A hybrid BAG–CARIMA–GPM model for climate-induced shifts in the solar power potential based on PV and CPV technologies

This study develops and applies a hybrid model combining bagging (BAG), controlled autoregressive integrated moving average (CARIMA) and a Gumbel probabilistic model (GPM), hereafter BAG–CARIMA–GPM to generate climate-smart forecasts of monthly mean daily global horizontal irradiation H (MJ m⁻² day⁻¹) and photovoltaic performance in two climatically contrasting Nigerian cities, Warri and Sokoto. Bias-corrected outputs from five CMIP6 models (GFDL-ESM4, HadGEM3-GC31, EC-Earth3, IITM-ESM4 and BCC-CSM2) are processed using quantile delta mapping (QDM), N-dimensional multivariate bias correction (MBCn) and generalized extreme value (GEV) tail adjustment to obtain dynamically consistent projections of radiation, temperature, humidity and cloud fraction under SSP1–2.6, SSP2–4.5 and SSP5–8.5. The BAG–CARIMA–GPM model couples controlled autoregressive time-series structure, ensemble learning and extreme-value statistics through a normalized insolation parameter that embeds sunshine fraction, cloudiness and thermal gradients. Thirty empirical and machine-learning configurations were benchmarked. The BAG–CARIMA–GPM model achieved a reported held-out test R 2 of 0.9718, compared with 0.9193 for CARIMA1, the strongest non-hybrid baseline. Climate-perturbed insolation is then propagated through published empirical PV/CPV response equations whose local calibration and electrical validation are not demonstrated for six flat-plate technologies (m-Si, p-Si, a-Si, h-Si, CdTe, CIGS) and dual-axis CPV. The results report Warri CPV changes of +24.1% in DJF, −25.6% in JJA and +7.1% annually under SSP5–8.5 for 2015–2099; under SSP1–2.6, the reported annual change is +4.0%, whereas small a-Si annual changes in Sokoto thereby establishing a robust gain or technology advantage in the estimates. The framework offers a transferable blueprint for technology–climate matching and risk-aware solar planning across data-sparse African regions, supporting investment decisions and policy pathways aligned with net-zero and energy-security objectives. By explicitly linking bias-corrected CMIP6 projections, hybrid forecasting and multi-technology performance envelopes, this study advances quantitative evidence for resilient solar infrastructure in rapidly warming, infrastructure-constrained tropical environments.