Disaster damage is spatial: buildings rarely fail in isolation. Yet using spatial context for damage classification remains surprisingly underexplored, and many pipelines still rely primarily on per-building appearance cues even when the dominant uncertainty is spatially structured. Complicating matters, the right neighbourhood is not the same across events. Floods, hurricanes, and wildfires can exhibit very different clustering behaviour, making spatial reasoning valuable but easy to misuse - naive context aggregation can improve visual coherence while oversmoothing boundaries or propagating structured errors. We study this tension on xBD (the dataset used in the xView2 challenge) in a controlled post-localization, classification-only setup: each building is represented by a pre/post combined (PPC) patch cropped from the provided polygons, and spatial context is modelled with GPS-derived building graphs. Our approach keeps local evidence "close" by preserving strong spatial relationships in disaster damage patterns, while bringing only the right neighbours "closer" through a disaster-type-conditioned graph model that injects a learnable multi-scale spatial kernel prior into attention, allowing the effective neighbourhood scale to adapt across disaster types rather than being learned as a single global smoothing rule. To discourage coherence-by-smoothing, we add a residual de-correlation loss that penalizes positive Moran's~I in prediction residuals. We evaluate the method under event and dataset shift with a leave-one-event-out (LOEO) protocol on xBD and cross-dataset transfer from xBD to Ida-BD. The model improves macro-F1 and substantially reduces residual spatial autocorrelation under zero-shot event shift, indicating better use of spatial context rather than naive smoothing and enabling more reliable transfer to unseen events within known disaster types.
Figures & tables
Rollup
Tiles (Train/Test)
Moran’s I mean ± std (Train/Test)
Median Moran’s I (Train/Test)
Frac. p<0.05 (Train/Test)
Buildings/tile (Train/Test)
earthquake
121 / 38
0.168 ± 0.222 / 0.356 ± 0.439
0.013 / 0.291
0.423 / 0.750
324.4 / 366.6
flood
898 / 80
0.364 ± 0.482 / 0.195 ± 0.522
0.339 / 0.071
0.453 / 0.090
75.6 / 46.7
hurricane
1219 / 387
0.306 ± 0.601 / 0.302 ± 0.589
0.226 / 0.203
0.233 / 0.260
72.5 / 75.0
tornado
719 / 93
0.437 ± 0.317 / 0.320 ± 0.533
0.313 / 0.249
0.376 / 0.433
93.2 / 89.5
tsunami
261 / 42
0.490 ± 0.407 / 0.484 ± 0.507
0.511 / 0.568
0.585 / 0.625
210.9 / 256.0
volcano
309 / 75
0.593 ± 0.369 / 0.398 ± 0.481
0.339 / 0.340
0.273 / 0.129
23.2 / 15.0
Table 1: xBD spatial summary by disaster rollup. Moran’s I statistics are computed over valid train/test tiles.
Figure 1: Method overview. PPC patches form graphs over fixed building instances; a disaster-conditioned DCMS kernel regularizes graph-attention logits, and training combines classification/ordinal losses with residual Moran de-correlation.
Figure 2: PPC patch protocol: pre/post building-aligned crops are extracted from xBD polygons and resized for classification.
Method
Macro-F1
Per-class F1 (U, Mi, Ma, D)
xView2 official solution #1 †
0.80446
(0.92344, 0.64445, 0.78591, 0.86403)
Patch-only encoder
0.82205
(0.92880, 0.67320, 0.81520, 0.87100)
Vanilla GAT on patch encoder
0.84102
(0.93160, 0.70050, 0.84030, 0.89168)
Ours (kernel + disaster + residual)
0.87288
(0.93400, 0.76200, 0.89200, 0.90352)
Table 2: xView2 holdout external reference. Macro-F1 is the arithmetic mean of class F1 over U/Mi/Ma/D; official solution #1 is shown only as a reference.
Variant
Kernel
Disaster
Residual
LOEO
Per-class F1
Residual
Severity
prior
cond.
loss
Macro-F1
(No, Mi, Ma, D)
Moran’s I ↓
MAE ↓
Patch-only (xView2 1st-place-style Res50 backend)
×
×
×
0.433
(0.61, 0.37, 0.41, 0.34)
0.248
0.562
Vanilla GAT (kNN graph baseline)
×
×
×
0.453
(0.57, 0.40, 0.44, 0.40)
0.256
0.541
+ Kernel prior only
✓
×
×
0.473
(0.60, 0.42, 0.46, 0.41)
0.182
0.507
+ Disaster conditioning only
×
✓
×
0.460
(0.59, 0.41, 0.45, 0.39)
0.205
0.521
+ Residual de-correlation only
×
×
✓
0.450
(0.58, 0.39, 0.44, 0.39)
0.139
0.529
Table 3: LOEO ablation study. Residual Moran’s I is computed on severity residuals (lower is better).
Graph sensitivity
Input sensitivity
k
Macro-F1
Moran’s I
Setting
Macro-F1
Mi/Ma F1
8
0.498
0.081
Post-only
0.828
0.762
16
0.503
0.079
Pre+Post (PPC)
0.873
0.827
32
0.500
0.084
Table 4: Sensitivity checks. Input sensitivity uses the fixed-instance holdout protocol; k -sensitivity uses LOEO. Mi/Ma averages Minor and Major F1.
Figure 3: LOEO confusion matrices. Row-normalized percentages under the LOEO protocol.
