ICESat-2 and Landsat cloud access and data integration#

This notebook (download) builds off of the icepyx IS2_cloud_data_access.ipynb and ICESat-2 Hackweek Data Integration 1 tutorials. It illustrates the use of icepyx for accessing ICESat-2 data currently available through the AWS (Amazon Web Services) us-west2 hub s3 data bucket as well as data integration with Landsat (cloud-optimized geotiff) and ATM (downloaded csv) datasets.

Learning Objectives


  • Identify and locate ICESat-2 and Landsat data

  • Acquire data from the cloud

  • Open data in pandas and xarray and basic functioning of DataFrames

Key Takeaway

By the end of this tutorial, you will be able to visualize Landsat Cloud Optimized Geotiffs with ICESat-2 and ATM data.


  1. ICESat-2 data became publicly available on the cloud on 29 September 2022. Thus, access methods and example workflows are still being developed by NSIDC, and the underlying code in icepyx will need to be updated now that these data (and the associated metadata) are available. We appreciate your patience and contributions (e.g. reporting bugs, sharing your code, etc.) during this transition!

  2. This example and the code it describes are part of ongoing development. Current limitations to using these features are described throughout the example, as appropriate.

  3. You MUST be working within an AWS instance. Otherwise, you will get a permissions error.

  4. Authentication is still more steps than we’d like. We’re working to address this - let us know if you’d like to join the conversation!

For this tutorial, feel free to run the code along with us as we live code by downsizing the zoom window and splitting your screen (or using two screens). Or you can simply watch the zoom walkthrough. Don’t worry if you fall behind on the code. The notebook is standalone and you can easily run the code at your own pace another time to catch anything you missed.

Computing environment#

We’ll be using the following open source Python libraries in this notebook:

# Suppress library deprecation warnings
import logging
import ipyleaflet
from ipyleaflet import Map, basemaps, basemap_to_tiles, Polyline

import ipywidgets
import datetime
import re
%matplotlib widget
import pystac_client
import geopandas as gpd
import h5py
import ast
import pandas as pd
import geoviews as gv
import hvplot.pandas
from ipywidgets import interact
from IPython.display import display, Image
import intake # if you've installed intake-STAC, it will automatically import alongside intake
import intake_stac
import xarray as xr
import matplotlib.pyplot as plt
import boto3
import rasterio as rio
from rasterio.session import AWSSession
from rasterio.plot import show
import rioxarray as rxr
from dask.utils import SerializableLock
import os
import fiona
import hvplot.xarray
import numpy as np
from pyproj import Proj, transform
import as ccrs

1. Identify and acquire the ICESat2 product(s) of interest#

Download ICESat-2 ATL06 data from desired region#

We are going to use icepyx to download some ICESat-2 ATL06 data over our region of interest.

import icepyx as ipx
# Specifying the necessary icepyx parameters
short_name = 'ATL06'
spatial_extent = 'hackweek_kml_jakobshavan.kml' # KML polygon centered on Sermeq Kujalleq
date_range = ['2019-04-01', '2019-04-30']
rgts = ['338'] # IS-2 RGT of interest

You may notice that we specified a RGT track. As seen below, a large number of ICESat-2 overpasses occur for Sermeq Kujalleq (briefly known as Jakobshavn Isbrae). In the interest of time (and computer memory), we are going to look at only one of these tracks.

# Open KML file for use
fiona.drvsupport.supported_drivers['LIBKML'] = 'rw' # enable KML support which is disabled by default
jk = gpd.read_file(spatial_extent)
# Setup the Query object
region = ipx.Query(short_name, spatial_extent, date_range, tracks=rgts)
# Visualize area of interest

Looks good! Now it’s time to acquire the data.

Get the granule s3 urls#

You must specify cloud=True to get the needed s3 urls. This function returns a list containing the list of the granule IDs and a list of the corresponding urls.

gran_ids = region.avail_granules(ids=True, cloud=True)

Log in to Earthdata and generate an s3 token#

You can use icepyx’s existing login functionality to generate your s3 data access token, which will be valid for one hour.

We currently do not have this set up to automatically renew, but earthaccess, which icepyx will soon be adopting for authentication, is working on handling the limits imposed by expiring s3 tokens. If you’re interested in working on helping icepyx and NSIDC (and DAACs more broadly) address these challenges, please get in touch or submit a PR. Documentation/example testers are always appreciated (so you don’t have to understand the code)!

