Extraction of 231 GS transformation trajectories and classification into five evolutionary types.

First, the LULC classification results from six periods were resampled to a spatial resolution of 5 m, and the land-cover types of GS for each period were extracted at the pixel level. For GS that consistently retained the morphological characteristics of the PDT period, these were identified as unaffected by urban construction. GS that experienced one or more episodes of development impact, such as engineering occupation, land leveling, or infrastructure construction, and reappeared in different forms after development were identified as GS affected by urban construction, yielding a total of 18,159,510 data points (3,026,585 pixels per period). Subsequently, focusing on GS that underwent transformation (Fig. 3), the transition characteristics of each GS pixel were identified for the five development stages of PDT – 2005, 2005 – 2010, 2010 – 2015, 2015 – 2020, and 2020 – 2025.
Second, based on five transitional features, each pixel is assigned a numerical value representing distinct characteristics. All transitional features possessed by each pixel are then encoded chronologically to generate independent transformation trajectories.
Finally, based on the full-cycle transformation trajectories, evolutionary types were classified by identifying the frequency of occurrence of each transition characteristic within the trajectory. For example, if the characteristic codes 3 and 4 did not appear in a trajectory, the evolutionary type was classified as Stabilization. All trajectories were grouped into the following five evolutionary types (Stabilization, Increase, Disappearance, Disturbance, and Fluctuation), each representing different degrees of impact of urban development on GS.