STAC Catalogue

Link Colab

Open In Colab

Nós vimos que é facil entrar em um provedor, pesquisar uma data e encontrar as images que queremos. Porém baixar tudo na mão parece ser extremamente trabalhoso, demandando muito tempo para prosseguir. Dessa forma, a comunidade engajada do mundo geoespacial criou o STAC (Spatial Temporal Asset Catalogue), que basicamente reune todos os provedores e dados em um sistema organizado que te permite buscar, localizar e baixar as images de sua preferência. Primeiramente vamos nos familiarizar um pouco com o STAC antes de prosseguirmos com operações mais avançadas de sensoriamento remoto.

# rode isso apenas se voce estiver no colab! 
# git clone https://github.com/ftl-brazil-2026/ebook.git

# rode isso apenas se voce estiver no colab! 
#!pip install -r /content/ebook/requirements.txt

from pystac_client import Client

Client?

Precisamos entao encontrar a nossa API para depois conseguir baixar os dados. Assim, o próprio site da STAC favorece uma lista de API STAC-API-catalogo

cbers_inpe = "https://data.inpe.br/bdc/stac/v1/"

client = Client.open(cbers_inpe)

# vamos listar todas as colecoes
for collection in client.get_collections():
    print(collection)

<CollectionClient id=GOES19-L2-CMI-1>
<CollectionClient id=S2_L2A-1>
<CollectionClient id=AMZ1-WFI-L2-DN-1>
<CollectionClient id=mosaic-cbers4a-paraiba-3m-1>
<CollectionClient id=mosaic-landsat-sp-6m-1>
<CollectionClient id=mosaic-s2-paraiba-3m-1>
<CollectionClient id=mosaic-s2-yanomami_territory-6m-1>
<CollectionClient id=modisa-ocsmart-rrs-monthly-1>
<CollectionClient id=landsat-2>
<CollectionClient id=s3-olci-rrs-water-1>
<CollectionClient id=AMZ1-WFI-L4-DN-1>
<CollectionClient id=s3-olci-pc-tiete-1>
<CollectionClient id=mod13q1-6.1>
<CollectionClient id=sentinel-1-grd-bundle-1>
<CollectionClient id=s2_msi_mq_wqp-1>
<CollectionClient id=sentinel-1-rtc-1>
<CollectionClient id=sentinel-3-olci-l1-bundle-1>
<CollectionClient id=CB4-MUX-L2-DN-1>
<CollectionClient id=s2_msi_mq_rrs-1>
<CollectionClient id=S2-16D-2>
<CollectionClient id=GOES-GL-DSWRF-Daily-1>
<CollectionClient id=mosaic-landsat-amazon-3m-1>
<CollectionClient id=CBERS4-WFI-16D-2>
<CollectionClient id=mosaic-s2-amazon-3m-1>
<CollectionClient id=landsat-tsirf-bimonthly-1>
<CollectionClient id=mosaic-landsat-brazil-6m-1>
<CollectionClient id=myd13q1-6.1>
<CollectionClient id=CB4-WFI-L4-SR-1>
<CollectionClient id=CB2B-HRC-L2-DN-1>
<CollectionClient id=CB4A-WPM-L2-DN-1>
<CollectionClient id=charter-wfi-1>
<CollectionClient id=topodata-1>
<CollectionClient id=mosaic-s2-cerrado-4m-1>
<CollectionClient id=modisa-ocsmart-biogeochemical-monthly-1>
<CollectionClient id=samet_hourly-1>
<CollectionClient id=mosaic-cbers4-brazil-3m-1>
<CollectionClient id=CBERS-WFI-8D-1>
<CollectionClient id=S2_L1C_BUNDLE-1>
<CollectionClient id=CB4A-MUX-L2-DN-1>
<CollectionClient id=landsat-tsirf-yearly-1>
<CollectionClient id=modisa-ocsmart-biogeochemical-daily-1>
<CollectionClient id=mosaic-s2-cerrado-2m-1>
<CollectionClient id=CB4-PAN10M-L2-DN-1>
<CollectionClient id=landsat-lgi-1>
<CollectionClient id=AMZ1-WFI-L4-SR-1>
<CollectionClient id=CB4-PAN5M-L4-DN-1>
<CollectionClient id=CBERS4-MUX-2M-1>
<CollectionClient id=CB4A-WFI-L4-SR-1>
<CollectionClient id=charter-mux-1>
<CollectionClient id=CB4-WFI-L2-DN-1>
<CollectionClient id=CB4A-WPM-L4-DN-1>
<CollectionClient id=mod11a2-6.1>
<CollectionClient id=CB4A-WFI-L2-DN-1>
<CollectionClient id=CB4-PAN10M-L4-DN-1>
<CollectionClient id=CB4-PAN5M-L2-DN-1>
<CollectionClient id=mosaic-s2-brazil-3m-1>
<CollectionClient id=mosaic-s2-amazon-1m-1>
<CollectionClient id=CB4A-MUX-L4-DN-1>
<CollectionClient id=CB4A-WFI-L4-DN-1>
<CollectionClient id=GOES13-L3-IMAGER-1>
<CollectionClient id=CB2B-CCD-L2-DN-1>
<CollectionClient id=myd11a2-6.1>
<CollectionClient id=CB2B-WFI-L2-DN-1>
<CollectionClient id=modisa-ocsmart-rrs-daily-1>
<CollectionClient id=charter-wpm-1>
<CollectionClient id=LANDSAT-16D-1>
<CollectionClient id=topodata_bundle-1>
<CollectionClient id=charter-pan-1>
<CollectionClient id=CB4A-MUX-L4-SR-1>
<CollectionClient id=CB4-MUX-L4-SR-1>
<CollectionClient id=prec_merge_hourly-1>
<CollectionClient id=CB4-WFI-L4-DN-1>
<CollectionClient id=CB4-MUX-L4-DN-1>
<CollectionClient id=samet_daily-1>
<CollectionClient id=CB4A-WPM-PCA-FUSED-1>
<CollectionClient id=mosaic-s2-amazon-3m-b08b04b03-1>
<CollectionClient id=EtaCCDay_CMIP5-1>
<CollectionClient id=GOES16-L2-CMI-1>
<CollectionClient id=CB2-CCD-L2-DN-1>
<CollectionClient id=prec_merge_daily-1>