# Add your EarthData credentials

Set up your s3 access using your credentials#

import s3fs
# Authenicate using your NASA Earth Data login credentials; enter your user id and password when prompted
credentials = region.s3login_credentials
s3 = s3fs.S3FileSystem(key=credentials['accessKeyId'],

Select an s3 url and access the data#

Data read in capabilities for cloud data are coming soon in icepyx (targeted Winter 2022-2023). Stay tuned and we’d love for you to join us and contribute!

Note: If you get a PermissionDenied Error when trying to read in the data, you may not be sending your request from an AWS hub in us-west2. We’re currently working on how to alert users if they will not be able to access ICESat-2 data in the cloud for this reason

# the first index, [1], gets us into the list of s3 urls
# the second index, [0], gets us the first entry in that list.
s3url = gran_ids[1][0]
# s3url =  's3://nsidc-cumulus-prod-protected/ATLAS/ATL03/004/2019/11/30/ATL03_20191130221008_09930503_004_01.h5'
# Open the file
%time f = h5py.File(,'rb'),'r')
CPU times: user 90 ms, sys: 28.1 ms, total: 118 ms
Wall time: 274 ms
# View its attributes

Reading the file with h5py allows us to open the entire file, but is not super intuitive for later analysis. Let’s use h5py with pandas to open the data into DataFrames in a way that is more convenient for our analyses.

# Load the ICESat-2 data. We will just look at the central beams (GT2R/L)
# is2_file = 'processed_ATL06_20190420093051_03380303_005_01_full.h5'
with h5py.File(,'rb'), 'r') as f:
    is2_gt2r = pd.DataFrame(data={'lat': f['gt2r/land_ice_segments/latitude'][:],
                                  'lon': f['gt2r/land_ice_segments/longitude'][:],
                                  'elev': f['gt2r/land_ice_segments/h_li'][:]})
    is2_gt2l = pd.DataFrame(data={'lat': f['gt2l/land_ice_segments/latitude'][:],
                                  'lon': f['gt2l/land_ice_segments/longitude'][:],
                                  'elev': f['gt2l/land_ice_segments/h_li'][:]})
lat lon elev
0 59.783826 -47.173056 -9.104559e+00
1 59.784362 -47.173166 -2.777765e+00
2 59.787225 -47.173696 3.402823e+38
3 59.787404 -47.173732 3.402823e+38
4 59.790085 -47.174286 3.402823e+38

We opened this data into a pandas DataFrame, which is a handy tool for Earth data exploration and analysis. The column names derive automatically from the first row of the h5 file and each row corresponds to an ICESat-2 measurement.

For a tutorial on how to use pandas on this data, check out the ICESat-2 Hackweek Data Integration I tutorial. You can learn more about pandas from this cookbook.

2. Acquire non-cloud data and open: ATM data access#

Now we show how we access Airborne Topographic Mapper (non-AWS) lidar spot measurements to co-register with the ICESat-2 data.

An airborne campaign called Operation IceBridge was flown across Sermeq Kujalleq as validation for ICESat-2. Onboard was the ATM, a lidar that works at both 532 nm (like ICESat-2) and 1064 nm (near-infrared). More information about Operation IceBridge and ATM may be found here: Because both data sets are rather large, this can be computationally expensive, so we will only consider one flight track with the ATM 532 nm beam.

Operation IceBridge data is not available on the cloud, so this data was downloaded directly from NSIDC. If you are interested in using IceBridge data, NSIDC has a useful data portal here:

Co-register ICESat-2 with ATM data#

# Load the ATM data into a DataFrame
atm_file = 'ILATM2_20190506_151600_smooth_nadir3seg_50pt.csv'
atm_l2 = pd.read_csv(atm_file)

UTC_Seconds_Of_Day Latitude(deg) Longitude(deg) WGS84_Ellipsoid_Height(m) South-to-North_Slope West-to-East_Slope RMS_Fit(cm) Number_Of_ATM_Measurments_Used Number_Of_ATM_Measurements_Removed Distance_Of_Block_To_The_Right_Of_Aircraft(m) Track_Identifier
0 54969.50 69.262002 310.351764 490.3974 0.077354 -0.069179 589.57 3723 5 78 1
1 54969.50 69.262065 310.353395 500.2330 -0.048777 0.006024 434.12 2185 21 14 2
2 54969.50 69.262128 310.355026 500.3090 0.068798 0.077559 777.80 3640 8 -51 3
3 54969.50 69.262079 310.353741 498.9152 -0.085600 -0.111001 472.64 2818 15 0 0
4 54969.75 69.261648 310.351873 487.1317 0.108085 -0.078827 520.83 3753 33 78 1

