Satellite Imagery Analysis — Core Concepts

Understand spectral bands, vegetation indices, cloud masking, and change detection — the building blocks of satellite analysis in Python.

Satellite imagery analysis extracts meaningful information from Earth-observation data. Python’s scientific stack — Rasterio for I/O, NumPy for math, and domain libraries for search and discovery — makes this workflow accessible on a laptop.

Spectral bands

Unlike a phone camera that captures red, green, and blue, satellites capture many spectral bands, including wavelengths invisible to the human eye:

Band name	Wavelength	What it reveals
Blue	450-510 nm	Water bodies, atmospheric scattering
Green	530-590 nm	Vegetation vigor
Red	640-670 nm	Chlorophyll absorption
Near-infrared (NIR)	780-900 nm	Plant cell structure, biomass
Short-wave IR (SWIR)	1,550-1,750 nm	Soil moisture, burn scars
Thermal IR	10,000-12,000 nm	Surface temperature

Each band is stored as a separate layer in the raster file. Analysis combines these layers with arithmetic to highlight specific features.

Vegetation indices

The most common calculation is NDVI — Normalized Difference Vegetation Index:

import rasterio
import numpy as np

with rasterio.open("sentinel2.tif") as src:
    red = src.read(4).astype(np.float32)
    nir = src.read(8).astype(np.float32)

ndvi = (nir - red) / (nir + red + 1e-10)
# Values: -1 to 1. Above 0.3 = healthy vegetation

Other useful indices:

NDWI (water): (green - nir) / (green + nir) — highlights water bodies
NBR (fire): (nir - swir) / (nir + swir) — detects burn scars
NDBI (built-up): (swir - nir) / (swir + nir) — highlights urban areas

Data discovery with STAC

SpatioTemporal Asset Catalogs (STAC) let you search for satellite imagery by location, date, and cloud cover:

from pystac_client import Client

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

search = catalog.search(
    collections=["sentinel-2-l2a"],
    bbox=[-73.99, 40.70, -73.95, 40.75],
    datetime="2025-06-01/2025-06-30",
    query={"eo:cloud_cover": {"lt": 10}},
)

items = list(search.items())
print(f"Found {len(items)} scenes")

STAC catalogs from AWS, Microsoft Planetary Computer, and Google Earth Engine index petabytes of free imagery.

Cloud masking

Clouds ruin satellite analysis. Most Level-2 products include a Scene Classification Layer (SCL) that labels each pixel:

with rasterio.open("SCL.tif") as src:
    scl = src.read(1)

# SCL values: 3=cloud shadow, 8=cloud medium, 9=cloud high
cloud_mask = np.isin(scl, [3, 8, 9])
ndvi_clean = np.where(cloud_mask, np.nan, ndvi)

Always mask clouds before computing statistics — a single cloudy pixel can skew an entire field’s average.

Change detection

Comparing two dates reveals what changed:

ndvi_before = compute_ndvi("2025-01-15.tif")
ndvi_after = compute_ndvi("2025-07-15.tif")

change = ndvi_after - ndvi_before
# Positive = greening (crop growth)
# Negative = browning (harvest, deforestation, fire)

Threshold the change map to create a binary mask of significant events.

Resolution and data sources

Satellite	Resolution	Revisit	Cost
Sentinel-2	10 m	5 days	Free
Landsat 8/9	30 m	16 days	Free
Planet Dove	3 m	Daily	Commercial
MODIS	250 m	Daily	Free

Sentinel-2 is the most popular choice for vegetation and land-use analysis because it balances resolution, revisit time, and cost.

Common misconception

People often think higher resolution is always better. In practice, coarser data like MODIS is superior for large-area, long-term trend analysis because it has daily coverage and a 20+ year archive. Resolution should match the question, not the ego.

The one thing to remember: Satellite analysis is band math — by combining visible and invisible light measurements from different spectral channels, Python can quantify vegetation health, water extent, urban growth, and fire damage from free, publicly available imagery.

pythonsatellite-imageryremote-sensinggeospatial