Statistical projections of the mean changes in intra-seasonal rainfall characteristics over Zambia from the reference period 1981-2010 to the future 2031-2060.

Nephat Mwanza , Suman Jain

Zambia's rain-fed agriculture depends less on how much rain falls in a season than on how that rain is distributed within it. This study projects changes in the longest dry spell (consecutive dry days, CDD) and the longest wet spell (consecutive wet days, CWD) over Zambia, together with the number of wet days, rainfall per wet day and seasonal total, for 2031–2060 relative to 1981–2010, using 25 CORDEX-Africa regional climate model simulations at 0.44° under RCP4.5 and RCP8.5. Both indices are computed in every grid cell from daily precipitation, separately for October–December (OND) and January–March (JFM), before any spatial averaging. The simulations are evaluated on the same indices against two daily observational products, CHIRPS and GPCC Full Data Daily, themselves checked against 18 rain gauges, and against reanalysis-driven runs of the same regional models. Because ensemble members share regional and driving models, significance is assessed over model families, and mapped tests control the false discovery rate. The observed OND dry spell is 17.5 days, which the simulations span from 12.7 to 34.7 days. The ensemble median projects the OND dry spell to lengthen by 2.23 days under RCP8.5 and 1.65 days under RCP4.5, with 95% of runs agreeing on the sign, and the number of OND wet days to fall by 2.7 and 2.0 days, while the seasonal total barely changes: early-season rain arrives on fewer days. The dry-spell lengthening survives false-discovery-rate control over 284 and 272 of 391 Zambian grid cells and is largest in Lusaka, Eastern, Muchinga and Central provinces. The OND wet spell shortens by 0.75 days (91% agreement), but the regional models disagree on it; rainfall per wet day rises in most runs but not robustly; and JFM changes are not robust. Treating the 22 runs as independent would give p < 0.001 for the headline result, against p = 0.062 when runs are first averaged within each regional model. An independent ensemble at 0.22° that shares no regional model with the primary one reproduces the observed dry spell more closely and projects a comparable change of 2.15 days. The results point to a more interrupted early rainy season and give no support for claims about the late season.