The ATM L2 file contains plenty of information, including surface height estimates and slope of the local topography. It also contains a track identifier - ATM takes measurements from multiple parts of the aircraft, namely starboard, port, and nadir. To keep things simple, we will filter the DataFrame to only look at the nadir track (Track_Identifier = 0).

atm_l2 = atm_l2[atm_l2['Track_Identifier']==0]

# Change the longitudes to be consistent with ICESat-2
atm_l2['Longitude(deg)'] -= 360


Let’s take a quick look at where ATM is relative to ICESat-2…

# Subset the ICESat-2 data to the ATM latitudes
is2_gt2r = is2_gt2r[(is2_gt2r['lat']<atm_l2['Latitude(deg)'].max()) & (is2_gt2r['lat']>atm_l2['Latitude(deg)'].min())]
is2_gt2l = is2_gt2l[(is2_gt2l['lat']<atm_l2['Latitude(deg)'].max()) & (is2_gt2l['lat']>atm_l2['Latitude(deg)'].min())]

# Set up a map with the flight tracks as overlays
m = Map(
    center=(69.25, 310.35-360),

gt2r_line = Polyline(
        [is2_gt2r['lat'].min(), is2_gt2r['lon'].max()],
        [is2_gt2r['lat'].max(), is2_gt2r['lon'].min()]
    color="green" ,

gt2l_line = Polyline(
        [is2_gt2l['lat'].min(), is2_gt2l['lon'].max()],
        [is2_gt2l['lat'].max(), is2_gt2l['lon'].min()]
    color="green" ,

atm_line = Polyline(
        [atm_l2['Latitude(deg)'].min(), atm_l2['Longitude(deg)'].max()],
        [atm_l2['Latitude(deg)'].max(), atm_l2['Longitude(deg)'].min()]
    color="orange" ,


Looks like ATM aligns very closely with the left beam (GT2L), so hopefully the two beams will agree. The terrain over this region is quite rough, so we may expect some differences between ATM and GT2R. ICESat-2 also flew over Sermeq Kujalleq 16 days before ATM, so there might be slight differences due to ice movement.

We have looked at how we can quickly access ICESat-2 and airborne lidar data, and process them using pandas.

3. Search and open (Landsat) raster imagery from the cloud#

Let’s now talk about a cloud-optimized approach that requires no downloading to search and access only the subsets of the data we want. Cloud-optimized formats (e.g., COG, zarr, parquet) make reading data two orders of magnitude faster than non-optimized formats.

We will be working with Cloud Optimized GeoTIFF (COG). A COG is a GeoTIFF file with an internal organization that enables more efficient workflows and prevents having to open the entire image (see more at

Here is the User Manual for more information about accessing Landsat S3.

Search for Landsat imagery#

To explore and access COG’s easily we will use a SpatioTemporal Asset Catalog (STAC). The STAC provides a common metadata format to make it easier to index and querry S3 buckets for geospatial data.

# Sets up AWS credentials for acquiring images through dask/xarray
os.environ["AWS_REQUEST_PAYER"] = "requester"

# Sets up proper AWS credentials for acquiring data through rasterio
aws_session = AWSSession(boto3.Session(), requester_pays=True)

Extract geometry bounds are extracted from the ICESat-2 KML file used above so that we can perform the Landsat spatial search.

# Extract geometry bounds
geom = jk.geometry[0]
(-51.3229009069365, 68.84029223511094, -48.20366423696812, 69.61656633135274)

We will search for imagery in STAC catalog using the pystac_client search tool.

# Search STAC API for Landsat images based on a bounding box, date and other metadata if desired

bbox = (geom.bounds[0], geom.bounds[1], geom.bounds[2], geom.bounds[3]) #(west, south, east, north) 

timeRange = '2019-05-06/2019-05-07'
url = ''
collection = 'landsat-c2l1' # Landsat Collection 2, Level 1
api =

items =
            bbox = bbox,
            datetime = timeRange,
            limit = 400, # This line not required
print(f'{len(items)} items')

# Write a json file that records our search output
gjson_outfile = f'/tmp/Landsat.geojson'
2 items

We can include property searches, such as path, row, cloud-cover, as well with the properties flag in the

