arxiv: 2604.26133 · v1 · submitted 2026-04-28 · 💻 cs.LG

Recognition: unknown

Spatially-constrained clustering of geospatial features for heat vulnerability assessment of favelas in Rio de Janeiro

Baptiste Clemence , Thomas Hallopeau , Vanderlei Pascoal De Matos , Laurent Demagistri , Joris Guerin

Authors on Pith no claims yet

Pith reviewed 2026-05-07 16:06 UTC · model grok-4.3

classification 💻 cs.LG

keywords favelasheat vulnerabilityspatially-constrained clusteringland surface temperatureRio de Janeiroinformal settlementsremote sensinggeospatial analysis

0 comments

The pith

Flat-terrain favelas in Rio experience 2-3°C higher temperatures than slope settlements during extreme heat events.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper creates a framework that groups favelas using their physical layout and location traits, then checks how those groups relate to measured surface temperatures. It separates recent flat-ground settlements that are easy to reach from older slope communities that have more vegetation and fewer connections. Across 16 heat waves the flat group runs 2 to 3 degrees warmer, tying visible settlement shape to greater heat exposure. This connection matters because informal settlements already carry high climate-health risks, and the method offers a repeatable way to spot which areas need cooling measures first.

Core claim

Applying spatially-constrained clustering to remote-sensing features of Rio favelas separates two settlement types. Cluster 0 contains recent, well-connected favelas on flat terrain. Cluster 1 contains historical, poorly-connected favelas on vegetated slopes. Temperature records from 16 extreme heat events show Cluster 0 favelas reach land-surface temperatures 2 to 3°C higher than Cluster 1, demonstrating that settlement morphology shapes heat vulnerability.

What carries the argument

Spatially-constrained clustering of geospatial features (terrain slope, vegetation, connectivity, settlement age) combined with land-surface temperature comparison across the resulting clusters.

Load-bearing premise

The observed temperature gap is caused by the morphological traits captured in the clusters rather than by other unmeasured factors such as building materials, density, or infrastructure quality.

What would settle it

Ground-level air temperature or health data collected inside the two cluster types during a heat wave that shows no significant difference would falsify the claim that the clusters reflect meaningful differences in heat exposure.

Figures

Figures reproduced from arXiv: 2604.26133 by Baptiste Clemence, Joris Guerin, Laurent Demagistri, Thomas Hallopeau, Vanderlei Pascoal De Matos.

**Figure 1.** Figure 1: Spatial distribution of favela clusters identified by COP-KMeans view at source ↗

**Figure 2.** Figure 2: Distribution of explanatory variables by cluster before normalization view at source ↗

**Figure 3.** Figure 3: Comparison of LST distribution between clusters during extreme heat events. view at source ↗

read the original abstract

Informal settlements face disproportionate exposure to climate-related health hazards. However, existing methodologies lack systematic approaches to link diverse settlement characteristics with environmental health outcomes. We develop a data-driven framework to assess heat vulnerability in Rio de Janeiro's favelas by combining spatially-constrained clustering with land surface temperature (LST) analysis. Using remote sensing and geospatial features, we identify two distinct favela typologies: recent, well-connected settlements on flat terrain (Cluster 0) and historical, poorly-connected communities on vegetated slopes (Cluster 1). Analysis of 16 extreme heat events reveals systematic temperature differences of 2--3$^\circ$C between clusters, with flat-terrain favelas experiencing significantly higher heat exposure. Our findings demonstrate that settlement morphology critically influences heat vulnerability, providing a replicable framework for targeted urban planning and public health interventions in informal settlements globally.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper finds 2-3°C LST differences between two favela clusters in Rio but the gap may trace back to terrain and vegetation already used to define the clusters rather than independent morphology effects.

read the letter

The core result is that flat-terrain favelas run 2-3°C hotter than slope ones during 16 heat events, identified via spatially constrained clustering on geospatial features. This gives a concrete, data-driven split between recent well-connected settlements and older poorly-connected ones on vegetated slopes, then ties the groups to satellite LST measurements. The framework is replicable for other informal settlements and pulls remote sensing into a practical vulnerability mapping exercise, which is a solid applied step even if the algorithm itself is standard.

Referee Report

2 major / 2 minor

Summary. The paper develops a data-driven framework that applies spatially-constrained clustering to geospatial features (including terrain, vegetation, and connectivity) derived from remote sensing to identify two favela typologies in Rio de Janeiro. Cluster 0 comprises recent, well-connected settlements on flat terrain; Cluster 1 comprises historical, poorly-connected communities on vegetated slopes. Analysis of land surface temperature (LST) during 16 extreme heat events reports systematic 2–3 °C differences, with flat-terrain favelas showing higher exposure. The central claim is that settlement morphology critically influences heat vulnerability and that the framework is replicable for targeted interventions.