Method
Pipeline / notes
Token
Macro-F1
Per-class F1 (No, Mi, Ma, D)
Siam-UNet [ 13 ]
D merged into Ma †
–
0.458
(0.916, 0.208, 0.251, –)
xView2 winner [ 21 ]
per-class weighted F1
–
0.268
(0.667, 0.211, 0.154, 0.041)
Patch-only (ours)
patch cls (polygons)
✓
0.275
(0.670, 0.190, 0.160, 0.080)
Vanilla GAT (ours)
patch cls + graph
✓
0.295
(0.690, 0.220, 0.180, 0.090)
Ours (Kernel + Disaster + Residual)
patch cls + graph
✓
0.335
(0.710, 0.270, 0.230, 0.130)
Table 5: Zero-shot transfer from xBD to Ida-BD. Models train on xBD and evaluate directly on Ida-BD.
Figure 4: Kernel and attention-distance diagnostics. Full rollup-specific grids and kernel parameter tables are provided in the supplementary material.
Figure 5: LOEO qualitative map on a high-Moran Hurricane-Harvey tile. Left-to-right: ground truth, patch-only, vanilla GAT, and ours.
Method
earthquake
flood
hurricane
tornado
tsunami
volcano
wildfire
Patch-only
.470/.160
.397/.313
.438/.263
.424/.368
.415/.427
.382/.359
.505/.260
Vanilla GAT
.482/.174
.416/.335
.459/.283
.442/.384
.429/.437
.401/.364
.542/.283
Ours
.548/.134
.474/.212
.517/.198
.502/.294
.488/.301
.445/.320
.547/.204
Table 6: Rollup LOEO Macro-F1 / residual Moran’s I. Higher F1 and lower Moran’s I are better.
Geospatial foundation models can support disaster damage assessment, but high accuracy on familiar disasters may hide shortcut learning: prediction from event or scene characteristics rather than evidence that transfers to a new disaster. We study embeddings from TerraMind, Prithvi-EO-2.0, DOFA, and DINOv3 on the multi-disaster datasets xView2 and BRIGHT. Holding out an entire disaster, instead of randomly splitting image tiles, reduces mean balanced accuracy from 0.807 to 0.638 on xView2 and from 0.818 to 0.568 on BRIGHT. Four sets of tests examine what information supports prediction: event and acquisition context, before/after image inputs, building count, and correct image pairing. Pre-event images remain predictive, and a classifier given only the annotated number of buildings reaches balanced accuracy of 0.629 and 0.653. Combining before/after features performs best on average on xView2, but does not improve over pre-only features on BRIGHT. Shuffling image pairs within the same disaster and damage class causes modest losses. Motivated by these findings, we test whether feature transformations learned only from training disasters improve generalization to unseen disasters. These transformations improve performance, with supervised contrastive learning performing best on xView2 and PCA on BRIGHT. Our results highlight the importance of evaluation on unseen domains for assessing model robustness and show that simple feature-level transformations can improve generalization. Code and experiments can be found at https://github.com/alishibli97/gfm-shortcut-learning.
Ali Shibli, Andrea Nascetti, Yifang Ban
Division of Geoinformatics KTH Royal Institute of Technology Stockholm, Sweden
Reliable post-disaster building damage assessment (BDA) from satellite imagery is hindered by severe class imbalance, background clutter, and domain shift across disaster types and geographies. In this work, we address these problems and explore ways to improve the MambaBDA, the BDA network of ChangeMamba architecture, one of the most successful BDA models. The approach enhances the MambaBDA with three modular components: (i) Focal Loss to mitigate class imbalance damage classification, (ii) lightweight Attention Gates to suppress irrelevant context, and (iii) a compact Alignment Module to spatially warp pre-event features toward post-event content before decoding. We experiment on multiple satellite imagery datasets, including xBD, Pakistan Flooding, Turkey Earthquake, and Ida Hurricane, and conduct in-domain and crossdataset tests. The proposed modular enhancements yield consistent improvements over the baseline model, with 0.8% to 5% performance gains in-domain, and up to 27% on unseen disasters. This indicates that the proposed enhancements are especially beneficial for the generalization capability of the system.
Alp Eren Gençoğlu, Hazım Kemal Ekenel
Department of Computer Engineering, Istanbul Technical University, Türkiye · Division of Engineering, New York University Abu Dhabi, UAE
Rapid structural damage assessment from remote sensing imagery is essential for timely disaster response. Within human-machine systems (HMS) for disaster management, automated damage detection provides decision-makers with actionable situational awareness. However, models trained on multi-disaster benchmarks often underperform in unseen geographic regions due to domain shift - a distributional mismatch between training and deployment data that undermines human trust in automated assessments. We explore a two-stage ensemble approach using supervised domain adaptation (SDA) for building damage classification across four severity classes. The pipeline adapts the xView2 first-place method to the Ida-BD dataset using SDA and systematically investigates the effect of individual augmentation components on classification performance. Comprehensive ablation experiments on the unseen Ida-BD test split demonstrate that SDA is indispensable: removing it causes damage detection to fail entirely. Our pipeline achieves the most robust performance using SDA with unsharp-enhanced RGB input, attaining a Macro-F1 of 0.5552. These results underscore the critical role of domain adaptation in building trustworthy automated damage assessment modules for HMS-integrated disaster response.
Asmae Mouradi, Shruti Kshirsagar
School of Computing, Wichita State University, Wichita, Kansas, USA