Large-scale photovoltaic (PV) deployment in arid lands is vital for decarbonisation, yet PV-induced cool island effect (CIE) and its ecological consequences remain unclear. We analysed eight PV plants in the arid regions of China using Landsat-8 land surface temperature, kernel normalised difference vegetation index, buffer analysis, and partial least squares structural equation modelling (PLS-SEM). PV plants exhibited a daytime CIE with a cooling intensity (XD) of 0.02-3.1 degrees C and a cooling distance (Dist) of 120-540 m. PLS-SEM indicated morphological complexity as the primary positive driver (beta = 0.92); plant scale exerted a suppressive effect (beta =-0.70), whereas climate contributed positively (beta = 0.35), explaining 63% of CIE variance (R2 = 0.63). Vegetation responses exhibited distinct spatial-seasonal heterogeneity, with their direction and intensity coordinatively regulated by background climate and XD. We proposed a geographically differentiated 'PV CIE-vegetation response' framework. Medium-scale, decentralised plants with superior shape complexity are preferable in relatively dry and warm regions; however, in cold, high-altitude areas, adjusting tilt and reducing panel density may mitigate vegetation risks. These findings quantify PV CIE characteristics, elucidate drivers, and inform coordinated energy development and ecological conservation.