Downloading Sentinel-2 archives from Google Cloud with sen2r

In short…

Retrieving Sentinel-2 data from official repositories (Copernicus Open Access Hub) for time series analysis has recently become particularly laborious, since products older than 30 days cannot be directly downloaded (users have to order them from Long Term Archive, with the chance to retrieve corrupted archives). To bypass these problems, the possibility to search and download Sentinel-2 data from Google Cloud was recently implemented in the R package sen2r. This functionality can be exploited without a paid Google Cloud account: it is sufficient to install the Google Cloud SDK and configure it with a free Google account.

Overview

Sentinel-2 SAFE archives are officially distributed through the ESA Copernicus Open Access Hub. The most recent products can be directly downloaded using two API protocols: DHUS or API Hub. Conversely, oldest products must be ordered from Long Term Archive, as described here: ordered products are made available online after more or less one day, then ZIP archives can be directly downloaded. Each user have a quota of orders which can submit, that is different among the two infrastructures (1 product / half hour using DHUS, 10 products / 12 hours using API Hub).

The retention period of online products was recently reduced from 12 or 18 months to 30 days, in order to increase the capabilities of the LTA ordering infrastructures (before this change, often requests could not have been submitted because of system saturation). This change, although not affecting the possibility to directly retrieve most recent images, made particularly difficult the analysis of Sentinel-2 time series (for which sen2r was designed): for example, to retrieve 3-years of data (2017-2020) on the sample Area Of Interest used for sen2r examples (which overlaps one Sentinel-2 tile and two orbits, and which can be obtained with the command system.file("extdata/vector/barbellino.geojson", package = "sen2r")) it is necessary to retrieve 435 archives, of which 433 are offline at the time this post was written; ordering them on DHUS using an automatised script (so to have the possibility to order exactly two every hour) would require 10 days (plus 3-4 additional days to have all of them made online).

In addition, there is the concrete chance that a substantial part of them would not be usable, due to a recently encountered problem (retrieval of corrupted SAFE archives from LTA) described here.

This background made urgent the needing to retrieve Sentinel-2 archives from an alternative data source. Among the available ones, Google Cloud Sentinel-2 bucket was chosen because it offers the possibility to download data for free and without any limitations. What is needed is:

  1. a Google account (no paid Google Cloud plans are required);
  2. Google Cloud SDK to be installed and configured.

Installation and configuration

The steps required to be able to automatically search and download Sentinel-2 data from Google Cloud through sen2r.

  1. Install (or update) the R package {sen2r} version >= 1.5.0:

    install.packages("sen2r")

    (note for Windows users: at the time this post was written, version 1.5.0 had just been released, so the source version of the package must be explicitly installed until the binary version 1.5.0 will be available).

  2. Install Google Cloud SDK following the official instructions.

  3. Configure sen2r to use Google Cloud SDK: this can be done running the following function:

    check_gcloud()

    which automatically retrieves the path of the binary gsutil (if the automatic retrieval would fail, i.e., because Google Cloud SDK was installed in a non-standard directory, arguments gsutil_dir and full_scan can be used – refer to the function documentation).

    Alternatively, check_sen2r_deps() can be used to launch a GUI which allows graphically configuring external dependencies (including Google Cloud SDK).

Usage

sen2r can be used as usual, remembering to set Google Cloud SDK as input source (Copernicus Hub remains the default choice).

Using sen2r from the GUI

The function sen2r() opens the sen2r GUI. In the first sheet, the section “SAFE options” was modified to include the selector “Input servers”, which allows keeping the default “ESA Hub”, replacing it with “Google Cloud” or leaving both (in which case products are retrieved from Google Cloud if available, or from Copernicus Hub otherwise). The selector is deactivated if Google Cloud SDK was not configured (or hidden if the offline mode was selected). If only Google Cloud was selected, it is not yet necessary to set SciHub credentials.

Using function sen2r() non-interactively

Users can also launch their processing chain in non-interactive mode as described in the vignette; in this case, they can set the argument server in one of these ways:

  • server = "gcloud" (search and download on Google Cloud exclusively);
  • server = c("gcloud", "scihub") (search and download on Google Cloud first, and on Copernicus Hub in case products were not found on Google Cloud);
  • server = c("scihub", "gcloud") (the same, but searching on SciHub first).

The function documentation can be accessed for additional details.

Using functions s2_list() and s2_download()

These two functions can be used to specifically search SAFE archives and download them. In this case, the argument server must be set in the function s2_list() in the same way seen for main function sen2r() (see also the function documentation). This function returns a SAFE list with the specific Google Cloud URLs.

