Abstract
Accurate short-term precipitation forecasts predominantly rely on denseweather-radar networks, limiting operational value in places most exposed toclimate extremes. We present TUPANN (Transferable and Universal Physics-AlignedNowcasting Network), a satellite-only model trained on GOES-16 RRQPE. Unlikemost deep learning models for nowcasting, TUPANN decomposes the forecast intophysically meaningful components: a variational encoder-decoder infers motionand intensity fields from recent imagery under optical-flow supervision, alead-time-conditioned MaxViT evolves the latent state, and a differentiableadvection operator reconstructs future frames. We evaluate TUPANN on bothGOES-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 metricsover 4-64 mm/h thresholds. Comparisons against optical-flow, deep learning andhybrid baselines show that TUPANN achieves the best or second-best skill inmost settings, with pronounced gains at higher thresholds. Training on multiplecities further improves performance, while cross-city experiments show modestdegradation and occasional gains for rare heavy-rain regimes. The modelproduces smooth, interpretable motion fields aligned with numerical opticalflow and runs in near real time due to the low latency of GOES-16. Theseresults indicate that physically aligned learning can provide nowcasts that areskillful, transferable and global.