Sergio Ivvan Valdez Peña

Education: Master’s and Ph.D. in Computer Science from the Centro de Investigación en Matemáticas A.C. (CIMAT). Research: optimization algorithms, numerical methods, and parallel computing, with applications to geospatial and engineering problems, including optimal design, time-series model fitting for pollutant and climatological variables, prediction of violent-event counts and land-use change. Publications: over 100 in total, of which 40 are journal articles. Graduated thesis students: 2 at the doctoral level, 11 at the master’s level, and 6 at the undergraduate level. Research projects: principal investigator on 2, collaborator on 7, and reviewer for SECIHTI-Mexico. Serves as reviewer for several international journals and conferences. Recipient of 3 best paper awards at international conferences; member of the National System of Researchers at Level 2. Teaching at the undergraduate, master’s, and doctoral levels. Currently a researcher within the Investigadoras e Investigadores por México programme of SECIHTI, stationed at CentroGeo at Querétaro, México.
https://orcid.org/0000-0002-5996-992X
https://publons.com/researcher/2104911/s-ivvan-valdez
https://scholar.google.com/citations?user=MG1jyREAAAAJ&hl
https://www.scopus.com/authid/detail.uri?authorId=57211028183


Sessions

11-11
12:45
15min
From Geographic Suitability Scores to Green Hydrogen Priority Regions: A GIS-Based Comparison of MCDA and GMM-Derived Favorability in Mexico
Juann Herrera, Sergio Ivvan Valdez Peña

Identifying suitable regions for sustainable green hydrogen projects requires integrating spatially distributed renewable resources, infrastructure accessibility, terrain conditions, territorial constraints, and expert-based decision criteria for energy planning. Although green hydrogen siting is often approached as a problem of renewable resource availability, territorial deployment also requires transparent geographic suitability representations and procedures for transforming pixel-level scores into compact regions that can support planning decisions. This is particularly relevant in national-scale assessments, where high-resource locations may not coincide with accessible, feasible, or spatially coherent development areas. In this sense, green hydrogen siting is not only an energy assessment problem, but also a GIScience problem involving spatial representation, multicriteria modeling, and spatial decision support.

This study compares alternative geographic suitability representations for green hydrogen territorial planning in Mexico. The objective is to evaluate how different GIS-based scoring approaches influence the selection of compact priority regions when the same input criteria and an approximately equal selected area are used. Three multicriteria outputs are compared: a baseline weighted multi-criteria decision analysis (MCDA) score, a production-oriented MCDA score with greater emphasis on hydrogen production benefit, and a Gaussian Mixture Model (GMM)-derived expected favorability score. The purpose is not to identify a universally best method, but to assess whether selected regions remain stable when explicit weighting is compared with model-based clustering.

GIS-based MCDA has been widely used for integrating heterogeneous spatial criteria into suitability analysis. Its main advantage is transparency, because criteria and weights can be explicitly defined, discussed, and adjusted according to planning priorities. In contrast, model-based clustering provides a probabilistic framework for identifying latent multicriteria profiles. Here, GMM-derived favorability is used as a complementary geographic representation that identifies recurrent spatial profiles among eligible pixels and translates component membership probabilities into an expected favorability score. This contrast allows the analysis to compare a normative weighting structure with a data-driven multicriteria representation.

The analysis uses a common eligible spatial domain and a shared set of normalized criteria: hydrogen production benefit, accessibility to roads, railways, and the electrical grid, as well as slope and terrain ruggedness suitability. All criteria are expressed on a normalized 0–1 scale, where higher values indicate greater suitability. The baseline MCDA score represents a balanced model combining hydrogen benefit, infrastructure access, and terrain conditions. The production-oriented MCDA score increases the weight of hydrogen benefit and removes direct terrain penalties from the score, while maintaining the same eligible spatial domain. Therefore, differences among methods reflect alternative representations of suitability rather than changes in input data or exclusions.

