Erandi Monterrubio Martínez
Erandi Monterrubio Martínez es bióloga de profesión con especialidad en Geomática, Maestra en Ciencias Ambientales y Doctora en Geociencias Aplicadas por el Instituto Potosino de Investigación Científica y Tecnológica (IPICYT). Su trayectoria se ha enfocado en el monitoreo ecológico mediante percepción remota, sistemas de información geográfica, análisis espacial, modelación ambiental e inteligencia artificial.
Su investigación doctoral se centró en el monitoreo de manglares en Veracruz mediante la integración de datos multifuente, incluyendo imágenes satelitales, drones, LiDAR, batimetría y modelos de aprendizaje automático, con el objetivo de analizar la distribución, composición y función ecohidrológica del manglar. Actualmente realiza una estancia posdoctoral SECIHTI en la División de Ciencias Ambientales del IPICYT, donde desarrolla modelos de inteligencia artificial y percepción remota para el monitoreo ecológico e identificación de zonas críticas en áreas naturales protegidas.
Sus intereses de investigación incluyen la conservación de ecosistemas, el análisis de cambios en la cobertura vegetal, la modelación espacial, el uso de sensores remotos y la generación de herramientas geoespaciales aplicadas a la toma de decisiones ambientales.
Sesión
Mangroves are coastal ecosystems of great importance given their role in coastal protection, hydrological regulation, carbon sequestration, and the maintenance of habitats for biodiversity. However, monitoring them remains a methodological challenge due to field accessibility and spatial heterogeneity, the hydrological dynamics that control their distribution, and the need to generate up-to-date, reliable, and replicable spatial information to support conservation and restoration efforts. In this context, this study proposes a GeoAI approach for the classification and monitoring of mangroves at priority sites in Veracruz, Mexico, through the integration of satellite data from optical sensors, products derived from remotely piloted aircraft systems, and artificial intelligence models.
The objective was to develop and evaluate deep learning models to determine the distribution of mangroves by classifying them at the binary and multi-class levels, distinguishing between mangrove and non-mangrove areas, as well as monospecific forest communities dominated by Rhizophora mangle, Avicennia germinans, and Laguncularia racemosa. The methodology was applied primarily at the priority site of La Mancha, Veracruz, and its performance was evaluated at Arroyo Moreno, with the aim of assessing its potential for spatial replicability in other mangrove ecosystems.
To this end, geospatial databases were constructed using Landsat and Sentinel-2 optical imagery, incorporating spectral bands in the visible, red edge, near-infrared, and short-wave infrared ranges. Additionally, reference masks generated using RPAS photogrammetry, field verification surveys, existing mangrove distribution maps, and local knowledge were utilized. This integration enabled the construction of spatially consistent training and validation datasets, suitable for representing the spectral variability of mangrove communities and reducing uncertainties associated with general coverage delineations.
Subsequently, artificial intelligence models were evaluated for mangrove classification. The models were trained using reflectance data and validated using training and test splits, as well as recent imagery. A comparison between Landsat and Sentinel-2 revealed differences in model performance and advantages associated with Sentinel-2’s spectral resolution, particularly due to the availability of red-edge bands, which are relevant for distinguishing vegetation communities with similar spectral responses.
The results show that integrating GeoAI and remote sensing improves the spatial identification of mangroves and enables a more detailed ecological classification than general land-cover delineation. Models based on deep learning and Sentinel-2 data performed best, with validation accuracies exceeding 0.94 for mangrove classification. Furthermore, spectral differences were observed between species, demonstrating the potential of the multi-class approach to characterize the composition, distribution, and condition of mangrove communities.
The main contribution of this study is a methodological proposal for ecological monitoring based on GeoAI, aimed at reducing field costs, leveraging open-access satellite data, and generating spatial products useful for decision-making. Beyond the case study, the approach can be adapted to other coastal wetlands and strategic ecosystems requiring monitoring, restoration, and assessment of changes. Thus, this study contributes to the development of reproducible geospatial methodologies, linking artificial intelligence, remote sensing, and conservation within a framework applicable to environmental and land management.