Landscape pattern determines the local distribution of resources and habitats, which has an important impact on a variety of ecological processes. Based on the full understand of the coupling relationship between landscape pattern and ecological processes, landscape pattern optimization is aimed at achieving the maximum ecological benefits through the adjustment of the landscape patches’ spatial distribution and size. In order to consider more factors in landscape pattern optimization and make the optimization results more scientific and reasonable, an SFLA-M-L model was built based on shuffled frog leaping algorithm (SFLA), logistic regression model and Markov model. The landscape pattern of Dengkou County, Bayannaoer City, Inner Mongolia was optimized to verify the model. Logistic regression model was used to analyze the landscape pattern suitability based on DEM, slope, under ground water depth, aridity index, NDVI and current landscape distribution. Markov model was used to build the landscape transition probability matrix. The objective function of SFLA-M-L was built based on the landscape suitability atlas and landscape aggregation index. Landscape pattern transition probability matrix was used to restrict the transfer of different landscape types. In the optimization results, the landscape aggregation index was 96.71%, which was 6.43 percentage points higher than the landscape pattern in 2016;landscape suitability index was 96.23%, which was 4.18 percentage points higher than the landscape pattern in 2016;the transfer area beyond the control of landscape pattern transition probability matrix was only 4.66km2，and the rationality of the optimization results was ensured.