OmniST: Topology-Invariant Spatiotemporal Forecasting with Guaranteed-Coverage Uncertainty at Unseen Sensor Locations
Authors: Praveg Vashishtha
Organizations: Department of Computer Science and Engineering, Indian Institute of Technology Patna, India · Sopra Banking Software India as Senior Software Development Engineer
Standard spatiotemporal graph convolutional networks need a fixed-size adjacency matrix, coupling model weights to a specific sensor count N; they cannot produce any output at a sensor absent from training. We present OmniST, a topology-invariant forecaster whose computation has zero structural dependency on the number of nodes, and which, composed with inductive split-conformal calibration, gives distribution-free, coverage-guaranteed prediction intervals at arbitrary (including never-seen) sensor locations, preserving that guarantee as sensors are added or removed. Topology invariance follows from coordinate-based tokenization, k-nearest-neighbor attention from geographic coordinates at each forward pass, and per-node temporal processing via a bidirectional GRU. On real Delhi air-quality data (37 stations, 2019-2023), OmniST attains a 30-day PM2.5 MAE of 31.0 ug/m3 and degrades at most 1.1 percent when 40 percent of stations are removed. Its inductive forecasts at unseen stations beat nearest-neighbor interpolation (about 12 percent) and a learned kriging GNN (IGNNK, about 21 percent), and, unlike either, carry a finite-sample coverage guarantee. The guarantee holds at 90 percent or above under 30-50 percent node dropout across four real datasets over three domains (Delhi and Beijing air quality, NOAA GHCN weather, Caltrans PeMSD7-M traffic), and still holds at 325 real sensors (full PeMS-BAY). We also fix a temporal-drift failure: static conformal under-covers over time (about 0.50 late), but a rolling adaptive-conformal estimator restores target coverage (0.89 overall, 0.96 late) at the cost of wider intervals.
Figures & tables
Month
OmniST
E-STGCN ∗
Note
Jan
57.3
56.1
Comparable
Mar
24.2
25.8
Comparable
Apr
22.0
25.4
OmniST better
May
30.9
29.0
Comparable
Jun
18.1
22.0
OmniST better
Jul
11.3
15.7
OmniST better
TABLE I: 30-day PM 2.5 MAE ( μ g/m 3 ) on Real Delhi CPCB Data (2023)
Active stations
MAE ( μ g/m 3 )
Degradation
37 (full network)
35.33
—
32 (5 failures)
35.72
+1.1%
27 (10 failures)
35.72
+1.1%
22 (15 failures)
35.58
+0.7%
TABLE II: Predictions Under Simulated Sensor Failure (Real CPCB Data)
Method
MAE ( μ g/m 3 )
Nearest-neighbor (copy closest training station)
30.3
IGNNK (learned inductive kriging [ 9 ] )
34.0 ± 1.3
OmniST + node-dropout training augmentation
27.1 ± 1.1
OmniST (coordinate-based)
26.7 ± 1.2
TABLE III: Inductive Point Prediction at Unseen Stations (Real CPCB, 5 held-out, 3 seeds)
Setting
Dataset
OmniST
E-STGCN
Transductive MAE (seen)
Delhi AQ
28.6 ± 1.2
33.5 ± 1.9
Weather
5.55 ± 0.14
5.30 ± 0.29
Inductive (unseen nodes)
Delhi AQ
MAE 26.7, cov 0.91
cannot predict
Weather
MAE 5.0, cov 0.92
cannot predict
Transductive = train and test on the same stations. On Delhi OmniST is better (paired p=0.03 ); on weather E-STGCN is marginally better ( p=0.14 , n.s.) — reported both ways. Inductive = predict at held-out stations: the fixed N×N training adjacency of E-STGCN ( 32×32 / 27×27 ) has no row/column for an unseen station, so it structurally produces no output; OmniST does, with a coverage guarantee.
Coverage under training-node dropout (OmniST + conformal):
30% stations dropped
0.914
—
Yes
50% stations dropped
0.913
—
Yes
This is the original 3-seed single-domain run. The 10-seed re-run (Table VI ) gives a consistent Delhi coverage of 0.914 (width 127.0); the small differences are seed/run variation, not a discrepancy.
TABLE VI: Inductive Coverage at Unseen Nodes Across Four Real Datasets / Three Domain Types (target 90%)
Fig. 1: Inductive coverage at unseen nodes across three physical domain types (four real datasets), at full network and under 30%/50% node dropout. OmniST holds the 90% target throughout.
Dataset
OmniST
NN
IDW
mean- k
GP
Delhi AQ
0.914
0.925
0.871
0.813
0.845
Beijing AQ
0.928
0.869
0.818
0.789
0.789
Weather
0.919
0.896
0.927
0.945
0.893
Traffic
0.911
0.887
0.905
0.906
0.922
All methods use identical split-conformal calibration. OmniST is the only one
≥0.90 on every dataset; interpolators under-cover (tighter but invalid) on several.
TABLE VII: OmniST vs Model-Free Interpolation Baselines (coverage @ target 90%, 4 real datasets)
Fig. 2: Calibration curves (empirical vs target coverage) across the four real datasets; all track the diagonal, i.e. the method hits whatever target it is set.
TABLE IX: Coverage Over Time Under Seasonal Drift (Real Delhi, target 90%, 3 seeds)
Fig. 3: Coverage over time under seasonal drift. Static conformal collapses in the late-drift window (0.50); the rolling-ACI online layer restores the 90% target (0.96 late-window).
Fig. 4: Conditional coverage: Mondrian (group-conditional) conformal lifts the worst-group coverage on the irregular traffic network from 0.80 toward target, leaving the already-tight air-quality domains unchanged.
Department of Computer Science and Engineering, Indian Institute of Technology Jodhpur, Jodhpur, Rajasthan, India · Department of Mathematics, Indian Institute of Technology Jodhpur, Jodhpur, Rajasthan, India