NowcastDiT: Diffusion Transformers are Effective Precipitation Nowcasters

Haoran Xu; Xingzhuo Guo; Yuchen Zhang; Jincheng Zhong; Jianmin Wang; Mingsheng Long

cs.LGSep 29, 2026

NowcastDiT: Diffusion Transformers are Effective Precipitation Nowcasters

Authors: Haoran Xu, Xingzhuo Guo, Yuchen Zhang, Jincheng Zhong, Jianmin Wang, Mingsheng Long

Organizations: School of Software, BNRist, Tsinghua University, China

Abstract

Precipitation nowcasting demands accurate short-term forecasts under strong spatiotemporal variability. Diffusion models are well suited to modeling complex precipitation distributions, yet existing approaches often introduce increasingly specialized designs, leaving the capability of a standard diffusion architecture underexplored. We show that a standard Diffusion Transformer already provides a simple and scalable foundation for precipitation nowcasting, with domain-specific requirements accommodated naturally within its design space. Based on this principle, we develop NowcastDiT and instantiate this flexibility through two complementary adaptations: a dynamics-aware noise prior for temporally coherent forecasts, and end-to-end reinforcement learning with timestep-aware rewards for meteorological skill. Experiments on SEVIR and MRMS benchmarks show that NowcastDiT achieves state-of-the-art performance in both perceptual quality and meteorological skill. These results suggest that standard DiT can serve as an effective foundation for precipitation nowcasting.

Figures & tables

Figure 1: Comparison of diffusion modeling paradigms for precipitation nowcasting. (a): Specialized diffusion nowcasters encode domain knowledge through task-specific models or losses. (b): A standard DiT is applicable for precipitation nowcasting. (c): NowcastDiT adds minimal precipitation-specific adaptation towards a standard DiT without changing its backbone.

Figure 2: Overview of NowcastDiT . Left: A standard Diffusion Transformer predicts future precipitation conditioned on encoded observations, with QK-Norm and 3D RoPE. Top right: The dynamics-aware noise prior (DyPro) schedules cross-frame noise correlation from temporally correlated at the noise endpoint to independent at the clean endpoint. Bottom right: Timestep-aware post-training shifts the reward emphasis from structure to detail along the denoising trajectory.

SEVIR							MRMS
Method	CSI ↑	CSI ${}_{181}^{\uparrow}$	CSI ${}_{219}^{\uparrow}$	HSS ↑	LPIPS ↓	SSIM ↑	Method	CSI ↑	CSI ${}_{16}^{\uparrow}$	CSI ${}_{32}^{\uparrow}$	HSS ↑	LPIPS ↓	SSIM ↑
ConvLSTM	0.2912	0.0684	0.0336	0.3571	0.2989	0.7257	ConvLSTM	0.2159	0.0514	0.0172	0.2877	0.3013	0.8978
PhyDNet	0.2874	0.0664	0.0211	0.3523	0.3081	0.7252	PhyDNet	0.2332	0.0635	0.0268	0.3111	0.3005	0.9001
Earthformer	0.2752	0.0496	0.0213	0.3391	0.3376	0.7142	Earthformer	0.2353	0.0691	0.0297	0.3152	0.3001	0.8994
SimVP	0.2951	0.0753	0.0416	0.3630	0.3113	0.7247	SimVP	0.2366	0.0657	0.0277	0.3155	0.2992	0.9006
AlphaPre	0.2980	0.0860	0.0436	0.3677	0.2897	0.7301	AlphaPre	0.2309	0.0587	0.0211	0.2933	0.3014	0.8944

Table 1: Performance comparison on SEVIR and MRMS.

\uparrow

indicates higher is better and

\downarrow

indicates lower is better. Bold and underlined values denote the best and second-best results, respectively.

Figure 3: Qualitative comparison on SEVIR. Forecasts are shown every 10 minutes up to 100 minutes. Colors indicate VIL values.

Figure 4: Qualitative comparison on MRMS. Forecasts are shown every 20 minutes up to 200 minutes. Colors indicate precipitation rate in mm h -1 .

Figure 5: Analysis on SEVIR. Bar lengths indicate CSI gains relative to the first setting in each panel; labels report absolute CSI values.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Architecture
Setting	SEVIR	MRMS
Parameters	82M
Input shape	$25\times 1\times 128\times 128$	$24\times 1\times 256\times 256$
Encoder stages	4
Residual blocks / encoder stage	2
Encoder channels	[128, 256, 512, 512]

Appendix

Table 2: VAE architecture and training configuration. Parameters include the encoder, decoder, and latent projections.

