Precipitation nowcasting of satellite data using physically-aligned neural networks

TL;DR

This work tackles the challenge of sparse radar coverage for short-term precipitation nowcasting by introducing TUPANN, a satellite-only neural network with physically aligned structure. TUPANN decomposes forecasting into motion inference via a variational encoder–decoder supervised by optical flow, latent evolution through a lead-time conditioned MaxViT transformer, and a differentiable warp for advection-based reconstruction. Across GOES-16 RRQPE and IMERG datasets in four climates, it achieves state-of-the-art CSI and HSS, especially at higher rainfall thresholds, and demonstrates robust cross-city and multi-city transferability with near real-time latency. The model provides interpretable motion fields and operational practicality, enabling radar-sparse regions to access skillful nowcasts, while future work will address uncertainty quantification and cross-platform generalization.

Abstract

Accurate short-term precipitation forecasts predominantly rely on dense weather-radar networks, limiting operational value in places most exposed to climate extremes. We present TUPANN (Transferable and Universal Physics-Aligned Nowcasting Network), a satellite-only model trained on GOES-16 RRQPE. Unlike most deep learning models for nowcasting, TUPANN decomposes the forecast into physically meaningful components: a variational encoder-decoder infers motion and intensity fields from recent imagery under optical-flow supervision, a lead-time-conditioned MaxViT evolves the latent state, and a differentiable advection operator reconstructs future frames. We evaluate TUPANN on both GOES-16 and IMERG data, in up to four distinct climates (Rio de Janeiro, Manaus, Miami, La Paz) at 10-180min lead times using the CSI and HSS metrics over 4-64 mm/h thresholds. Comparisons against optical-flow, deep learning and hybrid baselines show that TUPANN achieves the best or second-best skill in most settings, with pronounced gains at higher thresholds. Training on multiple cities further improves performance, while cross-city experiments show modest degradation and occasional gains for rare heavy-rain regimes. The model produces smooth, interpretable motion fields aligned with numerical optical flow and runs in near real time due to the low latency of GOES-16. These results indicate that physically aligned learning can provide nowcasts that are skillful, transferable and global.

Figures (11)

• Figure 1: Proportion of observations above different precipitation thresholds in the GOES‑16 RRQPE dataset for each of the four study regions. Manaus exhibits the highest frequency of heavy rainfall across thresholds, while Rio de Janeiro and La Paz show intermediate levels
• Figure 2: Accumulated precipitation in GOES‑16 RRQPE from January 2020 to December 2023 over each study region. Shaded areas denote training, validation and test splits. Seasonal variability differs markedly between regions, with pronounced dry and wet seasons in La Paz and Rio de Janeiro
• Figure 3: Statistics of IMERG data, highlighting differences with respect to the RRQPE dataset
• Figure 4: TUPANN architecture. The VED and MaxViT modules displayed are learned; motion fields and the final predictions are extrapolated through a warp function
• Figure 5: Ground truth motion fields are obtained using an optical flow method from a pair of past and future images. The past image is advected to obtain an intermediate one, $\tilde{X}_1$. Finally, ground truth intensity is the subtraction of $\tilde{X}_1$ from the future image