Estimating and mapping cultural ecosystem services in Yarlung Tsangpo Grand Canyon National Park using MLLMs and the SolVES model

National parks are essential for advancing ecological conservation and enhancing human well-being. However, the tension is growing between their stringent conservation requirements and the public’s increasing demand for recreation. Assessing Cultural Ecosystem Services (CES) within national parks can reveal their spatial distribution patterns and driving mechanisms, offering scientific support for resolving the tension. In this study, we proposed a novel CES assessment framework that combines multimodal large language models (MLLMs) with the SolVES (Social Values for Ecosystem Services) model. Using 62,188 geotagged photos, we quantitatively assessed CES in the Yarlung Tsangpo Grand Canyon National Park (YTGCNP). The results showed that CES values were ranked as follows: aesthetic > recreation > cultural heritage > species observation. Furthermore, the spatial patterns of CES values were primarily shaped by accessibility, specifically distance to attractions (DTA) and distance to roads (DTR), rather than by biophysical factors. High-value areas for all CES types were concentrated within 2 km of roads and 10 km of attractions. Notably, a significant discrepancy existed between public perceptions of the value of species observation and the area’s objective ecological significance, mainly attributable to accessibility constraints. These findings have direct implications for the optimization of functional zoning and conservation strategies in YTGCNP. They underscore the need for managers to improve infrastructure layout and to develop robust environmental education and ecological interpretation programs. Such efforts are critical for bridging this perceptual gap and for aligning ecological conservation with high-quality visitor experiences.