Organizations: Nanjing University of Information Science and Technology · Nanyang Technological University · Harbin Institute of Technology · Princeton University
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.
Figures & tables
Figure 1: Comparison of fixed-channel forecasting and FlexCast for arbitrary atmospheric field sets. (a) A fixed-channel model forecasts the complete 69-field state. (b) A changed input subset is incompatible with the fixed-channel model’s predefined layout. (c) FlexCast forecasts the supplied fields from valid subsets with a single model. Here, G denotes the 69-field registry, XG is the complete input, St⊆G is the available subset, N=∣St∣ , and Qt=St are the requested forecast fields.
Figure 2: Illustration of the proposed FlexCast framework. (a) Overall forecasting pipeline. Each field in the available set St is encoded using its spatial map and metadata. The adapted field features are fused and processed by a shared U-Transformer, while identity-aware queries produce forecasts for the requested set Qt=St , with N=∣St∣ . From left to right, the three addition nodes correspond to metadata conditioning, the gated low-rank feature update, and residual reconstruction using the de-standardized six-hour increment. (b) Metadata-conditioned modulation, where the shared projections D and U reduce and restore the feature width, and ⊙ denotes element-wise multiplication by the metadata gate. (c) Masked set fusion, where learned slots attend to valid field tokens and are refined by slot self-attention before projection to the spatial representation.
Model
6 h
1 day
4 days
7 days
10 days
FengWu
0.1252
0.2028
0.4860
0.6887
0.8143
Pangu-Weather
0.1182
0.1859
0.4509
0.6960
0.7582
FuXi
0.1258
0.2050
0.4742
0.6484
0.7831
FlexCast(Full-69)
0.1421
0.2269
0.4246
0.5626
0.6341
Table 1: Multi-lead forecasting results on the 2020 ERA5 test set. Values are nRMSE (lower is better) under the same evaluation protocol. The best result in each column is shown in bold, and the second best is underlined.
Condition
Anchors
nRMSE
Change
Anchor only
13
0.1943
–
+Same-variable context
9
0.1695
−10.2%
+Same-level context
9
0.1408
−21.0%
+Broad matched context
13
0.1269
−34.3%
+Broad mismatched context
13
0.2818
+124.5%
Table 2: Cross-field context comparison at 6 h on the 2020 ERA5 test set. Values are mean anchor-field nRMSE (lower is better). Relative changes are computed for each anchor and then averaged: the same-variable, same-level, and broad-matched conditions are compared with the corresponding anchor-only condition, whereas the broad-mismatched condition is compared with the broad-matched condition.
N
6 h
12 h
18 h
24 h
8
0.1627
0.2232
0.2659
0.3007
16
0.1662
0.2186
0.2539
0.2822
32
0.1570
0.2032
0.2331
0.2563
48
0.1566
0.2003
0.2278
0.2489
69
0.1421
0.1997
0.2163
0.2269
Table 3: Forecasting results for different field-set cardinalities on the 2020 ERA5 test set. N denotes the number of available fields. Values are nRMSE under same-set forecasting (lower is better). The best result is shown in bold, and the second best is underlined.
Method
6 h
1 day
4 days
7 days
10 days
FlexCast (Full-69)
0.1421
0.2269
0.4246
0.5626
0.6341
w/o Metadata Addition
0.2420
0.4906
0.9178
1.1734
1.3965
w/o Modulation
0.1972
0.3844
0.6798
0.7641
0.8072
Table 4: Ablation study of the metadata addition and modulation in FlexCast. Mean nRMSE (lower is better) is reported across different lead times. The best result in each column is shown in bold, and the second best is underlined. “w/o” denotes “without”.
Global medium-range weather forecasting requires modeling structured yet state-dependent interactions among heterogeneous atmospheric fields. Existing data-driven models largely learn these interactions implicitly, whereas equation-level physical constraints may inherit approximation and model-form biases. We present VeinCast, a physics-guided dynamic field graph and graph-conditioned fusion framework that jointly forecasts 69 surface and upper-air fields. Within each local window, its Physics-Guided Dynamic Field Graph combines predefined atmospheric relations with state-dependent Top-K residual edges and adapts Earth-window attention using the resulting graph context. Graph-Conditioned Latent Fusion further employs graph context and source-node centrality to guide field-to-latent aggregation, while bounded feedback preserves field-specific information. On the 1.5∘ ERA5 benchmark, VeinCast demonstrates competitive forecasting performance across all 69 meteorological fields at lead times of up to 14 days, compared with representative global weather forecasting models including FuXi, Pangu-Weather, GraphCast, FengWu, and ARROW. Ablations confirm that the two modules provide complementary gains, demonstrating the effectiveness of relational-level physical guidance for data-driven weather forecasting.
Zhisheng Chen, Jinhan Li, Yuxuan Li +6
1Nanyang Technological University · 2Tsinghua University · 3Peking University +1
We investigate the transferability of Earth weather foundation models to planetary atmospheres by adapting the GraphCast graph neural weather forecasting model to Mars. While GraphCast achieves state-of-the-art performance for terrestrial forecasting, its applicability to non-Earth environments remains unexplored. Using the Mars Climate Database (MCD), which provides global atmospheric fields across vertical altitude levels (similar to Earth pressure levels), we evaluate zero-shot and fine-tuned GraphCast predictions of Martian temperature and wind fields. Zero-shot forecasts produce a surprisingly accurate depiction of current conditions but fail to reproduce diurnal variability and rapidly decay toward climatological mean states. To address this limitation, we fine-tune GraphCast using MCD variables and top-of-atmosphere solar radiation forcing while holding humidity constant. Fine-tuning enables rapid learning of Martian thermal variability. Within as few as 10 training epochs, the model begins to capture the diurnal cycle and forecasts up to 10 days reproduce seasonal and vertical temperature structure. Prediction quality improves with training sample size and exhibits sensitivity to seasonal initialization. These results demonstrate that Earth-trained AI weather models can be adapted to simulate Martian atmospheric dynamics, providing a pathway toward rapid planetary weather prediction to support mission operations, dust storm risk mitigation, and future human exploration.
M. L. Carroll, J. Li, S. D. Guzewich +3
1) NASA Goddard Space Flight Center · 2) ADNet Systems, Inc.
Multi-station multivariate weather forecasting aims to forecast future weather variables at multiple weather stations from historical surface observations. Existing station forecasting models learn statistical dependencies among discrete stations, but lack explicit physical evolution. Meanwhile, PDE-based weather models provide interpretable physical dynamics, yet require continuous fields and upper-air variables unavailable in surface station data. To bridge this gap, we propose StationPDE, a station-oriented surface PDE learning model. StationPDE constructs a terrain-aware continuous surface field from discrete station observations and decomposes its physical evolution into surface wind transport and upper-air inference. Surface wind transport explicitly evolves observable weather variables, while upper-air inference uses learnable horizontal diffusion to approximate the missing influence of unavailable upper-air variables. A parallel data-driven diffusion branch captures complementary motion patterns, and an adaptive router integrates the two forecasts for station-level multivariate forecasting. Experiments on Weather2K and MeteoNet show that StationPDE consistently outperforms state-of-the-art baselines, reducing MSE by about 9.6% on average compared with the strongest baseline.