When Satellite and Street Views Disagree: Cross-View Complementarity in Disaster Damage Assessment

Yifan Yang; Lei Zou

cs.CVOct 3, 2026

When Satellite and Street Views Disagree: Cross-View Complementarity in Disaster Damage Assessment

Authors: Yifan Yang, Lei Zou

Organizations: Department of Geography, Texas A&M University, College Station, Texas, USA

Abstract

Satellite and street-view imagery see different parts of a damaged building: roofs from above, facades and debris from the ground. Average accuracy hides whether fusion methods use this complementary evidence. We study the conflict cases, the samples on which independently trained single-view models disagree, as a diagnostic subset. On paired collections from the 2025 Eaton wildfire and Hurricanes Ian and Milton, conflicts make up 9.6-33.2% of the test data. Fusion improves conflict accuracy over either view, yet an oracle that selects the correct single view per conflict remains 0.19-0.32 above the best evaluated fusion method. A linear fusion gate over building-visibility features, calibrated confidences, and disagreement recovers part of this headroom where ground photographs are aimed at the building: on Eaton it beats end-to-end fusion by +0.072 conflict accuracy p < 10^-4 and calibrated averaging by +0.051 p = 0.0001; on the panoramic hurricane datasets it is indistinguishable from end-to-end fusion. Cropping panoramas toward the building raises the conflict gain of fusion, whereas random crops of the same size do not. The spatial density of conflicts, computed without labels, correlates with tile-level wildfire damage Spearman r = 0.615, 25 tiles. Conflict cases thus show how much complementary evidence fusion uses, which view it carries under each hazard and capture setting, and where current methods fall short.

Figures & tables

Figure 1. Overview. Single-view models define the conflict cases; a linear fusion gate decides how much weight each view receives per sample (the implemented gate also mixes in the end-to-end fusion model, omitted here); conflict density is a label-free damage indicator (Eaton-fire example). Pipeline diagram: a street-view image feeds a street model and an overhead image feeds a remote model; where the two disagree, conflict cases pass to a linear fusion gate over building visibility, calibrated confidence and street-overhead disagreement, which outputs a gated prediction; the spatial density of conflicts gives a damage map.

	Eaton wildfire (conflict rate 9.6%)			Hurricane Ian (33.2%)			Hurricane Milton (25.7%)
Method	Acc.	Conflict acc.	Closure	Acc.	Conflict acc.	Closure	Acc.	Conflict acc.	Closure
street_only	0.931	0.486	—	0.703	0.529	—	0.759	0.557	—
remote_only	0.929	0.475	—	0.648	0.367	—	0.721	0.407	—
concat	0.930	0.674	0.40	0.709	0.588	0.16	0.766	0.622	0.16
crossview (end-to-end fusion)	0.940	0.699	0.45	0.735	0.624	0.26	0.776	0.649	0.23
calibrated probability averaging	0.954	0.732	0.52	0.731	0.617	0.24	0.776	0.626	0.17

Table 1. Test-set results (five-seed means; accuracy std

\leq 0.03

). Conflict accuracy is computed on each seed’s conflict subset. Closure follows Eq. ( 1 ) applied to the displayed means; the oracle reaches 0.961 / 0.896 / 0.964 conflict accuracy, i.e. 0.475 / 0.367 / 0.407 above the better single view.

Figure 2. Coefficients of the two-expert (street vs. remote) linear gate on Eaton (mean

\pm

std over five seeds). Positive values shift weight toward the street view; bars crossing zero have no stable sign across seeds. Horizontal bar chart of the eleven coefficients of the two-expert linear gate on the Eaton wildfire, with error bars over five seeds. Building centering has the largest positive coefficient; building pixel ratio and street entropy are negative; several error bars cross zero.

Dataset	Street-view variant	Street acc.	Conflict gain
Ian	original panorama	0.709	$+0.064$
Ian	building-centered crop	0.648	$\mathbf{+0.109}$
Ian	random crop (control)	0.574	$+0.042$
Milton	original panorama	0.740	$+0.067$
Milton	building-centered crop	0.693	$\mathbf{+0.149}$
Milton	random crop (control)	0.654	$+0.055$

Table 2. Field-of-view intervention (three seeds, three-epoch schedule; remote models unchanged). Conflict gain = crossview

-

best single view on each variant’s own conflict subset.

