AI 数据中心卫星监控可行性报告

MS-01 cron (monthly)
  -> aoi.yaml (10 AOIs)
  -> Earth Search STAC query: S2 L2A + S1 GRD + Landsat L2
  -> COG window read (no bulk download)
  -> monthly cloud-free composite
  -> metrics: NDBI, BSI, NDVI, NDWI, edge density, SAR dB delta
  -> Cloudflare D1 (observations table)
  -> PNG thumbnails + delta heatmap to R2
  -> Discord alert
  -> if score > threshold: trigger 50cm tasking quote

初始化

uv init ai-dc-monitor
cd ai-dc-monitor
uv add pystac-client rasterio rioxarray stackstac geopandas shapely \
       pyproj numpy pandas scikit-image opencv-python duckdb httpx pydantic rich

核心代码骨架

from pystac_client import Client
import numpy as np

STAC = "https://earth-search.aws.element84.com/v1"
cat = Client.open(STAC)

def search(aoi_geojson, start, end, collection="sentinel-2-l2a"):
    return cat.search(
        collections=[collection],
        intersects=aoi_geojson,
        datetime=f"{start}/{end}",
        query={"eo:cloud_cover": {"lt": 30}},
        max_items=20,
    ).item_collection()

def ndbi(nir, swir):
    return (swir - nir) / (swir + nir + 1e-6)

def bsi(red, nir, blue, swir):
    return ((swir + red) - (nir + blue)) / ((swir + red) + (nir + blue) + 1e-6)

def construction_score(prev, curr):
    return (
        0.35 * np.nanmean(curr["ndbi"] - prev["ndbi"])
        + 0.25 * np.nanmean(curr["bsi"] - prev["bsi"])
        - 0.20 * np.nanmean(curr["ndvi"] - prev["ndvi"])
        + 0.20 * np.nanmean(curr["edge"] - prev["edge"])
    )

D1 最小表

CREATE TABLE observations (
  month TEXT,
  aoi_id TEXT,
  ndbi_delta REAL,
  bsi_delta REAL,
  ndvi_delta REAL,
  sar_db_delta REAL,
  disturbed_ha REAL,
  roof_like_ha REAL,
  alert_score REAL,
  thumbnail_url TEXT,
  source_items TEXT,
  created_at TEXT,
  PRIMARY KEY (month, aoi_id)
);