Cities in arid and semi-arid region of northwest China face an acute contradiction between limited water resources and rapid urban green space expansion. Traditional methods for estimating urban green space water demand, such as the Landscape Coefficient Method (LCM), are widely used due to their applicability, but mainly focus on microclimatic conditions and do not fully consider the macro-scale urban-rural gradient caused by the urban heat island (UHI) effect. This study investigates 54 cities across Northwest China, analyzing their UHI characteristics and seasonal patterns of potential evapotranspiration (PET) across urban-rural gradients. Results show that Northwest China can be divided into 6 UHI clusters. Cities within each cluster display highly consistent PET characteristics and regular variations along the urban-rural gradient. The stronger the UHI effect, the greater the PET difference along the urban-rural gradient. In most cities, spring PET in urban cores is about 10% higher than in surrounding natural areas (K-URG approximate to 1.1), and in some cities the increase exceeds 20% (K-URG approximate to 1.2). Based on these findings, this study developed the Urban-Rural Gradient Correction Coefficient (K-URG) and provided seasonal reference values for each city. By introducing K-URG into the existing LCM, this study proposes a novel dual-scale water demand estimation framework that seamlessly integrates macro-scale regional thermal gradients with micro-scale landscape characteristics. Overall, the proposed framework provides a practical and scalable quantitative tool to support water-saving urban greening strategies and the refined management of urban landscapes.