Buscando imagens Sentinel-2 via STAC

# Descomente no Colab:
# !pip install -q pystac-client planetary-computer geopandas rioxarray matplotlib

import geopandas as gpd
import matplotlib.pyplot as plt
import planetary_computer
import pystac_client
import rioxarray  # noqa: F401  (registra o acessor .rio)
from shapely.geometry import box, shape

plt.style.use("dark_background")  # opcional: gráficos no estilo escuro do site

catalogo = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
    modifier=planetary_computer.sign_inplace,
)
s2 = catalogo.get_collection("sentinel-2-l2a")
print(s2.title)
print("Extensão temporal:", s2.extent.temporal.intervals[0][0].date(), "até hoje")

lat, lon = -12.55, -55.72
d = 0.05  # meio lado do quadrado, em graus (~5,5 km)
bbox = [lon - d, lat - d, lon + d, lat + d]

busca = catalogo.search(
    collections=["sentinel-2-l2a"],
    bbox=bbox,
    datetime="2025-06-01/2025-09-30",
    query={"eo:cloud_cover": {"lt": 20}},
)
itens = busca.item_collection()
print(f"{len(itens)} cenas encontradas")

gdf = gpd.GeoDataFrame.from_features(itens.to_dict(), crs="EPSG:4326")
gdf["id"] = [i.id for i in itens]
gdf["data"] = gdf["datetime"].str[:10]
gdf["tile"] = gdf["s2:mgrs_tile"]
tabela = gdf[["data", "platform", "tile", "eo:cloud_cover"]].sort_values("data")
tabela.head(10)

aoi = gpd.GeoDataFrame(geometry=[box(*bbox)], crs="EPSG:4326")

fig, ax = plt.subplots(figsize=(7, 6))
gdf.drop_duplicates("tile").plot(
    ax=ax,
    column="tile",
    alpha=0.35,
    edgecolor="white",
    legend=True,
    legend_kwds={"title": "Tile MGRS", "loc": "upper left"},
)
aoi.boundary.plot(ax=ax, color="#FF7A45", linewidth=2.5)
ax.set_title("Footprints dos tiles e área de interesse (laranja)")
ax.set_xlabel("Longitude")
ax.set_ylabel("Latitude")
plt.show()

por_data = gdf.groupby("data")["eo:cloud_cover"].min()

fig, ax = plt.subplots(figsize=(8, 3.2))
ax.bar(por_data.index, por_data.values, color="#22D3EE")
ax.axhline(5, color="#FF7A45", ls="--", lw=1, label="5%")
ax.set_ylabel("Nuvens (%)")
ax.set_title("Cobertura de nuvens por data")
ax.tick_params(axis="x", rotation=60, labelsize=8)
ax.legend()
plt.tight_layout()
plt.show()

area = box(*bbox)
cobrem = [i for i in itens if shape(i.geometry).contains(area)]
print(f"{len(cobrem)} de {len(itens)} cenas cobrem toda a área")

melhor = min(cobrem, key=lambda i: i.properties["eo:cloud_cover"])

print("Cena:", melhor.id)
print("Data:", melhor.datetime.date())
print(f"Nuvens: {melhor.properties['eo:cloud_cover']:.2f}%")
print("Assets disponíveis:", ", ".join(sorted(melhor.assets)))

visual = rioxarray.open_rasterio(melhor.assets["visual"].href)
print("Cena completa:", visual.shape, "pixels — CRS", visual.rio.crs)

recorte = visual.rio.clip_box(*bbox, crs="EPSG:4326")
print("Recorte:", recorte.shape)

fig, ax = plt.subplots(figsize=(8, 8))
recorte.plot.imshow(ax=ax)
ax.set_title(f"Sentinel-2 · cor verdadeira · {melhor.datetime.date()} · Sorriso (MT)")
ax.set_xlabel("x (m, UTM)")
ax.set_ylabel("y (m, UTM)")
ax.set_aspect("equal")
plt.show()

img = recorte.transpose("y", "x", "band").values
altura = img.shape[0]
faixa = img[int(altura * 0.2) : int(altura * 0.2) + int(img.shape[1] * 9 / 16)]  # recorte 16:9

fig = plt.figure(figsize=(8, 4.5), dpi=150)
ax = fig.add_axes([0, 0, 1, 1])
ax.imshow(faixa)
ax.axis("off")
fig.savefig("thumbnail.webp", dpi=150)
plt.close(fig)