Accurate, efficient, and timely yield estimation is critical for crop variety breeding and management optimization. However, systematically deciphering the contributions of different proximal sensing data characteristics (e.g., spectral, temporal, and spatial information) to yield estimation is an important and unresolved issue. Here, we collected long-term, hyper-temporal, and large-volume LiDAR and multispectral data to (i) seek the best machine learning method and prediction stage for yield estimation, (ii) decipher the contribution of multisource data fusion and the dynamic importance of structural and spectral traits, and (iii) elucidate the contribution of time-series data fusion and 3D spatial information for yield estimation. The results showed that wheat yield can be accurately (R2=0.891) and timely (approximately 2 months before harvest) estimated from the fused LiDAR and multispectral data. The ANN model and the flowering stage were always the best method and prediction stage, respectively. Spectral traits (e.g., CIgreen) dominate wheat yield estimation, especially in the early stage, while the importance of structural traits (e.g., height) is more stable in the late stage. Fusing the spectral and structural traits improved the estimation accuracy at the entire growth stage. Better yield estimation is from traits derived from complete 3D points than canopy surface points and from integrated multi-stage (especially from jointing to heading and flowering stages) data than single-stage data. To the best of our knowledge, this study proposed a novel perspective on deciphering the contribution of spectral, structural, and time-series information to wheat yield estimation and can guide accurate, efficient, and timely estimation of wheat yield.