A three-month ENSO change term did not lower monthly rain test error in three tropical regions in 1980-2025

We tested whether the three-month change in Nino 3.4 adds information about concurrent monthly rain after current ENSO strength and season are included. We used fixed NOAA GPCP V2.3 and ERSSTv5 records for 552 months in 1980-2025. The baseline had month-specific intercepts, linear and squared index slopes, and a common time trend. We excluded five-year blocks and one adjacent year on each side from fitting. Adding one common index-change slope raised mean squared error by 1.79% in the eastern Pacific, 2.06% in the central Pacific, and 0.36% over the Maritime Continent. The respective 98.33% three-year block intervals for error reduction were -5.28% to 0.22%, -4.47% to 0.04%, and -1.21% to 0.29%. These intervals condition on fitted predictions. Most planned checks also gave negative point gains. Some eastern and Maritime Continent checks changed sign. Thus, this term gave no clear added value in this test. The result does not rule out season-specific memory, spatial SST effects, or useful forecasts. Short records, retrospective source revisions, a common change slope, and conditional uncertainty limit inference. The package gives source hashes, monthly evidence, all predictions, uncertainty draws, code, checks and a figure.

ENSO · rainfall · event evolution · GPCP · blocked validation · null result