Harmonized Landsat Sentinel 2 (HLS): tiling configuration and rendering performance¶

Background¶

The Harmonized Landsat Sentinel (HLS) dataset provides surface reflectance data by combining observations from the Operational Land Imager (OLI) and Multi-Spectral Instrument (MSI) aboard the Landsat 8/9 and Sentinel-2 satellites, respectively. This dataset is processed to enable time series analyses with greater density of observations in time than is possible with any single global satellite data collection with similar spatial resolution.

Two HLS collections are available in CMR:

• HLSL30: Harmonized Landsat 8/9 data (30m resolution)
• HLSS30: Harmonized Sentinel-2 data (30m resolution)

These collections have specific characteristics that impact the performance of tile rendering:

1. Each granule covers one MGRS (Military Grid Reference System) tile
2. The data is stored as separate files per band (e.g., B01, B02, etc.)
3. Temporal coverage is not uniform across the globe

Performance Factors¶

Several factors affect tiling performance (in order of importance):

1. Zoom level: Lower zoom levels require rendering pixels from more assets
2. Temporal range: Larger time ranges require searching through and rendering more granules
3. Collection choice: HLSS30 has a more frequent revisit rate than HLSL30
   • HLSS30's higher revisit rate means that for a fixed area and datetime interval there will usually be more granules to process than for HLSL30
4. Band combination: More bands require more files to be accessed and processed

Tile rendering benchmark analysis¶

To assess tile rendering performance for the HLS collections we wrote a benchmarking process that measures the time it takes for the API to render tiles for a range of zoom levels, datetime interval widths, and number of band combinations. This notebook explores the trends and findings from the benchmark results. The following diagram shows the observed performance across all of these dimensions for the two collections (HLSL30 and HLSS30). Based on this analysis we recommend these configuration limits:

• minzoom=7: the titiler-cmr API should be able to return a full viewport's worth of tiles in ~10 seconds.
• use single-day datetime intervals to reduce the number of assets requested in a single tile (e.g. datetime=2024-08-10T00:00:01Z/2024-08-10T23:59:59)
  • this will give users an orbit-wise view of the data, which is probably the most intuitive way to view it
  • wider datetime intervals will work, but performance can substantially slower
• limiting the other factors do not seem to introduce enough variance in response time to warrant additional guidelines.

HLSL30 (Landsat) Parameters¶

Basic Parameters¶

Parameter	Value
concept_id	"C2021957657-LPCLOUD"
backend	"rasterio"
bands_regex	"B[0-9][0-9]"
minzoom	8
maxzoom	13

True Color RGB¶

Parameter	Value
bands	["B04", "B03", "B02"]
color_formula	"Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"

False Color NIR¶

Parameter	Value
bands	["B05", "B03", "B02"]
color_formula	"Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"

NDVI¶

Parameter	Value
bands	["B05", "B04"]
expression	"(B05-B04)/(B05+B04)"
colormap_name	"summer_r"
rescale	"-0.5,1"

HLSS30 (Sentinel-2) Parameters¶

Basic Parameters¶

Parameter	Value
concept_id	"C2021957295-LPCLOUD"
backend	"rasterio"
bands_regex	"B[0-9][0-9A-Za-z]"
minzoom	8
maxzoom	13

True Color RGB¶

Parameter	Value
bands	["B04", "B03", "B02"]
color_formula	"Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"

False Color NIR¶

Parameter	Value
bands	["B8A", "B03", "B02"]
color_formula	"Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"

NDVI¶

Parameter	Value
bands	["B8A", "B04"]
expression	"(B8A-B04)/(B8A+B04)"
colormap_name	"summer_r"
rescale	"-0.5,1"

Interactive map examples¶

This section demonstrates how you might parameterize tilejson requests (in Python) for each of the collections for variety of layers (true color, false color, NDVI).

