基于野外光谱辐射测量与可解释性机器学习的干旱区光伏电站地表反照率评估

Abstract As research on photovoltaic (PV)-induced climate effects continues to deepen, numerical simulation has become an essential approach. However, existing parameterization schemes remain limited, particularly in the representation of surface albedo. To address this gap, this study investigates the characteristics of spectral radiation and surface albedo variations based on observational data from April to August 2020 at a PV-Gobi composite surface in Wujiaqu, Xinjiang. The results demonstrate that the incident solar radiation exhibits a spectral hierarchy of near-infrared (NIR) > visible (VIS) > ultraviolet (UV), with respective contributions of 57.4 %, 38.4 %, and 4.1 % to the total shortwave radiation. All spectral bands showed synchronized fluctuations driven by weather processes. The albedo of the PV-Gobi composite surface was significantly lower than that of natural gobi terrain, with weighted mean values of 0.139 (global radiation, GR), 0.148 (NIR), 0.130 (VIS), and 0.081 (UV). Machine learning-based interpretation identified solar elevation angle (θ), relative humidity (RH), and photovoltaic module temperature (PT) as the dominant drivers of albedo dynamics. A parameterization model incorporating these three factors achieved high accuracy across weather scenarios and seasonal transitions, providing a multi-mechanism coupled framework to optimize PV albedo representation in geophysical models for simulating PV power plant-climate interactions.