# vamos ver a metadata
collection = client.get_collection('CB4-MUX-L4-SR-1')

collection

<CollectionClient id=CB4-MUX-L4-SR-1>

• type "Collection"
• id "CB4-MUX-L4-SR-1"
• stac_version "1.1.0"
• description "CBERS-4/MUX - Level-4 Surface Reflectance product over Brazil and part of South America. L4 SR product provides orthorectified surface reflectance images. This dataset is provided as Cloud Optimized GeoTIFF (COG)."
• links[] 5 items

  • 0

    • rel "self"
    • href "https://data.inpe.br/bdc/stac/v1/collections/CB4-MUX-L4-SR-1"
    • type "application/json"
    • title "Link to this document"

  • 1

    • rel "items"
    • href "https://data.inpe.br/bdc/stac/v1/collections/CB4-MUX-L4-SR-1/items"
    • type "application/json"
    • title "Items of the collection CB4-MUX-L4-SR-1"

  • 2

    • rel "parent"
    • href "https://data.inpe.br/bdc/stac/v1/"
    • type "application/json"
    • title "Link to catalog collections"

  • 3

    • rel "root"
    • href "https://data.inpe.br/bdc/stac/v1/"
    • type "application/json"
    • title "INPE STAC Server"

  • 4

    • rel "license"
    • href "https://creativecommons.org/licenses/by/4.0/"
    • title "Creative-Commons-Attribution-4.0-International"
• stac_extensions[] 4 items

  • 0 "https://stac-extensions.github.io/version/v1.0.0/schema.json"

  • 1 "https://stac-extensions.github.io/processing/v1.0.0/schema.json"

  • 2 "https://stac-extensions.github.io/item-assets/v1.0.0/schema.json"

  • 3 "https://stac-extensions.github.io/eo/v1.0.0/schema.json"
• version "1"
• deprecated False
• item_assets

  • BAND5

    • type "image/tiff; application=geotiff; profile=cloud-optimized"
    • roles[] 1 items

      • 0 "data"
    • title "Band blue"
  • BAND6

    • type "image/tiff; application=geotiff; profile=cloud-optimized"
    • roles[] 1 items

      • 0 "data"
    • title "Band green"
  • BAND7

    • type "image/tiff; application=geotiff; profile=cloud-optimized"
    • roles[] 1 items

      • 0 "data"
    • title "Band red"
  • BAND8

    • type "image/tiff; application=geotiff; profile=cloud-optimized"
    • roles[] 1 items

      • 0 "data"
    • title "Band nir"
  • CMASK

    • type "image/tiff; application=geotiff; profile=cloud-optimized"
    • roles[] 1 items

      • 0 "data"
    • title "Band cmask"
  • thumbnail

    • type "image/png"
    • bands

      • red "BAND7"
      • blue "BAND5"
      • green "BAND8"
    • roles[] 1 items

      • 0 "thumbnail"
    • title "Thumbnail"
• properties

  • renders

    • rgb

      • title "True Color"
      • assets[] 3 items

        • 0 "BAND7"

        • 1 "BAND6"

        • 2 "BAND5"
      • rescale[] 3 items

        • 0[] 2 items

          • 0 0

          • 1 3000

        • 1[] 2 items

          • 0 0

          • 1 3000

        • 2[] 2 items

          • 0 0

          • 1 3000
      • resampling "nearest"
  • sources[] 1 items

    • 0 "http://www.inpe.br"
  • bdc:visual

    • rgb[] 3 items

      • 0 "BAND7"

      • 1 "BAND6"

      • 2 "BAND5"
    • rescale[] 3 items

      • 0[] 2 items

        • 0 0

        • 1 3000

      • 1[] 2 items

        • 0 0

        • 1 3000

      • 2[] 2 items

        • 0 0

        • 1 3000
  • bdc:applications[] 1 items

    • 0 "BDCExplorer"
  • created "2024-03-05T17:06:15.788121Z"
  • updated "2024-03-05T17:06:15.788130Z"
  • eo:bands[] 5 items

    • 0

      • name "BAND5"
      • common_name "blue"
      • description "B5 blue band from 0.45 to 0.52 micrometers"
      • min 0.0
      • max 10000.0
      • nodata -9999.0
      • scale 0.0001
      • scale_add None
      • data_type "int16"
      • resolution_x 20.0
      • resolution_y 20.0
      • center_wavelength 0.485
      • full_width_half_max 0.07

    • 1

      • name "CMASK"
      • common_name "quality"
      • description "Cloud Mask Product"
      • min 0.0
      • max 4.0
      • nodata 255.0
      • scale 1.0
      • scale_add None
      • data_type "uint8"
      • resolution_x 20.0
      • resolution_y 20.0

    • 2

      • name "BAND7"
      • common_name "red"
      • description "B7 red band from 0.63 to 0.69 micrometers"
      • min 0.0
      • max 10000.0
      • nodata -9999.0
      • scale 0.0001
      • scale_add None
      • data_type "int16"
      • resolution_x 20.0
      • resolution_y 20.0
      • center_wavelength 0.66
      • full_width_half_max 0.06

    • 3

      • name "BAND8"
      • common_name "nir"
      • description "B8 nir band from 0.77 to 0.89 micrometers"
      • min 0.0
      • max 10000.0
      • nodata -9999.0
      • scale 0.0001
      • scale_add None
      • data_type "int16"
      • resolution_x 20.0
      • resolution_y 20.0
      • center_wavelength 0.83
      • full_width_half_max 0.12