In [2]:

import httpx
from folium import LayerControl, Map, TileLayer

titiler_endpoint = "https://staging.openveda.cloud/api/titiler-cmr"
minzoom = 7

import httpx from folium import LayerControl, Map, TileLayer titiler_endpoint = "https://staging.openveda.cloud/api/titiler-cmr" minzoom = 7

HLSL30¶

In [3]:

m = Map(location=(46.7653, -91.0321), zoom_start=minzoom + 1)
base_parameters = (
    ("concept_id", "C2021957657-LPCLOUD"),
    # Datetime in form of `start_date/end_date`
    # ("datetime", "2025-03-16T00:00:00Z/2025-03-16T23:59:59Z"),  # leaf-off
    ("datetime", "2024-08-06T00:00:00Z/2024-08-06T23:59:59Z"),  # leaf-on
    # We know that the HLS collection dataset is stored as File per Band
    # so we need to pass a `band_regex` option to assign `bands` to each URL
    ("bands_regex", "B[0-9][0-9]"),
    # titiler-cmr can work with both Zarr and COG dataset
    # but we need to tell the endpoints in advance which backend
    # to use
    ("backend", "rasterio"),
    # We need to set min/max zoom because we don't want to use lowerzoom level (e.g 0)
    # which will results in useless large scale query
    ("minzoom", 7),
    ("maxzoom", 13),
)

landsat_true_color_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # True Color Image B04,B03,B02
        ("bands", "B04"),
        ("bands", "B03"),
        ("bands", "B02"),
        # The data is in type of Uint16 so we need to apply some
        # rescaling/color_formula in order to create PNGs
        ("color_formula", "Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"),
    ),
    timeout=None,
).json()

landsat_true_color_tile_url = landsat_true_color_tilejson["tiles"][0]


TileLayer(
    tiles=landsat_true_color_tile_url,
    opacity=1,
    name="HLSL30 true color",
    overlay=True,
    show=True,
    control=True,
    attr="NASA",
).add_to(m)

landsat_false_color_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # False Color Image B05,B03,B02
        ("bands", "B05"),
        ("bands", "B03"),
        ("bands", "B02"),
        # The data is in type of Uint16 so we need to apply some
        # rescaling/color_formula in order to create PNGs
        ("color_formula", "Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"),
    ),
    timeout=None,
).json()

landsat_false_color_tile_url = landsat_false_color_tilejson["tiles"][0]

TileLayer(
    tiles=landsat_false_color_tile_url,
    opacity=1,
    name="HLSL30 false color",
    overlay=True,
    show=False,
    control=True,
    attr="NASA",
).add_to(m)

landsat_ndvi_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # NDVI: B05 (NIR), B04 (Red)
        ("bands", "B05"),
        ("bands", "B04"),
        # Calculate NDVI from the NIR and Red bands
        ("expression", "(B05-B04)/(B05+B04)"),
        # Rescale to NDVI scale
        ("rescale", "-0.5,1"),
        # Choose a colormap
        ("colormap_name", "summer_r"),
    ),
    timeout=None,
).json()

landsat_ndvi_tile_url = landsat_ndvi_tilejson["tiles"][0]

TileLayer(
    tiles=landsat_ndvi_tile_url,
    opacity=1,
    name="HLSL30 NDVI",
    overlay=True,
    show=False,
    control=True,
    attr="NASA",
).add_to(m)

LayerControl(position="topright", collapsed=False).add_to(m)

