A python package that extends Google Earth Engine
GitHub: https://github.com/davemlz/eemont
Documentation: https://eemont.readthedocs.io/
PyPI: https://pypi.org/project/eemont/
Conda-forge: https://anaconda.org/conda-forge/eemont
Tutorials: https://github.com/davemlz/eemont/tree/master/docs/tutorials
Paper: https://joss.theoj.org/papers/10.21105/joss.03168
Google Earth Engine is a cloud-based service for geospatial processing of vector and raster data. The Earth Engine platform has a JavaScript and a Python API with different methods to process geospatial objects. Google Earth Engine also provides a HUGE PETABYTE-SCALE CATALOG of raster and vector data that users can process online (e.g. Landsat Missions Image Collections, Sentinel Missions Image Collections, MODIS Products Image Collections, World Database of Protected Areas, etc.). The eemont package extends the Google Earth Engine Python API with pre-processing and processing tools for the most used satellite platforms by adding utility methods for different Earth Engine Objects that are friendly with the Python method chaining.
The eemont Python package can be found in the Earth Engine Community: Developer Resources together with other awesome resources such as geemap and rgee.
The eemont python package extends the following Earth Engine classes:
New utility methods and constructors are added to above-mentioned classes in order to create a more fluid code by being friendly with the Python method chaining. These methods are mandatory for some pre-processing and processing tasks (e.g. clouds masking, shadows masking, image scaling, spectral indices computation, etc.), and they are presented as simple functions that give researchers, students and analysts the chance to analyze data with far fewer lines of code.
Look at this simple example where a Sentinel-2 Surface Reflectance Image Collection is pre-processed and processed in just one step:
import ee, eemont
ee.Authenticate()
ee.Initialize()
point = ee.Geometry.PointFromQuery(
'Cali, Colombia',
user_agent = 'eemont-example'
) # Extended constructor
S2 = (ee.ImageCollection('COPERNICUS/S2_SR')
.filterBounds(point)
.closest('2020-10-15') # Extended (pre-processing)
.maskClouds(prob = 70) # Extended (pre-processing)
.scaleAndOffset() # Extended (pre-processing)
.spectralIndices(['NDVI','NDWI','BAIS2'])) # Extended (processing)And just like that, the collection was pre-processed, processed and ready to be analyzed!
Install the latest version from PyPI:
pip install eemont
Upgrade eemont by running:
pip install -U eemont
Install the latest version from conda-forge:
conda install -c conda-forge eemont
Install the latest dev version from GitHub by running:
pip install git+https://github.com/davemlz/eemont
Let's see some of the main features of eemont and how simple they are compared to the GEE Python API original methods:
The following operators are overloaded: +, -, *, /, //, %, **\ , <<, >>, &, |, <, <=,
==, !=, >, >=, -, ~. (and you can avoid the ee.Image.expression() method!)
| GEE Python API | eemont-style |
|---|---|
ds = 'COPERNICUS/S2_SR'
S2 = (ee.ImageCollection(ds)
.first())
def scaleImage(img):
scaling = img.select('B.*')
x = scaling.multiply(0.0001)
scaling = img.select(['AOT','WVP'])
scaling = scaling.multiply(0.001)
x = x.addBands(scaling)
notScaling = img.select([
'SCL',
'TCI.*',
'MSK.*',
'QA.*'
]))
return x.addBands(notScaling)
S2 = scaleImage(S2)
exp = '2.5*(N-R)/(N+(6*R)-(7.5*B)+1)'
imgDict = {
'N': S2.select('B8'),
'R': S2.select('B4'),
'B': S2.select('B2')
}
EVI = S2.expression(exp,imgDict) |
ds = 'COPERNICUS/S2_SR'
S2 = (ee.ImageCollection(ds)
.first()
.scale())
N = S2.select('B8')
R = S2.select('B4')
B = S2.select('B2')
EVI = 2.5*(N-R)/(N+(6*R)-(7.5*B)+1) |
Masking clouds and shadows can be done using eemont with just one method: maskClouds()!
| GEE Python API | eemont-style |
|---|---|
ds = 'LANDSAT/LC08/C01/T1_SR'
def maskCloudsShadows(img):
c = (1 << 3)
s = (1 << 5)
qa = 'pixel_qa'
qa = img.select(qa)
cm = qa.bitwiseAnd(c).eq(0)
sm = qa.bitwiseAnd(s).eq(0)
mask = cm.And(sm)
return img.updateMask(mask)
(ee.ImageCollection(ds)
.map(maskCloudsShadows)) |
ds = 'LANDSAT/LC08/C01/T1_SR'
(ee.ImageCollection(ds)
.maskClouds()) |
Scaling and offsetting can also be done using eemont with just one method: scale()!