    • 4

      • name "BAND6"
      • common_name "green"
      • description "B6 green band from 0.52 to 0.59 micrometers"
      • min 0.0
      • max 10000.0
      • nodata -9999.0
      • scale 0.0001
      • scale_add None
      • data_type "int16"
      • resolution_x 20.0
      • resolution_y 20.0
      • center_wavelength 0.555
      • full_width_half_max 0.07
• bdc:type "collection"
• bdc:public True
• bdc:grs "CBERS4_MUX_GRID"
• bdc:bands_quicklook[] 3 items

  • 0 "BAND7"

  • 1 "BAND8"

  • 2 "BAND5"
• title "CBERS-4/MUX - Level-4-SR - Cloud Optimized GeoTIFF"
• extent

  • spatial

    • bbox[] 1 items

      • 0[] 4 items

        • 0 -79.86115089672425

        • 1 -40.79283317617634

        • 2 100.42739064486346

        • 3 54.005907068086806
  • temporal

    • interval[] 1 items

      • 0[] 2 items

        • 0 "2016-01-01T00:00:00Z"

        • 1 "2026-08-29T00:00:00Z"
• license "Creative-Commons-Attribution-4.0-International"
• keywords[] 10 items

  • 0 "cbers"

  • 1 "cbers-4"

  • 2 "mux"

  • 3 "level-4"

  • 4 "surface reflectance"

  • 5 "cloud optimized geotiff"

  • 6 "cog"

  • 7 "earth observation"

  • 8 "brazil"

  • 9 "visible imagery"
• providers[] 1 items

  • 0

    • name "National Institute for Space Research (INPE)"
    • roles[] 4 items

      • 0 "producer"

      • 1 "licensor"

      • 2 "host"

      • 3 "processor"
    • url "https://data.inpe.br/big/"
    • processing:level "L4 SR"
    • processing:lineage "Generation of L4 in Cloud Optimized GeoTIFF"
    • processing:facility "LGI"
    • processing:software

      • rio-cogeo "3.0"
      • bdc-collection-builder "1.0"
• summaries

  • gsd[] 1 items

    • 0 20
  • platform[] 1 items

    • 0 "cbers-4"
  • instruments[] 1 items

    • 0 "mux"
  • constellation[] 1 items

    • 0 "cbers-4"
  • view:off_nadir[] 1 items

    • 0 0

Encontre a sua bbox por aqui:

bbox finder

search = client.search(
    max_items=10,
    collections=["sentinel-2-c1-l2a"],
    bbox=[-45.963435, -11.891593, -45.423607, -11.525831]
)

item_search = client.search(bbox=[-45.963435, -11.891593, -45.423607, -11.525831],
                             datetime='2024-08-01/2025-07-31',
                             collections=['CB4-MUX-L4-SR-1'],
                             query={
                                 "eo:cloud_cover": {"lt": 15}
                                 })

## voce pode passar mais filtros as query como
# query={
#         "tileId": {"in": ["22JHT", "22JGT", "22KHU", "22KGU"]}
# }
## nesse caso selecionando todos os tiles
item_search

<pystac_client.item_search.ItemSearch at 0x10e8f60d0>

Esse método retorna uma representação ItemSearch a qual possue métodos acessíveis para identificar os resultados

item_search.matched()

19

for item in item_search.items():
    print(item)

<Item id=CBERS_4_MUX_20250722_156_113_L4>
<Item id=CBERS_4_MUX_20250719_157_113_L4>
<Item id=CBERS_4_MUX_20250626_156_113_L4>
<Item id=CBERS_4_MUX_20250623_157_113_L4>
<Item id=CBERS_4_MUX_20250531_156_113_L4>
<Item id=CBERS_4_MUX_20250528_157_113_L4>
<Item id=CBERS_4_MUX_20250505_156_113_L4>
<Item id=CBERS_4_MUX_20250409_156_113_L4>
<Item id=CBERS_4_MUX_20250314_156_113_L4>
<Item id=CBERS_4_MUX_20250121_156_113_L4>
<Item id=CBERS_4_MUX_20241223_157_113_L4>
<Item id=CBERS_4_MUX_20241130_156_113_L4>
<Item id=CBERS_4_MUX_20241101_157_113_L4>
<Item id=CBERS_4_MUX_20241009_156_113_L4>
<Item id=CBERS_4_MUX_20241006_157_113_L4>
<Item id=CBERS_4_MUX_20240913_156_113_L4>
<Item id=CBERS_4_MUX_20240910_157_113_L4>
<Item id=CBERS_4_MUX_20240818_156_113_L4>
<Item id=CBERS_4_MUX_20240815_157_113_L4>

Isso é tudo? Claro que não, também podemos filtrar através do nosso querido geopandas

Toma essa pedrada aqui

import geobr

muni = geobr.read_municipality(year=2020)
muni = muni.loc[muni['abbrev_state']=='PA']

# convert to WGS84 (padrao internacional)
acara = muni.to_crs(4326).iloc[3]

geom = acara.geometry

bounds = geom.bounds

print(f"Geometria: {geom}")
print()
print(f"Tipo da geometria: {geom.geom_type}")
print()
print(f"Bounds: {bounds}")

