Atmospheric CO2 Forecasting
Time series analysis of Keeling Curve data with ARIMA/SARIMA modeling and 26-year forecast validation.
The Problem
Long-Range Climate Projections
Climate policy depends on accurate CO2 projections. The challenge was building forecasting models from historical Keeling Curve data and validating their long-term accuracy against decades of observed measurements.
The Approach
ARIMA/SARIMA Modeling
Analyzed 40 years of Mauna Loa CO2 data (1958-1997) using ARIMA/SARIMA models with seasonal decomposition and ADF stationarity testing. Validated forecasts against 26 years of actual observations (1998-2024) to quantify model drift.
Technologies & Methods
The Results
26-Year Forecast Validated
Model predicted 420 ppm by 2030, but actual levels crossed that threshold in 2022, eight years early. Simpler ARIMA specifications outperformed complex models (RMSE: 0.0211→0.0035), showing that accelerating emissions can push forecasts consistently off track.