| GEE Python API | eemont-style |
|---|---|
def scaleBands(img):
scaling = img.select([
'NDVI',
'EVI',
'sur.*'
])
x = scaling.multiply(0.0001)
scaling = img.select('.*th')
scaling = scaling.multiply(0.01)
x = x.addBands(scaling)
notScaling = img.select([
'DetailedQA',
'DayOfYear',
'SummaryQA'
])
return x.addBands(notScaling)
ds = 'MODIS/006/MOD13Q1'
(ee.ImageCollection(ds)
.map(scaleBands)) |
ds = 'MODIS/006/MOD13Q1'
(ee.ImageCollection(ds)
.scaleAndOffset()) |
The complete preprocessing workflow (Masking clouds and shadows, and image scaling and
offsetting) can be done using eemont with just one method: preprocess()!
| GEE Python API | eemont-style |
|---|---|
ds = 'LANDSAT/LC08/C01/T1_SR'
def maskCloudsShadows(img):
c = (1 << 3)
s = (1 << 5)
qa = 'pixel_qa'
qa = img.select(qa)
cm = qa.bitwiseAnd(c).eq(0)
sm = qa.bitwiseAnd(s).eq(0)
mask = cm.And(sm)
return img.updateMask(mask)
def scaleBands(img):
scaling = img.select('B[1-7]')
x = scaling.multiply(0.0001)
scaling = img.select([
'B10','B11'
])
scaling = scaling.multiply(0.1)
x = x.addBands(scaling)
notScaling = img.select([
'sr_aerosol',
'pixel_qa',
'radsat_qa'
])
return x.addBands(notScaling)
(ee.ImageCollection(ds)
.map(maskCloudsShadows)
.map(scaleBands)) |
ds = 'LANDSAT/LC08/C01/T1_SR'
(ee.ImageCollection(ds)
.preprocess()) |
Do you need to compute several spectral indices? Use the spectralIndices() method! The
indices are taken from Awesome Spectral Indices.
| GEE Python API | eemont-style |
|---|---|
ds = 'LANDSAT/LC08/C01/T1_SR'
def scaleImage(img):
scaling = img.select('B[1-7]')
x = scaling.multiply(0.0001)
scaling = img.select(['B10','B11'])
scaling = scaling.multiply(0.1)
x = x.addBands(scaling)
notScaling = img.select([
'sr_aerosol',
'pixel_qa',
'radsat_qa'
]))
return x.addBands(notScaling)
def addIndices(img):
x = ['B5','B4']
a = img.normalizedDifference(x)
a = a.rename('NDVI')
x = ['B5','B3']
b = img.normalizedDifference(x)
b = b.rename('GNDVI')
x = ['B3','B6']
c = img.normalizedDifference(x)
c = b.rename('NDSI')
return img.addBands([a,b,c])
(ee.ImageCollection(ds)
.map(scaleImage)
.map(addIndices)) |
ds = 'LANDSAT/LC08/C01/T1_SR'
(ee.ImageCollection(ds)
.scaleAndOffset()
.spectralIndices([
'NDVI',
'GNDVI',
'NDSI'])
) |
The list of available indices can be retrieved by running:
eemont.listIndices()Information about the indices can also be checked:
indices = eemont.indices()
indices.BAIS2.formula
indices.BAIS2.referenceStruggling to get the closest image to a specific date? Here is the solution: the
closest() method!
| GEE Python API | eemont-style |
|---|---|
ds = 'COPERNICUS/S5P/OFFL/L3_NO2'
xy = [-76.21, 3.45]
poi = ee.Geometry.Point(xy)
date = ee.Date('2020-10-15')
date = date.millis()
def setTimeDelta(img):
prop = 'system:time_start'
prop = img.get(prop)
prop = ee.Number(prop)
delta = prop.subtract(date)
delta = delta.abs()
return img.set(
'dateDist',
delta)
(ee.ImageCollection(ds)
.filterBounds(poi)
.map(setTimeDelta)
.sort('dateDist')
.first()) |
ds = 'COPERNICUS/S5P/OFFL/L3_NO2'
xy = [-76.21, 3.45]
poi = ee.Geometry.Point(xy)
(ee.ImageCollection(ds)
.filterBounds(poi)
.closest('2020-10-15')) |
The JavaScript API has a method for time series extraction (included in the ui.Chart
module), but this method is missing in the Python API... so, here it is!
