GHI-based PV yield assessment in high-altitude Andean sites: workflow performance and systematic limitations of irradiance decomposition models

This study develops and evaluates a practical workflow for estimating annual PV energy yield in remote, data-scarce Andean locations, and tests it at two climatically contrasting sites in southern Peru: Cota Cota (4240 m a.s.l.) and Pampas de Majes (1498 m a.s.l.). Three years of local irradiance and temperature measurements were combined with ERA5 reanalysis, using machine learning to correct systematic biases and generate a 25-year hourly climatology, from which irradiance components and PV energy yield were estimated across the full range of tilt and azimuth configurations. Bias correction performed well at both sites, and optimal annual yields reached approximately 1042 kWh per panel at Majes and 1018 kWh at Cota Cota, with low sensitivity to small deviations from the ideal orientation. The analysis also revealed a systematic, altitude-dependent divergence between two independent estimates of the diffuse fraction: at Cota Cota, the Engerer2 decomposition and the satellite-based Solargis product differ by over 25 percentage points, a gap considerably smaller at Majes. As both are model-based estimates and no in-situ component measurements were available, the true bias cannot be resolved from this comparison, and substituting the clear-sky model did not reduce the divergence. These findings highlight both the value and the current limitations of the workflow, and underscore the need for in-situ component measurements to validate and locally calibrate irradiance decomposition models.