To move from suitability scores to planning-relevant regions, each score is processed through a region-growing algorithm. This procedure starts from high-scoring pixels and expands through neighboring cells until a target area is reached. The process is applied independently to the states of the Mexican Republic and repeated across three rounds. Region-growing approaches are useful for converting local suitability values into spatially contiguous units. The comparison uses cumulative selected area, selected hydrogen potential, hydrogen density, and state-level pairwise differences. Mean scores are interpreted within each method and are not treated as direct cross-method performance indicators, because MCDA and GMM-derived scores are produced through different modeling procedures.

The results show broadly similar national-level outcomes under an approximately equal-area land allocation, but relevant differences emerge at the state level. After three region-growing rounds, the three methods select close to 9,605 km² nationally. The production-oriented MCDA produces the highest cumulative selected hydrogen potential, reaching 23,563.84 kt/yr over 9,604.83 km². The baseline MCDA follows with 23,483.64 kt/yr over 9,604.98 km², while the GMM-derived expected favorability score selects 23,171.93 kt/yr over 9,605.18 km². The production-oriented MCDA increases selected hydrogen potential by 80.21 kt/yr, or 0.34%, relative to the baseline. The GMM-derived score selects 311.70 kt/yr less than the baseline and 391.91 kt/yr less than the production-oriented MCDA.

Although national differences are modest, the state-level comparison reveals stronger methodological sensitivity. Some Mexican states are consistently prioritized across methods, indicating robust suitability signals, while others vary depending on the geographic representation used. The GMM-derived expected favorability score increases selected potential compared with the baseline MCDA in Morelos, Chiapas, Baja California Sur, Colima, and Yucatán, while it selects less potential in Veracruz, Tamaulipas, Nayarit, Quintana Roo, and Querétaro. These differences suggest that GMM-derived favorability does not simply reproduce the weighted MCDA ranking. Instead, it highlights alternative spatial profiles that may be more balanced across criteria but not necessarily the highest in hydrogen production.

This work contributes to GIScience by framing green hydrogen siting as a problem of geographic suitability representation and spatial decision support. MCDA provides a transparent and policy-oriented framework, GMM-derived favorability identifies recurrent multicriteria profiles without imposing a single weighting structure, and region growing translates continuous scores into compact priority regions. The main contribution is not the identification of a single optimal region, but the demonstration that methodological choices in suitability representation and spatial aggregation can alter state-level priority regions even when the selected national area remains approximately constant. The proposed framework can be adapted to other infrastructure and territorial planning problems where spatial representation influences decision outcomes.

Geographic Information Systems and Cartography
Multifuncional (Remoto)
11-12
11:45
15min
Typologies of land-cover loss to urbanization in Mexican metropolises between 2005 and 2020
Alberto García Robledo, Sergio Ivvan Valdez Peña, Joyce Valdovinos, Jorge Paredes

Typologies of land-cover loss to urbanization in Mexican metropolises between 2005 and 2020

Alberto Garcia-Robledo (corresponding), Jorge Paredes-Tavares, S. Ivvan Valdez-Pena, Joyce Valdovinos-Ortega. CentroGeo, SECIHTI, Mexico.

Introduction and contribution

Urban sprawl reshapes the Mexican metropolitan territory, with consequences for biodiversity, food production, water, and climate adaptation. The land absorbed by metropolitan growth is mostly rural: shrubland, forest, cropland, wetlands, and surface water, and its composition differs across metropolises. Existing Mexican studies (González and Larralde 2019) describe expansion as national or per-class totals, but do not compare metropolises by the composition of converted cover. We ask which of the 92 metropolises in the current official delimitation (SEDATU et al. 2024) followed similar trajectories of land-cover loss to urbanization. Cities of very different scale may share a similar mix of conversion pressures, and such typologies, if stable, are candidates for targeted planning. Our contribution is threefold: the first composition-based expansion typology of all 92 Mexican metropolises; absolute hectare footprints per typology, enabling environmental risk-mapping; and identification of outlier metropolises with no analogue.

