Quantification of landscape pattern and its dynamic is essential to monitor and evaluate the consequences of urban forests. The trees in urban area provide a wide range of ecosystem services, and are beneficial to the well-being of humans as well as the environment. Previous researches have demonstrated that tree canopy coverage is closely correlated with urban heat island effect, community microclimate, air quality, biodiversity, and quality of urban life and so forth. Tree canopy coverage in urban area benefits directly the community and can be recognized as a special landscape type of urban forest at fine scale. In this paper, a vector map of tree canopy coverage was obtained by a visual interpretation of aerial image in the central city of Guangzhou, and the spatial resolution of the image was 0.4m. A raster data transformed from the map was analyzed in FRAGSTATS 3.3 software. It is well known that the observed landscape pattern and its relationship with process depend upon the scale. In this regard, these studies integrated transect analyses to discuss the change of the scale effect, including grain and extent. The methods of Standard and Moving Window were chosen respectively, and different radius were tried to quantify the landscape structure. The window moves over one cell of the landscape at a time, calculating the selected metric within the window and assigning the value to the center cell. A series of new continuous grid maps for selected metrics, including PLAND, PD, MPS, LPI, ED, PARA_AM, ENN_MN, COHESION etc., were calculated on the class level, and spatial gradient maps were outputted in grid format. The value of each grid cell represented the selected metric within selected window, and landscape pattern was described in visualization form. Based on the moving window analysis, the result demonstrated that spatial heterogeneity of landscape pattern was significant in the study area, and the selected landscape metrics could be depicted in the smoothing images. Considering effects of the grain size and extent, 3-5m and 0.4-0.6km were the optimal grain size and the moving window radius in the study area, respectively. A series of landscape pattern gradient maps in study area were obtained with the grain size of 5m and the radius of moving window of 0.5km. In comparison with quantifying metrics on the traditional holistic landscape pattern, the approach of moving window made a significant contribution to our understanding of the landscape pattern and process, which depicted a set of continuous variable maps such as vegetable density and hydrograph and relief map and others, and it linked landscape pattern and process more effectively in local scale. The analyzed result presents a continuous surface and is recorded in raster format, it is conductive to directly reveal driving forces at local level, and it can be used to compare the landscape pattern visually. Combined with other surface data such as geography information and socioeconomic data in multivariate models, the result of landscape pattern gradient analysis provided a novel way for discerning the driving force of landscape dynamic and optimizing urban forest landscape pattern at the fine scale.