Estimation of extreme rainfall using intensity-duration-frequency curves for Ashanti Region, Ghana

This study addresses the challenge of constructing Intensity-Duration-Frequency (IDF) curves in
Ghana’s Ashanti Region, where sparse gauge networks limit high-resolution extreme rainfall
analysis. A 36-station-year annual maximum series (AMS) was developed by concatenating 14
sub-daily stations from the Kumasi Meso Rain Gauge Network (KMRGN), creating a continuous
regional reference series. The validity of this concatenation approach was supported by L-
moment-based regional frequency analysis, confirming statistical homogeneity of the network
despite localized variability in higher-order moments. Temporal consistency analysis further in-
dicates no statistically significant trends or change-points in the AMS, supporting the assumption of stationarity for extreme-value modelling. The AMS of the benchmark concatenated sub-daily series is modeled within a Generalized Extreme Value (GEV) framework, complemented by intercomparison with Gumbel, Pearson Type III, and Log-Pearson Type III distributions under a multi-criteria model selection approach. The distributions of each model are validated using Chisquare and Kolmogorov-Smirnov tests, and are applied to construct and compare IDF curves. ... mehrResults from the multi-criteria framework indicate that the Log-Pearson Type III distribution provides the most balanced performance in terms of goodness-of-fit, bias, and stability of extreme estimates. This benchmark is used to evaluate AMS derived from disaggregated daily records from the Ghana Meteorological Agency network. Results show that the concatenated sub-daily AMS reliably captures short-duration rainfall intensities while atte station-specific anomalies. However, disaggregated daily AMS underestimated short-duration (12 min-1 h) intensities by up to -18%, particularly during the convectively active minor rainy season, reflecting smoothening of peak rainfall. Performance improves at longer durations, and when concatenated into a 220 station-year homogeneous regional series, the disaggregated data closely reproduce the regional intensity structure across durations and return periods. Uncertainty analysis reveals strong duration dependence, with the highest sensitivity at short durations primarily controlled by the nonlinear scaling exponent of the disaggregation procedure. Comparison of disaggregated curves with existing 2016 IDF curves from the Ghana Meteorological Agency under the Gumbel distribution, similarly adopted by the agency, indicates systematic positive biases, ranging 20–40% on average across all durations for each return period and more than 50% increase in rainfall intensity at longer return periods across the respective durations and return periods, suggesting intensification of extremes. However, the absence of statistically significant trends in the AMS suggests that these increases are largely attributable to extended sampling, improved temporal representation, and methodological differences rather than definitive climatic non-stationarity. These findings establish the concatenated sub-daily series as a robust regional reference and highlight the necessity of high-resolution data or seasonally adjusted correction methods for reliable flood risk assessment, infrastructure design, and climate adaptation planning in the region.