Data and methods

We use the GLAD/GLCLUC2020 v2 land-cover product (Potapov et al. 2022), at 30 m with quinquennial layers from 2000 to 2020, reclassifying its 110 thematic classes into seven categories: four natural-vegetation classes crossing ecology (upland versus wetland) with growth form (low vegetation versus trees), plus cropland, surface water, and built-up land. The raster is clipped to the 92 metropolitan polygons. For each metropolis and each of four periods (2000-2005 through 2015-2020) we compute the area of every non-built origin class converted to built-up land. The six origin shares are L1-normalized within each period, and the four blocks are concatenated into a 24-dimensional profile; built-up persistence is excluded because it dominates roughly ninety percent of mass and would swamp the conversion signal. We compute pairwise cosine similarity between profiles, keep only reciprocal k-nearest-neighbour edges at k=12 (a density-robust filter; von Luxburg 2007), and partition the weighted graph by Louvain modularity-maximization (Blondel et al. 2008). We select k through a coverage-versus-stability sweep and validate the partition across 10 random seeds using the adjusted Rand index (ARI) and normalized mutual information (NMI). Membership clustering is tested with Moran's I on each community indicator, using a 6-nearest-neighbour spatial weights matrix over metropolitan centroids and a 999-permutation test.

Results

The partition yields nine communities at modularity 0.74, seven substantive (n>=3) and two singletons, with 98 percent of metropolises in the largest component and 331 edges. It is highly stable: mean inter-seed NMI 0.96 and ARI 0.92. Across 2005-2020 the metropolises absorbed roughly 820 thousand hectares of non-built land. The central-and-industrial community (C0, 18 metropolises, about 231 thousand hectares) is dominated by upland shrubland, about three quarters of gain, with cropland secondary. The agricultural-pressure community (C1, 13) draws nearly as much from cropland as shrubland (shares about 0.42 and 0.55) and concentrates in the Bajio corridor and the Sinaloa-Sonora irrigated valleys, absorbing 39 thousand hectares of cropland. The northern-arid community (C2, 13) is near-pure shrubland (share about 0.97). The multi-class coastal community (C3, 13) has no single secondary axis. The forest-loss community (C4, 12) is the only one where forest dominates (share about 0.50), absorbing 29.5 thousand hectares of tree cover in central Veracruz and the Yucatan Peninsula. A second shrubland community (C5, 11) shows even higher shrubland dominance (about 0.87); a transitional coastal community (C6, 10) adds notable tree-cover pressure (about 0.29). Two singletons, Moroleon-Uriangato (surface-water signature) and Villahermosa (wetland signature), have no reciprocal match. Five of seven communities show significant positive spatial autocorrelation, despite the typology being built without geographic information: C4 (I=0.38), C2 (I=0.22), C1 (I=0.17), C0 (I=0.13), and C3 (I=0.10), all at the 5 percent level.

Discussion and conclusion

The communities organize along two interpretable axes: the ecology of the absorbed cover (dry-land, forested, or multi-class coastal) and the secondary contributing class within the dry-land family (cropland, near-pure shrubland, or mixed). Each maps to a distinct environmental-risk and planning profile relevant to environmental and regional spatial analysis: forest-canopy loss with biodiversity, heat-island, and carbon implications in the tropical southeast; cropland conversion as a food-security concern in the Bajio and the northwestern irrigation districts; and arid-scrub intrusion across the north. That a typology built purely from transition composition recovers five geographically clustered communities indicates that urbanization profiles partly proxy the ecoregional constraints on what land is available, and that the typology offers an empirical basis for differentiated metropolitan planning. Limitations: cosine similarity ignores trajectory phase shifts; results inherit the underlying land-cover classification; and the Moran's I tests are uncorrected for multiple comparisons, with only C2 and C4 unambiguous under a Bonferroni threshold. The two outlier metropolises merit case-specific treatment.

Disclosure of AI tool use

An AI assistant helped write the analysis code, render the figures, and draft this abstract. All data, results, and interpretation were produced and verified by the authors; no AI tool was used to fabricate results, figures, or interpretation.

References

Blondel V. et al. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics.

