The relationship between built environment and metro ridership at zone-to-zone level is of great significance to urban rail transit network planning and passenger flow control. It's expected to help holistic capacity allocation for different zones at meso-level. Most of the existing studies associated the built environment with metro ridership at stations level. However, it's difficult to distinguish the interaction between stations due to the high density of stations. This paper identified the distinct station groups for zoning by taking into account the category of stations, geographical location and passenger flow destination proportion. After zoning, collecting travel impedance variables, demographics and land use attributes, study adopts the RF model to reveal the relative importance of multidimensional variables. Important features screened from RF are used as input to GBDT model to explore the nonlinear relationship of land use features and metro ridership at zone-to-zone level. The results illustrate that land use attributes on the origin side have a larger effect while we focus on the ridership from residential zone to office zone in morning rush hour. Almost all the land use variables show threshold effects on ridership. These findings help planners determine effective ranges of land use indicators and prioritize land use attributes.