Can We Forecast Mauna Loa's CO2

Research question

Does a seasonal time-series model improve on simple CO2 forecasts at each prediction horizon?

Describe long-term rise and seasonality in public Mauna Loa monthly CO2, and evaluate a compact forecasting model against simple baselines using chronological held-out targets. Distinguish current-vintage historical evaluation from real-time forecasting.

Data and comparison design

The comparison uses SARIMA with the seasonal-drift baseline separately at one, six and twelve months. Assess both point errors and interval coverage: a narrow interval is only useful if its stated probability matches how often later observations fall inside it.

Input or source	Role in this case	Source context
NOAA monthly Mauna Loa CO2	Frozen public monthly series; retain NOAA attribution and historical Scripps credit.	Retain attribution and source terms; this README grants no additional reuse rights
NOAA trends documentation	Definitions, preliminary-data caveats and measurement context.	Retain attribution and source terms; this README grants no additional reuse rights

Recorded findings

Forecast component	Recorded result	Interpretation
Annual mean change	315.98 ppm (1959) → 427.35 ppm (2025)	The station series, not a global emissions total.
One- and six-month error	SARIMA lower than seasonal drift	Advantage depends on horizon.
Twelve-month error	SARIMA slightly higher	Short-horizon improvement does not extend uniformly.
Nominal 80% twelve-month interval	33/49 targets covered	Coverage must be assessed separately from point error.

The selected analysis uses 814 monthly values from March 1958 through December 2025. Annual means rise from 315.9816667 ppm in 1959 to 427.3491667 ppm in 2025: +111.3675 ppm, or 35.2449245%. At 1/6/12-month horizons, common eligible targets are 60/55/49. SARIMA MAEs are 0.298604/0.445793/0.669625 ppm; seasonal-drift MAEs are 0.552486/0.600024/0.640619 ppm. The small 12-month drift advantage is descriptive, not a demonstrated general superiority. The 12-month 80% interval covers 33/49 targets (67.3%).

The original session records scientific execution and repeatability checks; this local preparation does not re-execute them. Scientific artifacts included in this selected export:

• analysis.R

These files are within session.science and its embedded archive; they are not separate files alongside this README. No empty inputs/expected/outputs/attachments directories or duplicate extracted datasets are created.

What the comparison shows

Short-horizon improvements do not imply a twelve-month advantage. The recorded interval coverage also shows why point accuracy and uncertainty calibration need separate evaluation.

Assumptions and limits

This is historical evaluation using a later data snapshot, not a real-time data-vintage backtest. Forecast errors overlap, intervals condition on fitted parameters, and NOAA records station substitution during the Mauna Loa shutdown. A single station is not a global emissions inventory.

Sources and research materials

The unchanged research package comes from the upstream archive. It retains the original conversation and included scientific artifacts, including analysis code, results and figures. The poster is an AI-generated thematic illustration, not study evidence. This introduction summarizes archived findings; the analysis was not rerun. Later bounded checks do not certify the entire study or mean the original package was corrected.