDataset(
    snippet=ee.Image("OpenLandMap/SOL/SOL_PH-H2O_USDA-4C1A2A_M/v02"),
    band='b0',
    name='soil bulk density'
)

Here is a false-color view of the dataset:

soil ph map

Note that if you zoom in, pixelation becomes visible, and the data does not always cover all the way to the coastline:

soil ph zoom

This is inherent to the raster format. The resolution of this dataset is 250 meters. You should be aware of the resolution you are working with and consider that in choosing a sample area radius and in reporting results. It's possible to reproject a raster dataset to a lower or higher resolution, but it's not clear if this is always the best thing to do. See the Earth Engine pages on scale and projections for more information.

Raster dataset series (Earth Engine ImageCollection datasets)

Some datasets in Earth Engine consist of a collection of many images. Usually these are overlapping images from different points in time, such as from satellite data.

The individual Images in a collection often do not fully cover the area of interest. So, you may need to experiment with different date ranges in order to get adequate coverage. For example, suppose we are interested in the Landsat 8 NDVI dataset. Here is the dataset coverage over the continental US over one day, one month, and one year:

date_range=('2019-01-01', '2019-01-02') date_range=('2019-01-01', '2019-02-01') date_range=('2019-01-01', '2019-12-31')

Let's say the one-month coverage is sufficient. Instantiate a RasterDataset using that date range:

RasterDataset(
    snippet=ee.ImageCollection("LANDSAT/LC08/C01/T1_8DAY_NDVI"),
    band='NDVI',
    date_range=('2019-01-01', '2019-02-01')
)

If the collection has more than one image in your chosen date range, they will be merged into one image using the ee.ImageCollection.mosaic() algorithm, and will be treated as one dataset. If you want to compare data from the same collection at different points in time, instantiate a RasterDataset for each date range. If you don't care so much about the time the imagery was taken, you can use a broad date range such as one year, which should provide good coverage.

You can also apply Earth Engine API functions to the dataset before passing it to the constructor. For example:

RasterDataset(
    snippet=ee.ImageCollection("LANDSAT/LC08/C01/T1_8DAY_NDVI").median(),
    band='NDVI',
    date_range=('2019-01-01', '2019-02-01')
)

Here we use the built-in median reducer function over the dataset, which returns an Image where each pixel's value is the median value across all images in the ImageCollection that cover that point. See the notebook for more examples.

Vector datasets (Earth Engine FeatureCollection datasets)

Less commonly, data is available in vector format as an ee.FeatureCollection object. For example, the EPA Level III Ecoregions dataset is in this format. To use it, instantiate a VectorDataset:

VectorDataset(
        snippet=ee.FeatureCollection('EPA/Ecoregions/2013/L3'),
        property='us_l3name'
    )

FeatureCollections have properties rather than bands.