Multi-modal retrieval has traditionally focused on combining diverse query in-puts, such as text and sketches, in remote sensing and computer vision. However, retrieval involving multi-modal target representations, such as paired rgband depth data, remains largely unaddressed. This work investigates whetherincorporating a depth modality can improve the performance of vision-languagemodels in the context of satellite image retrieval. To explore this, a novel dataset, rsitdd, was constructed, combining orthophotos and digital height models, andused to train a clip-based remote sensing depth encoder. Experimental results show that models augmented with a depth encoder outperform their text-image-only counterparts across multiple benchmark settings. These findings highlight the potential of depth-enhanced models for remote sensing applications and demonstrate that even simple fusion techniques can yield measurable performance improvements.