Architecture
Setting	SEVIR	MRMS
Parameters	130M
Frames (history / future)	5 / 20	4 / 20
Transformer blocks	12
Hidden dimension	768
Feedforward dimension	3072

Appendix

Table 3: Transformer architecture, flow-matching pretraining, and inference configuration. Parameters include the conditioner, time embedding, and output layer. Inference settings refer to the pretrained-model evaluation configurations.

Optimization
Setting	SEVIR	MRMS
Optimizer	AdamW, $(\beta_{1},\beta_{2})=(0.9,0.95)$
Learning rate	$10^{-5}$ , constant
Batch size	8
Policy ratio clipping range $\varepsilon_{c}$	$10^{-4}$
Advantage clipping	$[-5,5]$

Appendix

Table 4: Post-training, reward, and evaluation configurations. Reward thresholds are expressed in VIL values for SEVIR and

\mathrm{mm\,h}^{-1}

for MRMS.

Setting	SEVIR	MRMS
Data variable	VIL	Precipitation rate
Spatial / temporal res.	$1~\mathrm{km}$ / $5~\mathrm{min}$	$0.01^{\circ}$ / $10~\mathrm{min}$
Forecast frames	$5\rightarrow 20$	$4\rightarrow 20$
Frame size	$128\times 128$	$256\times 256$
Train / validation / test	59,530 / 28,145 / 7,220	6,807,528 / 12,000 / 12,000

Appendix

Table 5: Dataset configurations. Spatial resolution refers to the source radar grids, and frame size denotes the model input and output size. Forecast frames are shown as observed

\rightarrow

predicted; split sizes count sequences.

Method	CSI $\uparrow$	CSI-181 $\uparrow$	CSI-219 $\uparrow$	HSS $\uparrow$	LPIPS $\downarrow$	SSIM $\uparrow$
DiT	0.2926	0.0948	0.0555	0.3733	0.1545	0.7048
+ 3D RoPE, QK-Norm	0.3012	0.1082	0.0676	0.3647	0.1317	0.6812
+ DyPro	0.3195	0.1204	0.0752	0.4092	0.1486	0.7099
+ RL	0.3240	0.1241	0.0750	0.4148	0.1469	0.7197

Appendix

Table 6: Cumulative component ablation on SEVIR.

\uparrow

indicates higher is better and

\downarrow

indicates lower is better. Bold and underlined values denote the best and second-best results, respectively.

RL	Guidance	CSI $\uparrow$	CSI-181 $\uparrow$	CSI-219 $\uparrow$	HSS $\uparrow$	LPIPS $\downarrow$	SSIM $\uparrow$
✘	✘	0.2934	0.0956	0.0560	0.3749	0.1541	0.7041
✘	✔	0.3195	0.1204	0.0752	0.4092	0.1486	0.7099
✔	✘	0.3109	0.1050	0.0588	0.3961	0.1580	0.7156
✔	✔	0.3240	0.1241	0.0750	0.4148	0.1469	0.7197

Appendix

Table 7: Effects of RL and classifier-free guidance on SEVIR.

\uparrow

indicates higher is better and

\downarrow

indicates lower is better. Bold and underlined values denote the best and second-best results, respectively.

$\bm{\alpha}$	CSI $\uparrow$	CSI-181 $\uparrow$	CSI-219 $\uparrow$	HSS $\uparrow$	LPIPS $\downarrow$	SSIM $\uparrow$
0.00	0.3012	0.1082	0.0676	0.3647	0.1317	0.6812
0.25	0.3188	0.1184	0.0759	0.4075	0.1485	0.7090
0.33	0.3191	0.1192	0.0742	0.4080	0.1483	0.7097
0.50	0.3195	0.1204	0.0752	0.4092	0.1486	0.7099
0.66	0.3185	0.1194	0.0740	0.4077	0.1495	0.7098
0.75	0.3184	0.1184	0.0750	0.4084	0.1494	0.7092

Appendix

Table 8: DyPro correlation strength

\alpha

on SEVIR.

\alpha=0

corresponds to the standard i.i.d. Gaussian noise prior.

\uparrow

indicates higher is better and

\downarrow

indicates lower is better. Bold and underlined values denote the best and second-best results, respectively.

