WeatherBridge: Forecast-Aware Multi-Source Input Adaptation for Pretrained Weather Forecasting Models

Mutian Xi; Xinyu Su; Xiaohui Zhong; Hao Li

cs.AIOct 10, 2026

WeatherBridge: Forecast-Aware Multi-Source Input Adaptation for Pretrained Weather Forecasting Models

Authors: Mutian Xi, Xinyu Su, Xiaohui Zhong, Hao Li

Organizations: Artificial Intelligence Innovation and Incubation Institute, Fudan University · Shanghai Academy of AI for Science · FuXi Intelligent Computing Technology Co., Ltd.

Abstract

Data-driven weather forecasting models are commonly trained on ERA5, yet operational deployment must use timely initial states from physics-based numerical weather prediction (NWP) systems whose distributions differ substantially from ERA5. Direct initialization with these states can degrade autoregressive forecasts, motivating input adaptation that keeps the pretrained forecasting model unchanged. Here, we introduce WeatherBridge, a unified, source-conditioned adapter that maps heterogeneous operational NWP initial states to ERA5 reference states without updating the pretrained forecasting model. WeatherBridge predicts the structured residual corrections with a Swin-UNet and injects source embeddings across multiple spatial scales, allowing the model to capture shared atmospheric structure and source-specific discrepancies. Its training objective combines latitude-weighted ERA5 reconstruction with One-step Rollout Alignment (ORA), which backpropagates one-step forecast error through a frozen FuXi model to optimize the adapted states for both initial state fidelity and downstream forecast compatibility. We evaluate WeatherBridge using daily GFS, HRES, and CMA initializations from 1 January to 22 November 2025 across six variables and lead times up to 240 hours. Relative to raw source initialization of FuXi, WeatherBridge reduces the mean RMSE of FuXi over the 6- 240-hour rollout for 17 of 18 source-variable combinations, with a median reduction of 10.2% and a maximum reduction of 46.3%. The adapter trained with frozen FuXi also improves FengWu forecasts on GFS inputs without retraining either the adapter or FengWu. These results demonstrate that forecast-aware input adaptation can improve forecasts from heterogeneous operational NWP inputs and transfer across pretrained ERA5-based forecasting models.

Figures & tables

Figure 1: Effect of operational-source initialization on a frozen ERA5-trained forecaster. (a) Schematic illustration of source-dependent input distribution differences. (b) Initialization of the same frozen forecaster with ERA5 or raw operational fields, without adaptation. (c) T2M RMSE of FuXi initialized with ERA5, GFS, HRES, and CMA, with all forecasts verified against ERA5. The forecast errors vary across initialization sources.

Figure 2: Overview of WeatherBridge. (a) Forecast-aware training. Given a source state $x_{t}^{s}$ , WeatherBridge maps it to a reference-domain state $\hat{x}_{t}^{E}$ . The adapted state and the preceding ERA5 state $x_{t-1}^{E}$ are fed into the frozen weather forecasting model to predict $\hat{x}_{t+1}^{E}$ . The adapter is optimized using the reconstruction loss $\mathcal{L}_{\mathrm{rec}}$ at time $t$ and the one-step rollout loss $\mathcal{L}_{\mathrm{ORA}}$ at time $t+1$ , while the forecasting model remains frozen. (b) Inference stage. Two consecutive source states, $x_{t-1}^{s}$ and $x_{t}^{s}$ , are mapped by WeatherBridge with shared parameters. The resulting state pair is then passed to the frozen weather forecasting model to predict $\hat{x}_{t+1}^{E}$ . (c) WeatherBridge adapter. The source identity and valid time are encoded as embeddings and hierarchically injected into multiple downsampling stages of a residual Swin-UNet. The network predicts a source-aware and time-aware correction $\Delta_{\theta}(x_{t}^{s},s,t)$ , yielding $\hat{x}_{t}^{E}=x_{t}^{s}+\Delta_{\theta}(x_{t}^{s},s,t)$ .

Figure 3: Forecasting performance on FuXi over the 240-hour rollout for the six evaluation variables. Higher ACC and lower RMSE indicate better forecasting performance. Solid curves denote the WeatherBridge-aligned initialization, while dashed curves of the same color denote the raw-source initialization for GFS (orange), HRES (green), and CMA (purple). The gray dashed curve denotes the ERA5 reference initialization.

