Abstract: Urban forest parks play a pivotal role in maintaining urban ecological security. The dynamics in fractional vegetation coverage (FVC) are key indicators for evaluating their ecological restoration effectiveness and sustainable management. Based on normalized difference vegetation index data of Chengdu Longquan Mountain Urban Forest Park from Landsat satellites spanning from 1993 to 2023, we used a comprehensive suite of analytical methods including Theil-Sen median slope estimation, Median-Kendall trend tests, additive seasonal and trend breakpoint detection, and the optimal partial geo-detector model (OPGD) to analyze both linear and nonlinear spatiotemporal variations of FVC within the park as well as driving factors. During 1993-2023, the park’s FVC fluctuated from 0.484 to 0.677, with 30.7% of the areas showing improvement. Furthermore, the vegetation coverage grade shifted to predominantly high and medium-high levels. The mutation of FVC primarily occurred between 1999 and 2020, peaking during 2005-2009. Among the various nonlinear mutation types of FVC, the decrease to increase pattern demonstrated the highest frequency of occurrence, accounting for 14.2% of the total pixel area. This pattern was predominantly observed in towns including Baihe, Shanquan, Luodai, and Qingquan. The key factors influencing FVC were mean annual evapotranspiration, precipitation, and slope. Before 2010, annual precipitation had the strongest influence. After 2010, the influence of slope increased subsequently, and the impact of human activities declined. The interaction of precipitation, slope, and evapotranspiration had the strongest influence. During the research period, the FVC of the Chengdu Longquan Mountain Urban Forest Park showed an improving trend, but its change process exhibited complex nonlinear characteristics. The factors influencing FVC varied significantly among different time periods, with the effect of water being the most prominent.

Key words: fractional vegetation cover, nonlinear trend, mutation, BFAST model, geoDetector, Longquan Mountain