m

m = Map(location=(46.7653, -91.0321), zoom_start=minzoom + 1) base_parameters = ( ("concept_id", "C2021957657-LPCLOUD"), # Datetime in form of `start_date/end_date` # ("datetime", "2025-03-16T00:00:00Z/2025-03-16T23:59:59Z"), # leaf-off ("datetime", "2024-08-06T00:00:00Z/2024-08-06T23:59:59Z"), # leaf-on # We know that the HLS collection dataset is stored as File per Band # so we need to pass a `band_regex` option to assign `bands` to each URL ("bands_regex", "B[0-9][0-9]"), # titiler-cmr can work with both Zarr and COG dataset # but we need to tell the endpoints in advance which backend # to use ("backend", "rasterio"), # We need to set min/max zoom because we don't want to use lowerzoom level (e.g 0) # which will results in useless large scale query ("minzoom", 7), ("maxzoom", 13), ) landsat_true_color_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # True Color Image B04,B03,B02 ("bands", "B04"), ("bands", "B03"), ("bands", "B02"), # The data is in type of Uint16 so we need to apply some # rescaling/color_formula in order to create PNGs ("color_formula", "Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"), ), timeout=None, ).json() landsat_true_color_tile_url = landsat_true_color_tilejson["tiles"][0] TileLayer( tiles=landsat_true_color_tile_url, opacity=1, name="HLSL30 true color", overlay=True, show=True, control=True, attr="NASA", ).add_to(m) landsat_false_color_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # False Color Image B05,B03,B02 ("bands", "B05"), ("bands", "B03"), ("bands", "B02"), # The data is in type of Uint16 so we need to apply some # rescaling/color_formula in order to create PNGs ("color_formula", "Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"), ), timeout=None, ).json() landsat_false_color_tile_url = landsat_false_color_tilejson["tiles"][0] TileLayer( tiles=landsat_false_color_tile_url, opacity=1, name="HLSL30 false color", overlay=True, show=False, control=True, attr="NASA", ).add_to(m) landsat_ndvi_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # NDVI: B05 (NIR), B04 (Red) ("bands", "B05"), ("bands", "B04"), # Calculate NDVI from the NIR and Red bands ("expression", "(B05-B04)/(B05+B04)"), # Rescale to NDVI scale ("rescale", "-0.5,1"), # Choose a colormap ("colormap_name", "summer_r"), ), timeout=None, ).json() landsat_ndvi_tile_url = landsat_ndvi_tilejson["tiles"][0] TileLayer( tiles=landsat_ndvi_tile_url, opacity=1, name="HLSL30 NDVI", overlay=True, show=False, control=True, attr="NASA", ).add_to(m) LayerControl(position="topright", collapsed=False).add_to(m) m

Out[3]:

Make this Notebook Trusted to load map: File -> Trust Notebook

HLSS30¶

In [4]:

m = Map(location=(46.7653, -91.0321), zoom_start=minzoom + 1)

base_parameters = (
    ("concept_id", "C2021957295-LPCLOUD"),
    # Datetime in form of `start_date/end_date`
    ("datetime", "2024-08-13T00:00:00Z/2024-08-13T23:59:59Z"),  # leaf-on
    # ("datetime", "2025-03-16T00:00:00Z/2025-03-16T23:59:59Z"),  # leaf-off
    # We know that the HLS collection dataset is stored as File per Band
    # so we need to pass a `band_regex` option to assign `bands` to each URL
    ("bands_regex", "B[0-9][0-9A-Za-z]"),
    # titiler-cmr can work with both Zarr and COG dataset
    # but we need to tell the endpoints in advance which backend
    # to use
    ("backend", "rasterio"),
    # We need to set min/max zoom because we don't want to use lowerzoom level (e.g 0)
    # which will results in useless large scale query
    ("minzoom", 7),
    ("maxzoom", 13),
)

sentinel_true_color_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # True Color Image B04,B03,B02
        ("bands", "B04"),
        ("bands", "B03"),
        ("bands", "B02"),
        # The data is in type of Uint16 so we need to apply some
        # rescaling/color_formula in order to create PNGs
        ("color_formula", "Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"),
    ),
    timeout=None,
).json()

sentinel_true_color_tile_url = sentinel_true_color_tilejson["tiles"][0]


TileLayer(
    tiles=sentinel_true_color_tile_url,
    opacity=1,
    name="HLSS30 true color",
    overlay=True,
    show=True,
    control=True,
    attr="NASA",
).add_to(m)

sentinel_false_color_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # False Color Image B8A, B03, B02
        ("bands", "B8A"),
        ("bands", "B03"),
        ("bands", "B02"),
        # The data is in type of Uint16 so we need to apply some
        # rescaling/color_formula in order to create PNGs
        ("color_formula", "Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"),
    ),
    timeout=None,
).json()

sentinel_false_color_tile_url = sentinel_false_color_tilejson["tiles"][0]