We are given a pystac collection of items (images)


Load the geojson file into geopandas and inspect the items we want to collect

# Load the geojson file
gf = gpd.read_file(gjson_outfile)
id datetime eo:cloud_cover view:sun_azimuth view:sun_elevation platform instruments view:off_nadir landsat:cloud_cover_land landsat:wrs_type ... landsat:correction accuracy:geometric_x_bias accuracy:geometric_y_bias accuracy:geometric_x_stddev accuracy:geometric_y_stddev accuracy:geometric_rmse proj:epsg created updated geometry
0 LC08_L1TP_008012_20190507_20200829_02_T1 2019-05-07 14:54:18.866000+00:00 0.18 173.852645 38.463606 LANDSAT_8 [OLI, TIRS] 0 0.0 2 ... L1TP 0 0 3.431 3.144 4.654 32622 2022-06-28 20:15:52.467000+00:00 2022-06-28 20:15:52.467000+00:00 POLYGON ((-50.65493 69.41549, -52.45035 67.796...
1 LC08_L1TP_008011_20190507_20200828_02_T1 2019-05-07 14:53:54.971000+00:00 10.18 175.877442 37.193127 LANDSAT_8 [OLI, TIRS] 0 10.3 2 ... L1TP 0 0 3.409 4.025 5.275 32623 2022-06-28 23:23:03.741000+00:00 2022-06-28 23:23:03.741000+00:00 POLYGON ((-48.95109 70.75210, -50.97098 69.148...

2 rows × 25 columns

# Plot search area of interest and frames on a map using Holoviz Libraries (more on these later)
cols = gf.loc[:,('id','landsat:wrs_path','landsat:wrs_row','geometry')]
footprints = cols.hvplot(geo=True, line_color='k', hover_cols=['landsat:wrs_path','landsat:wrs_row'], alpha=0.3, title='Landsat 8 T1',tiles='ESRI')
tiles = gv.tile_sources.CartoEco.options(width=700, height=500) 
labels = gv.tile_sources.StamenLabels.options(level='annotation')
tiles * footprints * labels

Intake all scenes using the intake-STAC library#

Intake-STAC facilitates discovering, exploring, and loading spatio-temporal datasets by providing Intake Drivers for STAC catalogs. This provides a simple toolkit for working with STAC catalogs and for loading STAC assets as xarray objects.

catalog = intake_stac.catalog.StacItemCollection(items)

Let’s explore the metadata and keys for the first scene

sceneids = list(catalog)
item3 = catalog[sceneids[1]]
# item3.metadata
for keys in item3.keys():
    print (keys)

We can explore the metadata for any of these:

{'href': '',
 'type': 'image/vnd.stac.geotiff; cloud-optimized=true',
 'title': 'Blue Band (B2)',
 'description': 'Collection 2 Level-1 Blue Band (B2) Top of Atmosphere Radiance',
 'eo:bands': [{'name': 'B2',
   'common_name': 'blue',
   'gsd': 30,
   'center_wavelength': 0.48}],
 'alternate': {'s3': {'storage:platform': 'AWS',
   'storage:requester_pays': True,
   'href': 's3://usgs-landsat/collection02/level-1/standard/oli-tirs/2019/008/011/LC08_L1TP_008011_20190507_20200828_02_T1/LC08_L1TP_008011_20190507_20200828_02_T1_B2.TIF'}},
 'file:checksum': '1340f0a602019909fbb3e6bd909443ef7f9adfe7efabc86586f6e5f85e941aab30f03adbde671e000852b535ce627d864c1e6c3b33011a907b9a67fa1b7e595d9f42',
 'roles': ['data'],
 'plots': {'geotiff': {'kind': 'image',
   'x': 'x',
   'y': 'y',
   'frame_width': 500,
   'data_aspect': 1,
   'rasterize': True,
   'dynamic': True,
   'cmap': 'viridis'}},
 'catalog_dir': ''}
# Either of these codes provide the url needed to grab data from the S3 bucket using the intake-STAC catalog
print(['alternate']['s3']['href']) # must use item asset name (blue)
print (items[1].assets['blue'].extra_fields['alternate']['s3']['href'])

Open and visualize each image using RasterIO#

import rasterio as rio

# Retrieve first scene using rio
item_url = items[1].assets['blue'].extra_fields['alternate']['s3']['href']

# Read and plot with grid coordinates 
with rio.Env(aws_session):
    with as src:
        fig, ax = plt.subplots(figsize=(9,8))
        # To plot
        # To open data into a numpy array
        profile = src.profile
        arr =

We can open directly into xarray using rasterIO

Manipulating data in Xarray#

Pandas and xarray have very similar structures and ways of manipulating data, but pandas excels with 2-D data and xarray is ideal for higher dimension data. Xarray introduces labels in the form of dimensions, coordinates and attributes on top of Pandas-like DataFrames.

