Performance Degradation of Common Photovoltaic Module Temperature Models From a Lowland Site to a Highland Site in Western China

ABSTRACT Accurate estimation of photovoltaic (PV) module temperature is essential for reliable PV performance assessment because module temperature substantially influences system efficiency and thus power output. However, the performance of commonly used PV temperature models under real‐world array operating conditions, particularly in cold high‐altitude regions, remains insufficiently evaluated. This study assesses the performance of four PV module temperature models (Ross, Faiman, PVsyst, and SAPM) implemented in the widely used pvlib library, using observations from two PV power plants located in a lowland site in northwestern China (Wujiaqu) and a highland site on the eastern Tibetan Plateau (Xingchuan), respectively. All models exhibit systematic underestimation of module temperatures, with mean bias ranging from −0.8°C to −3.4°C at the lowland site and from −2.7°C to −5.2°C at the highland site. This bias can be amplified under strong solar radiation and snow‐cover conditions. Especially at the highland site, cold bias can frequently exceed 8°C, implying a considerable overestimation of PV power output if these models are used. Model parameters recalibrated at the lowland site power plant substantially improve local model performance but remain biased when directly applied to the highland site. This limited cross‐site transferability restricts the applicability of empirical models and their coefficients across different operating environments. These findings highlight the importance of site‐adapted parameterization and the need for more physically based modeling frameworks.