TileLayer(
    tiles=sentinel_false_color_tile_url,
    opacity=1,
    name="HLSS30 false color",
    overlay=True,
    show=False,
    control=True,
    attr="NASA",
).add_to(m)

sentinel_ndvi_tilejson = httpx.get(
    f"{titiler_endpoint}/WebMercatorQuad/tilejson.json",
    params=(
        *base_parameters,
        # NDVI: B8A (NIR), B04 (Red)
        ("bands", "B8A"),
        ("bands", "B04"),
        # Calculate NDVI from the NIR and Red bands
        ("expression", "(B8A-B04)/(B8A+B04)"),
        # Rescale to -1,1
        ("rescale", "-0.5,1"),
        # Choose a colormap
        ("colormap_name", "summer_r"),
    ),
    timeout=None,
).json()

sentinel_ndvi_tile_url = sentinel_ndvi_tilejson["tiles"][0]

TileLayer(
    tiles=sentinel_ndvi_tile_url,
    opacity=1,
    name="HLSS30 NDVI",
    overlay=True,
    show=False,
    control=True,
    attr="NASA",
).add_to(m)

LayerControl(position="topright", collapsed=False).add_to(m)

m

m = Map(location=(46.7653, -91.0321), zoom_start=minzoom + 1) base_parameters = ( ("concept_id", "C2021957295-LPCLOUD"), # Datetime in form of `start_date/end_date` ("datetime", "2024-08-13T00:00:00Z/2024-08-13T23:59:59Z"), # leaf-on # ("datetime", "2025-03-16T00:00:00Z/2025-03-16T23:59:59Z"), # leaf-off # We know that the HLS collection dataset is stored as File per Band # so we need to pass a `band_regex` option to assign `bands` to each URL ("bands_regex", "B[0-9][0-9A-Za-z]"), # titiler-cmr can work with both Zarr and COG dataset # but we need to tell the endpoints in advance which backend # to use ("backend", "rasterio"), # We need to set min/max zoom because we don't want to use lowerzoom level (e.g 0) # which will results in useless large scale query ("minzoom", 7), ("maxzoom", 13), ) sentinel_true_color_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # True Color Image B04,B03,B02 ("bands", "B04"), ("bands", "B03"), ("bands", "B02"), # The data is in type of Uint16 so we need to apply some # rescaling/color_formula in order to create PNGs ("color_formula", "Gamma RGB 3.5 Saturation 1.2 Sigmoidal RGB 15 0.35"), ), timeout=None, ).json() sentinel_true_color_tile_url = sentinel_true_color_tilejson["tiles"][0] TileLayer( tiles=sentinel_true_color_tile_url, opacity=1, name="HLSS30 true color", overlay=True, show=True, control=True, attr="NASA", ).add_to(m) sentinel_false_color_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # False Color Image B8A, B03, B02 ("bands", "B8A"), ("bands", "B03"), ("bands", "B02"), # The data is in type of Uint16 so we need to apply some # rescaling/color_formula in order to create PNGs ("color_formula", "Gamma RGB 2.5 Saturation 1.2 Sigmoidal RGB 10 0.35"), ), timeout=None, ).json() sentinel_false_color_tile_url = sentinel_false_color_tilejson["tiles"][0] TileLayer( tiles=sentinel_false_color_tile_url, opacity=1, name="HLSS30 false color", overlay=True, show=False, control=True, attr="NASA", ).add_to(m) sentinel_ndvi_tilejson = httpx.get( f"{titiler_endpoint}/WebMercatorQuad/tilejson.json", params=( *base_parameters, # NDVI: B8A (NIR), B04 (Red) ("bands", "B8A"), ("bands", "B04"), # Calculate NDVI from the NIR and Red bands ("expression", "(B8A-B04)/(B8A+B04)"), # Rescale to -1,1 ("rescale", "-0.5,1"), # Choose a colormap ("colormap_name", "summer_r"), ), timeout=None, ).json() sentinel_ndvi_tile_url = sentinel_ndvi_tilejson["tiles"][0] TileLayer( tiles=sentinel_ndvi_tile_url, opacity=1, name="HLSS30 NDVI", overlay=True, show=False, control=True, attr="NASA", ).add_to(m) LayerControl(position="topright", collapsed=False).add_to(m) m

Out[4]:

Make this Notebook Trusted to load map: File -> Trust Notebook