Multi-Sensor Mapping of Vulnerable Urban Settlements Using SAR, Multispectral, and Hyperspectral Imagery: A Case Study in Córdoba, Argentina

Luigi Russo; Anabella Ferral; Silvia Liberata Ullo; Paolo Gamba

cs.CVAug 26, 2026

Multi-Sensor Mapping of Vulnerable Urban Settlements Using SAR, Multispectral, and Hyperspectral Imagery: A Case Study in Córdoba, Argentina

Authors: Luigi Russo, Anabella Ferral, Silvia Liberata Ullo, Paolo Gamba

Organizations: Department of Electrical, Computer and Biomedical Engineering, University of Pavia, 27100 Pavia, Italy · Instituto Gulich, UNC-CONAE-CONICET, Córdoba, Argentina · Department of Engineering, University of Sannio, 82100 Benevento, Italy

Abstract

Informal settlements represent a major urban challenge in rapidly expanding cities, yet their identification from Earth Observation (EO) data remains difficult because of their heterogeneous appearance and incomplete official inventories. This work presents a multi-sensor deep learning (DL) framework for slum-likelihood mapping in Córdoba, Argentina, integrating high-resolution PlanetScope multispectral (MS) imagery, COSMO-SkyMed (CSK) Synthetic Aperture Radar (SAR) data, and medium-resolution PRISMA hyperspectral (HS) observations. The problem is formulated as a patch-level classification task using the official Registro Nacional de Barrios Populares (ReNaBaP) inventory as reference, and the models are evaluated through four geographically partitioned folds. SAR-only and MS-only baselines, their configurations with PRISMA HS support, and early fusion (EF), middle fusion (MF), and late fusion (LF) strategies are systematically compared. Results show that LF+HS provides the best overall balance between classification performance and spatial selectivity, while PRISMA contributes complementary spectral information alongside the higher-resolution MS and SAR representations. Beyond the standard evaluation against ReNaBaP, an external municipal vulnerability layer is used to interpret detections outside the official polygons, showing that several apparent false positives overlap broader vulnerable urban areas. Thermal analysis further shows that ReNaBaP settlements exhibit significantly higher surface temperatures than their immediate surroundings during a heatwave event, indicating localised surface-heat amplification. Taken together, these results suggest that multi-sensor EO fusion can support both the mapping of ReNaBaP settlements and the interpretation of broader urban vulnerability patterns.

Figures & tables

Fig. 1: Study area in Córdoba, Argentina, and EO data used in this work. The figure shows (a) PlanetScope RGB imagery acquired on 19 July 2025, (b) COSMO-SkyMed HH SAR backscatter acquired on 28 July 2025, (c) PRISMA false-colour composite generated from selected hyperspectral bands and acquired on 28 March 2021, and (d) the ReNaBaP polygons overlaid on PlanetScope RGB imagery. All panels refer to the same spatial extent and provide the basis for model training, prediction, and contextual evaluation.

Data source	Category	Acquisition date / version	Product / attributes	Spatial resolution / unit	Role in the workflow
COSMO-SkyMed	SAR image	28-07-2025	L1D GTC, single-pol HH backscatter	$\sim$ 3 m	High-resolution SAR input
PlanetScope	Multispectral image	19-07-2025	Surface reflectance, 8 bands	$\sim$ 3 m	High-resolution MS input
PRISMA	Hyperspectral image	28-03-2021	L2D surface reflectance, original 230-band cube	30 m	PCA-reduced spectral descriptors
Landsat-8 TIRS	Thermal image	03-01-2022	Collection 2 Level-2 Band 10 surface temperature product	100 m	Surface-temperature comparison
ReNaBaP	Reference polygons	2025	Official barrios populares inventory	Settlement polygons	Label derivation and reference-map comparison
CBA layer	Contextual polygons	2025	High-/low-vulnerability classes	Neighbourhood-scale units	Contextual interpretation

TABLE I: EO, reference and contextual datasets used in this study.

Fig. 2: Multi-sensor dataset construction workflow. Spatially aligned PS and CSK patches are extracted from the high-resolution data, while corresponding PRISMA subsets are cropped at native resolution and compressed into 16 PCA components. ReNaBaP polygons are then used for binary patch labelling, and the resulting samples are split into four geographically partitioned folds for OOF training and evaluation.

