ALTIS: Automated Loss Triage and Impact Scoring from Sentinel-1 SAR for Property-Level Flood Damage Assessment

Amogh Vinaykumar; Prem Kamasani

cs.CVMar 14, 2026

ALTIS: Automated Loss Triage and Impact Scoring from Sentinel-1 SAR for Property-Level Flood Damage Assessment

Authors: Amogh Vinaykumar, Prem Kamasani

Organizations: Altis Intelligence

Abstract

Floods are among the costliest natural catastrophes globally, yet the property and casualty insurance industry's post-event response remains heavily reliant on manual field inspection: slow, expensive, and geographically constrained. Satellite Synthetic Aperture Radar (SAR) offers cloud-penetrating, all-weather imaging uniquely suited to rapid post-flood assessment, but existing research evaluates SAR flood detection against academic benchmarks such as IoU and F1-score that do not capture insurance-workflow requirements. We present ALTIS: a five-stage pipeline transforming raw Sentinel-1 GRD and SLC imagery into property-level impact scores within 24-48 hours of flood peak. Unlike prior approaches producing pixel-level maps or binary outputs, ALTIS delivers a ranked, confidence-scored triage list consumable by claims platforms, integrating (i) multi-temporal SAR change detection using dual-polarization VV/VH intensity and InSAR coherence, (ii) physics-informed depth estimation fusing flood extent with high-resolution DEMs, (iii) property-level zonal statistics from parcel footprints, (iv) depth-damage calibration against NFIP claims, and (v) confidence-scored triage ranking. We formally define Insurance-Grade Flood Triage (IGFT) and introduce the Inspection Reduction Rate (IRR) and Triage Efficiency Score (TES). Using Hurricane Harvey (2017) across Harris County, Texas, we present preliminary analysis grounded in validated sub-components suggesting ALTIS is designed to achieve an IRR of approximately 0.52 at 90% recall of high-severity claims, potentially eliminating over half of unnecessary dispatches. By blending SAR flood intelligence with the realities of claims management, ALTIS establishes a methodological baseline for translating earth observation research into measurable insurance outcomes.

Figures & tables

Figure 1: Optical vs. SAR imagery over the Addicks sector, Harvey 2017. (a) Pre-event optical: clear conditions, parcel footprints visible. (b) Co-event optical: 100% cloud cover for five consecutive days. (c) Sentinel-2 multispectral attempt: severe cloud contamination. (d) Sentinel-1 VH SAR (30 Aug 2017): unambiguous flood signal penetrating cloud cover, parcels coloured by ALTIS severity score. The cloud occlusion in (b) motivates SAR as the only operationally viable modality for 24–48 hour post-event triage.

Figure 2: ALTIS five-stage pipeline. Sentinel-1 GRD products enter Stage 1 for preprocessing; the resulting $\sigma^{0}$ rasters and InSAR coherence layers feed Stage 2 change detection; flood extent drives Stage 3 kriging depth estimation; parcel intersection and HAZUS functions produce Stage 4 severity scores; Stage 5 ranks and tiers all FNOL properties for adjuster dispatch.

Figure 3: Sentinel-1 VH false-colour SAR composite over Harris County, Texas, Hurricane Harvey, 30 August 2017. Red areas indicate backscatter decrease consistent with inundation; cyan/grey areas are non-flooded; dark navy is permanent water. Key geographic features and the Addicks/Barker Reservoir overflow zones are annotated. This composite is the primary Stage 1 output and the visual basis for Stage 2 BCR computation.

Figure 4: Feature-space separability of ALTIS signal channels, Harvey 2017. (a) VH vs. VV BCR for flooded (blue), non-flooded (grey), and HAND-excluded (red) pixels. (b) Pre-event vs. co-event coherence. (c) HAND value vs. VH BCR showing the terrain gate at 10 m. (d) ALTIS severity score vs. NFIP claim payment. Each channel provides distinct discriminative information; their combination motivates the Bayesian fusion in Equation ( 19 ).

Figure 5: Stage 2 BCR signal decomposition, Harris County TX, 30 August 2017. (a–b) Pre- and co-event VH false-colour composites with Addicks Reservoir annotation box. (c–d) BCR map and BCR filtered by HAND mask (red hatching shows HAND-excluded pixels). (e–f) Addicks sector zoom: aerial optical context and BCR+HAND zoom confirming HAND elimination of reservoir dam artefacts. (g–h) Binary flood masks from BCR+HAND and full ALTIS BCR+CCI+HAND; the coherence integration recovers flooded urban blocks missed by amplitude-only detection.

Figure 6: Stage 3 flood depth estimation results, Harris County TX, Aug. 2017. Left: ALTIS-estimated depth map (blue-red, 0–4 m) with 90% kriging confidence interval bands. Right: ALTIS depth at HCAD parcel centroids vs. FEMA 3 m depth grid ( $R^{2}=0.81$ , RMSE $=0.38$ m, $n=12{,}400$ ); systematic underestimation in deeper zones ( $d>3$ m) near reservoir spillways is attributable to SAR geometric shadowing at peak inundation. Depth field and validation statistics are measured outputs of the executed Stage 3 pipeline.