Geometria: POLYGON ((-50.50552749999999 0.18421312, -50.487641394 0.182131646, -50.47088226999999 0.177517622, -50.453262471 0.168380503, -50.432880081 0.149561829, -50.412970485 0.11246927, -50.39569862999999 0.050498172, -50.392408435 0.043802715, -50.377884294 0.022813954, -50.33181063899999 -0.013135121, -50.303169569 -0.058106244, -50.298803919 -0.0682877, -50.29750065799999 -0.076907141, -50.29866601199999 -0.101589071, -50.195977641 -0.213524257, -50.107219092 -0.470861646, -50.114524226 -0.4647672029999998, -50.233640676 -0.312785947, -50.23793479799999 -0.314432649, -50.24891409999999 -0.328941307, -50.248903115 -0.331782522, -50.26709537499999 -0.343060697, -50.268934353999995 -0.347090542, -50.26434494699999 -0.349092106, -50.26432067199999 -0.3536532110000001, -50.272549680000004 -0.3563823700000001, -50.270107392 -0.362346172, -50.271754099 -0.3664660499999999, -50.27124171199999 -0.369450066, -50.272674329999994 -0.370605038, -50.270769655 -0.374336869, -50.27273425599999 -0.388825775, -50.27099588899999 -0.394919251, -50.271329635 -0.401021416, -50.27646656899999 -0.4053761619999999, -50.27706008100001 -0.416254322, -50.283704306 -0.4177852570000001, -50.286514186 -0.4270074689999999, -50.28729715299999 -0.442007743, -50.28523355499999 -0.455084404, -50.287532821000006 -0.460592909, -50.288298403 -0.471919366, -50.301986371999995 -0.470859188, -50.307638506 -0.472719031, -50.310181165 -0.4790833199999999, -50.317934920999996 -0.487018107, -50.322946865 -0.4956312580000001, -50.319260584 -0.500772729, -50.325793038 -0.5181105779999999, -50.326153420999994 -0.5228897879999999, -50.322618885999994 -0.528411348, -50.32019509399999 -0.5365424339999999, -50.328949257999994 -0.544015078, -50.330039681 -0.5507829179999999, -50.328881925 -0.555549062, -50.33328244799999 -0.557463329, -50.333877390999994 -0.564264892, -50.329313285 -0.568870252, -50.324134046 -0.5706019699999999, -50.322884561 -0.577940301, -50.31977519299999 -0.580010917, -50.319087241 -0.5821934969999999, -50.318832081 -0.586903864, -50.323915663 -0.5980117089999998, -50.324476396 -0.6030956329999999, -50.318305732999995 -0.613684691, -50.325268711 -0.6207799480000001, -50.327395939 -0.633144343, -50.334452106 -0.6382022349999998, -50.338979755 -0.636125328, -50.342803483999994 -0.629409833, -50.35157846799999 -0.627441748, -50.35802605399999 -0.6294282479999999, -50.358656763999996 -0.637645688, -50.362324468999994 -0.6419819929999999, -50.370151979 -0.6418981489999999, -50.385097142 -0.6384161800000001, -50.391231058 -0.6444433, -50.39913706699999 -0.648308953, -50.406532054 -0.6659411079999998, -50.404146363 -0.675465542, -50.406410551 -0.689295778, -50.404492082 -0.696236134, -50.40585441899999 -0.700865854, -50.41771023599999 -0.704873167, -50.432179221999995 -0.701867027, -50.436514166 -0.699820217, -50.441191151000005 -0.6925548149999999, -50.449436433 -0.6870985599999999, -50.462711301000006 -0.684298589, -50.467010671999994 -0.692465787, -50.469845116 -0.7123114669999999, -50.473062633 -0.7206367619999999, -50.47676213700001 -0.723834631, -50.477493443 -0.7209560479999999, -50.49059907 -0.710308105, -50.512800934999994 -0.6999079549999999, -50.518257145 -0.6953118319999999, -50.527515339 -0.68045534, -50.53135603199999 -0.676765208, -50.537101574 -0.6741712369999999, -50.541557489 -0.6724600440000001, -50.57699427699999 -0.669867023, -50.605155769 -0.6698608179999999, -50.60898339799999 -0.668087617, -50.612294926 -0.6661255549999999, -50.61682577199999 -0.6591861449999998, -50.619997794999996 -0.649761966, -50.624497212 -0.6435707689999999, -50.632086313 -0.639390639, -50.640867312 -0.6368965659999999, -50.659244967999996 -0.638012331, -50.663704722000006 -0.637149355, -50.684189377 -0.622449044, -50.686422399 -0.6189020770000001, -50.690996905999995 -0.5942915549999997, -50.701006869 -0.579534066, -50.71069481 -0.556963541, -50.728749392999994 -0.539773115, -50.737000992999995 -0.525772722, -50.746394105 -0.5154547189999998, -50.76122297199999 -0.5070380720000001, -50.77125400399999 -0.50881465, -50.792674124 -0.521835048, -50.80392162399999 -0.5252070229999999, -50.842681384 -0.531944163, -50.860085932 -0.5394353959999999, -50.913504018999994 -0.555401725, -50.958113889 -0.5744286869999999, -50.968225871 -0.581198295, -50.9858983 -0.596611301, -51.011059573 -0.6309225189999998, -51.02233707399999 -0.641487667, -51.044752748 -0.6533240160000001, -51.055047087000005 -0.6566700119999999, -51.079399981 -0.661832551, -51.094455161999996 -0.66155512, -51.099919922999995 -0.6570233319999998, -51.11789186199999 -0.649309436, -51.12356556799999 -0.6388722989999999, -51.121839247 -0.6251766429999999, -51.104643738 -0.594502246, -51.09292411300001 -0.5628769540000002, -51.095753338 -0.546356702, -51.101795921999994 -0.5410687569999999, -51.10443970099999 -0.5410684319999999, -51.12425473 -0.550700651, -51.144673454999996 -0.5478598399999999, -51.150720243 -0.545025568, -51.16715731699999 -0.5314234029999998, -51.17489231999999 -0.520164973, -51.180914406999996 -0.505630156, -51.18055914699999 -0.501468177, -51.178103252999996 -0.4990136149999998, -51.16109016699999 -0.492985153, -51.150338569999995 -0.4861773559999999, -51.14844981400001 -0.483156327, -51.14938797599999 -0.4682314609999999, -51.14844216699999 -0.4614347990000001, -51.142211851999996 -0.448974706, -51.14094799699999 -0.436973151, -51.143517591 -0.420682553, -51.14825141699999 -0.409256593, -51.155051188 -0.4024578489999999, -51.16529030699999 -0.4049593479999999, -51.19585597 -0.4202054969999999, -51.209096825 -0.4239807309999999, -51.229505763 -0.4290801839999999, -51.272033689999994 -0.431906376, -51.303847442 -0.4367483019999999, -51.33485433800001 -0.44534919, -51.35316300699999 -0.45334942, -51.363036859 -0.459991353, -51.371928521 -0.463975806, -51.427895557999996 -0.4764486929999999, -51.432479184 -0.4733425139999999, -51.42843357499999 -0.4619430110000001, -51.423669527 -0.436611184, -51.42072552999999 -0.428993181, -51.41667555 -0.424349896, -51.403771327 -0.4147374899999999, -51.36836991199999 -0.376257533, -51.362505289 -0.368107914, -51.362004016 -0.358820731, -51.354451847 -0.318524078, -51.34778282599999 -0.29961334, -51.339009671 -0.284576967, -51.332118322 -0.27378442, -51.323032110999996 -0.2637693280000001, -51.314950887 -0.2589056240000001, -51.306900666 -0.256091726, -51.302481611999994 -0.252642652, -51.28959545699999 -0.228076845, -51.259976644000005 -0.186993986, -51.257173959 -0.1805518989999999, -51.24098061499999 -0.156214057, -51.235036198 -0.143639152, -51.21636285899999 -0.118612786, -51.211565834 -0.118558104, -51.174898631 -0.104534773, -51.15136097399999 -0.103185898, -51.11320271499999 -0.096607927, -51.09972835899999 -0.091257984, -51.097034527 -0.088359381, -51.084918057 -0.087540136, -51.075930658999994 -0.083732497, -51.065674179 -0.076226305, -51.039642633999996 -0.043500375, -51.017338174 -0.010723634, -51.007475584 -0.000345649, -51.000473192 -0.000345649, -50.99691225099999 0.003451082, -50.985116033999994 0.029420072, -50.980962928 0.035405772, -50.95947258599999 0.05385794, -50.946239869 0.062709271, -50.921093250999995 0.076133569, -50.80351855 0.106323446, -50.76195613600001 0.120977995, -50.74350884099999 0.130485156, -50.718752435 0.146967632, -50.700979577999995 0.161606765, -50.668197496 0.185034472, -50.623601042999994 0.231776274, -50.606778257 0.245736404, -50.599667974 0.249357407, -50.59747472099999 0.244017858, -50.580159097 0.227188826, -50.558318697999994 0.212895155, -50.54415736599999 0.201131342, -50.52301344 0.188375297, -50.50552749999999 0.18421312))

