Determining an appropriate grain size is crucial in landscape pattern analysis and ecological research. In order gain deeper insights into the spatial diversity and dynamic characteristics of landscape pattern, based on the survey data of the 12 hm² large plot in Jiangxi Guanshan National Nature Reserve in 2015 and 2020, the grain size effect of landscape indices in eight different landscape types within the range of 5-50 meters were analyzed; the variation coefficients were used to reveal the changing characteristics of different landscape pattern indices as the grain size increased, and the optimal grain size was selected based on the inflection points of changes in various landscape indices. The results were as follows: (1) Based on the importance value analysis, eight different landscape types were identified in the arbor layer of the plot, including forest gap, bamboo forest, Cunninghamia lanceolata forest, Pinus massoniana forest, broad-leaved forest, bamboo and Cunninghamia lanceolata mixed forest, bamboo and Pinus massoniana mixed forest, Cunninghamia lanceolata and Pinus massoniana mixed forest. (2) The overall distribution of patches in the landscape was relatively balanced, and the distribution pattern remained relatively stable. At larger spatial grain size, the aggregation of each landscape type increased, leading to a higher probability of landscape merging. In contrast, at smaller granularity, there was a noticeable trend of fragmentation in landscape types, providing a more comprehensive display of the quantity, density, and shape of each landscape type. Patch density (PD), number of patches (NP), edge density (ED), landscape shape index (LSI), mean fractal dimension index (FRAC_MN), contagion index (CONTAG) exhibited significant negative correlations with increasing grain size, while mean patch area (AREA_MN), Shannon's diversity index (SHDI), Shannon's evenness index (SHEI) exhibited significant positive correlations with increasing grain size. (3) The coefficients of variation for landscape indices PD, NP, ED, LSI, and AREA_MN were the highest, combining landscape structural stability and diversity, the changes in different landscape indices were primarily concentrated around the inflection point at 5 meters. These results illustrate the diversity of landscape types, with the coefficient of variation reflecting the most sensitive characteristics of landscape pattern changes. This study suggests that a grain size of 5 meters is optimal for studying the landscape pattern of subtropical evergreen broad-leaved forests in Guanshan, Jiangxi. This study provides favorable evidence for the restoration of forest resources and stand stucture. [ABSTRACT FROM AUTHOR]