Characterising wind power extremes over Kenya using an enhanced process-based reanalysis-driven model

This study presents a robust framework for addressing systematic biases in the ERA5 wind speeds to model long-term, high-resolution wind energy and characterise wind power extremes in data-sparse regions. By integrating Weibull Quantile Mapping, hub-height extrapolation, and dynamic efficiency, the study models hourly output for three Kenyan wind farms: Lake Turkana Wind Power, Kipeto, and Ngong Hills. The model significantly reduced Mean Bias Error and Root Mean Square Error in the reanalysis while preserving temporal rank correlations. The reanalysis-driven model captures the fundamental variability of wind power generation. Persistence and ramp diagnostics using Threshold-Duration Frequency analysis reveal that: LTWP exhibits low variability, with high-output events (>80% Capacity Factor) sustained for durations exceeding 14 days, contrasting with the frequent multi-day droughts and pronounced ramping typical of mid-latitude wind turbine fleets. Kipeto and Ngong Hills sites exhibit strong diurnal cycling, necessitating short-term storage rather than seasonal balancing. While LTWP frequently undergoes large shifts (>60% Δ Capacity Factor) over diurnal cycles, extreme volatility at shorter timescales (3-h) is heavily damped. This framework demonstrates a transferable process for realistic wind power modelling in data-sparse environments, supporting regional energy planning and integration of renewables into developing power systems.