GEE Timeseries Exlorer

GEE Timeseries Exlorer is a QGIS plugin for interactive exploration of temporal raster data available via the Google Earth Engine (GEE) Data Catalog.

The GEE Timeseries Exlorer plugin adds a panel for selecting a GEE image collection and a plot widget for visualizing temporal profiles.

Installation

In QGIS select QGIS > Plugins > Manage and Install Plugins…, search for GEE Timeseries Exlorer and install the plugin.

Note: this plugin relies on the Earth Engine Python API. Search for Google Earth Engine and install this plugin as well. You may also want to have a look here in case of any issue whilst installing: https://gee-community.github.io/qgis-earthengine-plugin/ In order to access Earth Engine, you must first authorize your Google account for the use of the python API. This in turn requires a Google Account with access to Earth Engine.

  • In QGIS select QGIS > Settings > User Profiles > Open Active Profile Folder

  • Navigate to python > plugins > ee_plugin > extlibs_windows (_linux / _darwin on Linux/Mac systems respectively) and copy the full path to this directory

  • Open the OSGeo4W Shell / Terminal and run the following:

py3_env
python
import sys
sys.path.append(r"Paste path to directory here")
import ee
ee.Authenticate()

Follow the instructions in your browser and copy the authentication code into the OSGeo4W / Terminal.

Getting started

Overview

In the toolbar click icon to show the GEE Timeseries Exlorer panel.

_images/getting_started_1.PNG

There are two main panels: The Plot Window on the left and the Collection and Visualization settings on the right.

Load a Collection

The Collection tab on the right panel allows you to select a predefined Image Collection from the list and filter it by time interval and/or metadata properties. The subsection Collection Editor provides the python code used to access the imagery and enables full control on the code executed by the user, having access to the entire Earth Engine Data Catalogue (https://developers.google.com/earth-engine/datasets).

For a quick start, the code snippet for accessing the USGS Landsat 8 Surface Reflectance Tier 1 image collection is already prepared.

_images/getting_started_2.PNG

Click on the info button icon_info next to the collection list to open and inspect the USGS Landsat 8 Surface Reflectance Tier 1 description.

Visualize Time Series

This section displays the spectral time series of the user defined collection and time frame. Click on Activate point selection tool in the upper left corner of the plugin and select at least one spectral band from the list. Here we use the predefined collection Landsat TM-ETM-OLI Surface Reflectance Cloud Masked which combines the Landsat sensors into one collection and applies masking based on quality bands (hint: check the Collection Editor for details). Set Filter Date to 1984-01-01 to 2020-12-31 to make use of the entire Landsat archive. Furthermore it can be useful to edit Filter Property to only consider scenes with less than 70% cloud cover.

_images/getting_started_3.PNG

Then, click into the map canvas to select a point location and to read the temporal profile data.

_images/getting_started_4.PNG

When changing the date range or filtered metadata properties, click on icon_refresh Read point profile to re-read the current location.

In the top panel, click on data_icon to retrieve the displayed time series as raw data:

_images/getting_started_5.PNG

It provides access to the raw data with information on the unique image id, the geographic coordinates, the date of acquisition and the spectral values for all bands selected.

Visualize Single Images

Apart from plot based time series, you can visualize entire images and image aggregates. Open the Visualization tab on the right panel. In the subsection Temporal Binning you can specify a temporal window for which the visualisation will be rendered. By default, this corresponds to a single date, but can be expanded to a date range by increasing the Length parameters value and/or type (Day, Month, Year). The yellow vertical line (single date) or box (multi date) in the Plot Window illustrates the given choice graphically and can further be used to change the temporal window interactively by clicking inside the plot window.

Secondly, select the desired image visualization in the Band Rendering subsection. For RGB composites use Multiband as Render Type and select an input band for each color channel, e.g by selecting SWIR2 as red band, NIR as green band and RED as blue band. By default, the Min / Max values are estimated using percentiles (default: 2% to 98%). Furthermore, we checked Show full map canvas extent to visualize all scenes acquired on the specified date that fall into our current map canvas extent.

_images/getting_started_6.PNG

Lastly, click Apply to estimate the values and make sure the image visibility is toggled on icon_img.

_images/getting_started_7.PNG

Also try to use the getting_started_6 buttons in the upper right corner of the plugin to jump to the previous/next observation dates or time frames.

