How is global atmospheric CO₂ changing, and where are the strongest sources?

What you can answer

Global XCO₂ trend (~2.5 ppm/yr current rate, seasonal cycle, annual growth rate)
Regional gradients (highest over Asia industrial corridors, lowest over Southern Ocean)
Anthropogenic point sources via OCO-3 SAM mode (megacities, power plants — when sky is clear)
Biosphere uptake via SIF (forests vs croplands vs grasslands)
El Niño / La Niña impact on tropical land carbon (a major OCO finding)

What you can NOT answer with these alone

Direct flux (kg CO₂/sec) — requires inversion via atmospheric-transport model (use OCO MIP).
Per-facility emissions without high-fidelity wind + boundary-layer height (use MERRA-2 PBL + 850 hPa winds).
Methane, which is on TROPOMI, EMIT, GOSAT, MERLIN — not on OCO sensors.

Code template

import earthaccess
import xarray as xr

earthaccess.login(strategy="netrc")

# Year-scale trend: OCO-2 global XCO2 monthly mean 2015-2025
aoi = None  # global
window = ("2015-01-01", "2025-12-31")
oco2 = earthaccess.search_data(short_name="OCO2_L2_Lite_FP",
                               temporal=window)

# Open all granules, filter by xco2_quality_flag == 0 (good)
# Compute monthly mean per 2°×2° grid → trend line + seasonal harmonic fit

# Facility-scale: OCO-3 SAM over a specific city/plant
sam_aoi = (-95.5, 29.5, -95.0, 30.0)  # Houston area
oco3 = earthaccess.search_data(short_name="OCO3_L2_Lite_FP",
                               bounding_box=sam_aoi,
                               temporal=("2022-01-01", "2024-12-31"))
# Extract XCO2 + wind context from MERRA-2 → enhancement above background

Expected output

Global map: 2015 mean XCO₂ minus 2024 mean XCO₂ (annotated with major source regions)
Time-series: monthly global XCO₂ 2015–present (showing seasonal cycle + secular rise)
Facility raster: OCO-3 SAM mode dense XCO₂ over a megacity, with wind-corrected enhancement overlay
SIF map: spring vs fall growing-season comparison for a major biome

Caveats

Cloud screening removes 75-90% of footprints. Sparse coverage is the default; aggregate to monthly or coarser.
Quality flags are critical — xco2_quality_flag == 0 is the only thing you should trust unfiltered. QF=1 is questionable; QF=2 should be discarded.
Bias corrections (v11.2 vs v10) matter for trend analyses spanning the version change.
SIF is a photosynthesis proxy, not GPP directly — the relationship varies by ecosystem, water stress, species composition.
OCO-3 SAM is scheduler-driven — not every target gets observed on every overpass.