A Novel Framework for Non-Stationary Flood Frequency Analysis Under Climate Change for Nigerian River Basins

Flood frequency analysis in Nigeria has traditionally assumed statistical stationarity in historical streamflow records, an assumption that is increasingly violated under observed climate change and land use transformation in Nigerian river basins. This study develops a novel non-stationary flood frequency analysis framework specifically calibrated for the climatic and hydrological characteristics of Nigerian river basins. The methodological framework integrates time-varying extreme value distribution fitting with physically based covariate identification, using rainfall trend metrics, land use change indicators, and large-scale climate indices (ENSO, AMO) as non-stationarity drivers. The framework was applied to annual maximum flood series from 25 gauging stations across the six geopolitical zones, with records spanning 1960 to 2023. Stationarity testing combined Mann-Kendall tests, change point detection, and goodness-of-fit assessment under stationary versus non-stationary GEV models. Time-varying distribution parameters were estimated using maximum likelihood with penalised likelihood regularisation. The Akaike Information Criterion was used for covariate model selection. Results confirm significant non-stationarity in 64 percent of flood series, predominantly in urban catchments and basins experiencing significant deforestation. Non-stationary 100-year return period flood estimates exceed stationary equivalents by 15 to 68 percent depending on zone and trend severity. The study makes an original methodological contribution to flood frequency analysis by providing a framework calibrated and validated for Nigerian hydrological conditions, enabling more accurate flood design quantiles for infrastructure design and risk assessment.

Keywords: non-stationary flood frequency, climate change, extreme value analysis, Nigerian river basins, flood design