A series of urgent issues such as global warming, frequent natural disasters, and water shortages are largely result from heavy human interference and its induced changes of landscape structure. How the landscape pattern variation impacts the surface runoff remains an unsolved issue for watershed water management. In the past decades, the action of ecological environment protection and projects of returning farmland to forests and grasslands have significantly changed the landscape pattern in Fuhe Basin of Poyang Lake watershed, which will inevitably affect the hydrological cycle of the basin. In this study, the SWAT (Soil and Water Assessment Tool) model for the Fuhe Basin was firstly built to predict the runoff depth for the divided 31 subbasins during three periods from 1990 to 1999, 2000 to 2008 and 2009 to 2013. Meanwhile, dynamic changes of landscape pattern were analyzed Based on the land use and cover data in the year of 1990, 2000, and 2008. Finally, the correlations between different landscape indicators and surface runoff were quantified over subbasins using Spearman Correlation analysis.

Results showed that the SWAT model is applicable in Fuhe Basin with R² (correlation coefficient) and E_ns (Nash-Sutcliffe efficiency coefficient) larger than 0.85 for runoff prediction. Based on the land use and cover data in the year of 1990, 2000, and 2008, the landscape index were calculated. The fragmentation degree increased first and then decreased during the two transition periods. The landscape patches shape became more complicated and connectivity of different landscapes decreased and then increased, among which the forest and paddy field have greatest fragmentation degree. Forest has a larger dominance degree, which was highly connectable to patches of other various types of landscape. According to the changes of landscape pattern from 1990 to 2008, the smaller fragmentation degree, more complex patch shape, and more uneven distribution of landscape patches led to more precipitation interception and runoff reduction.

From the perspective of the landscape type, the landscape area ratio (PLAND) of dry land, paddy field, water area, urban and rural construction land patches has a significantly positive correlation with surface runoff, while PLAND and FRAC_AM (Area Weighted Patch Fractal Dimension) index of forestland have a significantly negative correlation with surface runoff. The complexity of forestland patch shape strengthened the interception of rainfall and plant transpiration of land surface, which would reduce water yield. The expansion of urban impervious area, aggravating landscape fragmentation, reduced runoff interception and increased surface runoff, which can be indicated by positive relation with PLAND and FRAC_AM index of urban construction. Deeply understanding surface runoff alternations, induced by landscape patterns changes, would provide theoretical support for solving the problems of sustainable utilization of water resources, but also offers best management practices (BMPs) for land management.