Fig. 3: PRISMA alignment and PCA-based compression. Each PS–CSK patch footprint is used to crop a native-resolution PRISMA subset, which is reduced from 230 spectral bands to 16 principal components.

Fig. 4: Patch labelling strategy based on the ReNaBaP-covered fraction within each patch. Patches with coverage $\geq$ 10% are labelled as ReNaBaP-positive patches, patches with 0% overlap are labelled as ReNaBaP-negative patches, and intermediate cases are discarded as ambiguous samples.

Fig. 5: Spatial CV folds used for the Córdoba study area. The four sectors define geographically partitioned folds used for OOF evaluation, while ReNaBaP polygons indicate the spatial distribution of reference settlements across the city.

Fold 1	Fold 2	Fold 3	Fold 4	Total
383	403	385	363	1534
24.97%	26.27%	25.10%	23.66%	100.00%

TABLE II: Spatial CV fold distribution.

Panel of Figure 8 — Fig. 6: Multi-sensor fusion architectures compared in this study. (a) EF stacks the PS and CSK inputs before a single ResNet-18 backbone. (b) MF processes the PS and CSK inputs with independent encoders and combines their feature maps before classification. (c) LF keeps the SAR and MS representations separate until feature-vector concatenation. In all “+HS” configurations, the PCA-reduced PRISMA subset is encoded as a compact spectral embedding and appended to the final feature representation before classification.

Fig. 7: Methodological and interpretation workflow following the dataset construction stage. The upper part summarises the patch-level learning framework, model-training ablation, spatial OOF inference and city-wide slum-likelihood mapping. The lower part summarises the subsequent validation and interpretation steps, including ReNaBaP-based metric computation, contextual analysis of detections outside official polygons, vulnerability-discrimination assessment and complementary thermal interpretation. Note: The step numbering starts from 7 because this workflow continues the dataset-construction sequence shown in Fig. 2 .

Metric	SAR- and MS-based configurations				Multi-sensor fusion strategies
Metric	SAR	SAR+HS	MS	MS+HS	EF	EF+HS	MF	MF+HS	LF	LF+HS
Precision	0.3590	0.2623	0.4103	0.4267	0.2426	0.2828	0.3913	0.4737	0.4000	0.5455
Recall	0.4746	0.5424	0.5424	0.5424	0.6942	0.6949	0.6102	0.3051	0.6441	0.6102
F1-score	0.4088	0.3536	0.4672	0.4776	0.3596	0.4020	0.4768	0.3711	0.4935	0.5760
AUROC	0.8198	0.8757	0.8634	0.8837	0.8676	0.8929	0.8880	0.8298	0.8826	0.8674
Accuracy	0.9472	0.9237	0.9524	0.9544	0.9048	0.9205	0.9485	0.9602	0.9492	0.9654

TABLE III: Patch-level classification performance of SAR- and MS-based configurations and multi-sensor fusion strategies under spatial OOF evaluation. PRISMA HS information is included as a PCA-reduced spectral embedding when “+HS” is indicated. The best value for each metric is highlighted in bold.

Fig. 8: City-wide slum-likelihood maps obtained with the different ResNet-18 fusion strategies, with and without PRISMA HS support. Warmer colours indicate higher slum-likelihood scores. The comparison highlights the different spatial behaviours of the EF, MF, and LF models.

Fig. 9: City-wide slum-likelihood map obtained with the ResNet-18 LF+HS configuration. ReNaBaP settlements are shown as yellow polygons. The insets show examples of detected official settlements, including (a) Barrio Villa La Tela and (b) Barrio 12 de Septiembre, and areas with high slum-likelihood scores outside the adopted ReNaBaP inventory, including (c) Barrio El Chingolo and (d) Barrio José Ignacio Díaz 2A Secc.

Fig. 10: LF+HS slum-likelihood map overlaid with the adopted ReNaBaP inventory and the Municipality of Córdoba vulnerability layer. ReNaBaP polygons are shown in blue, high-vulnerability CBA neighbourhoods in yellow and low-vulnerability neighbourhoods in green. Several areas with high slum-likelihood scores outside ReNaBaP fall within broader areas identified by the municipality as vulnerable. The CBA polygons are used only for contextual interpretation and do not constitute additional slum reference labels.

Fig. 11: Distribution of maximum LF+HS slum-likelihood values within CBA high- and low-vulnerability neighbourhoods.