Tipo da geometria: Polygon

Bounds: (-51.432479184, -0.723834631, -50.107219092, 0.249357407)

import folium
from shapely.geometry import box

bounds_polygon = box(*bounds)

centro_lat = (bounds[1] + bounds[3]) / 2
centro_lon = (bounds[0] + bounds[2]) / 2

m = folium.Map(location=[centro_lat, centro_lon], zoom_start=15)

folium.GeoJson(
    geom,
    name="Acara",
    style_function=lambda x: {
        'fillColor': 'green',
        'color': 'darkgreen',
        'weight': 2,
        'fillOpacity': 0.5
    }
).add_to(m)

folium.GeoJson(
    bounds_polygon.__geo_interface__,
    name="Bounds",
    style_function=lambda x: {
        'fillColor': 'none',
        'color': 'blue',
        'weight': 3,
        'fillOpacity': 0
    }
).add_to(m)

m.fit_bounds([[bounds[1], bounds[0]], [bounds[3], bounds[2]]])
m

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

item_search = Client.open('https://data.inpe.br/bdc/stac/v1/').search(intersects=geom,
                             datetime='2024-08-01/2025-07-31',
                             collections=['CB4-MUX-L4-SR-1'],
                             query={"eo:cloud_cover": {"lt": 15}})
print(item_search.matched())

7

item_list = list(item_search.items())
item_list

[<Item id=CBERS_4_MUX_20250727_163_101_L4>,
 <Item id=CBERS_4_MUX_20250724_164_101_L4>,
 <Item id=CBERS_4_MUX_20250628_164_101_L4>,
 <Item id=CBERS_4_MUX_20241011_164_100_L4>,
 <Item id=CBERS_4_MUX_20240915_164_100_L4>,
 <Item id=CBERS_4_MUX_20240915_164_101_L4>,
 <Item id=CBERS_4_MUX_20240817_165_100_L4>]

Assets

Assets é onde estão guardado as metadatas, thumbnails, imagens e bandas.

item_n = 6
print(item_list[item_n])
item = item_list[item_n]

<Item id=CBERS_4_MUX_20240817_165_100_L4>

item.assets

{'BAND5': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100_L4_BAND5_GRID_SURFACE.tif>,
 'BAND6': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100_L4_BAND6_GRID_SURFACE.tif>,
 'BAND7': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100_L4_BAND7_GRID_SURFACE.tif>,
 'BAND8': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100_L4_BAND8_GRID_SURFACE.tif>,
 'CMASK': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100_L4_CMASK_GRID_SURFACE.tif>,
 'thumbnail': <Asset href=https://data.inpe.br/bdc/data/CB4-MUX-L4-SR/2024_08/CBERS_4_MUX_DRD_2024_08_17.13_26_30_CB11/165_100_0/4_BC_UTM_WGS84/CBERS_4_MUX_20240817_165_100.png>}

for k in item.assets.keys():
    print(k)

BAND5
BAND6
BAND7
BAND8
CMASK
thumbnail

Rasterio e Numpy para acessar as images de satélite

import rasterio
import numpy as np
import matplotlib.pyplot as plt 

Monitorando o uso de RAM

Imagens de satélite em resolução completa ocupam bastante memória (uma única banda dessa cena tem quase 51 milhões de pixels!). Se seu computador tem 4GB de RAM, é fácil travar o Jupyter sem perceber. Vamos criar uma função com psutil para acompanhar o consumo de memória RAM ao longo do notebook.

