CT-NET: A cross-modal transformer network for satellite-borne GNSS-R soil moisture retrieval

Global Navigation Satellite System-Reflectometry (GNSS-R) technology has been widely applied in soil moisture retrieval due to its low cost, short revisit periods, and precise positioning. To fully exploit global contextual information and long-range dependencies in Delay-Doppler maps (DDMs) from the Cyclone Global Navigation Satellite System (CYGNSS) and related auxiliary data, this paper proposes a novel method—Cross-Modal Transformer Network (CT-NET). CT-NET uniquely combines Transformer's multi-head self-attention and cross-attention mechanisms with a complementary fusion. The self-attention mechanism captures intrinsic features of GNSS-R multimodal data, while cross-attention enhances interconnectivity between different modal data. The Yellow River Delta is selected as the experimental area to validate the proposed method using soil moisture observations from the Soil Moisture Active Passive (SMAP) satellite. Compared with three state-of-the-art models, experimental results show that CT-NET outperforms other methods, confirming the innovation and superiority of this approach for soil moisture retrieval.