González Arellano S. and Larralde Corona A. (2019). La forma urbana actual de las zonas metropolitanas en México. Estudios Demográficos y Urbanos.

Potapov P. et al. (2022). The Global 2000-2020 Land Cover and Land Use Change Dataset. Frontiers in Remote Sensing.

SEDATU, CONAPO, INEGI (2024). Metrópolis de México 2020.

Von Luxburg U. (2007). A tutorial on spectral clustering. Statistics and Computing.

Geographical Information Science and Society
Multifuncional (Remoto)
11-12
12:20
15min
Modeling classification thresholds for areas burned by forest fires considering phenological variability using Sentinel-2 images in Mexico.
Briones-Herrera Carlos Ivan, Sergio Ivvan Valdez Peña

Burned-area mapping from satellite imagery depends on the ability of spectral thresholds to separate burned and unburned surfaces under varying environmental conditions. This study evaluates the phenologically corrected Relative Burn Ratio (RBRc) as the spectral index for burned area classification, considering two threshold approaches using Sentinel-2 imagery and Google Earth Engine for 294 forest fire events in Mexico. A performance-adjusted threshold (τa) and an intersection threshold (τx), derived from the crossing of the burned and combined unburned-phenological distributions, were compared. A weighted f_UBPHEN^β(xi) distribution was used to represent different contributions of unburned and phenological responses. Values of β were evaluated in increments of 0.05, and Kernel Density Estimation was used to characterize the burned-area and f_UBPHEN^β(xi) RBRc distributions. The classification performance of both thresholds was evaluated using overall accuracy and sensitivity. Both thresholds showed similar classification performance, with high values for both metrics. Increasing the phenological contribution was associated with lower overall accuracy and sensitivity, consistent with greater spectral overlap between burned and phenological responses. The results indicate that the intersection threshold τx can provide classification performance comparable to the performance-adjusted threshold τa. The proposed f_UBPHEN^β(xi) approach provides a practical way to represent phenological influence and evaluate its effect on RBRc-based burned-area classification.

Artificial Intelligence and Geographical Information Science
Sala Geomática 7
11-12
12:35
15min
A graph theory approach to the geospatial analysis of femicide in Guanajuato
Fabricio Cruz Lopez, Sergio Ivvan Valdez Peña, Daniela Moctezuma