Figure 7: Projected ALTIS severity score distribution across the Harvey FNOL portfolio. Top: overlapping distributions by triage tier show a dominant low-severity cluster (Tier 3) and a high-severity tail (Tier 1) separated at the 0.40 EFL threshold; dashed lines mark tier boundaries. Bottom: cumulative distribution with dual-axis Inspection Reduction Rate, showing that the 90% recall operating point eliminates approximately 52% of dispatches.

Field	Type	Description
claim_id	str	FNOL claim identifier (passthrough)
parcel_id	str	HCAD / county parcel identifier
latitude	float	WGS-84 centroid latitude
longitude	float	WGS-84 centroid longitude
severity_score	float	EFL $s_{i}\in[0,1]$
confidence	float	$c_{i}\in[0,1]$

Table 1: ALTIS per-property output schema (CSV and GeoJSON).

Figure 8: Study area and ALTIS pipeline outputs, Harris County TX, Harvey 2017. Top: (a) geographic context with FEMA flood zone overlay and sub-regions A (Addicks/Barker), B (downtown Houston), C (Galveston Bay fringe); (b) Sentinel-1 VH false-colour composite; (c) validation data – USGS HWMs (diamonds, $n=47$ ) and NFIP claim locations (dots). Bottom: (d) ALTIS binary flood extent; (e) estimated depth field; (f) triage tier choropleth (red=Tier 1, orange=Tier 2, green=Tier 3). Sub-regions A/B/C are consistent across all panels.

Source	Product	Resolution	Role
ESA Copernicus	Sentinel-1 GRD IW	10 m	Pipeline input (SAR intensity)
ESA Copernicus	Sentinel-1 SLC IW	$\sim$ 20 m	Pipeline input (InSAR coherence)
USGS StratMap	LiDAR DEM (2017)	1 m	DEM for depth estimation
Copernicus	GLO-30 DEM	30 m	Fallback DEM; HAND derivation
FEMA	Harvey Depth Grid	3 m	Depth validation ground truth
USGS	High-Water Marks (HWM)	Point ( $n=47$ )	Independent depth validation

Table 2: Data sources for ALTIS Harvey evaluation.

Method	Precision	Recall	IoU	F1
BCR-only	0.708	0.793	0.598	0.748
BCR + CCI	0.741	0.831	0.640	0.783
ALTIS (full)	0.787	0.862	0.692	0.823

Table 3: Flood extent detection accuracy, Harris County, Harvey 2017, vs. FEMA flood inundation reference.

Validation against FEMA 3 m depth grid
Depth Range (m)	$n$	$R^{2}$	RMSE (m)	MAE (m)
All depths	12,400	0.81	0.38	0.29
0 – 0.5	3,214	0.72	0.22	0.17
0.5 – 1.5	4,887	0.83	0.31	0.24
1.5 – 3.0	2,941	0.80	0.44	0.35
$>$ 3.0	1,358	0.61	0.68	0.54

Table 4: Flood depth estimation accuracy. RMSE and MAE in metres.

∗ TES evaluated at severity score threshold $\tau^{*}=0.40$ , selected by maximising TES on the calibration partition subject to Recall $\geq 0.90$ ; here $k$ denotes dispatch count and $\tau$ denotes the continuous severity score threshold used to determine that count (see Section 7.7 ).
Method	IRR @ 90% Rec.	IRR @ 95% Rec.	TES ∗	AUIRC
BCR-only	0.38	0.24	0.28	0.51
BCR + CCI	0.45	0.31	0.35	0.58
ALTIS (full)	0.52	0.38	0.41	0.64
Triage estimates are grounded projections derived from measured E1/E2 sub-component accuracy composed with HAZUS depth-damage function characteristics and the Harvey NFIP claims portfolio distribution. Full end-to-end cross-validation is ongoing.

Table 5: Projected insurance triage performance on Harvey test partition (

n=24{,}563

properties;

n_{\mathrm{high}}=9{,}821

;

\theta_{\mathrm{damage}}=\

5{,}000$ ). Estimates derived from measured E1/E2 sub-component accuracy

Figure 9: Qualitative ALTIS pipeline results, Harvey 2017. Left: per-property pipeline progression for four representative FNOL properties spanning the full severity range. Each row shows (left to right) pre-event VH SAR, co-event VH SAR, BCR signal, HAND-filtered flood mask, kriging depth estimate, and final triage score. The bottom row shows a BCR-only false positive corrected by coherence and HAND constraints. Right: severity score choropleth across all FNOL-reporting Harris County parcels, with highest-severity assignments concentrated around Addicks/Barker and the Buffalo Bayou corridor.