Reward	Static	Timestep-aware
Mixed	$(M_{s}+M_{d})/2$	$(1-t)M_{s}+tM_{d}$
CSI	$(C_{L}+C_{H})/2$	$(1-t)C_{L}+tC_{H}$
Perceptual	$(0.1S-0.9P)/2$	$0.1(1-t)S-0.9tP$

Appendix

Table 9: Terminal rewards for the six ablation settings.

Figure 6: Forecast skill across lead times on MRMS and SEVIR. The horizontal axes show forecast lead time in minutes; higher CSI and HSS indicate better performance.

Figure 7: Additional reward ablations on SEVIR. LPIPS (left) and RMSE (right) during RL training for the rewards in Table 9 . The legend follows Figure 5 (d); the timestep-aware perceptual reward uses

0.1(1-t)\mathrm{SSIM}-0.9t\mathrm{LPIPS}

. Lower values indicate better performance for both metrics.

Figure 8: Qualitative comparison on SEVIR (case 802). Five input observations span $-20$ to $0$ minutes at 5-minute intervals. Ground truth and forecasts are shown every 10 minutes up to 100 minutes. Colors indicate VIL values.

Figure 9: Qualitative comparison on SEVIR (case 6155). Five input observations span $-20$ to $0$ minutes at 5-minute intervals. Ground truth and forecasts are shown every 10 minutes up to 100 minutes. Colors indicate VIL values.

Figure 10: Qualitative comparison on MRMS (case 239). Four input observations span $-30$ to $0$ minutes at 10-minute intervals. Ground truth and forecasts are shown every 20 minutes up to 200 minutes. Colors indicate precipitation rate in mm h -1 .

Figure 11: Qualitative comparison on MRMS (case 1149). Four input observations span $-30$ to $0$ minutes at 10-minute intervals. Ground truth and forecasts are shown every 20 minutes up to 200 minutes. Colors indicate precipitation rate in mm h -1 .

Explore similar work

Probabilistic Precipitation Nowcasting with Rectified Flow Transformers
May 29, 2026Johannes Schusterbauer, Jannik Wiese, Nick Stracke +2Precipitation Nowcasting Rectified Flow
Learning to Refine: Spectral-Decoupled Iterative Refinement Framework for Precipitation Nowcasting
Jun 1, 2026Yunlong Zhou, Chen Zhao, Danyang Peng +2Precipitation Nowcasting Turbulent Flows
School of Intelligence Science and Technology, Nanjing University, Suzhou, 215163, China. · State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, 210023, China.
A Space-Time Transformer for Precipitation Nowcasting
Nov 14, 2025Levi Harris, Tianlong ChenPrecipitation Nowcasting Artificial Intelligence Weather Models
UNC Chapel Hill

May 29, 2026, cs.CVJ/K move · Enter open · S save

Probabilistic Precipitation Nowcasting with Rectified Flow Transformers

Johannes Schusterbauer, Jannik Wiese, Nick Stracke, Timy Phan+1

Precipitation Nowcasting Rectified Flow Data-Driven Forecasting Data Compression Methods

Open paper AskPDF arXiv

Accurate weather forecasts are essential across various domains and are safety-critical in extreme weather conditions. Compared to simulation-based forecasting, data-driven approaches show greater efficiency, enabling short-term, high-resolution nowcasting. In particular, diffusion models proved effective in weather nowcasting due to their strong probabilistic foundation. However, existing methods rely on deterministic compression to reduce the complexity of high-dimensional weather data, limiting their ability to capture uncertainty in the decoding process. In this work, we introduce

\textbf{FREUD}

, a

\textbf{Fr}

ame-wise

\textbf{E}

ncoder and

\textbf{U}

nited

\textbf{D}

ecoder model based on rectified flow transformers for efficient compression of spatio-temporal weather data. Frame-wise encoding enables continuous forecast updates, while the unified video decoder ensures temporal consistency. Our uncertainty-preserving first stage allows us to capture aleatoric uncertainty via ensembling, which is particularly beneficial for extreme weather events with high decoding variability. We achieve state-of-the-art performance in precipitation nowcasting with a compact latent-space rectified flow transformer on the SEVIR benchmark and show further performance gains by model and test-time scaling. Code available here: https://github.com/CompVis/weather-rf

Abstract

Explore similar work

Probabilistic Precipitation Nowcasting with Rectified Flow Transformers

Learning to Refine: Spectral-Decoupled Iterative Refinement Framework for Precipitation Nowcasting

A Space-Time Transformer for Precipitation Nowcasting