Visualize Temporally Aggregated Images

In case of Landsat data it is usefull to not only visualize data at a specific date in time, but to aggregate multiple observations over a date range (e.g. the revisiting time of 16 days) and to also visualize observations from neighbouring overflight pathes at the same time. Furthermore, the aggregation of time series into pixel based statistics is usefull for preserving variance whilst reducing the dimensionality of the data. Pixel-based calculations over time are referred to as Reducers in Earth Engine. For each band, you can select from a variety of statistical reducers and visualize them accordingly.

For example, let’s visualize the median of the selected band combination for our entire map canvas using all imagery acquired in 2018:

_images/getting_started_8.PNG
_images/getting_started_9.PNG
Download Time Series Data

Often we need time series of spectral bands or indices for multiple locations in space. GEE-TSE allows to specify a point layer that is loaded in QGIS for which the time series of the selected bands and time frame can be downloaded to csv-files. To do so, select a point layer as input in the Point Browser panel at the bottom of the plugin and click cloud_download in the lower right corner. Specify a target folder in which the .csv-files for each feature of the point layer are stored.

_images/getting_started_10.PNG

Furthermore, you can select attributes of the point layer and use the list or the getting_started_6 buttons to reload the plot and image visualization for the given feature.

Advanced Usage

Collection Editor

In the Collection tab, the subsection Collection Editor provides the python code used to access the imagery and enables full control on the code executed by the user. For instance, we might want to filter the specified collection prior to plotting and visualization.

Have a look at the datasets available in the Earth Engine Catalogue by clicking on icon_catalogue. For now, consider the MOD13Q1 16-day vegetation index product (https://developers.google.com/earth-engine/datasets/catalog/MODIS_006_MOD13Q1). Open the Collection Editor and provide python code that loads the collection and applies a user-defined cloud masking function. Important: The variable name of the final ImageCollection must be imageCollection in order to be understood by the plugin.

import ee

# function to mask clouds
def maskClouds(image):
  QA = image.select('SummaryQA')
  return image.updateMask(QA.eq(0))

# define imgcol with selected bands and apply cloud masking
imageCollection = ee.ImageCollection('MODIS/006/MOD13Q1')\
     .select('NDVI','EVI','SummaryQA')\
     .map(maskClouds)

In the Collection Editor this then looks as follows:

_images/getting_started_11.PNG
More Collections

Sentinel-2

This collection returns the merged Sentinel-2A and -2B L2A (surface reflectance product) archives. It uses an aggressive cloud masking discarding images with >50% cloud cover, uses the scene classification layer to remove saturated pixels, clouds at all confidence levels, cirrus clouds, cloud shadows and snow. Furthermore, the cloud displacement index (CDI; see Frantz et al. 2018 https://doi.org/10.1016/j.rse.2018.04.046 for details) is used to improve cloud masking. The collection returns renamed bands and includes a NDVI band for convenience.

import ee

def maskS2scl(image):
    scl = image.select('SCL')
    sat = scl.neq(1)
    shadow = scl.neq(3)
    cloud_lo = scl.neq(7)
    cloud_md = scl.neq(8)
    cloud_hi = scl.neq(9)
    cirrus = scl.neq(10)
    snow = scl.neq(11)
    return image.updateMask(sat.eq(1)).updateMask(shadow.eq(1)).updateMask(cloud_lo.eq(1)).updateMask(cloud_md.eq(1)).updateMask(cloud_hi.eq(1)).updateMask(cirrus.eq(1)).updateMask(snow.eq(1)))

def maskS2cdi(image):
    cdi = ee.Algorithms.Sentinel2.CDI(image)
    return image.updateMask(cdi.gt(-0.8)).addBands(cdi)


bands = ee.List(['B2', 'B3', 'B4', 'B5', 'B6', 'B7', 'B8', 'B8A', 'B11', 'B12', 'QA60', 'SCL', 'cdi'])
band_names = ee.List(['blue', 'green', 'red', 'rededge1', 'rededge2', 'rededge3', 'nir', 'broadnir', 'swir1', 'swir2', 'QA60', 'SCL', 'CDI'])

sen = ee.ImageCollection('COPERNICUS/S2_SR')\
                .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 50))\
                .map(maskS2scl) \
                .map(maskS2cdi) \
                .select(bands, band_names)

sen = sen.map(lambda image: image.addBands(image.normalizedDifference(['broadnir', 'red'])\
                                            .multiply(10000).toInt16().rename('ndvi')))

imageCollection = sen

Contact

Please provide feedback to Andreas Rabe (andreas.rabe@geo.hu-berlin.de) or create an issue.