Figure 3. Conflict density as a label-free damage indicator (Eaton wildfire, $0.01^{\circ}$ grid). (a) Labeled tile damage. (b) Tile conflict density from model predictions. (c) Their rank correlation. (d) The single-view uncertainty control, which anti-correlates with damage. Point size reflects tile sample count; $p$ -values ignore spatial dependence. Four panels for the Eaton wildfire on a 0.01-degree grid: a tile map of labeled damage, a tile map of conflict density computed without labels, a scatter of density against damage with positive rank correlation, and the single-view uncertainty control with negative correlation.

Disaster Damage Assessment

Disaster damage assessment uses imagery and geospatial data to locate, classify, and quantify damage for emergency response and recovery.

See all 48 papers

DamageArbiter: A Multimodal Arbitration Framework for Disaster Damage Assessment from Street-View Imagery
Mar 16, 2026Yifan Yang, Lei Zou, Wenjing Gong +6Disaster Damage Assessment Multimodal Classification
Department of Geography, Texas A&M University, College Station, USA · Department of Landscape Architecture & Urban Planning, Texas A&M University, College Station, USA · Department of Industrial and Systems Engineering, University of Florida, Gainesville, USA +2
RAPID: A Reproducible Multi-Agent Pipeline for Interpretable Disaster Damage Assessment from Satellite and Street-View Imagery
Jun 20, 2026Yifan Yang, Wenjing Gong, Kaili Zhang +5Remote Sensing Image Understanding Disaster Damage Assessment
Texas A&M University College Station, Texas, USA · National University of Singapore Singapore, Singapore · The University of Alabama Tuscaloosa, Alabama, USA
Damage-TriageFormer: A Foundation-Model Framework for Typology-Based Building Damage Assessment from Mono-Temporal Imagery
Jun 10, 2026Yiming Xiao, Yu-Hsuan Ho, Sanjay Thasma +2Remote Sensing Image Understanding Disaster Damage Assessment
Texas A&M University · Resilitix Intelligence LLC · Institute for a Disaster Resilient Texas

Mar 16, 2026J/K move · Enter open · S save

DamageArbiter: A Multimodal Arbitration Framework for Disaster Damage Assessment from Street-View Imagery

Yifan Yang, Lei Zou, Wenjing Gong, Kani Fu+5

Department of Geography, Texas A&M University, College Station, USA · Department of Landscape Architecture & Urban Planning, Texas A&M University, College Station, USA · Department of Industrial and Systems Engineering, University of Florida, Gainesville, USA+2

Disaster Damage Assessment Multimodal Classification Multimodal Model Evaluation

Open paper AskPDF arXiv

Analyzing street-view imagery with computer vision models offers a promising approach for rapid, hyperlocal disaster damage assessment, but existing approaches typically rely on black-box pre-trained vision models, which lack interpretability and reliability. This study proposes DamageArbiter, a multimodal disagreement-driven arbitration framework designed to improve the accuracy and reliability of street-view-based damage assessment. DamageArbiter leverages the complementary strengths of unimodal and multimodal models and employs a lightweight logistic regression meta-classifier to arbitrate cases in which model predictions disagree. Using 2,556 post-disaster street-view images, paired with manually generated or large language model (LLM)-generated text descriptions, we systematically compared DamageArbiter with fine-tuned unimodal (image-only and text-only) models and CLIP-based multimodal models in terms of classification performance and overconfidence errors. Results show that DamageArbiter improved accuracy to 75.85% and the Matthews correlation coefficient (MCC) to 0.6188, compared with the best-performing text-only baseline (63.07% accuracy, 0.4126 MCC), image-only baseline (74.33% accuracy, 0.5947 MCC), and CLIP baseline (74.22% accuracy, 0.5915 MCC). The overconfidence analysis further reveals that DamageArbiter substantially reduced the overconfidence error from 70.58% for the best-performing baseline, the image-only ViT model, to 16.45%. Overall, this study demonstrates that accuracy alone is insufficient for evaluating disaster damage classification models and highlights the importance of measuring overconfidence errors as part of model reliability assessment. DamageArbiter thus offers a more reliable framework for rapid, hyperlocal disaster damage assessment from street-view imagery.

Abstract

Disaster Damage Assessment

Explore similar work

DamageArbiter: A Multimodal Arbitration Framework for Disaster Damage Assessment from Street-View Imagery

RAPID: A Reproducible Multi-Agent Pipeline for Interpretable Disaster Damage Assessment from Satellite and Street-View Imagery

Damage-TriageFormer: A Foundation-Model Framework for Typology-Based Building Damage Assessment from Mono-Temporal Imagery