Séminaire
TEMLI: Temperature Estimation with Machine Learning and Land Input over Morocco
Wiam Salih
Wiam Salih, doctorante à l’UM6P (Maroc), est en court séjour de recherche à l’IPSL pour travailler sur l’interface entre l’hydrologie et l’apprentissage automatique. Elle présentera ses travaux de thèse sur l’utilisation de méthodes ML pour la création d’un dataset de résolution kilométrique de température de l’air sur le Maroc.
Description
Air temperature (Tair) at high spatial resolution is essential for understanding climate dynamics and decision-making across various sectors. However, the scarcity of observed data presents significant challenges. This study introduces a comprehensive methodology for developing a robust Tair estimation model (i.e., TEMLI), utilizing Land Surface Temperature (LST), NDVI, NDWI, albedo, and wind speed products as inputs. By capturing the nonlinear relationships among these inputs and Tair, TEMLI enhances accuracy in Tair estimation.
Among the machine learning models evaluated within the TEMLI framework, the Multilayer Perceptron (MLP) model performed best, achieving an RMSE of less than 1.6°C. The model demonstrates exceptional performance in estimating extreme temperatures, particularly in high-altitude areas, highlighting the potential of deep learning techniques. Trained with Moroccan weather station data, this methodology provides a reliable and generalizable approach to Tair estimation, offering valuable insights into diverse climatic conditions.
This study suggests a scalable, novel approach to air temperature estimation using machine learning and satellite-derived data, advancing climate modeling and decision-making in data-sparse regions.
Informations supplémentaires
Lieu
Grande Salle LATMOS 4e étage
Visio
https://ird-fr.zoom.us/my/redouanelguensat?pwd=Mnp3NzMya3BnNzVScC91QlFKOGVpZz09
