September DMI gave no clear added skill for December rain share in an East Africa box in 2000-2025

We tested whether September Dipole Mode Index (DMI) adds forecast skill for December's share of October-December rain in 5 S-5 N, 35 E-40 E. Each test year in 2000-2025 used earlier years from 1979 for fitting. The baseline used September local rain, September Nino 3.4 and a time trend. Adding DMI increased mean squared error by 2.10%. The conditional 95% three-year block interval for error reduction was -4.43% to +1.01%. Root mean squared error rose from 13.67 to 13.81 percentage points. Both models had higher error than the earlier-year mean. Planned share checks also gave negative point gains. A monthly rain centroid gave no clear gain. The same model terms cut total seasonal rain error by 30.14%, which agrees with a prior campaign result. Thus, skill for seasonal amount did not extend to this monthly share in the tested models. This null result does not prove that the ocean has no effect on timing. A small region, 26 test years, revised and interim data, and conditional intervals limit the result. The package gives cell data, index values, forecasts, uncertainty draws, source hashes, code, a figure and independent checks.

East Africa · Indian Ocean Dipole · rainfall timing · December share · GPCP · ENSO · null result · seasonal forecast