Kenya national rainfall and equatorial bimodal seasonal rainfall reproducibility package, 1991–2024

Description This repository contains the derived rainfall input series, annual and seasonal analytical datasets, reproducible analysis code, quality-assurance manifests, provenance documentation and machine-readable statistical results supporting a source-era-separated analysis of national rainfall in Kenya for 1991–2024. The analytical design distinguishes two rainfall source periods: Historical reference period (1991–1999). The historical component comprises 108 national monthly rainfall observations, assembled from the governed historical national rainfall reference series. It is used as an exploratory historical context and is not spatially resolved to counties or raster cells in the analytical input. Primary analysis period (2000–2024). The CHIRPS component comprises 300 national monthly rainfall observations derived from the governed CHIRPS monthly raster series. Each monthly national value represents the mean across 18,896 valid pixels. The corresponding source raster identifiers are retained in the monthly input series. The package provides annual rainfall observations, annual derived measures, MAM (March–May) and OND (October–December) seasonal rainfall totals, paired annual seasonal records, signed MAM-minus-OND contrasts, individual seasonal trend estimates, paired seasonal trend-rate comparisons, season-specific lower-tail rainfall classifications, uncertainty estimates and sensitivity results. Trend analyses use Theil–Sen slope estimates and residual circular moving-block bootstrap uncertainty intervals. Heteroskedasticity- and autocorrelation-consistent (HAC) regression inference is reported alongside block-length and HAC-lag sensitivity analyses. Holm adjustment is applied to the two seasonal trend tests within each source era. The two source eras are analysed independently. The 1991–1999 historical estimates are exploratory because of the short nine-year record. No pooled 1991–2024 trend model, pooled seasonal threshold or formal cross-era hypothesis test is included. The package also preserves two distinct seasonal estimands: (1) the temporal trend in the annual signed MAM-minus-OND rainfall accumulation gap, and (2) the difference between the independently estimated MAM and OND Theil–Sen trend rates. These are not interchangeable. Season-specific P20 classifications use empirical lower-tail thresholds estimated separately for each season and source era. Their probabilities describe empirical rainfall-deficit frequencies, not health outcomes or probabilities of agricultural drought. The reported probability intervals are conditional on the estimated thresholds; threshold-estimation uncertainty is not propagated. Data and reproducibility scope The archived CHIRPS monthly input is a derived national summary series, not the original monthly raster collection. The 300 source GeoTIFFs and the complete upstream raster-to-national-mean extraction environment are not included in this deposit. Source raster identifiers and provenance information are retained to support traceability. The historical monthly input is national tabular data. No unverified upstream grid resolution is assigned to it. The repository includes the derived analytical inputs, analysis scripts, result tables, run summaries, QA manifests, source checksums and documentation required to inspect the deposited workflows and their reported results. Intended use The package supports reproducibility, independent inspection and further methodological evaluation of national annual and bimodal seasonal rainfall trends in Kenya. It should be interpreted within the temporal, source and spatial-support limitations documented in the accompanying methods and provenance files. keywords: Kenya; rainfall variability; CHIRPS; MAM; OND; bimodal rainfall; seasonal rainfall; annual rainfall; Theil–Sen estimator; block bootstrap; autocorrelation; drought; climate variability; reproducible research.