import psutil

def check_memory(label: str = "", warn_threshold: float = 85.0) -> None:
    """Imprime o uso atual de RAM e alerta se estiver perto do limite da máquina."""
    mem = psutil.virtual_memory()
    used_gb = mem.used / (1024 ** 3)
    total_gb = mem.total / (1024 ** 3)
    available_gb = mem.available / (1024 ** 3)

    prefix = f"[RAM] {label}: " if label else "[RAM] "
    print(f"{prefix}{used_gb:.2f} GB usados de {total_gb:.2f} GB ({mem.percent:.1f}%) | disponível: {available_gb:.2f} GB")

    if mem.percent >= warn_threshold:
        print(f"ATENÇÃO: uso de RAM acima de {warn_threshold:.0f}%! Considere reiniciar o kernel, "
              "fechar outros programas, ou usar leitura por janelas (Window), como veremos mais adiante.")

check_memory("linha de base")

[RAM] linha de base: 7.38 GB usados de 32.00 GB (69.2%) | disponível: 9.86 GB

assets = item.assets
with rasterio.open(assets['BAND5'].href) as blue_ds:
    blue = blue_ds.read(1)

blue.shape

(6945, 7281)

blue.mean()

np.float64(-2712.703948173639)

with rasterio.open(assets['BAND6'].href) as green_ds:
    green = green_ds.read(1)
with rasterio.open(assets['BAND7'].href) as blue_ds:
    red = blue_ds.read(1)

check_memory("depois de carregar 3 bandas em resolução total")

[RAM] depois de carregar 3 bandas em resolução total: 8.33 GB usados de 32.00 GB (71.6%) | disponível: 9.08 GB

fig, (ax1, ax2, ax3) = plt.subplots(1,3, figsize=(12, 4))
ax1.imshow(red, cmap='gray')
ax2.imshow(green, cmap='gray')
ax3.imshow(blue, cmap='gray')

vamos criar uma composição agora com as 3 bandas: RGB

def normalize(array, brightness_factor=1):
    """Normalizes numpy arrays with optional brightness adjustment"""
    array_min, array_max = array.min(), array.max()
    normalized = (array - array_min) / (array_max - array_min)
    return np.clip(normalized * brightness_factor, 0, 1)

brightness_factor = 1
rgb = np.dstack((
    normalize(red, brightness_factor), 
    normalize(green, brightness_factor), 
    normalize(blue, brightness_factor))
)
plt.imshow(rgb)

check_memory("depois de criar a composição RGB (arrays float64)")

[RAM] depois de criar a composição RGB (arrays float64): 8.95 GB usados de 32.00 GB (78.5%) | disponível: 6.88 GB

que tal uma composição de falsa cor? 

Lembrando que é NIR, G, B:

with rasterio.open(assets['BAND8'].href) as nir_ds:
    nir = nir_ds.read(1)

brightness_factor = 1
rgb = np.dstack((
    normalize(nir, brightness_factor), 
    normalize(green, brightness_factor), 
    normalize(blue, brightness_factor))
)
plt.imshow(rgb)

Rasterio - Window

Ao inves de acessar a imagem inteira (7000x7000 pixels), a gente pode utilizar uma função para só acessar um pedacinho dessa imagem, a nossa área de interesse

from rasterio.windows import Window

def read(uri: str, 
         window: Window,
        masked: bool = True):
    """Read raster window as numpy.ma.masked_array."""
    with rasterio.open(uri) as ds:
        return ds.read(1, window=window, masked=masked)

red = read(assets['BAND7'].href, window=Window(0, 0, 500, 500)) # Window(col_off, row_off, width, height)
green = read(assets['BAND6'].href, window=Window(0, 0, 500, 500))
blue = read(assets['BAND5'].href, window=Window(0, 0, 500, 500))

check_memory("depois de ler só uma janela (window) de 500x500 pixels")
# compare esse valor com o de "depois de carregar 3 bandas em resolução total" acima:
# ler por janelas é a estratégia recomendada se sua máquina tem pouca RAM (ex: 4GB)

[RAM] depois de ler só uma janela (window) de 500x500 pixels: 7.06 GB usados de 32.00 GB (76.2%) | disponível: 7.62 GB

ao invés de passar os valores dos posição de pixel, você poderia passar somente as coordenadas

from rasterio.windows import from_bounds

from pyproj import Transformer

def transform_bounds(bounds, 
                     from_crs='EPSG:4326', 
                     to_crs=None):
    transformer = Transformer.from_crs(from_crs, to_crs, always_xy=True)
    
    minx, miny = transformer.transform(bounds[0], bounds[1])
    maxx, maxy = transformer.transform(bounds[2], bounds[3])
    
    return (minx, miny, maxx, maxy)

green

masked_array(
  data=[[--, --, --, ..., --, --, --],
        [--, --, --, ..., --, --, --],
        [--, --, --, ..., --, --, --],
        ...,
        [--, --, --, ..., --, --, --],
        [--, --, --, ..., --, --, --],
        [--, --, --, ..., --, --, --]],
  mask=[[ True,  True,  True, ...,  True,  True,  True],
        [ True,  True,  True, ...,  True,  True,  True],
        [ True,  True,  True, ...,  True,  True,  True],
        ...,
        [ True,  True,  True, ...,  True,  True,  True],
        [ True,  True,  True, ...,  True,  True,  True],
        [ True,  True,  True, ...,  True,  True,  True]],
  fill_value=-9999,
  dtype=int16)

bounds

(-51.432479184, -0.723834631, -50.107219092, 0.249357407)

with rasterio.open(assets['BAND5'].href) as src:
    reprojected_bounds = transform_bounds(bounds, 'EPSG:4326', src.crs)
    window = from_bounds(
        reprojected_bounds[0],  # left
        reprojected_bounds[1],  # bottom  
        reprojected_bounds[2],  # right
        reprojected_bounds[3],  # top
        src.transform
    )
    window = window.round_offsets().round_lengths()
    
    blue = read(assets['BAND5'].href, window=window, masked=True)
    green = read(assets['BAND6'].href, window=window, masked=True)
    red = read(assets['BAND7'].href, window=window, masked=True)
print(red.shape)

(4781, 3878)

brightness_factor = 1
rgb = np.dstack((
    normalize(red, brightness_factor), 
    normalize(green, brightness_factor), 
    normalize(blue, brightness_factor))
)
plt.imshow(rgb)

Como baixar as imagens em seu computador

O procedimento é o mesmo de tudos que vimos até agora, a única diferença é que vamos baixar cada asset em nossa máquina.