Source	Var.	PSNR $\uparrow$			SSIM $\uparrow$			Mean RMSE $\downarrow$
Source	Var.	raw	WeatherBridge	$\Delta$	raw	WeatherBridge	$\Delta$	raw	WeatherBridge	$\Delta$
GFS	Z500	44.78	46.68	4.24%	0.9999	0.9999	0.00%	719.36	386.46	46.28%
	T2M	31.45	34.27	8.97%	0.9943	0.9971	0.28%	3.31	1.78	46.22%
	T850	29.61	33.68	13.75%	0.9922	0.9967	0.45%	3.60	2.11	41.39%
	WS10	21.89	23.77	8.59%	0.9409	0.9609	2.13%	3.09	2.08	32.69%
	WS850	23.50	24.18	2.89%	0.9529	0.9596	0.70%	5.06	3.51	30.63%

Table 1: Per-source and per-variable results on FuXi. PSNR and SSIM assess initial-state alignment with ERA5, while mean RMSE is averaged over the 40-step rollout up to 240 h. “

\uparrow

” and “

\downarrow

” indicate that higher and lower values are better, respectively.

\Delta

denotes the relative improvement of the WeatherBridge-aligned result over the raw-source result; negative values indicate degradation.

Var.	RMSE $\downarrow$			ACC $\uparrow$
Var.	Raw	Adapted	$\Delta$ (%)	Raw	Adapted	$\Delta$ (%)
Z500	1023.32	952.65	6.91	0.0732	0.1748	138.68
T2M	3.93	3.23	17.77	0.3584	0.4365	21.78
T850	4.81	4.08	15.25	0.0783	0.1717	119.27
WS10	3.88	3.86	0.61	0.0479	0.0907	89.50
WS850	6.29	5.91	6.08	0.0322	0.0873	171.17

Table 2: FengWu results on GFS at 240 h with the FuXi-trained adapter, neither retrained.

\uparrow

\downarrow

: higher/lower is better.

\Delta

: relative gain over raw-source initialization.

Figure 4: Reduction of the initial-field error against ERA5 obtained by the WeatherBridge-aligned initialization relative to the raw-source initialization, averaged over the 00z initializations of the test period. Rows are sources and columns are variables; red marks grid cells where the aligned initial field is closer to ERA5 than the raw-source field, blue marks cells where it is farther. The map compares initial fields rather than forecasts.

Figure 5: NRMSE of GFS-only w/o ORA vs. GFS-only; negative (blue) means ORA helps.

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

Type	Full name	Abbreviation	Role	Units	Channels
Upper-air variables	Geopotential	Z	Input and output	$\mathrm{m^{2}\,s^{-2}}$	13
	Temperature	T	Input and output	$\mathrm{K}$	13
	u component of wind	U	Input and output	$\mathrm{m\,s^{-1}}$	13
	v component of wind	V	Input and output	$\mathrm{m\,s^{-1}}$	13
	Relative humidity	R	Input and output	%	13
Surface variables	2-m temperature	T2M	Input and output	$\mathrm{K}$	1

Appendix

Table 3: A summary of the 70 channels used by WeatherBridge. The Units column gives the physical unit of each channel.

Channel	Per-channel RMSE
Z500	21.8412
Z850	22.2562
U200	1.0605
T2M	0.6205
U10M	0.4522
V10M	0.4659

Appendix

Table 4: Per-channel reference error

\mathrm{RMSE}_{\mathrm{ref}}^{g}

used to normalize the ORA loss: the RMSE of the frozen backbone initialized from ERA5 at the first inference step.

Variable	RMSE (6 h)	ACC (6 h)	RMSE (240 h)	ACC (240 h)
Z500	21.71	0.9996	745.90	0.5420
T2M	0.60	0.9825	2.67	0.5937
T850	0.38	0.9943	3.49	0.4757
WS10	0.46	0.9841	2.97	0.2774
WS850	0.66	0.9892	5.10	0.2853
MSL	32.02	0.9990	686.49	0.4975

Appendix

Table 5: ERA5 reference initialization: the frozen backbone driven by ERA5 fields, per evaluation variable, at 6 h and 240 h.

Source	Var	Mean RMSE improvement (%)	95% CI	Sig.
GFS	Z500	+46.11	$[+45.50,+46.75]$	yes
	T2M	+45.70	$[+45.14,+46.24]$	yes
	T850	+41.21	$[+40.76,+41.65]$	yes
	WS10	+32.60	$[+32.30,+32.91]$	yes
	WS850	+30.54	$[+30.20,+30.88]$	yes
	MSL	+41.75	$[+41.12,+42.40]$	yes

Appendix

Table 6: Paired bootstrap confidence intervals for the mean rollout RMSE improvement over the raw-source initialization. Each initialization’s mean over the 40 six-hourly lead steps (6 h to 240 h) is formed first and the improvement is then averaged over the 326 daily

00

z initializations; the last column marks whether the 95% percentile interval (10,000 resamples) excludes zero. The pooled mean in Table 1 weights every lead step equally instead, so the two averages differ slightly. ACC behaves the same way: 13 of the 18 pairs have an interval excluding zero.

