PrecipJEPA: JEPA-Regularized Future-State Prediction with Motion-Source Rendering for Precipitation Nowcasting

Yufeng Zhu; Dan Niu; Qiliang Wu; Weiwei Huang; Yixiao Liang; Yongchao Feng; Chunlei Shi

cs.MMSep 30, 2026

PrecipJEPA: JEPA-Regularized Future-State Prediction with Motion-Source Rendering for Precipitation Nowcasting

Authors: Yufeng Zhu, Dan Niu, Qiliang Wu, Weiwei Huang, Yixiao Liang, Yongchao Feng, Chunlei Shi

Organizations: Department of Automation, Southeast University, Nanjing 210096, China

Abstract

Long-term precipitation nowcasting requires modeling radar-echo evolution while preserving localized high-intensity structures. Recent radar-specific studies motivate location-aware prediction and separating echo displacement from intensity change. However existing encoders learn historical representations mainly from final forecast errors. We propose PrecipJEPA, which couples a structured forecasting path with an auxiliary path that enriches its encoder from observed radar history. In the forecasting path, an online encoder first converts the observations into spatiotemporal tokens. The Task-Driven Future-State Predictor (TFP) combines these tokens with a recent-dynamics summary and spatiotemporal queries to construct future radar states. The Parallel Motion-Source Renderer (PMSR) decodes these states into motion and source-sink fields that transform the latest observation into future frames. During joint training, the History-Masked JEPA (H-JEPA) operates on the auxiliary path to predict masked historical features from visible context, directly supervising the same online encoder from the observed sequence. Experiments on SEVIR and MeteoNet show that PrecipJEPA improves highest-threshold CSI by 118.6% and 35.1%, respectively, over the strongest baselines, while maintaining the highest mean CSI throughout the 3-hour forecast.

Figures & tables

Figure 1: Overview of PrecipJEPA. TFP maps historical representations to lead-specific future states, which PMSR renders into radar forecasts, while H-JEPA supervises the shared encoder through masked-latent prediction during training.

	SEVIR								MeteoNet
Method	CSI $\uparrow$	CSI-p4 $\uparrow$	CSI-p16 $\uparrow$	HSS $\uparrow$	LPIPS $\downarrow$	CSI-last $\uparrow$	CSI 181 $\uparrow$	CSI 219 $\uparrow$	CSI $\uparrow$	CSI-p4 $\uparrow$	CSI-p16 $\uparrow$	HSS $\uparrow$	LPIPS $\downarrow$	CSI-last $\uparrow$	CSI 24 $\uparrow$	CSI 32 $\uparrow$
SimVP [ 4 ]	0.2073	0.2277	0.2405	0.2518	0.4348	0.1362	0.0276	0.0147	0.2598	0.3251	0.3562	0.3589	0.3444	0.1290	0.1974	0.0674
Earthformer [ 6 ]	0.2006	0.2193	0.2384	0.2456	0.4434	0.1366	0.0168	0.0024	0.2398	0.2968	0.3393	0.3370	0.4130	0.1309	0.1924	0.0362
AlphaPre [ 10 ]	0.2162	0.2381	0.2495	0.2652	0.4192	0.1432	0.0311	0.0139	0.2708	0.3322	0.3624	0.3723	0.3308	0.1366	0.2191	0.0771
exPreCast [ 18 ]	0.2142	0.2365	0.2497	0.2726	0.3617	0.1472	0.0391	0.0161	0.2583	0.3230	0.3668	0.3659	0.3170	0.1532	0.2279	0.1043
CasCast [ 7 ]	0.1977	0.2105	0.2333	0.2549	0.4108	0.1341	0.0337	0.0145	0.2191	0.2299	0.2224	0.3131	0.3454	0.1596	0.2053	0.0824

Table 1: Quantitative comparison on SEVIR and MeteoNet. Aggregate metrics cover all thresholds and forecast times; CSI-last covers the final forecast hour. CSI 181 , CSI 219 , CSI 24 , and CSI 32 report the two highest standard thresholds of each dataset. Best results are in bold and second-best results are underlined .

Figure 2: Long-term evaluation on SEVIR: (a) mean CSI over the full 180-minute forecast; (b) example forecasts from different models.

