Recurrent flooding occurs in most years along different parts of the Gulf of Mexico coastline and the central and southeastern parts of Mexico. These events cause significant economic losses in the agricultural, livestock, and infrastructure sectors, and frequently involve loss of human life. Climate change has contributed to flooding events and their more frequent occurrence, even in areas where such events were previously rare. Satellite images have become valuable information sources to identify, precisely locate, and monitor flooding events. The machine learning models use remote sensing images pixels as input feature. In this paper, we report a study involving 16 combinations of Sentinel-1 SAR images, Sentinel-2 optical images, and digital elevation model (DEM) data, which were analyzed to evaluate the performance of two widely used machine learning algorithms, gradient boosting (GB) and random forest (RF), for providing information about flooding events. With machine learning models GB and RF, the input dataset (Sentinel-1, Sentinel-2, and DEM) was used to establish rules and classify the set in the categories specified by previous tags. Monitoring of flooding was performed by tracking the evolution of water bodies during the dry season (before the event) through to the occurrence of floods during the rainy season (during the event). For detection of bodies of water in the dry season, the metrics indicate that the best algorithm is GB with combination 15 (F1m = 0.997, AUC = 0.999, K = 0.994). In the rainy season, the GB algorithm had better metrics with combination 16 (F1m = 0.995, AUC = 0.999, Kappa = 0.994), and detected an extent of flooded areas of 1113.36 ha with depths of <1 m. The high classification performance shown by machine learning algorithms, particularly the so-called assembly algorithms, means that they should be considered capable of improving satellite image classification for detection of flooding over traditional methods, in turn leading to better monitoring of flooding at local, regional, and continental scales.

The information about where crops are distributed is useful for agri-environmental assessments, but is chiefly important for food security and agricultural policy managers. The quickness with which this information becomes available, especially over large areas, is important for decision makers. Methodologies have been proposed for the study of crops. Most of them require field survey for ground truth data and a single crop map is generated for the whole season at the end of the crop cycle and for the next crop cycle a new field survey is necessary. Here, we present models for recognizing maize (Zea mays L.), beans (Phaseolus vulgaris L.), and alfalfa (Medicago sativa L.) before the crop cycle ends without current-year field survey for ground truth data. The models were trained with an exhaustive field survey at plot level in a previous crop cycle. The field surveys begin since days before the emergence of crops to maturity. The algorithms used for classification were support vector machine (SVM) and bagged tree (BT), and the spectral information captured in the visible, red-edge, near infrared, and shortwave infrared regions bands of Sentinel 2 images was used. The models were validated within the next crop cycle each fifteen days before the mid-season. The overall accuracies range from 71.9% (38 days after the begin of cycle) to 87.5% (81 days after the begin cycle) and a kappa coefficient ranging from 0.53 at the beginning to 0.74 at mid-season

The heterogeneity of corn‐growing conditions in Mexico makes accurate predictions of yield ahead of harvest time difficult. Such predictions are needed by the government to estimate, ahead of time, the amount of corn required to be imported to meet the expected domestic shortfall. In this paper, therefore, a methodology for the estimation of corn yield ahead of harvest time is developed specifically for the corn‐growing conditions particular to Mexico. The method is based on the multi‐temporal analysis of NOAA‐AVHRR satellite images, and uses normalized difference vegetation indices (NDVIs), leaf area indices (LAIs), and degree‐days (DDs) to predict corn occurrence and yield. Results of the application of the methodology to successfully identify sites with corn, and to predict corn yield, for a 15,840 ha area of land in the State of Guanajuato, are presented and discussed.

Contexto: La optimización de máquinas de soporte vectorial (SVM, por su sigla en inglés) ha sido un tema recurrente en diversos foros y publicaciones relacionadas con la inteligencia artificial (IA) y el aprendizaje automático. A pesar de su relevancia, existe una notable ausencia de directrices claras para la selección y optimización de los hiperparámetros asociados a las distintas funciones kernel utilizadas en SVM, lo cual puede comprometer la eficacia de estas herramientas en tareas de clasificación y regresión.Objetivo: Este estudio tiene como objetivo general presentar un método detallado y riguroso para la optimización de los hiperparámetros en SVM, mediante el uso de la librería e1071 en R. Se busca, así, proporcionar una guía clara para futuros investigadores y practicantes en el campo, que les permita implementar de manera efectiva SVM ensus proyectos.Metodología: Se emplearon las librerías e1071 y scales en R para estandarizar los datos y ajustar los modelos SVM con diferentes tipos de kernel (lineal, radial, sigmoide y polinomial). Se procedió a la optimización de hiperparámetros mediante la función “tune”, con análisis de conjuntos de entrenamiento y prueba para verificar la bondad de ajuste. Se utilizó una metodología detallada para seleccionar los hiperparámetros óptimos de cada kernel, a partir del estudio de la eficacia de la clasificación y la minimización del error.Resultados: Los resultados destacan la correcta selección y optimización de los hiperparámetros en SVM, y demuestran mejoras significativas en la clasificación al aplicar el método propuesto. Se encontró que la elección de hiperparámetros óptimos varía considerablemente entre los diferentes tipos de kernel y que su adecuada optimización contribuye a la precisión del modelo.Conclusiones: La implementación detallada y el ajuste de los hiperparámetros en SVM son cruciales para maximizar su desempeño. Este artículo proporciona un código optimizado y una metodología clara para su implementaciónen R, lo cual facilita la tarea de investigadores y analistas en el campo del aprendizaje automático. Se enfatiza la necesidad de considerar la especificidad de cada conjunto de datos y la relevancia de la experiencia en la selección de hiperparámetros. También, se sugiere, para futuras investigaciones, explorar y expandir las fronteras de las SVM en la solución de problemas complejos de clasificación.