$\theta_{\mathrm{damage}}$	$n_{\mathrm{high}}$	IRR @ 90% Rec.	TES	AUIRC
$5,000	9,821	0.52	0.41	0.64
$10,000	7,304	0.49	0.38	0.61
$20,000	5,017	0.46	0.35	0.59
$50,000	2,389	0.37	0.27	0.53

Table 6: Triage performance sensitivity to high-severity damage threshold.

Configuration	BCR	CCI	HAND	IoU / F1	IRR / AUIRC
BCR only	✓			0.598 / 0.748	0.38 / 0.51
BCR + HAND	✓		✓	0.631 / 0.774	0.43 / 0.56
BCR + CCI	✓	✓		0.640 / 0.783	0.45 / 0.58
BCR + CCI + HAND	✓	✓	✓	0.692 / 0.823	0.52 / 0.64

Table 7: Ablation study: per-signal contribution to flood extent (IoU, F1) and projected triage performance (IRR @ 90% recall, AUIRC).

Disaster Damage Assessment

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

See all 49 papers

Beyond Backscatter: AlphaEarth Land-Cover Priors for Rapid SAR Flood Segmentation Across Foundation Backbones
Jun 28, 2026Sanjay Thasma, Yu-Hsuan Ho, Ali MostafaviRemote Sensing Image Segmentation Geospatial Foundation Models
Department of Computer Science and Engineering, Texas A&M University, College Station, TX · Urban Resilience.AI Lab, Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX
High-Resolution Flood Mapping With Sentinel-1 and Sentinel-2 via Misalignment-Robust Cross-Sensor Learning and Generative Despeckling
Jun 29, 2026David Ma, Jeremy Feinstein, Shreya Pandit +2Remote Sensing Image Segmentation Flood Inundation Mapping
Argonne National Laboratory, Lemont, IL 60439 USA · Ohlone College, Fremont, CA 94539 USA
Explainable Flood Segmentation on Sentinel-1 SAR1 Imagery Using CNN and Transformer Architectures
Jun 15, 2026Arundhuti Banerjee, David DaouRemote Sensing Image Segmentation Explainable Artificial Intelligence
United Nations University’s Institute for Environment and Human Security (UNU-EHS) · Bonn, Germany 53113 · United Nations University’s InstituteBonn,forGermanyEnvironment53113and Human Security (UNU-EHS)

Jun 28, 2026J/K move · Enter open · S save

Beyond Backscatter: AlphaEarth Land-Cover Priors for Rapid SAR Flood Segmentation Across Foundation Backbones

Sanjay Thasma, Yu-Hsuan Ho, Ali Mostafavi

Department of Computer Science and Engineering, Texas A&M University, College Station, TX · Urban Resilience.AI Lab, Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX

Remote Sensing Image Segmentation Geospatial Foundation Models Flood Inundation Mapping Remote Sensing Synthetic Aperture Radar

Open paper AskPDF arXiv

Rapid flood mapping is critical for emergency response, yet optical imagery is often unusable during major flooding and single-temporal SAR is ambiguous, since new inundation, permanent water, and other smooth surfaces produce similar backscatter. This study evaluates whether stable land-context priors can improve post-event SAR flood segmentation when a registered, seasonally matched pre-event acquisition is unavailable. Using the CONUS (Continental United States) subset of ImpactMesh-Flood, we compare four backbones spanning distinct pretraining regimes-a from-scratch CNN UNet, an ImageNet-pretrained UNet, the SAR-pretrained TerraMind Vision Transformer, and the optical-satellite-pretrained DINOv3 Vision Transformer-in SAR-only, SAR+DEM, and SAR+AlphaEarth configurations under an identical fusion design, training protocol, and event-stratified split. Models are selected on a validation flood event and evaluated separately on two held-out events, Hurricane Florence and the Louisiana floods, with three-seed reporting for auxiliary configurations. Both auxiliary priors improve over the observed SAR-only baselines across all backbones and test events. AlphaEarth exceeds DEM on the harder Florence event for every backbone and achieves the best Florence IoU, while DEM is competitive on Louisiana and produces the best result there. The seed analysis reveals a trade-off: DEM is more stable across initializations, whereas AlphaEarth offers higher peak performance and higher recall on the harder event. Cross-event differences track flood-class prevalence and similarity to the training distribution, underscoring the need for per-event evaluation. We reframe single-temporal SAR flood segmentation as an alignment between radar observations and stable land-surface priors, where learned and physical context offer complementary pathways to more reliable rapid flood mapping.

Abstract

Disaster Damage Assessment

Explore similar work

Beyond Backscatter: AlphaEarth Land-Cover Priors for Rapid SAR Flood Segmentation Across Foundation Backbones

High-Resolution Flood Mapping With Sentinel-1 and Sentinel-2 via Misalignment-Robust Cross-Sensor Learning and Generative Despeckling

Explainable Flood Segmentation on Sentinel-1 SAR1 Imagery Using CNN and Transformer Architectures