Significance. If the confounding concern is resolved, the work supplies a concrete, replicable pipeline that links clustering of open geospatial data with multi-event LST observations, yielding quantifiable temperature differentials. This is valuable for urban climate adaptation in data-scarce informal settlements. The use of 16 independent heat events and explicit cluster descriptions strengthens falsifiability and practical utility for public-health planning.

major comments (2)

[Methods / Results] Methods (geospatial feature selection and clustering): The input features explicitly encode terrain flatness, slope, vegetation (NDVI), and connectivity. The subsequent LST comparison (Results) attributes the 2–3 °C gap to “settlement morphology” without any regression, propensity-score matching, or variance decomposition that isolates morphology after controlling for the topographic and vegetative variables already used to define the clusters. This directly undermines the causal interpretation in the abstract and conclusion.
[Results] Results (temperature analysis): The manuscript reports “significantly higher” exposure for Cluster 0 but provides no detail on the statistical test, degrees of freedom, or adjustment for spatial autocorrelation and multiple-event dependence. Without these, the 2–3 °C difference cannot be confidently separated from the built-in terrain/vegetation effects.

minor comments (2)

[Abstract / Methods] Abstract and §3: The choice of k=2 is presented without justification, elbow-plot, or silhouette analysis; a brief sensitivity check to k=3 would strengthen robustness.
[Figures / Methods] Figure captions and text: LST values are given in °C but the satellite product (e.g., Landsat or MODIS) and retrieval algorithm are not named, hindering reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help clarify the scope and limitations of our analysis. We address the two major comments point by point below, proposing targeted revisions to improve statistical transparency and interpretive precision while preserving the core contribution of the typology-based framework.

read point-by-point responses

Referee: [Methods / Results] Methods (geospatial feature selection and clustering): The input features explicitly encode terrain flatness, slope, vegetation (NDVI), and connectivity. The subsequent LST comparison (Results) attributes the 2–3 °C gap to “settlement morphology” without any regression, propensity-score matching, or variance decomposition that isolates morphology after controlling for the topographic and vegetative variables already used to define the clusters. This directly undermines the causal interpretation in the abstract and conclusion.

Authors: We agree that the clusters are defined by the multivariate combination of terrain, vegetation, connectivity and related features, so the observed LST differences are inherently associated with these defining characteristics rather than an isolated effect of morphology. Our intent is to show that the resulting typologies exhibit systematic heat-exposure differentials relevant to vulnerability assessment, not to claim a causal mechanism independent of the input variables. In revision we will (i) replace causal phrasing such as “critically influences” in the abstract and conclusion with language of association with the identified typologies, (ii) add an explicit limitations paragraph discussing the confounding structure, and (iii) include a supplementary regression of LST on the original features plus cluster membership to illustrate the incremental contribution of the typology label. These changes will be marked as partial because a full propensity-score or variance-decomposition analysis would require additional methodological development beyond the scope of the current study. revision: partial
Referee: [Results] Results (temperature analysis): The manuscript reports “significantly higher” exposure for Cluster 0 but provides no detail on the statistical test, degrees of freedom, or adjustment for spatial autocorrelation and multiple-event dependence. Without these, the 2–3 °C difference cannot be confidently separated from the built-in terrain/vegetation effects.

Authors: We acknowledge the omission of methodological detail. The current manuscript will be revised to report the exact statistical procedure used for the cluster-wise LST comparison (including test name, test statistic, degrees of freedom, p-value, and any correction applied), together with the approach taken for spatial autocorrelation (e.g., effective sample size or spatial covariance adjustment) and for the 16 heat events (e.g., treating events as independent replicates or using a mixed-effects model). These additions will be placed in the Results section and the Methods subsection on temperature analysis, allowing readers to evaluate the robustness of the 2–3 °C differential relative to the terrain and vegetation covariates. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation proceeds by clustering on independent geospatial features (terrain, slope, vegetation, connectivity) then measuring LST separately from satellite imagery across 16 heat events. No equations, definitions, or self-citations make the observed 2–3 °C differences reduce to the clustering inputs by construction; the temperature data remain external to the feature set used for typology formation. The central claim that morphology influences vulnerability rests on this empirical separation rather than tautological fitting or imported uniqueness results.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions of clustering validity and the causal relevance of the chosen geospatial features; no new entities are postulated and only one obvious free parameter (cluster count) is implied.

free parameters (1)

number of clusters
Set to two to produce the reported distinct typologies; value chosen to match observed data structure rather than derived from first principles.

axioms (1)

domain assumption Spatially-constrained clustering on the selected geospatial features yields groups that meaningfully reflect settlement morphology relevant to heat exposure
Invoked when interpreting the two clusters as typologies that influence vulnerability; no independent validation of this mapping is described.

pith-pipeline@v0.9.0 · 5468 in / 1281 out tokens · 74609 ms · 2026-05-07T16:06:27.579296+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

12 extracted references · 10 canonical work pages

[1]

World urbanization prospects: The 2025 revision,

[Online]. Available: https: //unstats.un.org/sdgs/report/2024/The-Sustainable-Development-Goals-Report-2024.pdf ——, “World urbanization prospects: The 2025 revision,”

2024
[2]

Slums from space: 15 years of slum mapping using remote sensing,

[Online]. Available: https: //www.un.org/development/desa/pd/content/world-urbanization-prospects-2025-summary-results M. Kuffer, K. Pfeffer, and R. Sliuzas, “Slums from space: 15 years of slum mapping using remote sensing,” Remote Sensing, vol. 8, no. 6, p. 455,