As an example:

example_s2_list_scihub <- s2_list(
  tile = "32TNS", orbit = "065",
  time_interval = c("2021-05-01", "2021-05-15")
)
example_s2_list_scihub
A named vector with 3 SAFE archives.
                                     S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE 
"https://apihub.copernicus.eu/apihub/odata/v1/Products('e9f01beb-8978-428c-89e6-f4c71156526b')/$value" 
                                     S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE 
"https://apihub.copernicus.eu/apihub/odata/v1/Products('9eb7cf57-49ab-4dc5-bb6b-79882196c7d9')/$value" 
                                     S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE 
"https://apihub.copernicus.eu/apihub/odata/v1/Products('1b35b71c-804c-4b9f-931c-de8e291393a4')/$value" 
The following attributes are included: mission, level, id_tile, id_orbit, sensing_datetime, ingestion_datetime, clouds, footprint, uuid, online.

By default outputs are searched on Copernicus Hub (as it can be noticed by product URLs). At the time this post was written, 2 of the 3 archives were not available online:

safe_is_online(example_s2_list_scihub)
1 out of 3 products are online.
S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE 
                                                            FALSE 
S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE 
                                                            FALSE 
S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE 
                                                             TRUE 

To search them on Google Cloud, the argument server = "gcloud" can be set:

example_s2_list_gcloud <- s2_list(
  server = "gcloud",
  tile = "32TNS", orbit = "065",
  time_interval = c("2021-05-01", "2021-05-15")
)
example_s2_list_gcloud
A named vector with 3 SAFE archives.
                                                    S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE/" 
                                                    S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE/" 
                                                    S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE/" 
The following attributes are included: mission, level, id_tile, id_orbit, sensing_datetime, ingestion_datetime, clouds, footprint, uuid, online.

Now URLs refer to Google Cloud locations.

The “mixed” research can be performed in two ways:

example_s2_list_mixed1 <- s2_list(
  server = c("gcloud","scihub"), availability = "check",
  tile = "32TNS", orbit = "065",
  time_interval = c("2021-05-01", "2021-05-15")
)
example_s2_list_mixed1
A named vector with 3 SAFE archives.
                                                    S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE/" 
                                                    S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE/" 
                                                    S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE/" 
The following attributes are included: mission, level, id_tile, id_orbit, sensing_datetime, ingestion_datetime, clouds, footprint, uuid, online.

Giving priority to Google Cloud, products available both on ESA Hub and Google Cloud (all products, in the example above) are retrieved from Google Cloud. Conversely, giving priority to ESA Hub products are taken from Copernicus if they are available online, and from Google Cloud if they are on LTA:

example_s2_list_mixed2 <- s2_list(
  server = c("scihub","gcloud"),
  tile = "32TNS", orbit = "065",
  time_interval = c("2021-05-01", "2021-05-15")
)
example_s2_list_mixed2
A named vector with 3 SAFE archives.
                                                    S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2B_MSIL2A_20210501T101559_N0300_R065_T32TNS_20210501T135123.SAFE/" 
                                                    S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE 
"gs://gcp-public-data-sentinel-2/L2/tiles/32/T/NS/S2A_MSIL2A_20210506T102021_N0300_R065_T32TNS_20210506T132458.SAFE/" 
                                                    S2B_MSIL2A_20210511T101559_N0300_R065_T32TNS_20210511T134528.SAFE 
               "https://apihub.copernicus.eu/apihub/odata/v1/Products('1b35b71c-804c-4b9f-931c-de8e291393a4')/$value" 
The following attributes are included: mission, level, id_tile, id_orbit, sensing_datetime, ingestion_datetime, clouds, footprint, uuid, online.

In the example above the last product is available for direct download from ESA Hub and so it is retrieved from this source; first two ones are instead retrieved from Google Cloud.

Notice that, in this case (and only in this case), availability of SciHub products is checked inside s2_list().

At this point, s2_download() can be used as usual without the needing to set any additional arguments:

s2_download(example_s2_list_mixed2, outdir = tempdir())

Conclusions

Starting from version 1.5.0, sen2r is able to download Sentinel-2 products from Google Cloud, so to provide a way to download “old” SAFE archives without the needing to order them from Long Term Archive, wait for their availability and have the change to obtain corrupted products.

This sen2r functionality is experimental, so users could encounter bugs. In this case, check among GitHub issues if other users already encountered the same problem, and eventually report them opening a new issue (carefully following the indications provided in the template).

Currently, products are searched on Google Cloud by directly reading the bucket content: thank to this method it is possible perform free-of-charge searches and downloads, although searching products can be a very slow process. In a future release, the possibility to implement a method based on Google BigQuery will be evaluated.

Credits

sen2r was developed by Luigi Ranghetti and Lorenzo Busetto (IREA-CNR), is maintained by Luigi Ranghetti and it is released under the GNU GPL-3 license.

Using sen2r for production (including scientific products) requires to cite it (use this entry).

Avatar
Luigi Ranghetti
Data scientist

My interests include spatial data processing, remote sensing analysis, R programming.

Related