Poster De Conférence Année : 2025

Contribution of Random Forest and Deep Neural Network Algorithms with Environmental Covariates for the Spatial SOC Stock Modelling

Résumé

This study aims to compare the effectiveness of random forest (RF) and deep neural network (DNN) algorithms in conjunction with variable selection methods for modelling soil organic carbon (SOC) stocks in the Aix-Marseille-Provence (AMP) Metropolis. Using a total of 51 soil samples and 29 different environmental factors, which include climate data, parent material, topographic details, land cover, human impact, remote sensing information, and soil characteristics, the examination has demonstrated that the Deep Neural Network (DNN) generally performs better than Random Forest (RF), particularly when all variables or those chosen through the Boruta technique or the variance inflation factor (VIF) are integrated. Nevertheless, RF excels over DNN when only non-redundant variables are considered. The resulting visual representations illustrate a higher SOC Stock in hilly forested regions and lower levels in maritime wetlands and farming areas. These conclusions offer a valuable tool for sustainable soil management and spatial organization, helping in achieving AMP Metropolis’s climate goals.

Fichier principal
Vignette du fichier
Poster_SAGEO2025_final.pdf (2.09 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05088381 , version 1 (28-05-2025)

Licence

Identifiants

  • HAL Id : hal-05088381 , version 1

Citer

Mounir Oukhattar, Sébastien Gadal, Yannick Robert, Catherine Keller. Contribution of Random Forest and Deep Neural Network Algorithms with Environmental Covariates for the Spatial SOC Stock Modelling. SAGEO 2025, May 2025, Avignon, France. . ⟨hal-05088381⟩
171 Consultations
109 Téléchargements

Partager

  • More