Model accuracy (backtest)
Forecast
Which parameter is likely to rise over the next two years?
Based on NASA's Vital Signs data for October 2026, the parameter that is most likely to rise in next two years is atmospheric carbon dioxide (CO₂). The latest Mauna Loa reading is about 426 ppm (Sep 2026), and in April 2026 it touched a record of around 431 ppm. CO₂ has increased every year since 1958 and the rate of increase is also accelerating, from about 1 ppm/year in the 1960s to around 2.5–3 ppm/year now. Because CO₂ stays in atmosphere for hundreds of years, it will surely keep rising. Sea level (about +4.4 mm/year) and global temperature are also going up, but temperature changes a lot year to year because of El Niño and La Niña.
I developed a dashboard that tests different models (linear, quadratic, quadratic with seasonal cycle, and ensemble) on the last 2 years of data and automatically choose the most accurate one. For CO₂ the seasonal model had error of only 0.57 ppm, compared to 3.6 ppm for simple linear trend. It predicts CO₂ will reach about 436 ppm by March 2028.
How the prediction works: I have fitted each dataset with four models: a linear trend, a quadratic trend (to capture acceleration), the quadratic trend plus seasonal harmonics (12-month and 6-month cycles, which matter for CO₂), and an ensemble average. I hold out the most recent 24 months (5 years for annual sea level), train every model on the earlier data and score it on the held-out part (RMSE/MAE). The dashboard then automatically chooses the lowest-error model, and I refit it on all data to forecast the next two years. I base the 95% interval on the spread of the model's residuals.