Interpretable, Locally Trained Regression Models for Hourly Wind Speed Forecasting in Nigeria: Implications for Wind Resource Assessment in Data-Scarce Regions

Accurate short-term wind speed forecasting is essential for cost-effective wind power integration and hybrid renewable system design, particularly in data-scarce regions across sub-Saharan Africa. Because wind power density scales with the cube of wind speed (P oo V³), even modest systematic deviations in wind speed estimation can propagate into substantial uncertainty in energy yield and project viability. This study leverages 15 years (2008-2022) of hourly 10-m ground-measured wind speed data from Nigerian Meteorological Agency (NiMET) stations representing all six geopolitical zones to develop and evaluate transparent regression-based forecasting models. The best-performing robust linear model achieves normalized mean absolute errors (nMABE) of 2.6-10.9% and Nash-Sutcliffe efficiencies (NSE) of 0.84-0.99 on fully independent 2018-2022 test data. When benchmarked against NASA's MERRA-2 reanalysis at equivalent height, the locally trained model demonstrates substantially improved station-scale performance, reducing forecasting error by factors of 3.1-19.7 (median 8.2×) while correcting persistent low-wind bias observed in the benchmark dataset. Given the cubic dependence of wind power on velocity, such bias can translate into pronounced underestimation of wind energy potential and capacity factors in site-level and national-scale assessments. The resulting closed-form forecasting equation requires only latitude, longitude, day-of-year, and a prior-year hourly wind record; inputs typically accessible to rural mini-grid operators and planners. The formulation is computationally lightweight, fully interpretable, and directly implementable in spreadsheets or low-cost embedded systems without reliance on advanced computing infrastructure. By demonstrating that long-term ground observations combined with parsimonious regression structures can deliver high forecasting accuracy, this study provides a practical and scalable framework for improving wind resource assessment, turbine siting, and hybrid renewable system optimization in emerging and data-constrained energy markets.