Urban expansion through land use allocation directly affects landscape patterns and land cover conditions. Spatial allocation strategies may lead to divergent ecosystem risks. Assessing these differences can support land use planning under future urban growth. This study develops an integrated land allocation framework that combines machine learning-based development probability, policy-oriented suitability, and ecosystem quality. We compare how different spatial allocation strategies and growth rates affect potential ecosystem risks. The study area is coastal Georgia, US. Random Forest predicts development probability from historical land development patterns. Policy-oriented suitability integrates accessibility, environmental constraints, and local comprehensive plans. Ecosystem quality is measured using vegetation, land surface temperature, wetlands, core habitat, and distance to water. Four allocation schemes and three growth rates generate 12 scenarios from 2024 to 2050. A mixed linear model compares ecosystem quality among scenarios. Random Forest identifies future development potential with an AUC of 0.946. Compared with probability-driven allocation, policy-oriented allocation reduces the mean Ecosystem Quality Index (EQI) of allocated land by 0.043-0.053, while conservation-oriented allocation reduces it by 0.094-0.104. Changing growth rates alone produces no significant difference in mean EQI. Ecosystem risks vary across counties. In Effingham County, mean EQI decreases from approximately 0.44 (probability-driven) to 0.34 (policy-oriented) and 0.28 (conservation-oriented) during 2024-2030. Machine learning identifies development potential, but planning and conservation priorities can substantially change the ecological conditions of allocated land. Conflicts between suitability and ecosystem quality are notable in coastal counties, while inland counties may provide opportunities for regional coordination to reduce pressure on sensitive habitats.