We will only scratch the surface here on what xarray can do. To learn more, there are great xarray tutorials here:

RasterIO and RioXarray#

We can use rasterIO to easily open into an xarray DataSet:

rastxr = xr.open_dataset(item_url,engine='rasterio')
Dimensions:      (band: 1, x: 8741, y: 8771)
  * band         (band) int64 1
  * x            (x) float64 2.613e+05 2.613e+05 ... 5.235e+05 5.235e+05
  * y            (y) float64 7.856e+06 7.855e+06 ... 7.592e+06 7.592e+06
    spatial_ref  int64 ...
Data variables:
    band_data    (band, y, x) float32 ...

We can also open using rioxarray, which integrates rasterIO and xarray and is the most efficient way of opening using rasterIO:

import rioxarray as rxr

rastrxr = rxr.open_rasterio(item_url)
<xarray.DataArray (band: 1, y: 8771, x: 8741)>
[76667311 values with dtype=uint16]
  * band         (band) int64 1
  * x            (x) float64 2.613e+05 2.613e+05 ... 5.235e+05 5.235e+05
  * y            (y) float64 7.856e+06 7.855e+06 ... 7.592e+06 7.592e+06
    spatial_ref  int64 0
    AREA_OR_POINT:  Point
    _FillValue:     0
    scale_factor:   1.0
    add_offset:     0.0

We can see Attributes have been added to the xarray using the same url.

Beyond what xarray and rasterIO provide, rioxarray has these added benefits (plus others):

  • Supports multidimensional datasets such as netCDF

  • Loads in the CRS, transform, and nodata metadata in standard CF & GDAL locations

  • Supports masking and scaling data

  • Loads raster metadata into the attributes

For more info:


Another convenient means for opening a lot of raster data into xarray is using dask. Xarray integrates with Dask to support parallel computations and streaming computation on datasets that don’t fit into memory. So this is perfect when you want to process a lot of data.

Dask divides arrays into many small pieces, called chunks, each of which is presumed to be small enough to fit into memory.

Unlike NumPy, which has eager evaluation, operations on dask arrays are lazy. Operations queue up a series of tasks mapped over blocks, and no computation is performed until you actually ask values to be computed (e.g., to print results to your screen or write to disk). At that point, data is loaded into memory and computation proceeds in a streaming fashion, block-by-block.

To expand our xarray toolbox for working with larger data sets that we don’t necessarily want entirely in memory, we will start by reading in 3 bands of a Landsat scene to xarray using dask.

sceneid = items[1]
print (

band_names = ['red','green','blue']

bands = []
# Construct xarray for scene
for band_name in band_names:
    # Specify chunk size (x,y), Landsat COG is natively in 512 chunks so is best to use this or a multiple
    asset = sceneid.assets[band_name]
    href = asset.extra_fields['alternate']['s3']['href']
    band = xr.open_dataset(href, engine='rasterio', chunks=dict(band=1,x=512, y=512))
    band['band'] = [band_name]
scene = xr.concat(bands, dim='band')
Dimensions:      (band: 3, x: 8741, y: 8771)
  * band         (band) <U5 'red' 'green' 'blue'
  * x            (x) float64 2.613e+05 2.613e+05 ... 5.235e+05 5.235e+05
  * y            (y) float64 7.856e+06 7.855e+06 ... 7.592e+06 7.592e+06
    spatial_ref  int64 0
Data variables:
    band_data    (band, y, x) float32 dask.array<chunksize=(1, 512, 512), meta=np.ndarray>

Typically, it’s best to align dask chunks with the way image chunks (typically called “tiles”) are stored on disk or cloud storage buckets. The landsat data is stored on AWS S3 in a tiled Geotiff format where tiles are 512x512, so we should pick some multiple of that, and typically aim for chunk sizes of ~100Mb (although this is subjective).

In a way that is similar to pandas, we can explore variables easily in xarray. We will first work with coordinates (equivalent to indices in pandas). Here x might often be the longitude (it can be renamed to this actually):