PD: Actually, there are two methods that you can use: getTimeSeriesByRegion() and
getTimeSeriesByRegions()!
f1 = ee.Feature(ee.Geometry.Point([3.984770,48.767221]).buffer(50),{'ID':'A'})
f2 = ee.Feature(ee.Geometry.Point([4.101367,48.748076]).buffer(50),{'ID':'B'})
fc = ee.FeatureCollection([f1,f2])
S2 = (ee.ImageCollection('COPERNICUS/S2_SR')
.filterBounds(fc)
.filterDate('2020-01-01','2021-01-01')
.maskClouds()
.scaleAndOffset()
.spectralIndices(['EVI','NDVI']))
# By Region
ts = S2.getTimeSeriesByRegion(reducer = [ee.Reducer.mean(),ee.Reducer.median()],
geometry = fc,
bands = ['EVI','NDVI'],
scale = 10)
# By Regions
ts = S2.getTimeSeriesByRegions(reducer = [ee.Reducer.mean(),ee.Reducer.median()],
collection = fc,
bands = ['EVI','NDVI'],
scale = 10)Don't you have the coordinates of a place? You can construct them by using queries!
usr = 'my-eemont-query-example'
seattle_bbox = ee.Geometry.BBoxFromQuery('Seattle',user_agent = usr)
cali_coords = ee.Feature.PointFromQuery('Cali, Colombia',user_agent = usr)
amazonas_river = ee.FeatureCollection.MultiPointFromQuery('Río Amazonas',user_agent = usr)This is perhaps the most important feature in eeExtra! What if you could use a
JavaScript module (originally just useful for the Code Editor) in python or R? Well,
wait no more for it!
| JS (Code Editor) | Python (eemont) | R (rgee+) |
|---|---|---|
var usr = 'users/sofiaermida/'
var rep = 'landsat_smw_lst:'
var fld = 'modules/'
var fle = 'Landsat_LST.js'
var pth = usr+rep+fld+fle
var mod = require(pth)
var LST = mod.collection(
ee.Geometry.Rectangle([
-8.91,
40.0,
-8.3,
40.4
]),
'L8',
'2018-05-15',
'2018-05-31',
true
) |
import ee, eemont
ee.Initialize()
usr = 'users/sofiaermida/'
rep = 'landsat_smw_lst:'
fld = 'modules/'
fle = 'Landsat_LST.js'
pth = usr+rep+fld+fle
ee.install(pth)
mod = ee.require(pth)
LST = mod.collection(
ee.Geometry.Rectangle([
-8.91,
40.0,
-8.3,
40.4
]),
'L8',
'2018-05-15',
'2018-05-31',
True
) |
library(rgee)
library(rgeeExtra)
ee_Initialize()
usr <- 'users/sofiaermida/'
rep <- 'landsat_smw_lst:'
fld <- 'modules/'
fle <- 'Landsat_LST.js'
pth <- paste0(usr,rep,fld,fle)
mod <- ee$require(pth)
LST = mod$collection(
ee$Geometry$Rectangle(c(
-8.91,
40.0,
-8.3,
40.4
)),
'L8',
'2018-05-15',
'2018-05-31',
TRUE
) |
The project is licensed under the MIT license.
Do you like using eemont and think it is useful? Share the love by citing it!:
Montero, D., (2021). eemont: A Python package that extends Google Earth Engine.
Journal of Open Source Software, 6(62), 3168, https://doi.org/10.21105/joss.03168
If required, here is the BibTex!:
@article{Montero2021,
doi = {10.21105/joss.03168},
url = {https://doi.org/10.21105/joss.03168},
year = {2021},
publisher = {The Open Journal},
volume = {6},
number = {62},
pages = {3168},
author = {David Montero},
title = {eemont: A Python package that extends Google Earth Engine},
journal = {Journal of Open Source Software}
}
- David Montero Loaiza: Lead Developer of eemont and eeExtra.
- César Aybar: Lead Developer of rgee and eeExtra.
- Aaron Zuspan: Plus Codes Constructors and Methods, Panchromatic Sharpening and Histogram Matching Developer.
Special thanks to Justin Braaten for featuring eemont in tutorials and the GEE Community: Developer Resources Page, to César Aybar for the formidable help with Awesome Spectral Indices and to the JOSS Review Team (Katy Barnhart, Jayaram Hariharan, Qiusheng Wu and Patrick Gray) for the comments, suggestions and contributions!
