This study proposes a Compressive Sensing Convolutional Long Short-Term Memory (CSCL) model that aims to improve short-term (12–24 h) forecast accuracy compared to standard ConvLSTM. It is especially useful when subtle spatiotemporal variations complicate feature extraction. CSCL uses uniform sampling to partially mask spatiotemporal wave fields. The model training strategy integrates both complete and masked samples from pre- and post-sampling. This design encourages the network to learn and amplify subtle distributional differences. Consequently, small variations in convolutional responses become more informative for feature extraction. We considered the theoretical explanations for why this sampling-augmented training enhances sensitivity to minor signals and validated the approach experimentally. For the region 120–140° E and 20–40° N, a four-layer CSCL model using the first five moments as inputs achieved the best prediction performance. Compared to ConvLSTM, the R² for significant wave height improved by 2.2–43.8% and for mean wave period by 3.7–22.3%. A wave-energy case study confirmed the model’s practicality. CSCL may be extended to the prediction of extreme events (e.g., typhoons, tsunamis) and other oceanic variables such as wind, sea-surface pressure, and temperature.

Multi-instrument synergistic observation is vital for studying cloud and precipitation physics. However, using the nearest scan time for matching inevitably introduces temporal mismatches. Here we employ three advection correction methods for temporal matching in weather radar and spaceborne radar observations: Lucas–Kanade (LK), Variational Echo Tracking (VET), and Anisotropic Diffusion (AD). These methods calculate the movement speed of the storms using optical flow methods, and then determine their positions based on the elapsed time between instruments. Next, we conducted a quantitative assessment of the performance of these three methods based on the consistency of storm morphology and rainfall rates. Our results demonstrate that all three advection correction methods effectively reduce the discrepancies in morphology and rainfall rate among multi-source data. Without correction, the Coincidence Rate (CR) and Structural Similarity (SSIM) were 30.96% and 0.689 in the US and 29.44% and 0.670 in China, respectively. In comparison, applying the LK, VET, and AD methods increased those indices to 32.94%, 32.72%, 32.85% and 0.718, 0.715, 0.716 in the US, and 31.34%, 31.17%, 31.24% and 0.696, 0.694, 0.693 in China, respectively. The rainfall rate inconsistencies were also effectively reduced after advection correction. The performances among the three methods were similar. Overall, the LK method performed slightly better than AD, followed by VET.