Abstract: The dry density and initial moisture content of compacted loess significantly affect the process of water infiltration. Taking Lanzhou Loess as the research object, the immersion infiltration tests were carried out through filling the test model in indoor. The volumetric water content and pore air pressure of the soil at different depths in the process of infiltration were tested. Then variation laws of wetting front and pore air pressure during water infiltration were analyzed. The test results suggest that when the degree of soil compaction increases, the water infiltration rate decreases, and the time of the wetting front reaching the measuring point is prolonged. The degree of soil compaction increases from 0.83 to 0.93, and the infiltration rate at different depths decreases as high as 21.7%. When the initial water content increases, the pore water connectivity is enhanced and the water infiltration rate increases. The effect of dry density on water infiltration is more significant compared with the initial moisture content. The pore air pressure changes continuously during water infiltration, and the process can be divided into two stages: rapid change and slow change. In the process of water infiltration, the maximum peak air pressure and stable air pressure are positively correlated with the initial moisture content and dry density of soil. However, the infiltration rate decreases greatly with the increase of the infiltration depth, and the bigger the dry density is or the lower the volumetric water content is, the smaller the infiltration rate is. Considering the influence of dry density and initial water content on pore air pressure, the Green-Ampt model was modified. Finally, through comparison and analysis of the measured values with the calculated values of the infiltration depth, it is found that the calculated value of the modified model is in good agreement with the measured value of the test.

Key words: unsaturated loess, immersion infiltration, wetting front, law of water-air migration, model test