Variant	TFP	PMSR	H-J	CSI $\uparrow$	HSS $\uparrow$	CSI-last $\uparrow$	CSI 219 $\uparrow$	HSS 219 $\uparrow$
M0 (Base)	–	–	–	0.2082	0.2658	0.1477	0.0130	0.0239
M1 (w/o TFP)	–	✓	✓	0.2162	0.2761	0.1344	0.0305	0.0528
M2 (w/o PMSR)	✓	–	✓	0.2082	0.2662	0.1441	0.0135	0.0247
M3 (w/o H-J)	✓	✓	–	0.2215	0.2841	0.1374	0.0342	0.0592
PrecipJEPA	✓	✓	✓	0.2315	0.2974	0.1493	0.0376	0.0657

Table 2: Component ablation on SEVIR and MeteoNet. H-J denotes H-JEPA. In M0 and M1, TFP is replaced by the basic future-state predictor; a dash under PMSR indicates use of the residual head. Best results are in bold .

Figure 3: H-JEPA training strategies on (a) SEVIR and (b) MeteoNet. M3 trains without H-JEPA, M4 uses it for encoder pretraining before forecast-only training, and PrecipJEPA jointly trains H-JEPA with the forecasting path.

Explore similar work

FreCast: Refining Radar Echo Intensity via Phase-Preserving Amplitude Residual Diffusion for Precipitation Nowcasting
Aug 9, 2026Heping Fang, Zihuai Yin, Kaicheng Mao +2Precipitation Nowcasting Radar
Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, China · Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China · Guangdong Provincial Key Laboratory of Brain-Inspired Intelligent Computation, Department of Computer Science and Engineering and the Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, China
Temporal Context Conditioning for Seasonality-Aware Precipitation Nowcasting of High-Intensity Rainfall
Jun 8, 2026Gijs van Nieuwkoop, Siamak MehrkanoonPrecipitation Nowcasting Rainfall
Department of Information and Computing Sciences, Utrecht University, Utrecht, The Netherlands
VMU-Diff: A Coarse-to-fine Multi-source Data Fusion Framework for Precipitation Nowcasting
May 14, 2026Chunlei Shi, Hao Li, Yufeng Zhu +6Precipitation Nowcasting Rainfall
Department of Automation, Southeast University, Nanjing 210096, China

Aug 9, 2026, cs.CVJ/K move · Enter open · S save

FreCast: Refining Radar Echo Intensity via Phase-Preserving Amplitude Residual Diffusion for Precipitation Nowcasting

Heping Fang, Zihuai Yin, Kaicheng Mao, Peiguang Zhang+1

Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, China · Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China · Guangdong Provincial Key Laboratory of Brain-Inspired Intelligent Computation, Department of Computer Science and Engineering and the Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, China

Precipitation Nowcasting Radar Residual Conditional Diffusion Model Frequency Spatiotemporal Deep Learning

Open paper AskPDF arXiv

Precipitation nowcasting predicts the spatiotemporal evolution of future radar echoes from historical radar echo sequences, thereby estimating the occurrence, development, and movement of precipitation over the near term. In recent years, deep learning has become an important approach to precipitation nowcasting. Although state-of-the-art models can generally capture the overall spatial distribution of future precipitation, their predictions still exhibit substantial biases in radar echo intensity at individual locations. This observation motivates a more targeted strategy for reducing forecast errors. Instead of regenerating an entire radar echo sequence without spatial constraints, the predicted precipitation structure can be used to guide the refinement of echo intensities at individual locations. This structure-guided refinement directly targets echo intensity biases. Accordingly, we propose FreCast, a two-stage framework for radar echo prediction. The first stage generates an initial forecast of future radar echoes. The second stage uses the spatial structure of the initial forecast as a constraint to further correct intensity biases at individual locations in the first-stage prediction. Experiments on three datasets demonstrate that FreCast achieves consistent improvements across forecast skill metrics. Qualitative results further show that FreCast better preserves rainband continuity and intense precipitation structures at longer lead times.

Abstract

Explore similar work

FreCast: Refining Radar Echo Intensity via Phase-Preserving Amplitude Residual Diffusion for Precipitation Nowcasting

Temporal Context Conditioning for Seasonality-Aware Precipitation Nowcasting of High-Intensity Rainfall

VMU-Diff: A Coarse-to-fine Multi-source Data Fusion Framework for Precipitation Nowcasting