Na próxima célula, deixo uma função que permite vocês baixarem

Uma imagem completa pode ter em 2 a 3 GB

import os
from urllib.parse import urlparse

import requests
from pystac import Asset
from tqdm import tqdm

# def download(asset: Asset, directory: str = None, chunk_size: int = 1024 * 16, **request_options) -> str:
#     """Smart download STAC Item asset.

#     This method uses a checksum validation and a progress bar to monitor download status.
#     """
#     if directory is None:
#         directory = ''

#     response = requests.get(asset.href, stream=True, **request_options)
#     output_file = os.path.join(directory, urlparse(asset.href)[2].split('/')[-1])
#     os.makedirs(directory, exist_ok=True)
#     total_bytes = int(response.headers.get('content-length', 0))
#     with tqdm.wrapattr(open(output_file, 'wb'), 'write', miniters=1, total=total_bytes, desc=os.path.basename(output_file)) as fout:
#         for chunk in response.iter_content(chunk_size=chunk_size):
#             fout.write(chunk)

# download(assets['BAND5'], 'img')

# para baixar todas as bandas de uma imagem, entao:
# for asset in assets.values():
#     download(asset, 'images')

Painel Interativo

Vamos fechar essa aula com um painel interativo, aquele painel que você vê por aí no LinkedIn ou como Google Earth Engine webgis. Vamos montar o mesmo efeito com o leafmap, usando o seu backend baseado em folium (leafmap.foliumap).

Fluxo de ponta a ponta: 1. Escolher uma área em qualquer lugar do Brasil; 2. Buscar no STAC duas cenas da mesma área, em anos diferentes; 3. Ler as bandas por janela (bbox) e montar duas composições RGB; 4. Salvar cada composição como um GeoTIFF local (com a georreferência da janela); 5. Servir os dois GeoTIFFs com leafmap.foliumap.Map().split_map(...), que já vem com o controle de swipe.

Vamos usar como exemplo o avanço do desmatamento na Amazônia, na região de Novo Progresso (PA), ao longo da BR-163 — uma área com um recorte antes/depois bem marcante.

import os

import leafmap.foliumap as leafmap
from rasterio.windows import transform as window_transform

1. Buscando as cenas para mesma área

O bbox abaixo pode ser trocado por qualquer área do Brasil (use o bbox finder lá em cima). Vamos buscar cenas CBERS-4/MUX com poucas nuvens em dois anos bem separados, para o desmatamento ficar evidente.

# Novo Progresso (PA), fronteira de desmatamento ao longo da BR-163
bbox = [-55.30, -8.35, -55.05, -8.15]

client = Client.open("https://data.inpe.br/bdc/stac/v1/")

for ano in ["2017-01-01/2017-12-31", "2023-01-01/2023-12-31"]:
    s = client.search(collections=["CB4-MUX-L4-SR-1"], 
                      bbox=bbox, 
                      datetime=ano,
                       query={"eo:cloud_cover": {"lt": 15}}
                       )
    print(ano, "- imagens encontradas:", s.matched())
    for it in s.items():
        print(" ", it.id, "- nuvens:", round(it.properties["eo:cloud_cover"], 1), "%")
    print()

2017-01-01/2017-12-31 - imagens encontradas: 10
  CBERS_4_MUX_20171103_167_109_L4 - nuvens: 13.1 %
  CBERS_4_MUX_20171103_167_110_L4 - nuvens: 10.5 %
  CBERS_4_MUX_20170912_167_109_L4 - nuvens: 1.0 %
  CBERS_4_MUX_20170912_167_110_L4 - nuvens: 0.4 %
  CBERS_4_MUX_20170626_167_109_L4 - nuvens: 0.0 %
  CBERS_4_MUX_20170626_167_110_L4 - nuvens: 0.0 %
  CBERS_4_MUX_20170505_167_109_L4 - nuvens: 1.7 %
  CBERS_4_MUX_20170505_167_110_L4 - nuvens: 4.1 %
  CBERS_4_MUX_20170409_167_109_L4 - nuvens: 4.1 %
  CBERS_4_MUX_20170409_167_110_L4 - nuvens: 13.5 %

2023-01-01/2023-12-31 - imagens encontradas: 13
  CBERS_4_MUX_20231027_168_109_L4 - nuvens: 10.4 %
  CBERS_4_MUX_20231001_168_109_L4 - nuvens: 2.2 %
  CBERS_4_MUX_20230908_167_109_L4 - nuvens: 6.7 %
  CBERS_4_MUX_20230905_168_109_L4 - nuvens: 2.8 %
  CBERS_4_MUX_20230813_167_109_L4 - nuvens: 0.8 %
  CBERS_4_MUX_20230813_167_110_L4 - nuvens: 0.7 %
  CBERS_4_MUX_20230810_168_110_L4 - nuvens: 1.9 %
  CBERS_4_MUX_20230718_167_109_L4 - nuvens: 0.0 %
  CBERS_4_MUX_20230718_167_110_L4 - nuvens: 0.0 %
  CBERS_4_MUX_20230622_167_109_L4 - nuvens: 0.0 %
  CBERS_4_MUX_20230622_167_110_L4 - nuvens: 0.3 %
  CBERS_4_MUX_20230527_167_109_L4 - nuvens: 0.1 %
  CBERS_4_MUX_20230527_167_110_L4 - nuvens: 1.4 %