Fig. 12: ROC curve obtained using the maximum LF+HS slum-likelihood value within each CBA neighbourhood to discriminate between high- and low-vulnerability classes.

Fig. 13: Thermal analysis setup for ReNaBaP settlements and their surrounding areas. Official ReNaBaP polygons are shown in purple, while yellow areas indicate the 250 m buffer zones used for local temperature comparison after excluding the ReNaBaP polygons.

Remote Sensing Image Understanding

Remote sensing image understanding uses AI to interpret satellite and aerial imagery for objects, land cover, change, hazards, and spatial relationships.

See all 208 papers

Context-Aware Slum Mapping in Sub-Saharan Africa Using Sentinel-1 Texture and Local Climate Zones
Jul 8, 2026Peterson Chepkilot, Babak Memar, Paolo GambaRemote Sensing Image Understanding Camera-Radar Fusion
Department of Civil, Building and Environmental Engineering, Sapienza University of Rome, 00184 Rome, Italy · Department of Electrical, Computer and Biomedical Engineering, University of Pavia, 27100 Pavia, Italy
Slum Detection and Density Mapping with AlphaEarth Foundations: A Representation Learning Evaluation Across 12 Global Cities
May 11, 2026Shuyang Hou, Ziqi Liu, Haoyue Jiao +7Geospatial Foundation Models Geospatial Representation Learning
State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, Wuhan, China
Heterogeneous SAR-optical fusion for near-real-time land use and land cover mapping under cloud contamination: A novel framework and global benchmark dataset
Jun 16, 2026Jiangong Xu, Weibao Xue, Xiaoyu Yu +3Land Cover Classification Remote Sensing
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China · Oriental Space Port Research Institute, Yantai, 265100, China · Hubei Luojia Laboratory, Wuhan, 430079, China +3

Jul 8, 2026J/K move · Enter open · S save

Context-Aware Slum Mapping in Sub-Saharan Africa Using Sentinel-1 Texture and Local Climate Zones

Peterson Chepkilot, Babak Memar, Paolo Gamba

Department of Civil, Building and Environmental Engineering, Sapienza University of Rome, 00184 Rome, Italy · Department of Electrical, Computer and Biomedical Engineering, University of Pavia, 27100 Pavia, Italy

Remote Sensing Image Understanding Camera-Radar Fusion Land Cover Classification Remote Sensing Synthetic Aperture Radar

Open paper AskPDF arXiv

Accurate mapping of informal settlements remains a major challenge in Sub-Saharan African (SSA) cities because optical imagery often fails to distinguish Informal Settlements (defined here as LCZ 7) from spectrally similar formal Compact Low-Rise areas (LCZ 3). This study presents a context-aware, reproducible Optical-SAR framework that improves informal settlement delineation using Sentinel-2 spectral features and Sentinel-1 structural information within an adapted Local Climate Zone (LCZ) taxonomy. We implement a three-tier SAR integration strategy: calibrated backscatter, GLCM textures, and a physics-guided feature engineered to capture the high structural disorder and weak radar return characteristic of SSA informal settlements. Using reference data across Nairobi and Eldoret (Kenya), we evaluate performance via a stratified hold-out protocol and a season-aware ablation study. Results show that SAR textures provide the dominant performance gain for LCZ 7 detection. The Optical-SAR model achieves overall accuracy of 0.816 (dry) and 0.807 (wet), significantly outperforming the WUDAPT baseline (OA 0.704) and reducing the critical LCZ 3 - LCZ 7 confusion to 7%. Seasonal analysis indicates that while optical-only separability varies with phenology, SAR-derived textures stabilize informal settlement mapping across seasons. These findings demonstrate that the incorporation of SAR-derived features yields consistent improvements for urban morphology mapping in data-scarce environments across seasons and across the evaluated source cities, while cross-city transfer remains limited without local adaptation strategies.

Abstract

Remote Sensing Image Understanding

Explore similar work

Context-Aware Slum Mapping in Sub-Saharan Africa Using Sentinel-1 Texture and Local Climate Zones

Slum Detection and Density Mapping with AlphaEarth Foundations: A Representation Learning Evaluation Across 12 Global Cities

Heterogeneous SAR-optical fusion for near-real-time land use and land cover mapping under cloud contamination: A novel framework and global benchmark dataset