Introduction. Traditional geospatial analysis of crime incidence is usually based on kernel density models and physical contiguity analysis to explain the displacement of violence, under the assumption of the so-called "neighborhood effect". However, this approach is insufficient to capture the nature of complex systemic phenomena where violence does not necessarily spread through proximity, but rather through nodes of functional connectivity and flows of human mobility. Recent literature on the geography of crime has begun to suggest that spatial models based solely on proximity omit the underlying dynamics that interconnect non-contiguous territories. This study proposes a comprehensive methodology based on network analysis to evaluate the geospatial dynamics of femicide in the 46 municipalities of the state of Guanajuato, Mexico. This approach allows us to transcend the static view of the territory and visualize the state as an interconnected system of dynamic risks, offering a perspective that integrates temporal synchronization variables on the spatial structure.
Methodology. The research starts from the construction of a topological network where the nodes represent the municipalities and the edges symbolize the intensity of the Hurdle temporal correlation (r) between their respective historical series of femicides. To filter out statistical noise and focus on strong relationships, a dynamic threshold of statistical synchrony (|r| > 0.4) was applied to municipalities with more than 3 feminicides. Subsequently, the Louvain community detection algorithm was implemented, a heuristic method designed to optimize network modularity and segment the territory into clusters based on synchronous criminal behaviors, regardless of their geographical location. To facilitate the exploration of this data, a dynamic geovisualization tool was developed (implemented using advanced web visualization libraries). This platform allows for real-time interaction: users can adjust correlation thresholds, visualize the weight of nodes (based on the cumulative total of cases), and examine the network topology through an intuitive interface. This functionality allows a transition from traditional static analysis to high-level visual analytics, facilitating the identification of patterns hidden from the naked eye. Results. An interesting disconnect is revealed between territorial proximity and criminal dynamics, questioning the hegemony of the spatial contagion model. Three distinct criminal ecologies were identified through network analysis: 1) The Isolation of Industrial Epicenters: This geospatial anomaly is maintained in the absolute epicenters of violence: León, Celaya and Irapuato. These municipalities, which concentrate a high volume of accumulated cases, operate under an apparent endemic temporary isolation. Topological analysis (output-input degree of the nodes) shows that these centers, despite being focal points of greater nominal incidence, maintain weak or negative correlations with their immediate periphery. This suggests that femicidal violence in these centers responds to structural, demographic and internal security factors specific to the urban core, characterized by a constant saturation that does not always fluctuate in sync with the rest of the state. 2) The "Strategic Periphery" and the Territorial Conflict: A critical finding is the identification of an extensive cluster that connects municipalities that, historically, are not part of the conventional industrial corridor, but are located on the borders of the state. This group includes Comonfort, Apaseo el Grande, Cortazar, Guanajuato and San Miguel de Allende, exhibiting robust internal connectivity. This network extends towards Tarimoro, San José Iturbide and Salvatierra. The configuration of this cluster is revealing: being located on the state peripheries, these municipalities have been, for years, the scene of an intense territorial dispute between criminal organizations (cartel war). The synchronicity in criminal behavior in this bloc is not accidental, but appears to be a byproduct of territorial control dynamics, where instability derived from high-impact conflict escalates to levels of gender violence, creating an ecosystem of shared insecurity in these border corridors. 3) Functional and Proximity Correlations: Finally, the model captures correlations that, while following contiguity logics, reinforce the validity of the network model by reflecting transport dynamics and population flow. The Salamanca-Yuriria and Acámbaro-Pénjamo axes stand out as "sensitive" or intuitive correlations. These connections demonstrate how administrative and regional mobility dynamics influence the spread of violence, allowing us to observe that the risk of femicide is not always random, but is channeled through established communication axes. Taken together, these findings demonstrate that femicide risk operates under a hybrid network logic: influenced by contiguity in functional communication axes, but dominated by a geopolitics of violence in border areas where conflict between criminal actors alters the municipal social fabric. Conclusions. The structuring of public prevention policies should not be limited to containment in geographically adjacent areas nor should it focus exclusively on the municipalities with the highest absolute volume of cases. Geospatial analysis, enhanced by analysis through graph theory algorithms, as well as interactive visualization, allows the discovery of non-contiguous corridors of violence (temporal clusters with r > 0.4). This evidence compels us to redefine intervention strategies towards an asymmetric and anticipatory model, where prevention is coordinated simultaneously in municipalities that, although distant on the map, share an identical criminal rate. This approach allows for the optimization of resources and action on the nodes of the network that function as precursors to violence in the state territory.

Geographic Information Systems and Cartography
Multifuncional (Remoto)
11-12
13:05
15min
Spatiotemporal quantification of crop cover changes using remote sensing in metropolitan areas of Mexico (2000–2020) and its relationship with agricultural production
David Leonardo Sanabria García, Sergio Ivvan Valdez Peña

Introduction: Soil is the physical and biological support for food production systems, and its transformations, loss of natural cover, expansion or contraction of the agricultural frontier directly affect food security, regional economy, and territorial environmental sustainability (Camacho-Sanabria et al., 2017; Cotler et al., n.d.).

Satellite remote sensing has consolidated itself as the most suitable tool for monitoring these processes, as it allows for the reconstruction of land cover time series at a regional scale.

Over the last two decades, Mexico's metropolitan areas have experienced accelerated land-use change processes, where urban sprawl competes for space with peri-urban crop areas, fragmenting the agricultural landscape, reducing the availability of fertile soils near consumption centers, and putting pressure on remnants of natural vegetation and wetlands (Herrera & Santos Cerquera, 2012; Nava Bernal et al., 2024; Ramírez-Pérez et al., 2025).