Source	Setting	SSIM	PSNR	RMSE (6 h) $\downarrow$						ORA loss
Source	Setting	SSIM	PSNR	Z500	T2M	T850	WS10	WS850	MSL	ORA loss
ORA weight ablation (GFS only)
GFS	$\lambda_{f}=8\times 10^{-3}$ (default)	0.9839	32.70	42.20	1.04	0.91	0.88	1.51	70.98	1.8491
GFS	$\lambda_{f}=0$	0.9844	32.57	45.43	1.08	0.90	0.90	1.49	81.67	1.9702
Source-embedding ablation (GFS, HRES and CMA)
GFS	Full WeatherBridge	0.9851	32.99	42.07	1.01	0.88	0.88	1.49	69.20	1.7574

Appendix

Table 7: Ablation study. Each variant is compared with the full model trained on the same sources: the ORA-weight ablation is trained on GFS only, and the source-embedding ablation is trained on GFS, HRES and CMA.

Source	Training mix	PSNR	SSIM	RMSE (6 h) $\downarrow$
Source	Training mix	PSNR	SSIM	Z500	T2M	T850	WS10	WS850	MSL
GFS	Balanced	32.99	0.9851	42.07	1.01	0.88	0.88	1.49	69.20
	Imbalanced	32.62	0.9840	42.04	1.03	0.91	0.89	1.50	71.13
	Imbalanced, w/o src emb	31.94	0.9823	43.55	1.12	0.97	0.91	1.53	79.29
HRES	Balanced	35.81	0.9904	29.39	0.81	0.70	0.72	1.19	46.80
	Imbalanced	35.71	0.9900	28.89	0.82	0.71	0.72	1.20	46.97

Appendix

Table 8: Source-imbalance stress test: balanced full-source training, imbalanced training (HRES 100%, GFS 50%, CMA 10%), and imbalanced training without the source embedding. PSNR/SSIM describe the WeatherBridge-aligned initial field against ERA5, so they do not depend on lead time and are listed once, in the upper block; the upper block then gives the RMSE of the six evaluation variables at the first step (6 h) and the lower block the same RMSE at 240 h, where forecast error has largely saturated.

Figure 6: Sensitivity to the ORA weight $\lambda_{f}$ (default $8\times 10^{-3}$ ; source: GFS). Each panel shows one metric against the weight: initial-field SSIM and PSNR, first-step Z500 RMSE and ACC, and the ORA loss; the star marks the default setting.

Eval source	Training setup	PSNR	SSIM	RMSE (6 h) $\downarrow$
Eval source	Training setup	PSNR	SSIM	Z500	T2M	T850	WS10	WS850	MSL
GFS	GFS-only	32.70	0.9839	42.20	1.04	0.91	0.88	1.51	70.98
	GFS+CMA	32.67	0.9842	42.51	1.04	0.91	0.89	1.51	71.32
	GFS+HRES	32.82	0.9844	41.35	1.03	0.89	0.88	1.49	69.60
HRES	HRES-only	35.67	0.9898	29.01	0.82	0.72	0.72	1.21	47.06
	GFS+HRES	35.72	0.9901	29.28	0.82	0.72	0.72	1.20	47.17

Appendix

Table 9: Multi-source training structure: single-source models and pair-trained adapters, each evaluated on every source.

Training samples	SSIM	PSNR	RMSE (6 h) $\downarrow$						ORA loss
Training samples	SSIM	PSNR	Z500	T2M	T850	WS10	WS850	MSL	ORA loss
25%	0.9795	31.12	44.45	1.20	1.03	0.92	1.54	83.97	2.0340
50%	0.9803	31.45	44.46	1.17	1.00	0.92	1.54	83.57	2.0160
100%	0.9839	32.70	42.20	1.04	0.91	0.88	1.51	70.98	1.8491

Appendix

Table 10: Data-scale study (default source GFS): training on 25%, 50% and 100% of the paired samples.

Model	SSIM	PSNR	RMSE (6 h) $\downarrow$						ORA loss
Model	SSIM	PSNR	Z500	T2M	T850	WS10	WS850	MSL	ORA loss
Lite	0.9839	32.70	42.20	1.04	0.91	0.88	1.51	70.98	1.8491
Mid	0.9836	32.50	42.65	1.08	0.92	0.89	1.52	71.89	1.8793
Deep	0.9831	32.33	43.25	1.11	0.94	0.90	1.52	74.35	1.8986

Appendix

Table 11: Model-scale study (default source: GFS).