Vamos escolher duas cenas do mesmo tile (167_109), uma de cada ano, ambas com 0% de nuvens.

item_2017 = list(client.search(collections=["CB4-MUX-L4-SR-1"], 
                               ids=["CBERS_4_MUX_20170626_167_109_L4"]).items())[0]
item_2023 = list(client.search(collections=["CB4-MUX-L4-SR-1"], 
                               ids=["CBERS_4_MUX_20230718_167_109_L4"]).items())[0]

print(item_2017)
print(item_2023)

<Item id=CBERS_4_MUX_20170626_167_109_L4>
<Item id=CBERS_4_MUX_20230718_167_109_L4>

2. Lendo por bbox e montando as composições RGB

Vamos reaproveitar o transform_bounds e o from_bounds que já vimos, e o check_memory da aula, para ler só a nossa área de interesse em cada cena.

def read_bbox(href: str, bbox: list, masked: bool = True):
    """Lê a janela do raster correspondente ao bbox (em EPSG:4326), com a georreferência da janela."""
    with rasterio.open(href) as src:
        reprojected_bounds = transform_bounds(bbox, "EPSG:4326", src.crs)
        window = from_bounds(*reprojected_bounds, src.transform).round_offsets().round_lengths()
        data = read(href, window=window, masked=masked)
        return data, window_transform(window, src.transform), src.crs


def read_rgb(item, bbox):
    assets = item.assets
    red, transform, crs = read_bbox(assets["BAND7"].href, bbox)
    green, _, _ = read_bbox(assets["BAND6"].href, bbox)
    blue, _, _ = read_bbox(assets["BAND5"].href, bbox)
    return red.astype("float64"), green.astype("float64"), blue.astype("float64"), transform, crs


red_2017, green_2017, blue_2017, transform_2017, crs_2017 = read_rgb(item_2017, bbox)
#check_memory("depois de ler a cena de 2017")

red_2023, green_2023, blue_2023, transform_2023, crs_2023 = read_rgb(item_2023, bbox)
#check_memory("depois de ler a cena de 2023")

print("shape:", red_2017.shape)

shape: (1100, 1382)

Para essa cena vamos usar um contraste por percentil ao invés de min/max que usamos antes. Esse constrate por percentil evitar de a sensibilidade de pixels ficarem muito claros ou muito escuros..

def stretch(array, 
            low_pct=2, 
            high_pct=98):
    """Normaliza um array usando contraste por percentil."""
    data = np.ma.filled(array, np.nan)
    valid = data[np.isfinite(data)]
    p_lo, p_hi = np.percentile(valid, [low_pct, high_pct])
    return np.clip((array - p_lo) / (p_hi - p_lo), 0, 1)

# essa funcao converte em int 8 bits of quantization
def make_rgb_uint8(red, green, blue):
    rgb = np.dstack((stretch(red), stretch(green), stretch(blue)))
    return (np.ma.filled(rgb, 0) * 255).astype("uint8")


rgb_2017 = make_rgb_uint8(red_2017, green_2017, blue_2017)
rgb_2023 = make_rgb_uint8(red_2023, green_2023, blue_2023)

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 6))
ax1.imshow(rgb_2017)
ax1.set_title("2017")
ax1.axis("off")
ax2.imshow(rgb_2023)
ax2.set_title("2023")
ax2.axis("off")
plt.tight_layout()
plt.show()

3. Salvando como GeoTIFF

Aqui vamos salvar cada composição como um GeoTIFF em disco, usando a transformação e o CRS da própria janela que lemos. Esse raster processado pode ser facilmente aberto no QGIS.

output_dir = "outputs"
os.makedirs(output_dir, exist_ok=True)


def write_rgb_geotiff(red, green, blue, transform, crs, out_path):
    """Salva uma composição RGB (uint8) como GeoTIFF local."""
    rgb_uint8 = make_rgb_uint8(red, green, blue)  # (linhas, colunas, bandas)
    rgb_uint8 = np.transpose(rgb_uint8, (2, 0, 1))  # rasterio espera (bandas, linhas, colunas)
    with rasterio.open(
        out_path, "w",
        driver="GTiff",
        height=rgb_uint8.shape[1],
        width=rgb_uint8.shape[2],
        count=3,
        dtype="uint8",
        crs=crs,
        transform=transform,
    ) as dst:
        dst.write(rgb_uint8)
    return out_path


tif_2017 = write_rgb_geotiff(red_2017, green_2017, blue_2017, transform_2017, crs_2017,
                              os.path.join(output_dir, "novo_progresso_2017.tif"))
tif_2023 = write_rgb_geotiff(red_2023, green_2023, blue_2023, transform_2023, crs_2023,
                              os.path.join(output_dir, "novo_progresso_2023.tif"))

print(tif_2017)
print(tif_2023)

outputs/novo_progresso_2017.tif
outputs/novo_progresso_2023.tif

4. Montando o painel com leafmap

leafmap.foliumap.Map é uma subclasse de folium.Map, então ela renderiza como HTML puro. O split_map recebe os dois caminhos de arquivo e cuida de servir cada GeoTIFF como tiles (via localtileserver) e de montar o controle de swipe.

m = leafmap.Map()
m.split_map(
    left_layer=tif_2017,
    right_layer=tif_2023,
    left_label="2017",
    right_label="2023",
)
m

Nota: diferente da tentativa anterior com uma imagem estática, agora as duas camadas são servidas como tiles de verdade pelo localtileserver (um servidor local que roda junto com o kernel), então elas não pixelam ao aproximar o zoom. Isso também significa que o mapa só funciona enquanto o kernel do notebook estiver rodando na sua máquina.

Referências

• Brazil Data Cube - Python
• E-book - BDC
• Apostila INPE - INPE