Este artículo analiza cómo los datos textuales de redes sociales pueden transformarse en información geográfica para comprender fenómenos sociales en el espacio. Se propone una metodología que integra Procesamiento de Lenguaje Natural (PLN), en particular el Reconocimiento de Entidades Nombradas (REN) y la detección automática de emociones (AE), con herramientas de análisis espacial y cartografía digital. El caso de estudio se centra en Colombia, empleando una base de datos con aproximadamente 27 millones de tuits. A partir de ellos, se identifican entidades de localización y emociones expresadas en los mensajes, para ello se desarrollan dos corpus etiquetados con datos de localizaciones y emociones, junto con modelos especializados y adaptados al contexto del español de Colombia , y se aplican métodos de estadística espacial (Moran’s I, densidades de Kernel) para reconocer patrones espaciales de distribución emocional. Como resultado se obtiene una base de datos geolocalizada de 3.8 millones de tuits con referencias al espacio, en donde se observa una correlación espacial moderada (>0.01) por cada una de las emociones (ira, tristeza, felicidad, disgusto, miedo, sorpresa) en Colombia. Adicionalmente se presenta una mejora en la exactitud de los modelos REN (44.60% → 97.26%) y AE (41.82% → 72.66%). La investigación aporta evidencia empírica al campo de la geografía de las emociones, mostrando el potencial de integrar PLN y geo tecnologías en estudios socioespaciales. Asimismo, se ofrecen aportes metodológicos replicables en otros contextos, donde la convergencia de datos no estructurados y análisis espacial constituye una oportunidad para nuevas aproximaciones interdisciplinarias en la geografía contemporánea.

Modern geographic information systems and artificial intelligence generate powerful analytical products for risk assessment especially in agriculture and farming area. However, their practical value remains limited without effective communication to end-users. This research focuses on a critical yet often overlooked aspect: the art of visual risk communication for non-expert audiences, specifically smallholder farmers in regions like the Congo Basin.

The goal is to develop and validate intuitive cartographic products that transform complex risk data—such as flood hazards, soil degradation, or drought—into clear and actionable visual messages. A key methodology is the iterative usability testing of map designs with the direct involvement of farmers. The study tests and compares various visualization approaches:

Color Schemes: The intuitiveness of standard schemes (e.g., "red = danger" vs. "blue = danger" for water-related threats) within the local cultural context.

Legend Design: The effectiveness of symbolic legends (pictograms) compared to more traditional interval-based legends, as well as the simplicity of language and terminology.

Level of Detail: The optimal balance between information richness and visual clutter, enabling the user to make quick decisions.

The project's outcome will be a set of empirically validated design principles for creating risk maps. These principles will ensure that critical information derived from AI and remote sensing technologies is correctly understood by farmers, ultimately empowering them to make better-informed decisions to protect their livelihoods and enhance their resilience to climate risks.

The Monterrey Metropolitan Area (MMA), characterized by complex lithology, rugged topography, intense rainfall, and increasing anthropogenic pressures, faces increasing landslide hazards. This study applies a quantitative approach using the weight of evidence (WoE) method to assess landslide susceptibility across the MMA. A total of 292 historical landslide events were mapped using aerial imagery and archival data, with a 70/30 split for model training and validation. Twelve conditioning factors—including slope, lithology, elevation, hydrology, and land use—were analyzed to determine their influence on landslide occurrence. The resulting susceptibility map was classified into five risk categories using the Natural Breaks method. Model validation using the Receiver Operating Characteristic (ROC) curve yielded an Area Under the Curve (AUC) value of 0.77, indicating good predictive accuracy. These results demonstrate the effectiveness of the WoE method in landslide susceptibility mapping and provide a valuable tool for risk management and territorial planning in the region.