Var	RMSE 6 h $\downarrow$			ACC 6 h $\uparrow$			RMSE 240 h $\downarrow$			ACC 240 h $\uparrow$
Var	raw	WeatherBridge	$\Delta$	raw	WeatherBridge	$\Delta$	raw	WeatherBridge	$\Delta$	raw	WeatherBridge	$\Delta$
Z500	226.94	56.41	75.14%	0.9621	0.9976	3.70%	1023.32	952.65	6.91%	0.0732	0.1748	138.68%
T2M	2.28	1.22	46.24%	0.7485	0.9268	23.82%	3.93	3.23	17.77%	0.3584	0.4365	21.78%
T850	6.14	0.93	84.83%	0.2846	0.9645	238.90%	4.81	4.08	15.25%	0.0783	0.1717	119.27%
WS10	1.65	0.90	45.49%	0.7880	0.9381	19.05%	3.88	3.86	0.61%	0.0479	0.0907	89.50%
WS850	2.77	1.56	43.57%	0.7948	0.9383	18.06%	6.29	5.91	6.08%	0.0322	0.0873	171.17%

Appendix

Table 12: Naive and WeatherBridge-aligned initialization on FengWu for the GFS source: RMSE and ACC at 6 h and 240 h per variable, with the raw-source value followed by the WeatherBridge-aligned value and by

\Delta

, the relative improvement of the WeatherBridge-aligned result over the raw-source result in percent, computed before rounding; “

\uparrow

” and “

\downarrow

” indicate that higher and lower values are better, respectively.

Explore similar work

FlexCast: Adaptive Weather Forecasting from Arbitrary Field Sets
Oct 4, 2026Yuang Zhang, Chen Hui, Weisi Lin +4Weather Forecasting
Nanjing University of Information Science and Technology · Nanyang Technological University · Harbin Institute of Technology +1
FuXiWeather2: Learning accurate atmospheric state estimation for operational global weather forecasting
Mar 16, 2026Xiaoze Xu, Xiuyu Sun, Songling Zhu +5Weather Forecasting Data Assimilation
Shanghai Academy of Artificial Intelligence for Science, Shanghai, & 200232, China. · Artificial Intelligence Innovation and Incubation Institute, Fudan University, Shanghai & 200433, China. · FuXi Intelligent Computing Technology Co., Ltd., Shanghai & 200233, China.
AGCD: Agent-Guided Cross-Modal Decoding for Weather Forecasting
Mar 16, 2026Jing Wu, Yang Liu, Lin Zhang +13Weather Forecasting
Sun Yat-sen University, Zhuhai, China · The Chinese University of Hong Kong, Hong Kong, China · Jiangxi Science and Technology Normal University, Nanchang, China +5

Oct 4, 2026J/K move · Enter open · S save

FlexCast: Adaptive Weather Forecasting from Arbitrary Field Sets

Yuang Zhang, Chen Hui, Weisi Lin, Haiqi Zhu+3

Nanjing University of Information Science and Technology · Nanyang Technological University · Harbin Institute of Technology+1

Weather Forecasting

Open paper AskPDF arXiv

Most deep learning weather models assign a fixed set of variables and pressure levels to predefined channels, limiting transfer across atmospheric field configurations. This dependence on a fixed field set limits the transferability of trained models across atmospheric field configurations. We propose FlexCast, a field-adaptive weather forecasting model that uses a single set of parameters to produce identity-aligned forecasts for variable-cardinality subsets drawn from a 69-field ERA5 registry. Specifically, a metadata-conditioned adapter the first encodes variable identity, pressure level, and field type and combines them with spatial features. Then, shared rank-16 projec?tions are modulated by metadata-dependent gates to produce field?specific features, while masked set fusion aggregates the available fields into a fixed-width representation. Subsequently, a multiscale U-Transformer processes the fused atmospheric features, while an identity-aware query decoder produces forecasts for the requested fields. Finally, FlexCast learns a standardized six-hour increment and applies it recursively to generate forecasts at longer lead times. Experiments on the 2020 ERA5 test set demonstrate that FlexCast operates across varying field configurations. Compatible cross-field context is associated with lower forecast errors, whereas mismatched context increases them.

Abstract

Explore similar work

FlexCast: Adaptive Weather Forecasting from Arbitrary Field Sets

FuXiWeather2: Learning accurate atmospheric state estimation for operational global weather forecasting

AGCD: Agent-Guided Cross-Modal Decoding for Weather Forecasting