The systematical exploration of how two-dimensional (2D) and three-dimensional (3D) features of urban landscapes influence land surface temperature (LST) is still limited. Therefore, we investigated the influence of three main urban landscapes—urban green space, impervious land, and water bodies on LST, with a particular focus on the 3D vegetation metrics of green volume (GV) and leaf area index (LAI). We used Yinchuan City, China, as a case study. We quantified the impacts of various 2D/3D metrics of the three landscape types on LST using a random forest analysis with multiple sources, including Unmanned Aerial Vehicle (UAV) and remote sensing images. We then generated a Bayesian Network (BN) model to identify the optimal configurations for each landscape type. We found that using 11 of the 31 metrics considered, our model could explain 81.8% of the observed variance in LST of Yinchuan City. Among those, water body metrics were the most important, followed by vegetation abundance, impervious land metrics, and landscape pattern of urban green space. The mean classification error of the BN model was only 22.9%. We suggest that this makes the BN model a promising support tool for urban planning with a view to urban heat island mitigation. Our findings also stress the importance of considering both 2D and 3D features when considering urban cooling strategies.