2025
[3]

Informal settlements and human health,

[Online]. Available: http://dx.doi.org/10.3390/rs8060455 J. Corburn and A. Sverdlik, “Informal settlements and human health,” inIntegrating Human Health Into Urban and Transport Planning: A Framework, M. Nieuwenhuijsen and H. Khreis, Eds. Cham: Springer International Publishing, 2019, pp. 155–171. [Online]. Available: http: //dx.doi.org/10.1007/978-3-319-...

work page doi:10.3390/rs8060455 2019
[4]

Building resilience in urban slums: Exploring urban poverty and policy responses amid crises,

[Online]. Available: https://www.academia.edu/34378816/LE_CLIMAT_URBAIN_ET_LA_SANT%C3%89_LES_ CHANGEMENTS_CLIMATIQUES_ET_LA_DENGUE_DANS_LES_VILLES_BR%C3%89SILIENNES Z. Kaiser, A. Sakil, R. Baikady, A. Deb, and M. Hossain, “Building resilience in urban slums: Exploring urban poverty and policy responses amid crises,”Discover Global Society, vol. 3, no. 1, p. 8,

work page arXiv
[5]

The urban heat island in rio de janeiro, brazil, in the last 30 years using remote sensing data,

[Online]. Available: http://dx.doi.org/10.1007/s44282-025-00142-3 L. de Faria Peres, A. J. de Lucena, O. C. Rotunno Filho, and J. R. de Almeida França, “The urban heat island in rio de janeiro, brazil, in the last 30 years using remote sensing data,”International Journal of Applied Earth Observation and Geoinformation, vol. 64, pp. 104–116,

work page doi:10.1007/s44282-025-00142-3
[6]

Mapping slums with medium resolution satellite imagery: A comparative analysis of multi-spectral data and grey-level co-occurrence matrix techniques,

[Online]. Available: https://doi.org/10.1016/j.jag.2017.08.012 A. C. De Mattos, G. McArdle, and M. Bertolotto, “Mapping slums with medium resolution satellite imagery: A comparative analysis of multi-spectral data and grey-level co-occurrence matrix techniques,”arXiv,

work page doi:10.1016/j.jag.2017.08.012 2017
[7]

Image based characterization of formal and informal neighborhoods in an urban landscape,

[Online]. Available: https://arxiv.org/pdf/2106.11395 J. Graesser, A. Cheriyadat, R. R. Vatsavai, V . Chandola, J. Long, and E. Bright, “Image based characterization of formal and informal neighborhoods in an urban landscape,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 5, no. 4, pp. 1164–1176,

work page arXiv
[8]

Addressing data imbalance in urban informal settlement mapping from earth observation using ensemble learning: A case study in rio de janeiro,

[Online]. Available: https://ieeexplore.ieee.org/document/6236225 T. Hallopeau, Y . Fouzai, L. Demagistri, J. Guérin, V . P. de Matos, R. Gracie, and N. Dessay, “Addressing data imbalance in urban informal settlement mapping from earth observation using ensemble learning: A case study in rio de janeiro,”Science of Remote Sensing, p. 100273,

work page arXiv
[9]

Constrained k-means clustering with background knowledge,

[Online]. Available: http://dx.doi.org/10.1016/j.apgeog.2012.11.016 K. Wagstaff, C. Cardie, S. Rogers, and S. Schrödl, “Constrained k-means clustering with background knowledge,” inProceedings of the 18th International Conference on Machine Learning (ICML-2001), vol. 1, June 2001, pp. 577–584. [Online]. Available: https://web.cse.msu.edu/~cse802/notes/Con...

work page doi:10.1016/j.apgeog.2012.11.016 2012
[10]

On the influence of density and morphology on the urban heat island intensity,

[Online]. Available: https://www.sciencedirect.com/science/article/pii/0377042787901257 Y . Li, S. Schubert, J. P. Kropp, and D. Rybski, “On the influence of density and morphology on the urban heat island intensity,”Nature Communications, vol. 11, p. 2647,

work page arXiv
[11]

Painel unificador covid-19 nas favelas: metodologia para dar visibilidade a territórios periféricos,

[Online]. Available: https://www.mdpi.com/2073-4433/8/2/18 5 R. Gracie, A. S. d. A. Silva, C. Bigler, G. Douglass-Jaimes, E. M. Campos, and T. Williamson, “Painel unificador covid-19 nas favelas: metodologia para dar visibilidade a territórios periféricos,” inCovid-19 no Brasil: cenários epidemiológicos e vigilância em saúde - Série Informação para a ação...

work page doi:10.7476/9786557081211.0016 2073
[12]

[Online]. Available: https://doi.org/10.1371/journal.pone.0295766 A OPTIMAL NUMBER OF CLUSTERS Figure A.1: Elbow method and Silhouette score The selection of the number of clusters is performed with elbow method and silhouette analysis Figure A.1. The decrease in inertia in the elbow method suggests 2 to 3 clusters. The absence of a local maximum on the s...

work page doi:10.1371/journal.pone.0295766