Identifying spatiotemporal dynamics and driving factors of landscape pattern in Giant Panda National Park (GPNP) using an interpretable machine learning technology

• Interpretable XGBoost–SHAP framework quantifies nonlinear effects of 16 drivers. • Clustering analysis identifies landscapes with distinct ecological functions. • Major habitat transitions occurred in 2005–2010 from restoration and disturbances. • Precipitation and NDVI dominant GPNP’s long-term landscape patterns. • Integrated LULC analysis and machine learning enhance conservation strategies. Giant Panda National Park (GPNP) is one of China’s landmark conservation initiatives within its national park system and serves as a critical stronghold for global biodiversity. The spatiotemporal changes in its landscape patterns profoundly influence giant panda habitat quality and ecosystem resilience. In this study, we developed a comprehensive framework to analyze landscape-pattern dynamics and their driving mechanisms in GPNP. The framework leverages multi-temporal (1990–2020) land use/cover remote-sensing data combined with landscape pattern metrics to quantify habitat type transitions and capture spatiotemporal pattern changes. The influence of environmental and anthropogenic drivers was quantified, and the complex interactions shaping landscape dynamics were revealed with an interpretable XGBoost-SHAP model. The results show that the most significant habitat transitions occurred during 2005–2010, accounting for 50.6 % of the total transition area, largely driven by ecological restoration and natural disturbances. Landscape connectivity steadily increased, reflecting the positive effects of ecological restoration policies targeting GPNP’s environment. The XGBoost–SHAP framework achieved strong predictive performance (accuracy = 0.74, AUC = 0.89), enabling reliable interpretation of landscape transitions. Among sixteen drivers, precipitation, NDVI, temperature, GPP, terrain ruggedness (TRI), and distance to roads (Lrdl) were identified as the most influential in shaping long-term landscape patterns in GPNP. These findings suggest that effective monitoring of GPNP’s landscape pattern dynamics can provide a scientific basis for the conservation and management of giant panda habitats. Moreover, the proposed framework offers a transferable approach for analyzing landscape changes and driving mechanisms in similar wildlife ecosystems.