nephatmwanza/zambia-cmip6-evaluation

# Can CMIP6 models reproduce Zambia's rainy season? *Evaluation of 29 CMIP6 models against CHIRPS observations, October–March, 1984/85–2013/14. Third of three companion studies — see Related work.* The observational rainfall study found that what matters for Zambian agriculture is not the seasonal total but the **dry spell inside the season**: a three-week gap in January fails a maize crop in a year of entirely normal rainfall. So before any climate model is used to plan Zambia's future, one question comes first — **can these models reproduce the present?** Mostly they cannot. And the reason turns out to be specific, measurable, and the same in almost every model. *Each dot is one model; the star is the observed climate. Almost every model sits to the right of it — raining on far more days than actually happens.* --- ## The finding **25 of 29 models rain on too many days.** Zambia's rainy season is wet on **55.6%** of days. The ensemble median is **71.3%**, and the worst model rains on **91%** of days — practically every day of the season. | | Observed (CHIRPS) | Ensemble median | Ensemble range | |---|---|---|---| | Mean rainfall | 5.22 mm/day | 6.46 | 2.54 – 9.25 | | **Wet-day frequency** | **55.6% of days** | **71.3%** | 44.7 – 91.3 | | Wet-day intensity | 9.31 mm/wet day | 9.00 | 4.31 – 11.25 | | **Longest dry spell** | **18.8 days** | **17.1** | **7.3 – 34.3** | This is the well-known tropical *drizzle problem* — models produce rain too often and too lightly. What this study adds is the consequence for Zambia specifically, and it is not a small one. ## Why it matters: the frequency error destroys dry spells A model that rains on 85% of days **cannot** produce a three-week dry spell. There is no room left in the calendar. If that mechanism is real, then the frequency error should predict the dry-spell error across the ensemble. It does, strongly: | | correlation with dry-spell error | variance explained | |---|---|---| | **Wet-day frequency bia …