Google Earth Engine – Calling ‘Mean’ on ImageCollection Produces Scalar Value

google-earth-enginegoogle-earth-engine-python-api

I define the following ImageCollection, which is 24 hourly images from a single day:

collection = ee.ImageCollection("ECMWF/ERA5_LAND/HOURLY").select("total_precipitation")
collection = collection.filterDate("2020-01-01", "2020-01-02")

Then I download the first image in the collection:

region = ee.Geometry.Polygon([
    (-119.69680109161911, 37.9180436245507),
    (-119.23634853488707, 37.9180436245507),
    (-119.23634853488707, 37.57830484843708),
    (-119.69680109161911, 37.57830484843708),
    (-119.69680109161911, 37.9180436245507)
])
img = collection.first()
task = ee.batch.Export.image.toDrive(
    img.toFloat(),
    description="sample_hourly_image",
    fileNamePrefix="sample_hourly_image",
    region=region,
    fileFormat="GeoTIFF")
task.start()

After downloading I can open this image and convert to a numpy array, which looks as follows – with 4 rows and 6 columns:

Then I want to take the mean of all 24 hourly images in the collection to produce a single image of mean temperature for that day. This is what I tried:

daily_img = collection.mean()
task = ee.batch.Export.image.toDrive(
    daily_img.toFloat(),
    description="sample_daily_image",
    fileNamePrefix="sample_daily_image",
    region=region,
    fileFormat="GeoTIFF",
)
task.start()

However this time when I download the image and convert to a numpy array it only has a single scalar value:

What am I doing wrong here?

Best Answer

You should set the scale and crs (or crsTransform and crs) parameters in the export call. When you reduce an image collection the result has a default projection that has 1-degree pixel size in WGS84 EPSG:4326 coordinate reference system. It does this because it is possible that images in the collection have different projections. Currently, because your region is pretty small and you are accepting defaults in the export, your exported image has a single 1-degree pixel; set the crs and scale to that of the original data. Read more about the default projection at The default projection.

Related Solutions

Google Earth Engine – How to Download Entire Image Collection of a Specific Month from 1990-2020

You can try converting the image collection to bands in a single image. I just tried the following code and it took around 10 minutes to export to Google Drive.

var era5_2mt = ee.ImageCollection('ECMWF/ERA5/DAILY')
                   .select('minimum_2m_air_temperature')
                   .filter(ee.Filter.calendarRange(1990,2020,'year'))
                   .filter(ee.Filter.calendarRange(11,11,'month'))
                   // clip to your study area
                   .map(function(image) {return image.clip(table)});


// if you actually need to subtract 273.15, you can map over the image collection, rather than a list of images
// uncomment the following block

// var study_area = era5_2mt.map(function(img){
//   var s_ts = img.get('system:time_start')    // get the 'system:time start' property from the image
//   img = ee.Image(img).subtract(273.15)
//                      .set({'system:time_start': s_ts});   // set the 'system:time_start' property from the output image
//   return(img);
  
// });


// flatten collection to a single image
var merged = era5_2mt.toBands();
// var merged = study_area.toBands();  // use this if you need the calculated images


// export image to Google Drive
Export.image.toDrive({
  image: merged,
  description: 'era5_2mt',
  scale: 20000,
  region: table
});

Google Earth Engine ImageCollection Merge – Merging a List of ImageCollections into a Single ImageCollection Using Google Earth Engine Python API

First, a general note: Always try to avoid getInfo() calls whenever you can. It evaluates the value server-side, and returns a client-side object. This often makes your script significantly slower than it needs to be. You can read up on it here.

You have a list where each element is an ImageCollection that you want to merge into a single ImageCollection. Below are two approaches for doing this.

First approach uses iterate(). It manually accumulates the images into a single collection. It's quite explicit in what it does while being quite verbose, due to all the casts needed.

Second approach uses a trick where the list of collections are put into a FeatureCollection, which then is flattened. More compact, but maybe a bit magical.

import ee

ee.Initialize()

admin = ee.FeatureCollection('FAO/GAUL_SIMPLIFIED_500m/2015/level2')
terraclimate = ee.ImageCollection('IDAHO_EPSCOR/TERRACLIMATE')

province_names = admin \
  .filter(ee.Filter.eq('ADM0_NAME', 'Indonesia')) \
  .aggregate_array('ADM1_NAME') \
  .distinct() # Remove duplicates directly in EE. Avoids getInfo() call

def get_province(province):
    img_col = terraclimate.filterBounds(admin.filter(ee.Filter.eq('ADM1_NAME', province)))
    return img_col.set({'province':province})


# Option 1 - iterate
new_img_col = province_names.map(get_province)
collection1 = ee.ImageCollection( # Cast result of iterate() to ImageCollection
  new_img_col.iterate(
    lambda proviceCollection, acc:
      ee.ImageCollection(acc).merge( # Both arguments must be cast too
        ee.ImageCollection(proviceCollection)
      ),
    ee.ImageCollection([])
  )
)

# Option 2 - flatten
collection2 = ee.ImageCollection( # Cast result to ImageCollection
  ee.FeatureCollection( # Create an intermediate FeatureCollection containing ImageCollections
    province_names.map(get_province)
  ).flatten() # Remove the intermediate feature collection with the merged image collections
)

Best Answer

Related Solutions

Google Earth Engine – How to Download Entire Image Collection of a Specific Month from 1990-2020

Google Earth Engine ImageCollection Merge – Merging a List of ImageCollections into a Single ImageCollection Using Google Earth Engine Python API

Related Question