Localized Smart Persistence model for intra-hour solar irradiance forecasting across diverse climates

Intra-hour solar irradiance forecasting is crucial for photovoltaic integration and grid operations. Smart Persistence (SP) is a widely used baseline, but its theoretical clear-sky scaling ratio may deviate from actual irradiance evolution under cloudy conditions and near sunrise/sunset. This study introduces Localized Smart Persistence (LSP), which replaces the clear-sky scaling ratio with a site-specific ratio derived from temporally averaged measured Global Horizontal Irradiance (GHI) profiles. The profiles are constructed by grouping historical daytime GHI measurements by time of day at the native temporal resolution of each dataset, and averaging them over a reference period. LSP was evaluated using temporal resolutions ranging from 30 s to 5 min and forecast horizons of 10 – 50 min across seven sites representing tropical, predominantly clear-sky, and high-latitude climatic conditions. Forecast performance was assessed using RMSE, MAE, R 2 , and MBE, with nRMSE and the NICE Σ additionally evaluated at the common 5-minute resolution to facilitate cross-site comparisons. The results show that the effectiveness of LSP is strongly site-dependent. LSP generally reduced RMSE and MAE relative to SP at tropical and high-latitude sites, particularly at longer forecast horizons, while providing more representative irradiance evolution under low solar elevation conditions. In contrast, under predominantly persistence-dominated conditions, climatological scaling provided limited practical benefit. Approximately one year of historical irradiance data was generally sufficient to construct a representative climatological GHI profile for most sites. Overall, LSP should be regarded as a site-dependent extension of SP rather than a universally superior replacement, with potential extension to plane-of-array irradiance forecasting.