Understanding the magnitude, direction, and speed of these changes through Earth observation data is relevant not only from an environmental perspective but also from a productive one. Despite the relevance of the topic, remote sensing change detection studies in Mexico are mostly concentrated on deforestation, urban expansion, or ecosystem degradation. Works applying multitemporal satellite series to the specific behavior of crop cover within the country's metropolitan areas are scarce, and even fewer link that signal to the dynamics of agricultural production recorded in those same territories (False Color Composite In Remote Sensing For Urban Areas, 2025; Mgohele et al., 2024; Xu & Guo, 2015). This absence of evidence limits the ability of decision-makers to anticipate food risk scenarios associated with urban growth and leaves a central question unresolved: does the increase or decrease in cultivated area detected by remote sensing temporally coincide with changes in agricultural production volumes?

Objective: Spatiotemporal quantification of crop cover changes using remote sensing in metropolitan municipalities of Mexico (2000–2020) and their relationship with agricultural production.

Methods: Land cover classification series derived from Landsat satellite images (GLAD product, ~30 m spatial resolution) were processed for the years 2000, 2005, 2010, 2015, and 2020, structured into eight thematic classes: short and woody vegetation on firm land, short and woody vegetation in wetlands, surface water, crops, built-up area, and no data. The workflow was implemented in the R environment and delimited to a select set of municipalities in Mexico by filtering and spatially matching State-Municipality pairs with national vector cartography. After determining the actual pixel area based on raster geometry, an exact zonal extraction was performed to calculate the surface area (ha) per class and year, weighting the coverage fraction of each cell to eliminate biases at administrative boundaries. Change detection was executed through pixel-by-pixel cross-tabulation for each consecutive period and for the entire horizon (2000–2020), generating spatial transition matrices from which flows exceeding a 1% change area threshold were isolated and structured into Sankey diagrams. The analysis was consolidated into an automated repository of individual outputs per municipality, composed of tabulated databases and evolution charts.

Results: Temporal evolution charts and Sankey diagrams synthesize the spatial transition matrices of the 208 municipalities and show that territorial dynamics are not homogeneous: at the national level, 43.2% of the surface that changed class between 2000 and 2020 ended up as built-up land, transitioning from firm land with short vegetation or cultivable soil, while the reverse conversion is practically nil (0.02%), evidence of the irreversibility of the process. This pattern of anthropization coexists with opposing local dynamics: Gómez Palacio, Matamoros, and Pénjamo concentrated the largest expansions of the agricultural frontier (5,489, 4,158, and 4,045 ha; 8%–19%), gained almost entirely from native vegetation, while León, Celaya, Tlajomulco de Zúñiga, and Zapopan recorded the sharpest crop losses due to direct conversion to urban sprawl (2,772–4,725 ha).

To contrast whether the surface change translated into real production changes, the coverage change was cross-referenced with agricultural volume series available in 130 of the 208 municipalities. In 57.7% of them, area and production moved in the same direction; in the remaining 42.3%, production changed independently of the surface area: it fell despite the area being maintained, or it grew without gaining land.

Conclusions: The results suggest partial, non-generalizable temporal patterns: in the Guadalajara Metropolitan Area, while Tlajomulco and Zapopan show sustained crop loss in favor of urban sprawl, both municipalities exhibit specific five-year periods (Zapopan 2010–2015, Tonalá 2015–2020) with a net rebound in the Crop class, which does not coincide with a unique trajectory of agricultural loss within the conurbation itself. Therefore, a unique or generalizable pattern is not identified for the set of metropolitan areas in the country; behavior depends on local conditions such as the dominant crop type, urban pressure, and the current land-use policy in each region. Thus, the relationship between agricultural cover change and production requires zone-by-zone reading.

Artificial Intelligence Use Statement: In compliance with the congress guidelines, the authors declare that a generative AI tool was used as support in the writing and debugging of the code (R language) used for processing the raster series, calculating the transition matrices, and generating the visualizations for this study. The conceptual design of the analysis, the selection of data sources, the interpretation of the results, and the conclusions are the sole responsibility of the authors.

Geographic Information Systems and Cartography
Multifuncional (Remoto)